feat: script gate for cron jobs (port from nanoclaw #1232)

Add optional 'script' field to cron jobs. Before waking the agent, the
script runs as bash with a 30s timeout. Its last stdout line is parsed
as JSON:
  {"wakeAgent": false}          -> skip the agent entirely
  {"wakeAgent": true, "data": ...} -> prepend data to agent prompt

This enables frequent cron schedules (every 1-10 min) without API cost.
E.g., a script checks if a GitHub PR has new comments — only wakes the
agent when there's actually something to do.

Graceful fallback: script errors, timeouts, or invalid JSON all log a
warning and proceed normally (never block the agent).

Inspired by qwibitai/nanoclaw#1232.

15 new tests covering all paths.

.dockerignore

@@ -0,0 +1,13 @@
+# Git
+.git
+.gitignore
+.gitmodules
+
+# Dependencies
+node_modules
+
+# CI/CD
+.github
+
+# Environment files
+.env

.env.example

@@ -59,12 +59,25 @@ OPENCODE_ZEN_API_KEY=
 # OpenCode Go provides access to open models (GLM-5, Kimi K2.5, MiniMax M2.5)
 # $10/month subscription. Get your key at: https://opencode.ai/auth
 OPENCODE_GO_API_KEY=
+
+# =============================================================================
+# LLM PROVIDER (Hugging Face Inference Providers)
+# =============================================================================
+# Hugging Face routes to 20+ open models via unified OpenAI-compatible endpoint.
+# Free tier included ($0.10/month), no markup on provider rates.
+# Get your token at: https://huggingface.co/settings/tokens
+# Required permission: "Make calls to Inference Providers"
+HF_TOKEN=
 # OPENCODE_GO_BASE_URL=https://opencode.ai/zen/go/v1  # Override default base URL
 
 # =============================================================================
 # TOOL API KEYS
 # =============================================================================
 
+# Exa API Key - AI-native web search and contents
+# Get at: https://exa.ai
+EXA_API_KEY=
+
 # Parallel API Key - AI-native web search and extract
 # Get at: https://parallel.ai
 PARALLEL_API_KEY=
@@ -85,7 +98,7 @@ FAL_KEY=
 HONCHO_API_KEY=
 
 # =============================================================================
-# TERMINAL TOOL CONFIGURATION (mini-swe-agent backend)
+# TERMINAL TOOL CONFIGURATION
 # =============================================================================
 # Backend type: "local", "singularity", "docker", "modal", or "ssh"
 # Terminal backend is configured in ~/.hermes/config.yaml (terminal.backend).

.github/workflows/docker-publish.yml

@@ -0,0 +1,61 @@
+name: Docker Build and Publish
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+concurrency:
+  group: docker-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  build-and-push:
+    runs-on: ubuntu-latest
+    timeout-minutes: 30
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+        with:
+          submodules: recursive
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Build image
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          file: Dockerfile
+          load: true
+          tags: nousresearch/hermes-agent:test
+          cache-from: type=gha
+          cache-to: type=gha,mode=max
+
+      - name: Test image starts
+        run: |
+          docker run --rm \
+            -v /tmp/hermes-test:/opt/data \
+            --entrypoint /opt/hermes/docker/entrypoint.sh \
+            nousresearch/hermes-agent:test --help
+
+      - name: Log in to Docker Hub
+        if: github.event_name == 'push' && github.ref == 'refs/heads/main'
+        uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
+
+      - name: Push image
+        if: github.event_name == 'push' && github.ref == 'refs/heads/main'
+        uses: docker/build-push-action@v6
+        with:
+          context: .
+          file: Dockerfile
+          push: true
+          tags: |
+            nousresearch/hermes-agent:latest
+            nousresearch/hermes-agent:${{ github.sha }}
+          cache-from: type=gha
+          cache-to: type=gha,mode=max

AGENTS.md

@@ -210,6 +210,10 @@ registry.register(
 
 The registry handles schema collection, dispatch, availability checking, and error wrapping. All handlers MUST return a JSON string.
 
+**Path references in tool schemas**: If the schema description mentions file paths (e.g. default output directories), use `display_hermes_home()` to make them profile-aware. The schema is generated at import time, which is after `_apply_profile_override()` sets `HERMES_HOME`.
+
+**State files**: If a tool stores persistent state (caches, logs, checkpoints), use `get_hermes_home()` for the base directory — never `Path.home() / ".hermes"`. This ensures each profile gets its own state.
+
 **Agent-level tools** (todo, memory): intercepted by `run_agent.py` before `handle_function_call()`. See `todo_tool.py` for the pattern.
 
 ---
@@ -358,8 +362,69 @@ in config.yaml (or `HERMES_BACKGROUND_NOTIFICATIONS` env var):
 
 ---
 
+## Profiles: Multi-Instance Support
+
+Hermes supports **profiles** — multiple fully isolated instances, each with its own
+`HERMES_HOME` directory (config, API keys, memory, sessions, skills, gateway, etc.).
+
+The core mechanism: `_apply_profile_override()` in `hermes_cli/main.py` sets
+`HERMES_HOME` before any module imports. All 119+ references to `get_hermes_home()`
+automatically scope to the active profile.
+
+### Rules for profile-safe code
+
+1. **Use `get_hermes_home()` for all HERMES_HOME paths.** Import from `hermes_constants`.
+   NEVER hardcode `~/.hermes` or `Path.home() / ".hermes"` in code that reads/writes state.
+   ```python
+   # GOOD
+   from hermes_constants import get_hermes_home
+   config_path = get_hermes_home() / "config.yaml"
+
+   # BAD — breaks profiles
+   config_path = Path.home() / ".hermes" / "config.yaml"
+   ```
+
+2. **Use `display_hermes_home()` for user-facing messages.** Import from `hermes_constants`.
+   This returns `~/.hermes` for default or `~/.hermes/profiles/<name>` for profiles.
+   ```python
+   # GOOD
+   from hermes_constants import display_hermes_home
+   print(f"Config saved to {display_hermes_home()}/config.yaml")
+
+   # BAD — shows wrong path for profiles
+   print("Config saved to ~/.hermes/config.yaml")
+   ```
+
+3. **Module-level constants are fine** — they cache `get_hermes_home()` at import time,
+   which is AFTER `_apply_profile_override()` sets the env var. Just use `get_hermes_home()`,
+   not `Path.home() / ".hermes"`.
+
+4. **Tests that mock `Path.home()` must also set `HERMES_HOME`** — since code now uses
+   `get_hermes_home()` (reads env var), not `Path.home() / ".hermes"`:
+   ```python
+   with patch.object(Path, "home", return_value=tmp_path), \
+        patch.dict(os.environ, {"HERMES_HOME": str(tmp_path / ".hermes")}):
+       ...
+   ```
+
+5. **Gateway platform adapters should use token locks** — if the adapter connects with
+   a unique credential (bot token, API key), call `acquire_scoped_lock()` from
+   `gateway.status` in the `connect()`/`start()` method and `release_scoped_lock()` in
+   `disconnect()`/`stop()`. This prevents two profiles from using the same credential.
+   See `gateway/platforms/telegram.py` for the canonical pattern.
+
+6. **Profile operations are HOME-anchored, not HERMES_HOME-anchored** — `_get_profiles_root()`
+   returns `Path.home() / ".hermes" / "profiles"`, NOT `get_hermes_home() / "profiles"`.
+   This is intentional — it lets `hermes -p coder profile list` see all profiles regardless
+   of which one is active.
+
 ## Known Pitfalls
 
+### DO NOT hardcode `~/.hermes` paths
+Use `get_hermes_home()` from `hermes_constants` for code paths. Use `display_hermes_home()`
+for user-facing print/log messages. Hardcoding `~/.hermes` breaks profiles — each profile
+has its own `HERMES_HOME` directory. This was the source of 5 bugs fixed in PR #3575.
+
 ### DO NOT use `simple_term_menu` for interactive menus
 Rendering bugs in tmux/iTerm2 — ghosting on scroll. Use `curses` (stdlib) instead. See `hermes_cli/tools_config.py` for the pattern.
 
@@ -375,6 +440,19 @@ Tool schema descriptions must not mention tools from other toolsets by name (e.g
 ### Tests must not write to `~/.hermes/`
 The `_isolate_hermes_home` autouse fixture in `tests/conftest.py` redirects `HERMES_HOME` to a temp dir. Never hardcode `~/.hermes/` paths in tests.
 
+**Profile tests**: When testing profile features, also mock `Path.home()` so that
+`_get_profiles_root()` and `_get_default_hermes_home()` resolve within the temp dir.
+Use the pattern from `tests/hermes_cli/test_profiles.py`:
+```python
+@pytest.fixture
+def profile_env(tmp_path, monkeypatch):
+    home = tmp_path / ".hermes"
+    home.mkdir()
+    monkeypatch.setattr(Path, "home", lambda: tmp_path)
+    monkeypatch.setenv("HERMES_HOME", str(home))
+    return home
+```
+
 ---
 
 ## Testing

Dockerfile

@@ -0,0 +1,20 @@
+FROM debian:13.4
+
+RUN apt-get update
+RUN apt-get install -y nodejs npm python3 python3-pip ripgrep ffmpeg gcc python3-dev libffi-dev
+
+COPY . /opt/hermes
+WORKDIR /opt/hermes
+
+RUN pip install -e ".[all]" --break-system-packages
+RUN npm install
+RUN npx playwright install --with-deps chromium
+WORKDIR /opt/hermes/scripts/whatsapp-bridge
+RUN npm install
+
+WORKDIR /opt/hermes
+RUN chmod +x /opt/hermes/docker/entrypoint.sh
+
+ENV HERMES_HOME=/opt/data
+VOLUME [ "/opt/data" ]
+ENTRYPOINT [ "/opt/hermes/docker/entrypoint.sh" ]

RELEASE_v0.5.0.md

@@ -0,0 +1,348 @@
+# Hermes Agent v0.5.0 (v2026.3.28)
+
+**Release Date:** March 28, 2026
+
+> The hardening release — Hugging Face provider, /model command overhaul, Telegram Private Chat Topics, native Modal SDK, plugin lifecycle hooks, tool-use enforcement for GPT models, Nix flake, 50+ security and reliability fixes, and a comprehensive supply chain audit.
+
+---
+
+## ✨ Highlights
+
+- **Nous Portal now supports 400+ models** — The Nous Research inference portal has expanded dramatically, giving Hermes Agent users access to over 400 models through a single provider endpoint
+
+- **Hugging Face as a first-class inference provider** — Full integration with HF Inference API including curated agentic model picker that maps to OpenRouter analogues, live `/models` endpoint probe, and setup wizard flow ([#3419](https://github.com/NousResearch/hermes-agent/pull/3419), [#3440](https://github.com/NousResearch/hermes-agent/pull/3440))
+
+- **Telegram Private Chat Topics** — Project-based conversations with functional skill binding per topic, enabling isolated workflows within a single Telegram chat ([#3163](https://github.com/NousResearch/hermes-agent/pull/3163))
+
+- **Native Modal SDK backend** — Replaced swe-rex dependency with native Modal SDK (`Sandbox.create.aio` + `exec.aio`), eliminating tunnels and simplifying the Modal terminal backend ([#3538](https://github.com/NousResearch/hermes-agent/pull/3538))
+
+- **Plugin lifecycle hooks activated** — `pre_llm_call`, `post_llm_call`, `on_session_start`, and `on_session_end` hooks now fire in the agent loop and CLI/gateway, completing the plugin hook system ([#3542](https://github.com/NousResearch/hermes-agent/pull/3542))
+
+- **Improved OpenAI Model Reliability** — Added `GPT_TOOL_USE_GUIDANCE` to prevent GPT models from describing intended actions instead of making tool calls, plus automatic stripping of stale budget warnings from conversation history that caused models to avoid tools across turns ([#3528](https://github.com/NousResearch/hermes-agent/pull/3528))
+
+- **Nix flake** — Full uv2nix build, NixOS module with persistent container mode, auto-generated config keys from Python source, and suffix PATHs for agent-friendliness ([#20](https://github.com/NousResearch/hermes-agent/pull/20), [#3274](https://github.com/NousResearch/hermes-agent/pull/3274), [#3061](https://github.com/NousResearch/hermes-agent/pull/3061)) by @alt-glitch
+
+- **Supply chain hardening** — Removed compromised `litellm` dependency, pinned all dependency version ranges, regenerated `uv.lock` with hashes, added CI workflow scanning PRs for supply chain attack patterns, and bumped deps to fix CVEs ([#2796](https://github.com/NousResearch/hermes-agent/pull/2796), [#2810](https://github.com/NousResearch/hermes-agent/pull/2810), [#2812](https://github.com/NousResearch/hermes-agent/pull/2812), [#2816](https://github.com/NousResearch/hermes-agent/pull/2816), [#3073](https://github.com/NousResearch/hermes-agent/pull/3073))
+
+- **Anthropic output limits fix** — Replaced hardcoded 16K `max_tokens` with per-model native output limits (128K for Opus 4.6, 64K for Sonnet 4.6), fixing "Response truncated" and thinking-budget exhaustion on direct Anthropic API ([#3426](https://github.com/NousResearch/hermes-agent/pull/3426), [#3444](https://github.com/NousResearch/hermes-agent/pull/3444))
+
+---
+
+## 🏗️ Core Agent & Architecture
+
+### New Provider: Hugging Face
+- First-class Hugging Face Inference API integration with auth, setup wizard, and model picker ([#3419](https://github.com/NousResearch/hermes-agent/pull/3419))
+- Curated model list mapping OpenRouter agentic defaults to HF equivalents — providers with 8+ curated models skip live `/models` probe for speed ([#3440](https://github.com/NousResearch/hermes-agent/pull/3440))
+- Added glm-5-turbo to Z.AI provider model list ([#3095](https://github.com/NousResearch/hermes-agent/pull/3095))
+
+### Provider & Model Improvements
+- `/model` command overhaul — extracted shared `switch_model()` pipeline for CLI and gateway, custom endpoint support, provider-aware routing ([#2795](https://github.com/NousResearch/hermes-agent/pull/2795), [#2799](https://github.com/NousResearch/hermes-agent/pull/2799))
+- Removed `/model` slash command from CLI and gateway in favor of `hermes model` subcommand ([#3080](https://github.com/NousResearch/hermes-agent/pull/3080))
+- Preserve `custom` provider instead of silently remapping to `openrouter` ([#2792](https://github.com/NousResearch/hermes-agent/pull/2792))
+- Read root-level `provider` and `base_url` from config.yaml into model config ([#3112](https://github.com/NousResearch/hermes-agent/pull/3112))
+- Align Nous Portal model slugs with OpenRouter naming ([#3253](https://github.com/NousResearch/hermes-agent/pull/3253))
+- Fix Alibaba provider default endpoint and model list ([#3484](https://github.com/NousResearch/hermes-agent/pull/3484))
+- Allow MiniMax users to override `/v1` → `/anthropic` auto-correction ([#3553](https://github.com/NousResearch/hermes-agent/pull/3553))
+- Migrate OAuth token refresh to `platform.claude.com` with fallback ([#3246](https://github.com/NousResearch/hermes-agent/pull/3246))
+
+### Agent Loop & Conversation
+- **Improved OpenAI model reliability** — `GPT_TOOL_USE_GUIDANCE` prevents GPT models from describing actions instead of calling tools + automatic budget warning stripping from history ([#3528](https://github.com/NousResearch/hermes-agent/pull/3528))
+- **Surface lifecycle events** — All retry, fallback, and compression events now surface to the user as formatted messages ([#3153](https://github.com/NousResearch/hermes-agent/pull/3153))
+- **Anthropic output limits** — Per-model native output limits instead of hardcoded 16K `max_tokens` ([#3426](https://github.com/NousResearch/hermes-agent/pull/3426))
+- **Thinking-budget exhaustion detection** — Skip useless continuation retries when model uses all output tokens on reasoning ([#3444](https://github.com/NousResearch/hermes-agent/pull/3444))
+- Always prefer streaming for API calls to prevent hung subagents ([#3120](https://github.com/NousResearch/hermes-agent/pull/3120))
+- Restore safe non-streaming fallback after stream failures ([#3020](https://github.com/NousResearch/hermes-agent/pull/3020))
+- Give subagents independent iteration budgets ([#3004](https://github.com/NousResearch/hermes-agent/pull/3004))
+- Update `api_key` in `_try_activate_fallback` for subagent auth ([#3103](https://github.com/NousResearch/hermes-agent/pull/3103))
+- Graceful return on max retries instead of crashing thread ([untagged commit](https://github.com/NousResearch/hermes-agent))
+- Count compression restarts toward retry limit ([#3070](https://github.com/NousResearch/hermes-agent/pull/3070))
+- Include tool tokens in preflight estimate, guard context probe persistence ([#3164](https://github.com/NousResearch/hermes-agent/pull/3164))
+- Update context compressor limits after fallback activation ([#3305](https://github.com/NousResearch/hermes-agent/pull/3305))
+- Validate empty user messages to prevent Anthropic API 400 errors ([#3322](https://github.com/NousResearch/hermes-agent/pull/3322))
+- GLM reasoning-only and max-length handling ([#3010](https://github.com/NousResearch/hermes-agent/pull/3010))
+- Increase API timeout default from 900s to 1800s for slow-thinking models ([#3431](https://github.com/NousResearch/hermes-agent/pull/3431))
+- Send `max_tokens` for Claude/OpenRouter + retry SSE connection errors ([#3497](https://github.com/NousResearch/hermes-agent/pull/3497))
+- Prevent AsyncOpenAI/httpx cross-loop deadlock in gateway mode ([#2701](https://github.com/NousResearch/hermes-agent/pull/2701)) by @ctlst
+
+### Streaming & Reasoning
+- **Persist reasoning across gateway session turns** with new schema v6 columns (`reasoning`, `reasoning_details`, `codex_reasoning_items`) ([#2974](https://github.com/NousResearch/hermes-agent/pull/2974))
+- Detect and kill stale SSE connections ([untagged commit](https://github.com/NousResearch/hermes-agent))
+- Fix stale stream detector race causing spurious `RemoteProtocolError` ([untagged commit](https://github.com/NousResearch/hermes-agent))
+- Skip duplicate callback for `<think>`-extracted reasoning during streaming ([#3116](https://github.com/NousResearch/hermes-agent/pull/3116))
+- Preserve reasoning fields in `rewrite_transcript` ([#3311](https://github.com/NousResearch/hermes-agent/pull/3311))
+- Preserve Gemini thought signatures in streamed tool calls ([#2997](https://github.com/NousResearch/hermes-agent/pull/2997))
+- Ensure first delta is fired during reasoning updates ([untagged commit](https://github.com/NousResearch/hermes-agent))
+
+### Session & Memory
+- **Session search recent sessions mode** — Omit query to browse recent sessions with titles, previews, and timestamps ([#2533](https://github.com/NousResearch/hermes-agent/pull/2533))
+- **Session config surfacing** on `/new`, `/reset`, and auto-reset ([#3321](https://github.com/NousResearch/hermes-agent/pull/3321))
+- **Third-party session isolation** — `--source` flag for isolating sessions by origin ([#3255](https://github.com/NousResearch/hermes-agent/pull/3255))
+- Add `/resume` CLI handler, session log truncation guard, `reopen_session` API ([#3315](https://github.com/NousResearch/hermes-agent/pull/3315))
+- Clear compressor summary and turn counter on `/clear` and `/new` ([#3102](https://github.com/NousResearch/hermes-agent/pull/3102))
+- Surface silent SessionDB failures that cause session data loss ([#2999](https://github.com/NousResearch/hermes-agent/pull/2999))
+- Session search fallback preview on summarization failure ([#3478](https://github.com/NousResearch/hermes-agent/pull/3478))
+- Prevent stale memory overwrites by flush agent ([#2687](https://github.com/NousResearch/hermes-agent/pull/2687))
+
+### Context Compression
+- Replace dead `summary_target_tokens` with ratio-based scaling ([#2554](https://github.com/NousResearch/hermes-agent/pull/2554))
+- Expose `compression.target_ratio`, `protect_last_n`, and `threshold` in `DEFAULT_CONFIG` ([untagged commit](https://github.com/NousResearch/hermes-agent))
+- Restore sane defaults and cap summary at 12K tokens ([untagged commit](https://github.com/NousResearch/hermes-agent))
+- Preserve transcript on `/compress` and hygiene compression ([#3556](https://github.com/NousResearch/hermes-agent/pull/3556))
+- Update context pressure warnings and token estimates after compaction ([untagged commit](https://github.com/NousResearch/hermes-agent))
+
+### Architecture & Dependencies
+- **Remove mini-swe-agent dependency** — Inline Docker and Modal backends directly ([#2804](https://github.com/NousResearch/hermes-agent/pull/2804))
+- **Replace swe-rex with native Modal SDK** for Modal backend ([#3538](https://github.com/NousResearch/hermes-agent/pull/3538))
+- **Plugin lifecycle hooks** — `pre_llm_call`, `post_llm_call`, `on_session_start`, `on_session_end` now fire in the agent loop ([#3542](https://github.com/NousResearch/hermes-agent/pull/3542))
+- Fix plugin toolsets invisible in `hermes tools` and standalone processes ([#3457](https://github.com/NousResearch/hermes-agent/pull/3457))
+- Consolidate `get_hermes_home()` and `parse_reasoning_effort()` ([#3062](https://github.com/NousResearch/hermes-agent/pull/3062))
+- Remove unused Hermes-native PKCE OAuth flow ([#3107](https://github.com/NousResearch/hermes-agent/pull/3107))
+- Remove ~100 unused imports across 55 files ([#3016](https://github.com/NousResearch/hermes-agent/pull/3016))
+- Fix 154 f-strings, simplify getattr/URL patterns, remove dead code ([#3119](https://github.com/NousResearch/hermes-agent/pull/3119))
+
+---
+
+## 📱 Messaging Platforms (Gateway)
+
+### Telegram
+- **Private Chat Topics** — Project-based conversations with functional skill binding per topic, enabling isolated workflows within a single Telegram chat ([#3163](https://github.com/NousResearch/hermes-agent/pull/3163))
+- **Auto-discover fallback IPs via DNS-over-HTTPS** when `api.telegram.org` is unreachable ([#3376](https://github.com/NousResearch/hermes-agent/pull/3376))
+- **Configurable reply threading mode** ([#2907](https://github.com/NousResearch/hermes-agent/pull/2907))
+- Fall back to no `thread_id` on "Message thread not found" BadRequest ([#3390](https://github.com/NousResearch/hermes-agent/pull/3390))
+- Self-reschedule reconnect when `start_polling` fails after 502 ([#3268](https://github.com/NousResearch/hermes-agent/pull/3268))
+
+### Discord
+- Stop phantom typing indicator after agent turn completes ([#3003](https://github.com/NousResearch/hermes-agent/pull/3003))
+
+### Slack
+- Send tool call progress messages to correct Slack thread ([#3063](https://github.com/NousResearch/hermes-agent/pull/3063))
+- Scope progress thread fallback to Slack only ([#3488](https://github.com/NousResearch/hermes-agent/pull/3488))
+
+### WhatsApp
+- Download documents, audio, and video media from messages ([#2978](https://github.com/NousResearch/hermes-agent/pull/2978))
+
+### Matrix
+- Add missing Matrix entry in `PLATFORMS` dict ([#3473](https://github.com/NousResearch/hermes-agent/pull/3473))
+- Harden e2ee access-token handling ([#3562](https://github.com/NousResearch/hermes-agent/pull/3562))
+- Add backoff for `SyncError` in sync loop ([#3280](https://github.com/NousResearch/hermes-agent/pull/3280))
+
+### Signal
+- Track SSE keepalive comments as connection activity ([#3316](https://github.com/NousResearch/hermes-agent/pull/3316))
+
+### Email
+- Prevent unbounded growth of `_seen_uids` in EmailAdapter ([#3490](https://github.com/NousResearch/hermes-agent/pull/3490))
+
+### Gateway Core
+- **Config-gated `/verbose` command** for messaging platforms — toggle tool output verbosity from chat ([#3262](https://github.com/NousResearch/hermes-agent/pull/3262))
+- **Background review notifications** delivered to user chat ([#3293](https://github.com/NousResearch/hermes-agent/pull/3293))
+- **Retry transient send failures** and notify user on exhaustion ([#3288](https://github.com/NousResearch/hermes-agent/pull/3288))
+- Recover from hung agents — `/stop` hard-kills session lock ([#3104](https://github.com/NousResearch/hermes-agent/pull/3104))
+- Thread-safe `SessionStore` — protect `_entries` with `threading.Lock` ([#3052](https://github.com/NousResearch/hermes-agent/pull/3052))
+- Fix gateway token double-counting with cached agents — use absolute set instead of increment ([#3306](https://github.com/NousResearch/hermes-agent/pull/3306), [#3317](https://github.com/NousResearch/hermes-agent/pull/3317))
+- Fingerprint full auth token in agent cache signature ([#3247](https://github.com/NousResearch/hermes-agent/pull/3247))
+- Silence background agent terminal output ([#3297](https://github.com/NousResearch/hermes-agent/pull/3297))
+- Include per-platform `ALLOW_ALL` and `SIGNAL_GROUP` in startup allowlist check ([#3313](https://github.com/NousResearch/hermes-agent/pull/3313))
+- Include user-local bin paths in systemd unit PATH ([#3527](https://github.com/NousResearch/hermes-agent/pull/3527))
+- Track background task references in `GatewayRunner` ([#3254](https://github.com/NousResearch/hermes-agent/pull/3254))
+- Add request timeouts to HA, Email, Mattermost, SMS adapters ([#3258](https://github.com/NousResearch/hermes-agent/pull/3258))
+- Add media download retry to Mattermost, Slack, and base cache ([#3323](https://github.com/NousResearch/hermes-agent/pull/3323))
+- Detect virtualenv path instead of hardcoding `venv/` ([#2797](https://github.com/NousResearch/hermes-agent/pull/2797))
+- Use `TERMINAL_CWD` for context file discovery, not process cwd ([untagged commit](https://github.com/NousResearch/hermes-agent))
+- Stop loading hermes repo AGENTS.md into gateway sessions (~10k wasted tokens) ([#2891](https://github.com/NousResearch/hermes-agent/pull/2891))
+
+---
+
+## 🖥️ CLI & User Experience
+
+### Interactive CLI
+- **Configurable busy input mode** + fix `/queue` always working ([#3298](https://github.com/NousResearch/hermes-agent/pull/3298))
+- **Preserve user input on multiline paste** ([#3065](https://github.com/NousResearch/hermes-agent/pull/3065))
+- **Tool generation callback** — streaming "preparing terminal…" updates during tool argument generation ([untagged commit](https://github.com/NousResearch/hermes-agent))
+- Show tool progress for substantive tools, not just "preparing" ([untagged commit](https://github.com/NousResearch/hermes-agent))
+- Buffer reasoning preview chunks and fix duplicate display ([#3013](https://github.com/NousResearch/hermes-agent/pull/3013))
+- Prevent reasoning box from rendering 3x during tool-calling loops ([#3405](https://github.com/NousResearch/hermes-agent/pull/3405))
+- Eliminate "Event loop is closed" / "Press ENTER to continue" during idle — three-layer fix with `neuter_async_httpx_del()`, custom exception handler, and stale client cleanup ([#3398](https://github.com/NousResearch/hermes-agent/pull/3398))
+- Fix status bar shows 26K instead of 260K for token counts with trailing zeros ([#3024](https://github.com/NousResearch/hermes-agent/pull/3024))
+- Fix status bar duplicates and degrades during long sessions ([#3291](https://github.com/NousResearch/hermes-agent/pull/3291))
+- Refresh TUI before background task output to prevent status bar overlap ([#3048](https://github.com/NousResearch/hermes-agent/pull/3048))
+- Suppress KawaiiSpinner animation under `patch_stdout` ([#2994](https://github.com/NousResearch/hermes-agent/pull/2994))
+- Skip KawaiiSpinner when TUI handles tool progress ([#2973](https://github.com/NousResearch/hermes-agent/pull/2973))
+- Guard `isatty()` against closed streams via `_is_tty` property ([#3056](https://github.com/NousResearch/hermes-agent/pull/3056))
+- Ensure single closure of streaming boxes during tool generation ([untagged commit](https://github.com/NousResearch/hermes-agent))
+- Cap context pressure percentage at 100% in display ([#3480](https://github.com/NousResearch/hermes-agent/pull/3480))
+- Clean up HTML error messages in CLI display ([#3069](https://github.com/NousResearch/hermes-agent/pull/3069))
+- Show HTTP status code and 400 body in API error output ([#3096](https://github.com/NousResearch/hermes-agent/pull/3096))
+- Extract useful info from HTML error pages, dump debug on max retries ([untagged commit](https://github.com/NousResearch/hermes-agent))
+- Prevent TypeError on startup when `base_url` is None ([#3068](https://github.com/NousResearch/hermes-agent/pull/3068))
+- Prevent update crash in non-TTY environments ([#3094](https://github.com/NousResearch/hermes-agent/pull/3094))
+- Handle EOFError in sessions delete/prune confirmation prompts ([#3101](https://github.com/NousResearch/hermes-agent/pull/3101))
+- Catch KeyboardInterrupt during `flush_memories` on exit and in exit cleanup handlers ([#3025](https://github.com/NousResearch/hermes-agent/pull/3025), [#3257](https://github.com/NousResearch/hermes-agent/pull/3257))
+- Guard `.strip()` against None values from YAML config ([#3552](https://github.com/NousResearch/hermes-agent/pull/3552))
+- Guard `config.get()` against YAML null values to prevent AttributeError ([#3377](https://github.com/NousResearch/hermes-agent/pull/3377))
+- Store asyncio task references to prevent GC mid-execution ([#3267](https://github.com/NousResearch/hermes-agent/pull/3267))
+
+### Setup & Configuration
+- Use explicit key mapping for returning-user menu dispatch instead of positional index ([#3083](https://github.com/NousResearch/hermes-agent/pull/3083))
+- Use `sys.executable` for pip in update commands to fix PEP 668 ([#3099](https://github.com/NousResearch/hermes-agent/pull/3099))
+- Harden `hermes update` against diverged history, non-main branches, and gateway edge cases ([#3492](https://github.com/NousResearch/hermes-agent/pull/3492))
+- OpenClaw migration overwrites defaults and setup wizard skips imported sections — fixed ([#3282](https://github.com/NousResearch/hermes-agent/pull/3282))
+- Stop recursive AGENTS.md walk, load top-level only ([#3110](https://github.com/NousResearch/hermes-agent/pull/3110))
+- Add macOS Homebrew paths to browser and terminal PATH resolution ([#2713](https://github.com/NousResearch/hermes-agent/pull/2713))
+- YAML boolean handling for `tool_progress` config ([#3300](https://github.com/NousResearch/hermes-agent/pull/3300))
+- Reset default SOUL.md to baseline identity text ([#3159](https://github.com/NousResearch/hermes-agent/pull/3159))
+- Reject relative cwd paths for container terminal backends ([untagged commit](https://github.com/NousResearch/hermes-agent))
+- Add explicit `hermes-api-server` toolset for API server platform ([#3304](https://github.com/NousResearch/hermes-agent/pull/3304))
+- Reorder setup wizard providers — OpenRouter first ([untagged commit](https://github.com/NousResearch/hermes-agent))
+
+---
+
+## 🔧 Tool System
+
+### API Server
+- **Idempotency-Key support**, body size limit, and OpenAI error envelope ([#2903](https://github.com/NousResearch/hermes-agent/pull/2903))
+- Allow Idempotency-Key in CORS headers ([#3530](https://github.com/NousResearch/hermes-agent/pull/3530))
+- Cancel orphaned agent + true interrupt on SSE disconnect ([#3427](https://github.com/NousResearch/hermes-agent/pull/3427))
+- Fix streaming breaks when agent makes tool calls ([#2985](https://github.com/NousResearch/hermes-agent/pull/2985))
+
+### Terminal & File Operations
+- Handle addition-only hunks in V4A patch parser ([#3325](https://github.com/NousResearch/hermes-agent/pull/3325))
+- Exponential backoff for persistent shell polling ([#2996](https://github.com/NousResearch/hermes-agent/pull/2996))
+- Add timeout to subprocess calls in `context_references` ([#3469](https://github.com/NousResearch/hermes-agent/pull/3469))
+
+### Browser & Vision
+- Handle 402 insufficient credits error in vision tool ([#2802](https://github.com/NousResearch/hermes-agent/pull/2802))
+- Fix `browser_vision` ignores `auxiliary.vision.timeout` config ([#2901](https://github.com/NousResearch/hermes-agent/pull/2901))
+- Make browser command timeout configurable via config.yaml ([#2801](https://github.com/NousResearch/hermes-agent/pull/2801))
+
+### MCP
+- MCP toolset resolution for runtime and config ([#3252](https://github.com/NousResearch/hermes-agent/pull/3252))
+- Add MCP tool name collision protection ([#3077](https://github.com/NousResearch/hermes-agent/pull/3077))
+
+### Auxiliary LLM
+- Guard aux LLM calls against None content + reasoning fallback + retry ([#3449](https://github.com/NousResearch/hermes-agent/pull/3449))
+- Catch ImportError from `build_anthropic_client` in vision auto-detection ([#3312](https://github.com/NousResearch/hermes-agent/pull/3312))
+
+### Other Tools
+- Add request timeouts to `send_message_tool` HTTP calls ([#3162](https://github.com/NousResearch/hermes-agent/pull/3162)) by @memosr
+- Auto-repair `jobs.json` with invalid control characters ([#3537](https://github.com/NousResearch/hermes-agent/pull/3537))
+- Enable fine-grained tool streaming for Claude/OpenRouter ([#3497](https://github.com/NousResearch/hermes-agent/pull/3497))
+
+---
+
+## 🧩 Skills Ecosystem
+
+### Skills System
+- **Env var passthrough** for skills and user config — skills can declare environment variables to pass through ([#2807](https://github.com/NousResearch/hermes-agent/pull/2807))
+- Cache skills prompt with shared `skill_utils` module for faster TTFT ([#3421](https://github.com/NousResearch/hermes-agent/pull/3421))
+- Avoid redundant file re-read for skill conditions ([#2992](https://github.com/NousResearch/hermes-agent/pull/2992))
+- Use Git Trees API to prevent silent subdirectory loss during install ([#2995](https://github.com/NousResearch/hermes-agent/pull/2995))
+- Fix skills-sh install for deeply nested repo structures ([#2980](https://github.com/NousResearch/hermes-agent/pull/2980))
+- Handle null metadata in skill frontmatter ([untagged commit](https://github.com/NousResearch/hermes-agent))
+- Preserve trust for skills-sh identifiers + reduce resolution churn ([#3251](https://github.com/NousResearch/hermes-agent/pull/3251))
+- Agent-created skills were incorrectly treated as untrusted community content — fixed ([untagged commit](https://github.com/NousResearch/hermes-agent))
+
+### New Skills
+- **G0DM0D3 godmode jailbreaking skill** + docs ([#3157](https://github.com/NousResearch/hermes-agent/pull/3157))
+- **Docker management skill** added to optional-skills ([#3060](https://github.com/NousResearch/hermes-agent/pull/3060))
+- **OpenClaw migration v2** — 17 new modules, terminal recap for migrating from OpenClaw to Hermes ([#2906](https://github.com/NousResearch/hermes-agent/pull/2906))
+
+---
+
+## 🔒 Security & Reliability
+
+### Security Hardening
+- **SSRF protection** added to `browser_navigate` ([#3058](https://github.com/NousResearch/hermes-agent/pull/3058))
+- **SSRF protection** added to `vision_tools` and `web_tools` (hardened) ([#2679](https://github.com/NousResearch/hermes-agent/pull/2679))
+- **Restrict subagent toolsets** to parent's enabled set ([#3269](https://github.com/NousResearch/hermes-agent/pull/3269))
+- **Prevent zip-slip path traversal** in self-update ([#3250](https://github.com/NousResearch/hermes-agent/pull/3250))
+- **Prevent shell injection** in `_expand_path` via `~user` path suffix ([#2685](https://github.com/NousResearch/hermes-agent/pull/2685))
+- **Normalize input** before dangerous command detection ([#3260](https://github.com/NousResearch/hermes-agent/pull/3260))
+- Make tirith block verdicts approvable instead of hard-blocking ([#3428](https://github.com/NousResearch/hermes-agent/pull/3428))
+- Remove compromised `litellm`/`typer`/`platformdirs` from deps ([#2796](https://github.com/NousResearch/hermes-agent/pull/2796))
+- Pin all dependency version ranges ([#2810](https://github.com/NousResearch/hermes-agent/pull/2810))
+- Regenerate `uv.lock` with hashes, use lockfile in setup ([#2812](https://github.com/NousResearch/hermes-agent/pull/2812))
+- Bump dependencies to fix CVEs + regenerate `uv.lock` ([#3073](https://github.com/NousResearch/hermes-agent/pull/3073))
+- Supply chain audit CI workflow for PR scanning ([#2816](https://github.com/NousResearch/hermes-agent/pull/2816))
+
+### Reliability
+- **SQLite WAL write-lock contention** causing 15-20s TUI freeze — fixed ([#3385](https://github.com/NousResearch/hermes-agent/pull/3385))
+- **SQLite concurrency hardening** + session transcript integrity ([#3249](https://github.com/NousResearch/hermes-agent/pull/3249))
+- Prevent recurring cron job re-fire on gateway crash/restart loop ([#3396](https://github.com/NousResearch/hermes-agent/pull/3396))
+- Mark cron session as ended after job completes ([#2998](https://github.com/NousResearch/hermes-agent/pull/2998))
+
+---
+
+## ⚡ Performance
+
+- **TTFT startup optimizations** — salvaged easy-win startup improvements ([#3395](https://github.com/NousResearch/hermes-agent/pull/3395))
+- Cache skills prompt with shared `skill_utils` module ([#3421](https://github.com/NousResearch/hermes-agent/pull/3421))
+- Avoid redundant file re-read for skill conditions in prompt builder ([#2992](https://github.com/NousResearch/hermes-agent/pull/2992))
+
+---
+
+## 🐛 Notable Bug Fixes
+
+- Fix gateway token double-counting with cached agents ([#3306](https://github.com/NousResearch/hermes-agent/pull/3306), [#3317](https://github.com/NousResearch/hermes-agent/pull/3317))
+- Fix "Event loop is closed" / "Press ENTER to continue" during idle sessions ([#3398](https://github.com/NousResearch/hermes-agent/pull/3398))
+- Fix reasoning box rendering 3x during tool-calling loops ([#3405](https://github.com/NousResearch/hermes-agent/pull/3405))
+- Fix status bar shows 26K instead of 260K for token counts ([#3024](https://github.com/NousResearch/hermes-agent/pull/3024))
+- Fix `/queue` always working regardless of config ([#3298](https://github.com/NousResearch/hermes-agent/pull/3298))
+- Fix phantom Discord typing indicator after agent turn ([#3003](https://github.com/NousResearch/hermes-agent/pull/3003))
+- Fix Slack progress messages appearing in wrong thread ([#3063](https://github.com/NousResearch/hermes-agent/pull/3063))
+- Fix WhatsApp media downloads (documents, audio, video) ([#2978](https://github.com/NousResearch/hermes-agent/pull/2978))
+- Fix Telegram "Message thread not found" killing progress messages ([#3390](https://github.com/NousResearch/hermes-agent/pull/3390))
+- Fix OpenClaw migration overwriting defaults ([#3282](https://github.com/NousResearch/hermes-agent/pull/3282))
+- Fix returning-user setup menu dispatching wrong section ([#3083](https://github.com/NousResearch/hermes-agent/pull/3083))
+- Fix `hermes update` PEP 668 "externally-managed-environment" error ([#3099](https://github.com/NousResearch/hermes-agent/pull/3099))
+- Fix subagents hitting `max_iterations` prematurely via shared budget ([#3004](https://github.com/NousResearch/hermes-agent/pull/3004))
+- Fix YAML boolean handling for `tool_progress` config ([#3300](https://github.com/NousResearch/hermes-agent/pull/3300))
+- Fix `config.get()` crashes on YAML null values ([#3377](https://github.com/NousResearch/hermes-agent/pull/3377))
+- Fix `.strip()` crash on None values from YAML config ([#3552](https://github.com/NousResearch/hermes-agent/pull/3552))
+- Fix hung agents on gateway — `/stop` now hard-kills session lock ([#3104](https://github.com/NousResearch/hermes-agent/pull/3104))
+- Fix `_custom` provider silently remapped to `openrouter` ([#2792](https://github.com/NousResearch/hermes-agent/pull/2792))
+- Fix Matrix missing from `PLATFORMS` dict ([#3473](https://github.com/NousResearch/hermes-agent/pull/3473))
+- Fix Email adapter unbounded `_seen_uids` growth ([#3490](https://github.com/NousResearch/hermes-agent/pull/3490))
+
+---
+
+## 🧪 Testing
+
+- Pin `agent-client-protocol` < 0.9 to handle breaking upstream release ([#3320](https://github.com/NousResearch/hermes-agent/pull/3320))
+- Catch anthropic ImportError in vision auto-detection tests ([#3312](https://github.com/NousResearch/hermes-agent/pull/3312))
+- Update retry-exhaust test for new graceful return behavior ([#3320](https://github.com/NousResearch/hermes-agent/pull/3320))
+- Add regression tests for null metadata frontmatter ([untagged commit](https://github.com/NousResearch/hermes-agent))
+
+---
+
+## 📚 Documentation
+
+- Update all docs for `/model` command overhaul and custom provider support ([#2800](https://github.com/NousResearch/hermes-agent/pull/2800))
+- Fix stale and incorrect documentation across 18 files ([#2805](https://github.com/NousResearch/hermes-agent/pull/2805))
+- Document 9 previously undocumented features ([#2814](https://github.com/NousResearch/hermes-agent/pull/2814))
+- Add missing skills, CLI commands, and messaging env vars to docs ([#2809](https://github.com/NousResearch/hermes-agent/pull/2809))
+- Fix api-server response storage documentation — SQLite, not in-memory ([#2819](https://github.com/NousResearch/hermes-agent/pull/2819))
+- Quote pip install extras to fix zsh glob errors ([#2815](https://github.com/NousResearch/hermes-agent/pull/2815))
+- Unify hooks documentation — add plugin hooks to hooks page, add `session:end` event ([untagged commit](https://github.com/NousResearch/hermes-agent))
+- Clarify two-mode behavior in `session_search` schema description ([untagged commit](https://github.com/NousResearch/hermes-agent))
+- Fix Discord Public Bot setting for Discord-provided invite link ([#3519](https://github.com/NousResearch/hermes-agent/pull/3519)) by @mehmoodosman
+- Revise v0.4.0 changelog — fix feature attribution, reorder sections ([untagged commit](https://github.com/NousResearch/hermes-agent))
+
+---
+
+## 👥 Contributors
+
+### Core
+- **@teknium1** — 157 PRs covering the full scope of this release
+
+### Community Contributors
+- **@alt-glitch** (Siddharth Balyan) — 2 PRs: Nix flake with uv2nix build, NixOS module, and persistent container mode ([#20](https://github.com/NousResearch/hermes-agent/pull/20)); auto-generated config keys and suffix PATHs for Nix builds ([#3061](https://github.com/NousResearch/hermes-agent/pull/3061), [#3274](https://github.com/NousResearch/hermes-agent/pull/3274))
+- **@ctlst** — 1 PR: Prevent AsyncOpenAI/httpx cross-loop deadlock in gateway mode ([#2701](https://github.com/NousResearch/hermes-agent/pull/2701))
+- **@memosr** (memosr.eth) — 1 PR: Add request timeouts to `send_message_tool` HTTP calls ([#3162](https://github.com/NousResearch/hermes-agent/pull/3162))
+- **@mehmoodosman** (Osman Mehmood) — 1 PR: Fix Discord docs for Public Bot setting ([#3519](https://github.com/NousResearch/hermes-agent/pull/3519))
+
+### All Contributors
+@alt-glitch, @ctlst, @mehmoodosman, @memosr, @teknium1
+
+---
+
+**Full Changelog**: [v2026.3.23...v2026.3.28](https://github.com/NousResearch/hermes-agent/compare/v2026.3.23...v2026.3.28)

acp_adapter/entry.py

@@ -74,7 +74,7 @@ def main() -> None:
 
     agent = HermesACPAgent()
     try:
-        asyncio.run(acp.run_agent(agent))
+        asyncio.run(acp.run_agent(agent, use_unstable_protocol=True))
     except KeyboardInterrupt:
         logger.info("Shutting down (KeyboardInterrupt)")
     except Exception:

acp_adapter/server.py

@@ -25,6 +25,9 @@ from acp.schema import (
     NewSessionResponse,
     PromptResponse,
     ResumeSessionResponse,
+    SetSessionConfigOptionResponse,
+    SetSessionModelResponse,
+    SetSessionModeResponse,
     ResourceContentBlock,
     SessionCapabilities,
     SessionForkCapabilities,
@@ -94,11 +97,14 @@ class HermesACPAgent(acp.Agent):
 
     async def initialize(
         self,
-        protocol_version: int,
+        protocol_version: int | None = None,
         client_capabilities: ClientCapabilities | None = None,
         client_info: Implementation | None = None,
         **kwargs: Any,
     ) -> InitializeResponse:
+        resolved_protocol_version = (
+            protocol_version if isinstance(protocol_version, int) else acp.PROTOCOL_VERSION
+        )
         provider = detect_provider()
         auth_methods = None
         if provider:
@@ -111,7 +117,11 @@ class HermesACPAgent(acp.Agent):
             ]
 
         client_name = client_info.name if client_info else "unknown"
-        logger.info("Initialize from %s (protocol v%s)", client_name, protocol_version)
+        logger.info(
+            "Initialize from %s (protocol v%s)",
+            client_name,
+            resolved_protocol_version,
+        )
 
         return InitializeResponse(
             protocol_version=acp.PROTOCOL_VERSION,
@@ -471,7 +481,7 @@ class HermesACPAgent(acp.Agent):
 
     async def set_session_model(
         self, model_id: str, session_id: str, **kwargs: Any
-    ):
+    ) -> SetSessionModelResponse | None:
         """Switch the model for a session (called by ACP protocol)."""
         state = self.session_manager.get_session(session_id)
         if state:
@@ -489,4 +499,37 @@ class HermesACPAgent(acp.Agent):
             )
             self.session_manager.save_session(session_id)
             logger.info("Session %s: model switched to %s", session_id, model_id)
+            return SetSessionModelResponse()
+        logger.warning("Session %s: model switch requested for missing session", session_id)
         return None
+
+    async def set_session_mode(
+        self, mode_id: str, session_id: str, **kwargs: Any
+    ) -> SetSessionModeResponse | None:
+        """Persist the editor-requested mode so ACP clients do not fail on mode switches."""
+        state = self.session_manager.get_session(session_id)
+        if state is None:
+            logger.warning("Session %s: mode switch requested for missing session", session_id)
+            return None
+        setattr(state, "mode", mode_id)
+        self.session_manager.save_session(session_id)
+        logger.info("Session %s: mode switched to %s", session_id, mode_id)
+        return SetSessionModeResponse()
+
+    async def set_config_option(
+        self, config_id: str, session_id: str, value: str, **kwargs: Any
+    ) -> SetSessionConfigOptionResponse | None:
+        """Accept ACP config option updates even when Hermes has no typed ACP config surface yet."""
+        state = self.session_manager.get_session(session_id)
+        if state is None:
+            logger.warning("Session %s: config update requested for missing session", session_id)
+            return None
+
+        options = getattr(state, "config_options", None)
+        if not isinstance(options, dict):
+            options = {}
+        options[str(config_id)] = value
+        setattr(state, "config_options", options)
+        self.session_manager.save_session(session_id)
+        logger.info("Session %s: config option %s updated", session_id, config_id)
+        return SetSessionConfigOptionResponse(config_options=[])

agent/anthropic_adapter.py

@@ -35,6 +35,54 @@ ADAPTIVE_EFFORT_MAP = {
     "minimal": "low",
 }
 
+# ── Max output token limits per Anthropic model ───────────────────────
+# Source: Anthropic docs + Cline model catalog.  Anthropic's API requires
+# 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 4.6
+    "claude-opus-4-6":   128_000,
+    "claude-sonnet-4-6":  64_000,
+    # Claude 4.5
+    "claude-opus-4-5":    64_000,
+    "claude-sonnet-4-5":  64_000,
+    "claude-haiku-4-5":   64_000,
+    # Claude 4
+    "claude-opus-4":      32_000,
+    "claude-sonnet-4":    64_000,
+    # Claude 3.7
+    "claude-3-7-sonnet": 128_000,
+    # Claude 3.5
+    "claude-3-5-sonnet":   8_192,
+    "claude-3-5-haiku":    8_192,
+    # Claude 3
+    "claude-3-opus":       4_096,
+    "claude-3-sonnet":     4_096,
+    "claude-3-haiku":      4_096,
+}
+
+# For any model not in the table, assume the highest current limit.
+# Future Anthropic models are unlikely to have *less* output capacity.
+_ANTHROPIC_DEFAULT_OUTPUT_LIMIT = 128_000
+
+
+def _get_anthropic_max_output(model: str) -> int:
+    """Look up the max output token limit for an Anthropic model.
+
+    Uses substring matching against _ANTHROPIC_OUTPUT_LIMITS so date-stamped
+    model IDs (claude-sonnet-4-5-20250929) and variant suffixes (:1m, :fast)
+    resolve correctly.  Longest-prefix match wins to avoid e.g. "claude-3-5"
+    matching before "claude-3-5-sonnet".
+    """
+    m = model.lower()
+    best_key = ""
+    best_val = _ANTHROPIC_DEFAULT_OUTPUT_LIMIT
+    for key, val in _ANTHROPIC_OUTPUT_LIMITS.items():
+        if key in m and len(key) > len(best_key):
+            best_key = key
+            best_val = val
+    return best_val
+
 
 def _supports_adaptive_thinking(model: str) -> bool:
     """Return True for Claude 4.6 models that support adaptive thinking."""
@@ -818,9 +866,15 @@ def build_anthropic_kwargs(
     tool_choice: Optional[str] = None,
     is_oauth: bool = False,
     preserve_dots: bool = False,
+    context_length: Optional[int] = None,
 ) -> Dict[str, Any]:
     """Build kwargs for anthropic.messages.create().
 
+    When *max_tokens* is None, the model's native output limit is used
+    (e.g. 128K for Opus 4.6, 64K for Sonnet 4.6).  If *context_length*
+    is provided, the effective limit is clamped so it doesn't exceed
+    the context window.
+
     When *is_oauth* is True, applies Claude Code compatibility transforms:
     system prompt prefix, tool name prefixing, and prompt sanitization.
 
@@ -831,7 +885,12 @@ def build_anthropic_kwargs(
     anthropic_tools = convert_tools_to_anthropic(tools) if tools else []
 
     model = normalize_model_name(model, preserve_dots=preserve_dots)
-    effective_max_tokens = max_tokens or 16384
+    effective_max_tokens = max_tokens or _get_anthropic_max_output(model)
+
+    # Clamp to context window if the user set a lower context_length
+    # (e.g. custom endpoint with limited capacity).
+    if context_length and effective_max_tokens > context_length:
+        effective_max_tokens = max(context_length - 1, 1)
 
     # ── OAuth: Claude Code identity ──────────────────────────────────
     if is_oauth:

agent/auxiliary_client.py

@@ -1458,6 +1458,29 @@ def _resolve_task_provider_model(
     return "auto", resolved_model, None, None
 
 
+_DEFAULT_AUX_TIMEOUT = 30.0
+
+
+def _get_task_timeout(task: str, default: float = _DEFAULT_AUX_TIMEOUT) -> float:
+    """Read timeout from auxiliary.{task}.timeout in config, falling back to *default*."""
+    if not task:
+        return default
+    try:
+        from hermes_cli.config import load_config
+        config = load_config()
+    except ImportError:
+        return default
+    aux = config.get("auxiliary", {}) if isinstance(config, dict) else {}
+    task_config = aux.get(task, {}) if isinstance(aux, dict) else {}
+    raw = task_config.get("timeout")
+    if raw is not None:
+        try:
+            return float(raw)
+        except (ValueError, TypeError):
+            pass
+    return default
+
+
 def _build_call_kwargs(
     provider: str,
     model: str,
@@ -1515,7 +1538,7 @@ def call_llm(
     temperature: float = None,
     max_tokens: int = None,
     tools: list = None,
-    timeout: float = 30.0,
+    timeout: float = None,
     extra_body: dict = None,
 ) -> Any:
     """Centralized synchronous LLM call.
@@ -1533,7 +1556,7 @@ def call_llm(
         temperature: Sampling temperature (None = provider default).
         max_tokens: Max output tokens (handles max_tokens vs max_completion_tokens).
         tools: Tool definitions (for function calling).
-        timeout: Request timeout in seconds.
+        timeout: Request timeout in seconds (None = read from auxiliary.{task}.timeout config).
         extra_body: Additional request body fields.
 
     Returns:
@@ -1598,10 +1621,12 @@ def call_llm(
                 f"No LLM provider configured for task={task} provider={resolved_provider}. "
                 f"Run: hermes setup")
 
+    effective_timeout = timeout if timeout is not None else _get_task_timeout(task)
+
     kwargs = _build_call_kwargs(
         resolved_provider, final_model, messages,
         temperature=temperature, max_tokens=max_tokens,
-        tools=tools, timeout=timeout, extra_body=extra_body,
+        tools=tools, timeout=effective_timeout, extra_body=extra_body,
         base_url=resolved_base_url)
 
     # Handle max_tokens vs max_completion_tokens retry
@@ -1616,6 +1641,62 @@ def call_llm(
         raise
 
 
+def extract_content_or_reasoning(response) -> str:
+    """Extract content from an LLM response, falling back to reasoning fields.
+
+    Mirrors the main agent loop's behavior when a reasoning model (DeepSeek-R1,
+    Qwen-QwQ, etc.) returns ``content=None`` with reasoning in structured fields.
+
+    Resolution order:
+      1. ``message.content`` — strip inline think/reasoning blocks, check for
+         remaining non-whitespace text.
+      2. ``message.reasoning`` / ``message.reasoning_content`` — direct
+         structured reasoning fields (DeepSeek, Moonshot, Novita, etc.).
+      3. ``message.reasoning_details`` — OpenRouter unified array format.
+
+    Returns the best available text, or ``""`` if nothing found.
+    """
+    import re
+
+    msg = response.choices[0].message
+    content = (msg.content or "").strip()
+
+    if content:
+        # Strip inline think/reasoning blocks (mirrors _strip_think_blocks)
+        cleaned = re.sub(
+            r"<(?:think|thinking|reasoning|REASONING_SCRATCHPAD)>"
+            r".*?"
+            r"</(?:think|thinking|reasoning|REASONING_SCRATCHPAD)>",
+            "", content, flags=re.DOTALL | re.IGNORECASE,
+        ).strip()
+        if cleaned:
+            return cleaned
+
+    # Content is empty or reasoning-only — try structured reasoning fields
+    reasoning_parts: list[str] = []
+    for field in ("reasoning", "reasoning_content"):
+        val = getattr(msg, field, None)
+        if val and isinstance(val, str) and val.strip() and val not in reasoning_parts:
+            reasoning_parts.append(val.strip())
+
+    details = getattr(msg, "reasoning_details", None)
+    if details and isinstance(details, list):
+        for detail in details:
+            if isinstance(detail, dict):
+                summary = (
+                    detail.get("summary")
+                    or detail.get("content")
+                    or detail.get("text")
+                )
+                if summary and summary not in reasoning_parts:
+                    reasoning_parts.append(summary.strip() if isinstance(summary, str) else str(summary))
+
+    if reasoning_parts:
+        return "\n\n".join(reasoning_parts)
+
+    return ""
+
+
 async def async_call_llm(
     task: str = None,
     *,
@@ -1627,7 +1708,7 @@ async def async_call_llm(
     temperature: float = None,
     max_tokens: int = None,
     tools: list = None,
-    timeout: float = 30.0,
+    timeout: float = None,
     extra_body: dict = None,
 ) -> Any:
     """Centralized asynchronous LLM call.
@@ -1688,10 +1769,12 @@ async def async_call_llm(
                 f"No LLM provider configured for task={task} provider={resolved_provider}. "
                 f"Run: hermes setup")
 
+    effective_timeout = timeout if timeout is not None else _get_task_timeout(task)
+
     kwargs = _build_call_kwargs(
         resolved_provider, final_model, messages,
         temperature=temperature, max_tokens=max_tokens,
-        tools=tools, timeout=timeout, extra_body=extra_body,
+        tools=tools, timeout=effective_timeout, extra_body=extra_body,
         base_url=resolved_base_url)
 
     try:

agent/context_compressor.py

@@ -141,7 +141,7 @@ class ContextCompressor:
             "last_prompt_tokens": self.last_prompt_tokens,
             "threshold_tokens": self.threshold_tokens,
             "context_length": self.context_length,
-            "usage_percent": (self.last_prompt_tokens / self.context_length * 100) if self.context_length else 0,
+            "usage_percent": min(100, (self.last_prompt_tokens / self.context_length * 100)) if self.context_length else 0,
             "compression_count": self.compression_count,
         }
 
@@ -347,7 +347,7 @@ Write only the summary body. Do not include any preamble or prefix."""
                 "messages": [{"role": "user", "content": prompt}],
                 "temperature": 0.3,
                 "max_tokens": summary_budget * 2,
-                "timeout": 45.0,
+                # timeout resolved from auxiliary.compression.timeout config by call_llm
             }
             if self.summary_model:
                 call_kwargs["model"] = self.summary_model

agent/context_references.py

@@ -286,12 +286,16 @@ def _expand_git_reference(
     args: list[str],
     label: str,
 ) -> tuple[str | None, str | None]:
-    result = subprocess.run(
-        ["git", *args],
-        cwd=cwd,
-        capture_output=True,
-        text=True,
-    )
+    try:
+        result = subprocess.run(
+            ["git", *args],
+            cwd=cwd,
+            capture_output=True,
+            text=True,
+            timeout=30,
+        )
+    except subprocess.TimeoutExpired:
+        return f"{ref.raw}: git command timed out (30s)", None
     if result.returncode != 0:
         stderr = (result.stderr or "").strip() or "git command failed"
         return f"{ref.raw}: {stderr}", None
@@ -449,9 +453,12 @@ def _rg_files(path: Path, cwd: Path, limit: int) -> list[Path] | None:
             cwd=cwd,
             capture_output=True,
             text=True,
+            timeout=10,
         )
     except FileNotFoundError:
         return None
+    except subprocess.TimeoutExpired:
+        return None
     if result.returncode != 0:
         return None
     files = [Path(line.strip()) for line in result.stdout.splitlines() if line.strip()]

agent/display.py

@@ -284,11 +284,11 @@ class KawaiiSpinner:
         The CLI already drives a TUI widget (_spinner_text) for spinner display,
         so KawaiiSpinner's \\r-based animation is redundant under StdoutProxy.
         """
-        out = self._out
-        # StdoutProxy has a 'raw' attribute (bool) that plain file objects lack.
-        if hasattr(out, 'raw') and type(out).__name__ == 'StdoutProxy':
-            return True
-        return False
+        try:
+            from prompt_toolkit.patch_stdout import StdoutProxy
+            return isinstance(self._out, StdoutProxy)
+        except ImportError:
+            return False
 
     def _animate(self):
         # When stdout is not a real terminal (e.g. Docker, systemd, pipe),
@@ -699,7 +699,7 @@ def format_context_pressure(
         threshold_percent: Compaction threshold as a fraction of context window.
         compression_enabled: Whether auto-compression is active.
     """
-    pct_int = int(compaction_progress * 100)
+    pct_int = min(int(compaction_progress * 100), 100)
     filled = min(int(compaction_progress * _BAR_WIDTH), _BAR_WIDTH)
     bar = _BAR_FILLED * filled + _BAR_EMPTY * (_BAR_WIDTH - filled)
 
@@ -729,7 +729,7 @@ def format_context_pressure_gateway(
     No ANSI — just Unicode and plain text suitable for Telegram/Discord/etc.
     The percentage shows progress toward the compaction threshold.
     """
-    pct_int = int(compaction_progress * 100)
+    pct_int = min(int(compaction_progress * 100), 100)
     filled = min(int(compaction_progress * _BAR_WIDTH), _BAR_WIDTH)
     bar = _BAR_FILLED * filled + _BAR_EMPTY * (_BAR_WIDTH - filled)
 

agent/model_metadata.py

@@ -113,6 +113,15 @@ DEFAULT_CONTEXT_LENGTHS = {
     "glm": 202752,
     # Kimi
     "kimi": 262144,
+    # Hugging Face Inference Providers — model IDs use org/name format
+    "Qwen/Qwen3.5-397B-A17B": 131072,
+    "Qwen/Qwen3.5-35B-A3B": 131072,
+    "deepseek-ai/DeepSeek-V3.2": 65536,
+    "moonshotai/Kimi-K2.5": 262144,
+    "moonshotai/Kimi-K2-Thinking": 262144,
+    "MiniMaxAI/MiniMax-M2.5": 204800,
+    "XiaomiMiMo/MiMo-V2-Flash": 32768,
+    "zai-org/GLM-5": 202752,
 }
 
 _CONTEXT_LENGTH_KEYS = (

agent/models_dev.py

@@ -15,6 +15,8 @@ import time
 from pathlib import Path
 from typing import Any, Dict, Optional
 
+from utils import atomic_json_write
+
 import requests
 
 logger = logging.getLogger(__name__)
@@ -64,12 +66,10 @@ def _load_disk_cache() -> Dict[str, Any]:
 
 
 def _save_disk_cache(data: Dict[str, Any]) -> None:
-    """Save models.dev data to disk cache."""
+    """Save models.dev data to disk cache atomically."""
     try:
         cache_path = _get_cache_path()
-        cache_path.parent.mkdir(parents=True, exist_ok=True)
-        with open(cache_path, "w", encoding="utf-8") as f:
-            json.dump(data, f, separators=(",", ":"))
+        atomic_json_write(cache_path, data, indent=None, separators=(",", ":"))
     except Exception as e:
         logger.debug("Failed to save models.dev disk cache: %s", e)
 

agent/prompt_builder.py

@@ -18,6 +18,7 @@ from typing import Optional
 from agent.skill_utils import (
     extract_skill_conditions,
     extract_skill_description,
+    get_all_skills_dirs,
     get_disabled_skill_names,
     iter_skill_index_files,
     parse_frontmatter,
@@ -169,6 +170,25 @@ SKILLS_GUIDANCE = (
     "Skills that aren't maintained become liabilities."
 )
 
+TOOL_USE_ENFORCEMENT_GUIDANCE = (
+    "# Tool-use enforcement\n"
+    "You MUST use your tools to take action — do not describe what you would do "
+    "or plan to do without actually doing it. When you say you will perform an "
+    "action (e.g. 'I will run the tests', 'Let me check the file', 'I will create "
+    "the project'), you MUST immediately make the corresponding tool call in the same "
+    "response. Never end your turn with a promise of future action — execute it now.\n"
+    "Keep working until the task is actually complete. Do not stop with a summary of "
+    "what you plan to do next time. If you have tools available that can accomplish "
+    "the task, use them instead of telling the user what you would do.\n"
+    "Every response should either (a) contain tool calls that make progress, or "
+    "(b) deliver a final result to the user. Responses that only describe intentions "
+    "without acting are not acceptable."
+)
+
+# Model name substrings that trigger tool-use enforcement guidance.
+# Add new patterns here when a model family needs explicit steering.
+TOOL_USE_ENFORCEMENT_MODELS = ("gpt", "codex")
+
 PLATFORM_HINTS = {
     "whatsapp": (
         "You are on a text messaging communication platform, WhatsApp. "
@@ -425,16 +445,23 @@ def build_skills_system_prompt(
          mtime/size manifest — survives process restarts
 
     Falls back to a full filesystem scan when both layers miss.
+
+    External skill directories (``skills.external_dirs`` in config.yaml) are
+    scanned alongside the local ``~/.hermes/skills/`` directory.  External dirs
+    are read-only — they appear in the index but new skills are always created
+    in the local dir.  Local skills take precedence when names collide.
     """
     hermes_home = get_hermes_home()
     skills_dir = hermes_home / "skills"
+    external_dirs = get_all_skills_dirs()[1:]  # skip local (index 0)
 
-    if not skills_dir.exists():
+    if not skills_dir.exists() and not external_dirs:
         return ""
 
     # ── Layer 1: in-process LRU cache ─────────────────────────────────
     cache_key = (
         str(skills_dir.resolve()),
+        tuple(str(d) for d in external_dirs),
         tuple(sorted(str(t) for t in (available_tools or set()))),
         tuple(sorted(str(ts) for ts in (available_toolsets or set()))),
     )
@@ -521,6 +548,56 @@ def build_skills_system_prompt(
             category_descriptions,
         )
 
+    # ── External skill directories ─────────────────────────────────────
+    # Scan external dirs directly (no snapshot caching — they're read-only
+    # and typically small).  Local skills already in skills_by_category take
+    # precedence: we track seen names and skip duplicates from external dirs.
+    seen_skill_names: set[str] = set()
+    for cat_skills in skills_by_category.values():
+        for name, _desc in cat_skills:
+            seen_skill_names.add(name)
+
+    for ext_dir in external_dirs:
+        if not ext_dir.exists():
+            continue
+        for skill_file in iter_skill_index_files(ext_dir, "SKILL.md"):
+            try:
+                is_compatible, frontmatter, desc = _parse_skill_file(skill_file)
+                if not is_compatible:
+                    continue
+                entry = _build_snapshot_entry(skill_file, ext_dir, frontmatter, desc)
+                skill_name = entry["skill_name"]
+                if skill_name in seen_skill_names:
+                    continue
+                if entry["frontmatter_name"] in disabled or skill_name in disabled:
+                    continue
+                if not _skill_should_show(
+                    extract_skill_conditions(frontmatter),
+                    available_tools,
+                    available_toolsets,
+                ):
+                    continue
+                seen_skill_names.add(skill_name)
+                skills_by_category.setdefault(entry["category"], []).append(
+                    (skill_name, entry["description"])
+                )
+            except Exception as e:
+                logger.debug("Error reading external skill %s: %s", skill_file, e)
+
+        # External category descriptions
+        for desc_file in iter_skill_index_files(ext_dir, "DESCRIPTION.md"):
+            try:
+                content = desc_file.read_text(encoding="utf-8")
+                fm, _ = parse_frontmatter(content)
+                cat_desc = fm.get("description")
+                if not cat_desc:
+                    continue
+                rel = desc_file.relative_to(ext_dir)
+                cat = "/".join(rel.parts[:-1]) if len(rel.parts) > 1 else "general"
+                category_descriptions.setdefault(cat, str(cat_desc).strip().strip("'\""))
+            except Exception as e:
+                logger.debug("Could not read external skill description %s: %s", desc_file, e)
+
     if not skills_by_category:
         result = ""
     else:

agent/skill_commands.py

@@ -128,7 +128,11 @@ def _build_skill_message(
                         supporting.append(rel)
 
     if supporting and skill_dir:
-        skill_view_target = str(skill_dir.relative_to(SKILLS_DIR))
+        try:
+            skill_view_target = str(skill_dir.relative_to(SKILLS_DIR))
+        except ValueError:
+            # Skill is from an external dir — use the skill name instead
+            skill_view_target = skill_dir.name
         parts.append("")
         parts.append("[This skill has supporting files you can load with the skill_view tool:]")
         for sf in supporting:
@@ -158,38 +162,49 @@ def scan_skill_commands() -> Dict[str, Dict[str, Any]]:
     _skill_commands = {}
     try:
         from tools.skills_tool import SKILLS_DIR, _parse_frontmatter, skill_matches_platform, _get_disabled_skill_names
-        if not SKILLS_DIR.exists():
-            return _skill_commands
+        from agent.skill_utils import get_external_skills_dirs
         disabled = _get_disabled_skill_names()
-        for skill_md in SKILLS_DIR.rglob("SKILL.md"):
-            if any(part in ('.git', '.github', '.hub') for part in skill_md.parts):
-                continue
-            try:
-                content = skill_md.read_text(encoding='utf-8')
-                frontmatter, body = _parse_frontmatter(content)
-                # Skip skills incompatible with the current OS platform
-                if not skill_matches_platform(frontmatter):
+        seen_names: set = set()
+
+        # Scan local dir first, then external dirs
+        dirs_to_scan = []
+        if SKILLS_DIR.exists():
+            dirs_to_scan.append(SKILLS_DIR)
+        dirs_to_scan.extend(get_external_skills_dirs())
+
+        for scan_dir in dirs_to_scan:
+            for skill_md in scan_dir.rglob("SKILL.md"):
+                if any(part in ('.git', '.github', '.hub') for part in skill_md.parts):
                     continue
-                name = frontmatter.get('name', skill_md.parent.name)
-                # Respect user's disabled skills config
-                if name in disabled:
+                try:
+                    content = skill_md.read_text(encoding='utf-8')
+                    frontmatter, body = _parse_frontmatter(content)
+                    # Skip skills incompatible with the current OS platform
+                    if not skill_matches_platform(frontmatter):
+                        continue
+                    name = frontmatter.get('name', skill_md.parent.name)
+                    if name in seen_names:
+                        continue
+                    # Respect user's disabled skills config
+                    if name in disabled:
+                        continue
+                    description = frontmatter.get('description', '')
+                    if not description:
+                        for line in body.strip().split('\n'):
+                            line = line.strip()
+                            if line and not line.startswith('#'):
+                                description = line[:80]
+                                break
+                    seen_names.add(name)
+                    cmd_name = name.lower().replace(' ', '-').replace('_', '-')
+                    _skill_commands[f"/{cmd_name}"] = {
+                        "name": name,
+                        "description": description or f"Invoke the {name} skill",
+                        "skill_md_path": str(skill_md),
+                        "skill_dir": str(skill_md.parent),
+                    }
+                except Exception:
                     continue
-                description = frontmatter.get('description', '')
-                if not description:
-                    for line in body.strip().split('\n'):
-                        line = line.strip()
-                        if line and not line.startswith('#'):
-                            description = line[:80]
-                            break
-                cmd_name = name.lower().replace(' ', '-').replace('_', '-')
-                _skill_commands[f"/{cmd_name}"] = {
-                    "name": name,
-                    "description": description or f"Invoke the {name} skill",
-                    "skill_md_path": str(skill_md),
-                    "skill_dir": str(skill_md.parent),
-                }
-            except Exception:
-                continue
     except Exception:
         pass
     return _skill_commands

agent/skill_utils.py

@@ -158,6 +158,73 @@ def _normalize_string_set(values) -> Set[str]:
     return {str(v).strip() for v in values if str(v).strip()}
 
 
+# ── External skills directories ──────────────────────────────────────────
+
+
+def get_external_skills_dirs() -> List[Path]:
+    """Read ``skills.external_dirs`` from config.yaml and return validated paths.
+
+    Each entry is expanded (``~`` and ``${VAR}``) and resolved to an absolute
+    path.  Only directories that actually exist are returned.  Duplicates and
+    paths that resolve to the local ``~/.hermes/skills/`` are silently skipped.
+    """
+    config_path = get_hermes_home() / "config.yaml"
+    if not config_path.exists():
+        return []
+    try:
+        parsed = yaml_load(config_path.read_text(encoding="utf-8"))
+    except Exception:
+        return []
+    if not isinstance(parsed, dict):
+        return []
+
+    skills_cfg = parsed.get("skills")
+    if not isinstance(skills_cfg, dict):
+        return []
+
+    raw_dirs = skills_cfg.get("external_dirs")
+    if not raw_dirs:
+        return []
+    if isinstance(raw_dirs, str):
+        raw_dirs = [raw_dirs]
+    if not isinstance(raw_dirs, list):
+        return []
+
+    local_skills = (get_hermes_home() / "skills").resolve()
+    seen: Set[Path] = set()
+    result: List[Path] = []
+
+    for entry in raw_dirs:
+        entry = str(entry).strip()
+        if not entry:
+            continue
+        # Expand ~ and environment variables
+        expanded = os.path.expanduser(os.path.expandvars(entry))
+        p = Path(expanded).resolve()
+        if p == local_skills:
+            continue
+        if p in seen:
+            continue
+        if p.is_dir():
+            seen.add(p)
+            result.append(p)
+        else:
+            logger.debug("External skills dir does not exist, skipping: %s", p)
+
+    return result
+
+
+def get_all_skills_dirs() -> List[Path]:
+    """Return all skill directories: local ``~/.hermes/skills/`` first, then external.
+
+    The local dir is always first (and always included even if it doesn't exist
+    yet — callers handle that).  External dirs follow in config order.
+    """
+    dirs = [get_hermes_home() / "skills"]
+    dirs.extend(get_external_skills_dirs())
+    return dirs
+
+
 # ── Condition extraction ──────────────────────────────────────────────────
 
 

agent/title_generator.py

@@ -19,7 +19,7 @@ _TITLE_PROMPT = (
 )
 
 
-def generate_title(user_message: str, assistant_response: str, timeout: float = 15.0) -> Optional[str]:
+def generate_title(user_message: str, assistant_response: str, timeout: float = 30.0) -> Optional[str]:
     """Generate a session title from the first exchange.
 
     Uses the auxiliary LLM client (cheapest/fastest available model).

cli-config.yaml.example

@@ -7,6 +7,7 @@
 # =============================================================================
 model:
   # Default model to use (can be overridden with --model flag)
+  # Both "default" and "model" work as the key name here.
   default: "anthropic/claude-opus-4.6"
   
   # Inference provider selection:
@@ -401,6 +402,15 @@ skills:
   # Set to 0 to disable.
   creation_nudge_interval: 15
 
+  # External skill directories — share skills across tools/agents without
+  # copying them into ~/.hermes/skills/.  Each path is expanded (~ and ${VAR})
+  # and resolved to an absolute path.  External dirs are read-only: skill
+  # creation always writes to ~/.hermes/skills/.  Local skills take precedence
+  # when names collide.
+  # external_dirs:
+  #   - ~/.agents/skills
+  #   - /home/shared/team-skills
+
 # =============================================================================
 # Agent Behavior
 # =============================================================================

cli.py

@@ -70,7 +70,7 @@ _COMMAND_SPINNER_FRAMES = ("⠋", "⠙", "⠹", "⠸", "⠼", "⠴", "⠦", "⠧
 
 # Load .env from ~/.hermes/.env first, then project root as dev fallback.
 # User-managed env files should override stale shell exports on restart.
-from hermes_constants import get_hermes_home, OPENROUTER_BASE_URL
+from hermes_constants import get_hermes_home, display_hermes_home, OPENROUTER_BASE_URL
 from hermes_cli.env_loader import load_hermes_dotenv
 
 _hermes_home = get_hermes_home()
@@ -1078,12 +1078,12 @@ class HermesCLI:
         # authoritative.  This avoids conflicts in multi-agent setups where
         # env vars would stomp each other.
         _model_config = CLI_CONFIG.get("model", {})
-        _config_model = _model_config.get("default", "") if isinstance(_model_config, dict) else (_model_config or "")
+        _config_model = (_model_config.get("default") or _model_config.get("model") or "") if isinstance(_model_config, dict) else (_model_config or "")
         _FALLBACK_MODEL = "anthropic/claude-opus-4.6"
         self.model = model or _config_model or _FALLBACK_MODEL
         # Auto-detect model from local server if still on fallback
         if self.model == _FALLBACK_MODEL:
-            _base_url = _model_config.get("base_url", "") if isinstance(_model_config, dict) else ""
+            _base_url = (_model_config.get("base_url") or "") if isinstance(_model_config, dict) else ""
             if "localhost" in _base_url or "127.0.0.1" in _base_url:
                 from hermes_cli.runtime_provider import _auto_detect_local_model
                 _detected = _auto_detect_local_model(_base_url)
@@ -1182,9 +1182,13 @@ class HermesCLI:
         self._provider_require_params = pr.get("require_parameters", False)
         self._provider_data_collection = pr.get("data_collection")
         
-        # Fallback model config — tried when primary provider fails after retries
-        fb = CLI_CONFIG.get("fallback_model") or {}
-        self._fallback_model = fb if fb.get("provider") and fb.get("model") else None
+        # Fallback provider chain — tried in order when primary fails after retries.
+        # Supports new list format (fallback_providers) and legacy single-dict (fallback_model).
+        fb = CLI_CONFIG.get("fallback_providers") or CLI_CONFIG.get("fallback_model") or []
+        # Normalize legacy single-dict to a one-element list
+        if isinstance(fb, dict):
+            fb = [fb] if fb.get("provider") and fb.get("model") else []
+        self._fallback_model = fb
 
         # Optional cheap-vs-strong routing for simple turns
         self._smart_model_routing = CLI_CONFIG.get("smart_model_routing", {}) or {}
@@ -3594,7 +3598,7 @@ class HermesCLI:
             print("  To start the gateway:")
             print("    python cli.py --gateway")
             print()
-            print("  Configuration file: ~/.hermes/config.yaml")
+            print(f"  Configuration file: {display_hermes_home()}/config.yaml")
             print()
             
         except Exception as e:
@@ -3604,7 +3608,7 @@ class HermesCLI:
             print("    1. Set environment variables:")
             print("       TELEGRAM_BOT_TOKEN=your_token")
             print("       DISCORD_BOT_TOKEN=your_token")
-            print("    2. Or configure settings in ~/.hermes/config.yaml")
+            print(f"    2. Or configure settings in {display_hermes_home()}/config.yaml")
             print()
     
     def process_command(self, command: str) -> bool:
@@ -3811,7 +3815,7 @@ class HermesCLI:
                 plugins = mgr.list_plugins()
                 if not plugins:
                     print("No plugins installed.")
-                    print("Drop plugin directories into ~/.hermes/plugins/ to get started.")
+                    print(f"Drop plugin directories into {display_hermes_home()}/plugins/ to get started.")
                 else:
                     print(f"Plugins ({len(plugins)}):")
                     for p in plugins:
@@ -4034,6 +4038,17 @@ class HermesCLI:
                     provider_data_collection=self._provider_data_collection,
                     fallback_model=self._fallback_model,
                 )
+                # Silence raw spinner; route thinking through TUI widget when no foreground agent is active.
+                bg_agent._print_fn = lambda *_a, **_kw: None
+
+                def _bg_thinking(text: str) -> None:
+                    # Concurrent bg tasks may race on _spinner_text; acceptable for best-effort UI.
+                    if not self._agent_running:
+                        self._spinner_text = text
+                        if self._app:
+                            self._app.invalidate()
+
+                bg_agent.thinking_callback = _bg_thinking
 
                 result = bg_agent.run_conversation(
                     user_message=prompt,
@@ -4096,6 +4111,9 @@ class HermesCLI:
                 _cprint(f"  ❌ Background task #{task_num} failed: {e}")
             finally:
                 self._background_tasks.pop(task_id, None)
+                # Clear spinner only if no foreground agent owns it
+                if not self._agent_running:
+                    self._spinner_text = ""
                 if self._app:
                     self._invalidate(min_interval=0)
 
@@ -4326,7 +4344,7 @@ class HermesCLI:
                 source = f" ({s['source']})" if s["source"] == "user" else ""
                 print(f"   {marker} {s['name']}{source} — {s['description']}")
             print("\n  Usage: /skin <name>")
-            print("  Custom skins: drop a YAML file in ~/.hermes/skins/\n")
+            print(f"  Custom skins: drop a YAML file in {display_hermes_home()}/skins/\n")
             return
 
         new_skin = parts[1].strip().lower()
@@ -4506,7 +4524,7 @@ class HermesCLI:
         compressor = agent.context_compressor
         last_prompt = compressor.last_prompt_tokens
         ctx_len = compressor.context_length
-        pct = (last_prompt / ctx_len * 100) if ctx_len else 0
+        pct = min(100, (last_prompt / ctx_len * 100)) if ctx_len else 0
         compressions = compressor.compression_count
 
         msg_count = len(self.conversation_history)
@@ -5534,6 +5552,13 @@ class HermesCLI:
             except Exception as e:
                 logging.debug("@ context reference expansion failed: %s", e)
 
+        # Sanitize surrogate characters that can arrive via clipboard paste from
+        # rich-text editors (Google Docs, Word, etc.).  Lone surrogates are invalid
+        # UTF-8 and crash JSON serialization in the OpenAI SDK.
+        if isinstance(message, str):
+            from run_agent import _sanitize_surrogates
+            message = _sanitize_surrogates(message)
+
         # Add user message to history
         self.conversation_history.append({"role": "user", "content": message})
 
@@ -5891,10 +5916,22 @@ class HermesCLI:
             else:
                 duration_str = f"{seconds}s"
             
+            # Look up session title for resume-by-name hint
+            session_title = None
+            if self._session_db:
+                try:
+                    session_title = self._session_db.get_session_title(self.session_id)
+                except Exception:
+                    pass
+
             print("Resume this session with:")
             print(f"  hermes --resume {self.session_id}")
+            if session_title:
+                print(f"  hermes -c \"{session_title}\"")
             print()
             print(f"Session:        {self.session_id}")
+            if session_title:
+                print(f"Title:          {session_title}")
             print(f"Duration:       {duration_str}")
             print(f"Messages:       {msg_count} ({user_msgs} user, {tool_calls} tool calls)")
         else:
@@ -5911,6 +5948,9 @@ class HermesCLI:
         ``normal_prompt`` is the full ``branding.prompt_symbol``.
         ``state_suffix`` is what special states (sudo/secret/approval/agent)
         should render after their leading icon.
+
+        When a profile is active (not "default"), the profile name is
+        prepended to the prompt symbol: ``coder ❯`` instead of ``❯``.
         """
         try:
             from hermes_cli.skin_engine import get_active_prompt_symbol
@@ -5919,6 +5959,15 @@ class HermesCLI:
             symbol = "❯ "
 
         symbol = (symbol or "❯ ").rstrip() + " "
+
+        # Prepend profile name when not default
+        try:
+            from hermes_cli.profiles import get_active_profile_name
+            profile = get_active_profile_name()
+            if profile not in ("default", "custom"):
+                symbol = f"{profile} {symbol}"
+        except Exception:
+            pass
         stripped = symbol.rstrip()
         if not stripped:
             return "❯ ", "❯ "
@@ -6070,7 +6119,7 @@ class HermesCLI:
             from honcho_integration.client import HonchoClientConfig
             from agent.display import honcho_session_line, write_tty
             hcfg = HonchoClientConfig.from_global_config()
-            if hcfg.enabled and hcfg.api_key and hcfg.explicitly_configured:
+            if hcfg.enabled and (hcfg.api_key or hcfg.base_url) and hcfg.explicitly_configured:
                 sname = hcfg.resolve_session_name(session_id=self.session_id)
                 if sname:
                     write_tty(honcho_session_line(hcfg.workspace_id, sname) + "\n")
@@ -6157,10 +6206,18 @@ class HermesCLI:
         set_approval_callback(self._approval_callback)
         set_secret_capture_callback(self._secret_capture_callback)
 
-        # Ensure tirith security scanner is available (downloads if needed)
+        # Ensure tirith security scanner is available (downloads if needed).
+        # Warn the user if tirith is enabled in config but not available,
+        # so they know command security scanning is degraded.
         try:
             from tools.tirith_security import ensure_installed
-            ensure_installed(log_failures=False)
+            tirith_path = ensure_installed(log_failures=False)
+            if tirith_path is None:
+                security_cfg = self.config.get("security", {}) or {}
+                tirith_enabled = security_cfg.get("tirith_enabled", True)
+                if tirith_enabled:
+                    _cprint(f"  {_DIM}⚠ tirith security scanner enabled but not available "
+                            f"— command scanning will use pattern matching only{_RST}")
         except Exception:
             pass  # Non-fatal — fail-open at scan time if unavailable
         
@@ -6447,6 +6504,24 @@ class HermesCLI:
             self._should_exit = True
             event.app.exit()
 
+        @kb.add('c-z')
+        def handle_ctrl_z(event):
+            """Handle Ctrl+Z - suspend process to background (Unix only)."""
+            import sys
+            if sys.platform == 'win32':
+                _cprint(f"\n{_DIM}Suspend (Ctrl+Z) is not supported on Windows.{_RST}")
+                event.app.invalidate()
+                return
+            import os, signal as _sig
+            from prompt_toolkit.application import run_in_terminal
+            from hermes_cli.skin_engine import get_active_skin
+            agent_name = get_active_skin().get_branding("agent_name", "Hermes Agent")
+            msg = f"\n{agent_name} has been suspended. Run `fg` to bring {agent_name} back."
+            def _suspend():
+                os.write(1, msg.encode())
+                os.kill(0, _sig.SIGTSTP)
+            run_in_terminal(_suspend)
+
         # Voice push-to-talk key: configurable via config.yaml (voice.record_key)
         # Default: Ctrl+B (avoids conflict with Ctrl+R readline reverse-search)
         # Config uses "ctrl+b" format; prompt_toolkit expects "c-b" format.
@@ -6636,6 +6711,7 @@ class HermesCLI:
         # Paste collapsing: detect large pastes and save to temp file
         _paste_counter = [0]
         _prev_text_len = [0]
+        _prev_newline_count = [0]
         _paste_just_collapsed = [False]
 
         def _on_text_changed(buf):
@@ -6644,18 +6720,27 @@ class HermesCLI:
             When bracketed paste is available, handle_paste collapses
             large pastes directly.  This handler is a fallback for
             terminals without bracketed paste support.
+
+            Two heuristics (either triggers collapse):
+            1. Many characters added at once (chars_added > 1) — works
+               when the terminal delivers the paste in one event-loop tick.
+            2. Newline count jumped by 4+ in a single text-change event —
+               catches terminals that feed characters individually but
+               still batch newlines.  Alt+Enter only adds 1 newline per
+               event so it never triggers this.
             """
             text = buf.text
             chars_added = len(text) - _prev_text_len[0]
             _prev_text_len[0] = len(text)
             if _paste_just_collapsed[0]:
                 _paste_just_collapsed[0] = False
+                _prev_newline_count[0] = text.count('\n')
                 return
             line_count = text.count('\n')
-            # Heuristic: a real paste adds many characters at once (not just a
-            # single newline from Alt+Enter) AND the result has 5+ lines.
-            # Fallback for terminals without bracketed paste support.
-            if line_count >= 5 and chars_added > 1 and not text.startswith('/'):
+            newlines_added = line_count - _prev_newline_count[0]
+            _prev_newline_count[0] = line_count
+            is_paste = chars_added > 1 or newlines_added >= 4
+            if line_count >= 5 and is_paste and not text.startswith('/'):
                 _paste_counter[0] += 1
                 # Save to temp file
                 paste_dir = _hermes_home / "pastes"
@@ -6663,6 +6748,7 @@ class HermesCLI:
                 paste_file = paste_dir / f"paste_{_paste_counter[0]}_{datetime.now().strftime('%H%M%S')}.txt"
                 paste_file.write_text(text, encoding="utf-8")
                 # Replace buffer with compact reference
+                _paste_just_collapsed[0] = True
                 buf.text = f"[Pasted text #{_paste_counter[0]}: {line_count + 1} lines \u2192 {paste_file}]"
                 buf.cursor_position = len(buf.text)
 

cron/jobs.py

@@ -327,7 +327,20 @@ def load_jobs() -> List[Dict[str, Any]]:
         with open(JOBS_FILE, 'r', encoding='utf-8') as f:
             data = json.load(f)
             return data.get("jobs", [])
-    except (json.JSONDecodeError, IOError):
+    except json.JSONDecodeError:
+        # Retry with strict=False to handle bare control chars in string values
+        try:
+            with open(JOBS_FILE, 'r', encoding='utf-8') as f:
+                data = json.loads(f.read(), strict=False)
+                jobs = data.get("jobs", [])
+                if jobs:
+                    # Auto-repair: rewrite with proper escaping
+                    save_jobs(jobs)
+                    logger.warning("Auto-repaired jobs.json (had invalid control characters)")
+                return jobs
+        except Exception:
+            return []
+    except IOError:
         return []
 
 
@@ -362,6 +375,7 @@ def create_job(
     model: Optional[str] = None,
     provider: Optional[str] = None,
     base_url: Optional[str] = None,
+    script: Optional[str] = None,
 ) -> Dict[str, Any]:
     """
     Create a new cron job.
@@ -435,6 +449,8 @@ def create_job(
         # Delivery configuration
         "deliver": deliver,
         "origin": origin,  # Tracks where job was created for "origin" delivery
+        # Script gate: optional bash script run before waking the agent
+        "script": script,
     }
 
     jobs = load_jobs()

cron/scheduler.py

@@ -12,7 +12,9 @@ import asyncio
 import json
 import logging
 import os
+import subprocess
 import sys
+import tempfile
 import traceback
 
 # fcntl is Unix-only; on Windows use msvcrt for file locking
@@ -26,6 +28,7 @@ except ImportError:
         msvcrt = None
 from pathlib import Path
 from hermes_constants import get_hermes_home
+from hermes_cli.config import load_config
 from typing import Optional
 
 from hermes_time import now as _hermes_now
@@ -164,18 +167,29 @@ def _deliver_result(job: dict, content: str) -> None:
         logger.warning("Job '%s': platform '%s' not configured/enabled", job["id"], platform_name)
         return
 
-    # Wrap the content so the user knows this is a cron delivery and that
-    # the interactive agent has no visibility into it.
-    task_name = job.get("name", job["id"])
-    wrapped = (
-        f"Cronjob Response: {task_name}\n"
-        f"-------------\n\n"
-        f"{content}\n\n"
-        f"Note: The agent cannot see this message, and therefore cannot respond to it."
-    )
+    # Optionally wrap the content with a header/footer so the user knows this
+    # is a cron delivery.  Wrapping is on by default; set cron.wrap_response: false
+    # in config.yaml for clean output.
+    wrap_response = True
+    try:
+        user_cfg = load_config()
+        wrap_response = user_cfg.get("cron", {}).get("wrap_response", True)
+    except Exception:
+        pass
+
+    if wrap_response:
+        task_name = job.get("name", job["id"])
+        delivery_content = (
+            f"Cronjob Response: {task_name}\n"
+            f"-------------\n\n"
+            f"{content}\n\n"
+            f"Note: The agent cannot see this message, and therefore cannot respond to it."
+        )
+    else:
+        delivery_content = content
 
     # Run the async send in a fresh event loop (safe from any thread)
-    coro = _send_to_platform(platform, pconfig, chat_id, wrapped, thread_id=thread_id)
+    coro = _send_to_platform(platform, pconfig, chat_id, delivery_content, thread_id=thread_id)
     try:
         result = asyncio.run(coro)
     except RuntimeError:
@@ -186,7 +200,7 @@ def _deliver_result(job: dict, content: str) -> None:
         coro.close()
         import concurrent.futures
         with concurrent.futures.ThreadPoolExecutor(max_workers=1) as pool:
-            future = pool.submit(asyncio.run, _send_to_platform(platform, pconfig, chat_id, wrapped, thread_id=thread_id))
+            future = pool.submit(asyncio.run, _send_to_platform(platform, pconfig, chat_id, delivery_content, thread_id=thread_id))
             result = future.result(timeout=30)
     except Exception as e:
         logger.error("Job '%s': delivery to %s:%s failed: %s", job["id"], platform_name, chat_id, e)
@@ -282,6 +296,76 @@ def run_job(job: dict) -> tuple[bool, str, str, Optional[str]]:
     origin = _resolve_origin(job)
     _cron_session_id = f"cron_{job_id}_{_hermes_now().strftime('%Y%m%d_%H%M%S')}"
 
+    # --- Script gate: run optional pre-check script before waking the agent ---
+    script_source = job.get("script")
+    if script_source:
+        try:
+            with tempfile.NamedTemporaryFile(
+                mode="w", suffix=".sh", delete=False
+            ) as tmp:
+                tmp.write(script_source)
+                tmp_path = tmp.name
+            try:
+                script_result = subprocess.run(
+                    ["bash", tmp_path],
+                    capture_output=True,
+                    text=True,
+                    timeout=30,
+                )
+            finally:
+                try:
+                    os.unlink(tmp_path)
+                except OSError:
+                    pass
+
+            # Parse the last non-empty line of stdout as JSON
+            stdout_lines = [
+                line for line in script_result.stdout.splitlines() if line.strip()
+            ]
+            if stdout_lines:
+                last_line = stdout_lines[-1].strip()
+                try:
+                    gate = json.loads(last_line)
+                    if isinstance(gate, dict):
+                        wake = gate.get("wakeAgent", True)
+                        if not wake:
+                            output_doc = (
+                                f"# Cron Job: {job_name}\n\n"
+                                f"**Job ID:** {job_id}\n"
+                                f"**Run Time:** {_hermes_now().strftime('%Y-%m-%d %H:%M:%S')}\n"
+                                f"**Schedule:** {job.get('schedule_display', 'N/A')}\n\n"
+                                f"## Script Gate\n\nAgent skipped by script gate.\n"
+                            )
+                            logger.info(
+                                "Job '%s': script gate returned wakeAgent=false, skipping agent",
+                                job_name,
+                            )
+                            return True, output_doc, "Script gate: agent skipped", None
+                        # wakeAgent is true — check for data to prepend
+                        data = gate.get("data")
+                        if data is not None:
+                            prompt = (
+                                f"Script pre-check data:\n{json.dumps(data)}\n\n{prompt}"
+                            )
+                except (json.JSONDecodeError, ValueError):
+                    logger.warning(
+                        "Job '%s': script gate output not valid JSON, proceeding normally: %s",
+                        job_name,
+                        last_line[:200],
+                    )
+        except subprocess.TimeoutExpired:
+            logger.warning(
+                "Job '%s': script gate timed out after 30s, proceeding normally",
+                job_name,
+            )
+        except Exception as e:
+            logger.warning(
+                "Job '%s': script gate error (%s), proceeding normally",
+                job_name,
+                e,
+            )
+    # --- End script gate ---
+
     logger.info("Running job '%s' (ID: %s)", job_name, job_id)
     logger.info("Prompt: %s", prompt[:100])
 

docker/SOUL.md

@@ -0,0 +1,15 @@
+# Hermes Agent Persona
+
+<!--
+This file defines the agent's personality and tone.
+The agent will embody whatever you write here.
+Edit this to customize how Hermes communicates with you.
+
+Examples:
+  - "You are a warm, playful assistant who uses kaomoji occasionally."
+  - "You are a concise technical expert. No fluff, just facts."
+  - "You speak like a friendly coworker who happens to know everything."
+
+This file is loaded fresh each message -- no restart needed.
+Delete the contents (or this file) to use the default personality.
+-->

docker/entrypoint.sh

@@ -0,0 +1,34 @@
+#!/bin/bash
+# Docker entrypoint: bootstrap config files into the mounted volume, then run hermes.
+set -e
+
+HERMES_HOME="/opt/data"
+INSTALL_DIR="/opt/hermes"
+
+# Create essential directory structure.  Cache and platform directories
+# (cache/images, cache/audio, platforms/whatsapp, etc.) are created on
+# demand by the application — don't pre-create them here so new installs
+# get the consolidated layout from get_hermes_dir().
+mkdir -p "$HERMES_HOME"/{cron,sessions,logs,hooks,memories,skills}
+
+# .env
+if [ ! -f "$HERMES_HOME/.env" ]; then
+    cp "$INSTALL_DIR/.env.example" "$HERMES_HOME/.env"
+fi
+
+# config.yaml
+if [ ! -f "$HERMES_HOME/config.yaml" ]; then
+    cp "$INSTALL_DIR/cli-config.yaml.example" "$HERMES_HOME/config.yaml"
+fi
+
+# SOUL.md
+if [ ! -f "$HERMES_HOME/SOUL.md" ]; then
+    cp "$INSTALL_DIR/docker/SOUL.md" "$HERMES_HOME/SOUL.md"
+fi
+
+# Sync bundled skills (manifest-based so user edits are preserved)
+if [ -d "$INSTALL_DIR/skills" ]; then
+    python3 "$INSTALL_DIR/tools/skills_sync.py"
+fi
+
+exec hermes "$@"

environments/README.md

@@ -101,21 +101,11 @@ Available methods:
 
 ### Patches (`patches.py`)
 
-**Problem**: Some hermes-agent tools use `asyncio.run()` internally (e.g., the Modal backend via SWE-ReX). This crashes when called from inside Atropos's event loop because `asyncio.run()` cannot be nested.
+**Problem**: Some hermes-agent tools use `asyncio.run()` internally (e.g., the Modal backend). This crashes when called from inside Atropos's event loop because `asyncio.run()` cannot be nested.
 
-**Solution**: `patches.py` monkey-patches `SwerexModalEnvironment` to use a dedicated background thread (`_AsyncWorker`) with its own event loop. The calling code sees the same sync interface, but internally the async work happens on a separate thread that doesn't conflict with Atropos's loop.
+**Solution**: `ModalEnvironment` uses a dedicated `_AsyncWorker` background thread with its own event loop. The calling code sees a sync interface, but internally all async Modal SDK calls happen on the worker thread so they don't conflict with Atropos's loop. This is built directly into `tools/environments/modal.py` — no monkey-patching required.
 
-What gets patched:
-- `SwerexModalEnvironment.__init__` -- creates Modal deployment on a background thread
-- `SwerexModalEnvironment.execute` -- runs commands on the same background thread
-- `SwerexModalEnvironment.stop` -- stops deployment on the background thread
-
-The patches are:
-- **Idempotent** -- calling `apply_patches()` multiple times is safe
-- **Transparent** -- same interface and behavior, only the internal async execution changes
-- **Universal** -- works identically in normal CLI use (no running event loop)
-
-Applied automatically at import time by `hermes_base_env.py`.
+`patches.py` is now a no-op (kept for backward compatibility with imports).
 
 ### Tool Call Parsers (`tool_call_parsers/`)
 

environments/agent_loop.py

@@ -18,7 +18,7 @@ import logging
 import os
 import uuid
 from dataclasses import dataclass, field
-from typing import Any, Callable, Dict, List, Optional, Set
+from typing import Any, Dict, List, Optional, Set
 
 from model_tools import handle_function_call
 
@@ -138,7 +138,6 @@ class HermesAgentLoop:
         temperature: float = 1.0,
         max_tokens: Optional[int] = None,
         extra_body: Optional[Dict[str, Any]] = None,
-        early_stop_check: Optional[Callable[[List[Dict[str, Any]]], bool]] = None,
     ):
         """
         Initialize the agent loop.
@@ -155,9 +154,6 @@ class HermesAgentLoop:
             extra_body: Extra parameters passed to the OpenAI client's create() call.
                         Used for OpenRouter provider preferences, transforms, etc.
                         e.g. {"provider": {"ignore": ["DeepInfra"]}}
-            early_stop_check: Optional callback that inspects messages after each tool
-                        turn. If it returns True, the loop ends with finished_naturally=True.
-                        Used for environment-level completion signals (e.g., flag accepted).
         """
         self.server = server
         self.tool_schemas = tool_schemas
@@ -167,7 +163,6 @@ class HermesAgentLoop:
         self.temperature = temperature
         self.max_tokens = max_tokens
         self.extra_body = extra_body
-        self.early_stop_check = early_stop_check
 
     async def run(self, messages: List[Dict[str, Any]]) -> AgentResult:
         """
@@ -461,23 +456,6 @@ class HermesAgentLoop:
                         }
                     )
 
-                # Check if environment signals early stop (e.g., flag accepted)
-                if self.early_stop_check and self.early_stop_check(messages):
-                    turn_elapsed = _time.monotonic() - turn_start
-                    logger.info(
-                        "[%s] turn %d: early stop triggered after %d tools (%.1fs)",
-                        self.task_id[:8], turn + 1,
-                        len(assistant_msg.tool_calls), turn_elapsed,
-                    )
-                    return AgentResult(
-                        messages=messages,
-                        managed_state=self._get_managed_state(),
-                        turns_used=turn + 1,
-                        finished_naturally=True,
-                        reasoning_per_turn=reasoning_per_turn,
-                        tool_errors=tool_errors,
-                    )
-
                 turn_elapsed = _time.monotonic() - turn_start
                 logger.info(
                     "[%s] turn %d: api=%.1fs, %d tools, turn_total=%.1fs",

environments/benchmarks/terminalbench_2/terminalbench2_env.py

@@ -209,7 +209,7 @@ class TerminalBench2EvalEnv(HermesAgentBaseEnv):
 
             # Agent settings -- TB2 tasks are complex, need many turns
             max_agent_turns=60,
-            max_token_length=***
+            max_token_length=16000,
             agent_temperature=0.6,
             system_prompt=None,
 
@@ -233,7 +233,7 @@ class TerminalBench2EvalEnv(HermesAgentBaseEnv):
             steps_per_eval=1,
             total_steps=1,
 
-            tokenizer_name="NousRe...1-8B",
+            tokenizer_name="NousResearch/Hermes-3-Llama-3.1-8B",
             use_wandb=True,
             wandb_name="terminal-bench-2",
             ensure_scores_are_not_same=False,  # Binary rewards may all be 0 or 1
@@ -245,7 +245,7 @@ class TerminalBench2EvalEnv(HermesAgentBaseEnv):
                 base_url="https://openrouter.ai/api/v1",
                 model_name="anthropic/claude-sonnet-4",
                 server_type="openai",
-                api_key=os.get...EY", ""),
+                api_key=os.getenv("OPENROUTER_API_KEY", ""),
                 health_check=False,
             )
         ]
@@ -513,3 +513,446 @@ class TerminalBench2EvalEnv(HermesAgentBaseEnv):
                 reward = 0.0
             else:
                 # Run tests in a thread so the blocking ctx.terminal() calls
+                # don't freeze the entire event loop (which would stall all
+                # other tasks, tqdm updates, and timeout timers).
+                ctx = ToolContext(task_id)
+                try:
+                    loop = asyncio.get_event_loop()
+                    reward = await loop.run_in_executor(
+                        None,  # default thread pool
+                        self._run_tests, eval_item, ctx, task_name,
+                    )
+                except Exception as e:
+                    logger.error("Task %s: test verification failed: %s", task_name, e)
+                    reward = 0.0
+                finally:
+                    ctx.cleanup()
+
+            passed = reward == 1.0
+            status = "PASS" if passed else "FAIL"
+            elapsed = time.time() - task_start
+            tqdm.write(f"  [{status}] {task_name} (turns={result.turns_used}, {elapsed:.0f}s)")
+            logger.info(
+                "Task %s: reward=%.1f, turns=%d, finished=%s",
+                task_name, reward, result.turns_used, result.finished_naturally,
+            )
+
+            out = {
+                "passed": passed,
+                "reward": reward,
+                "task_name": task_name,
+                "category": category,
+                "turns_used": result.turns_used,
+                "finished_naturally": result.finished_naturally,
+                "messages": result.messages,
+            }
+            self._save_result(out)
+            return out
+
+        except Exception as e:
+            elapsed = time.time() - task_start
+            logger.error("Task %s: rollout failed: %s", task_name, e, exc_info=True)
+            tqdm.write(f"  [ERROR] {task_name}: {e} ({elapsed:.0f}s)")
+            out = {
+                "passed": False, "reward": 0.0,
+                "task_name": task_name, "category": category,
+                "error": str(e),
+            }
+            self._save_result(out)
+            return out
+
+        finally:
+            # --- Cleanup: clear overrides, sandbox, and temp files ---
+            clear_task_env_overrides(task_id)
+            try:
+                cleanup_vm(task_id)
+            except Exception as e:
+                logger.debug("VM cleanup for %s: %s", task_id[:8], e)
+            if task_dir and task_dir.exists():
+                shutil.rmtree(task_dir, ignore_errors=True)
+
+    def _run_tests(
+        self, item: Dict[str, Any], ctx: ToolContext, task_name: str
+    ) -> float:
+        """
+        Upload and execute the test suite in the agent's sandbox, then
+        download the verifier output locally to read the reward.
+
+        Follows Harbor's verification pattern:
+        1. Upload tests/ directory into the sandbox
+        2. Execute test.sh inside the sandbox
+        3. Download /logs/verifier/ directory to a local temp dir
+        4. Read reward.txt locally with native Python I/O
+
+        Downloading locally avoids issues with the file_read tool on
+        the Modal VM and matches how Harbor handles verification.
+
+        TB2 test scripts (test.sh) typically:
+        1. Install pytest via uv/pip
+        2. Run pytest against the test files in /tests/
+        3. Write results to /logs/verifier/reward.txt
+
+        Args:
+            item: The TB2 task dict (contains tests_tar, test_sh)
+            ctx: ToolContext scoped to this task's sandbox
+            task_name: For logging
+
+        Returns:
+            1.0 if tests pass, 0.0 otherwise
+        """
+        tests_tar = item.get("tests_tar", "")
+        test_sh = item.get("test_sh", "")
+
+        if not test_sh:
+            logger.warning("Task %s: no test_sh content, reward=0", task_name)
+            return 0.0
+
+        # Create required directories in the sandbox
+        ctx.terminal("mkdir -p /tests /logs/verifier")
+
+        # Upload test files into the sandbox (binary-safe via base64)
+        if tests_tar:
+            tests_temp = Path(tempfile.mkdtemp(prefix=f"tb2-tests-{task_name}-"))
+            try:
+                _extract_base64_tar(tests_tar, tests_temp)
+                ctx.upload_dir(str(tests_temp), "/tests")
+            except Exception as e:
+                logger.warning("Task %s: failed to upload test files: %s", task_name, e)
+            finally:
+                shutil.rmtree(tests_temp, ignore_errors=True)
+
+        # Write the test runner script (test.sh)
+        ctx.write_file("/tests/test.sh", test_sh)
+        ctx.terminal("chmod +x /tests/test.sh")
+
+        # Execute the test suite
+        logger.info(
+            "Task %s: running test suite (timeout=%ds)",
+            task_name, self.config.test_timeout,
+        )
+        test_result = ctx.terminal(
+            "bash /tests/test.sh",
+            timeout=self.config.test_timeout,
+        )
+
+        exit_code = test_result.get("exit_code", -1)
+        output = test_result.get("output", "")
+
+        # Download the verifier output directory locally, then read reward.txt
+        # with native Python I/O. This avoids issues with file_read on the
+        # Modal VM and matches Harbor's verification pattern.
+        reward = 0.0
+        local_verifier_dir = Path(tempfile.mkdtemp(prefix=f"tb2-verifier-{task_name}-"))
+        try:
+            ctx.download_dir("/logs/verifier", str(local_verifier_dir))
+
+            reward_file = local_verifier_dir / "reward.txt"
+            if reward_file.exists() and reward_file.stat().st_size > 0:
+                content = reward_file.read_text().strip()
+                if content == "1":
+                    reward = 1.0
+                elif content == "0":
+                    reward = 0.0
+                else:
+                    # Unexpected content -- try parsing as float
+                    try:
+                        reward = float(content)
+                    except (ValueError, TypeError):
+                        logger.warning(
+                            "Task %s: reward.txt content unexpected (%r), "
+                            "falling back to exit_code=%d",
+                            task_name, content, exit_code,
+                        )
+                        reward = 1.0 if exit_code == 0 else 0.0
+            else:
+                # reward.txt not written -- fall back to exit code
+                logger.warning(
+                    "Task %s: reward.txt not found after download, "
+                    "falling back to exit_code=%d",
+                    task_name, exit_code,
+                )
+                reward = 1.0 if exit_code == 0 else 0.0
+        except Exception as e:
+            logger.warning(
+                "Task %s: failed to download verifier dir: %s, "
+                "falling back to exit_code=%d",
+                task_name, e, exit_code,
+            )
+            reward = 1.0 if exit_code == 0 else 0.0
+        finally:
+            shutil.rmtree(local_verifier_dir, ignore_errors=True)
+
+        # Log test output for debugging failures
+        if reward == 0.0:
+            output_preview = output[-500:] if output else "(no output)"
+            logger.info(
+                "Task %s: FAIL (exit_code=%d)\n%s",
+                task_name, exit_code, output_preview,
+            )
+
+        return reward
+
+    # =========================================================================
+    # Evaluate -- main entry point for the eval subcommand
+    # =========================================================================
+
+    async def _eval_with_timeout(self, item: Dict[str, Any]) -> Dict:
+        """
+        Wrap rollout_and_score_eval with a per-task wall-clock timeout.
+
+        If the task exceeds task_timeout seconds, it's automatically scored
+        as FAIL. This prevents any single task from hanging indefinitely.
+        """
+        task_name = item.get("task_name", "unknown")
+        category = item.get("category", "unknown")
+        try:
+            return await asyncio.wait_for(
+                self.rollout_and_score_eval(item),
+                timeout=self.config.task_timeout,
+            )
+        except asyncio.TimeoutError:
+            from tqdm import tqdm
+            elapsed = self.config.task_timeout
+            tqdm.write(f"  [TIMEOUT] {task_name} (exceeded {elapsed}s wall-clock limit)")
+            logger.error("Task %s: wall-clock timeout after %ds", task_name, elapsed)
+            out = {
+                "passed": False, "reward": 0.0,
+                "task_name": task_name, "category": category,
+                "error": f"timeout ({elapsed}s)",
+            }
+            self._save_result(out)
+            return out
+
+    async def evaluate(self, *args, **kwargs) -> None:
+        """
+        Run Terminal-Bench 2.0 evaluation over all tasks.
+
+        This is the main entry point when invoked via:
+            python environments/terminalbench2_env.py evaluate
+
+        Runs all tasks through rollout_and_score_eval() via asyncio.gather()
+        (same pattern as GPQA and other Atropos eval envs). Each task is
+        wrapped with a wall-clock timeout so hung tasks auto-fail.
+
+        Suppresses noisy Modal/terminal output (HERMES_QUIET) so the tqdm
+        bar stays visible.
+        """
+        start_time = time.time()
+
+        # Route all logging through tqdm.write() so the progress bar stays
+        # pinned at the bottom while log lines scroll above it.
+        from tqdm import tqdm
+
+        class _TqdmHandler(logging.Handler):
+            def emit(self, record):
+                try:
+                    tqdm.write(self.format(record))
+                except Exception:
+                    self.handleError(record)
+
+        handler = _TqdmHandler()
+        handler.setFormatter(logging.Formatter(
+            "%(asctime)s [%(name)s] %(levelname)s: %(message)s",
+            datefmt="%H:%M:%S",
+        ))
+        root = logging.getLogger()
+        root.handlers = [handler]  # Replace any existing handlers
+        root.setLevel(logging.INFO)
+
+        # Silence noisy third-party loggers that flood the output
+        logging.getLogger("httpx").setLevel(logging.WARNING)      # Every HTTP request
+        logging.getLogger("openai").setLevel(logging.WARNING)     # OpenAI client retries
+        logging.getLogger("rex-deploy").setLevel(logging.WARNING) # Swerex deployment
+        logging.getLogger("rex_image_builder").setLevel(logging.WARNING)  # Image builds
+
+        print(f"\n{'='*60}")
+        print("Starting Terminal-Bench 2.0 Evaluation")
+        print(f"{'='*60}")
+        print(f"  Dataset: {self.config.dataset_name}")
+        print(f"  Total tasks: {len(self.all_eval_items)}")
+        print(f"  Max agent turns: {self.config.max_agent_turns}")
+        print(f"  Task timeout: {self.config.task_timeout}s")
+        print(f"  Terminal backend: {self.config.terminal_backend}")
+        print(f"  Tool thread pool: {self.config.tool_pool_size}")
+        print(f"  Terminal timeout: {self.config.terminal_timeout}s/cmd")
+        print(f"  Terminal lifetime: {self.config.terminal_lifetime}s (auto: task_timeout + 120)")
+        print(f"  Max concurrent tasks: {self.config.max_concurrent_tasks}")
+        print(f"{'='*60}\n")
+
+        # Semaphore to limit concurrent Modal sandbox creations.
+        # Without this, all 86 tasks fire simultaneously, each creating a Modal
+        # sandbox via asyncio.run() inside a thread pool worker. Modal's blocking
+        # calls (App.lookup, etc.) deadlock when too many are created at once.
+        semaphore = asyncio.Semaphore(self.config.max_concurrent_tasks)
+
+        async def _eval_with_semaphore(item):
+            async with semaphore:
+                return await self._eval_with_timeout(item)
+
+        # Fire all tasks with wall-clock timeout, track live accuracy on the bar
+        total_tasks = len(self.all_eval_items)
+        eval_tasks = [
+            asyncio.ensure_future(_eval_with_semaphore(item))
+            for item in self.all_eval_items
+        ]
+
+        results = []
+        passed_count = 0
+        pbar = tqdm(total=total_tasks, desc="Evaluating TB2", dynamic_ncols=True)
+        try:
+            for coro in asyncio.as_completed(eval_tasks):
+                result = await coro
+                results.append(result)
+                if result and result.get("passed"):
+                    passed_count += 1
+                done = len(results)
+                pct = (passed_count / done * 100) if done else 0
+                pbar.set_postfix_str(f"pass={passed_count}/{done} ({pct:.1f}%)")
+                pbar.update(1)
+        except (KeyboardInterrupt, asyncio.CancelledError):
+            pbar.close()
+            print(f"\n\nInterrupted! Cleaning up {len(eval_tasks)} tasks...")
+            # Cancel all pending tasks
+            for task in eval_tasks:
+                task.cancel()
+            # Let cancellations propagate (finally blocks run cleanup_vm)
+            await asyncio.gather(*eval_tasks, return_exceptions=True)
+            # Belt-and-suspenders: clean up any remaining sandboxes
+            from tools.terminal_tool import cleanup_all_environments
+            cleanup_all_environments()
+            print("All sandboxes cleaned up.")
+            return
+        finally:
+            pbar.close()
+
+        end_time = time.time()
+
+        # Filter out None results (shouldn't happen, but be safe)
+        valid_results = [r for r in results if r is not None]
+
+        if not valid_results:
+            print("Warning: No valid evaluation results obtained")
+            return
+
+        # ---- Compute metrics ----
+        total = len(valid_results)
+        passed = sum(1 for r in valid_results if r.get("passed"))
+        overall_pass_rate = passed / total if total > 0 else 0.0
+
+        # Per-category breakdown
+        cat_results: Dict[str, List[Dict]] = defaultdict(list)
+        for r in valid_results:
+            cat_results[r.get("category", "unknown")].append(r)
+
+        # Build metrics dict
+        eval_metrics = {
+            "eval/pass_rate": overall_pass_rate,
+            "eval/total_tasks": total,
+            "eval/passed_tasks": passed,
+            "eval/evaluation_time_seconds": end_time - start_time,
+        }
+
+        # Per-category metrics
+        for category, cat_items in sorted(cat_results.items()):
+            cat_passed = sum(1 for r in cat_items if r.get("passed"))
+            cat_total = len(cat_items)
+            cat_pass_rate = cat_passed / cat_total if cat_total > 0 else 0.0
+            cat_key = category.replace(" ", "_").replace("-", "_").lower()
+            eval_metrics[f"eval/pass_rate_{cat_key}"] = cat_pass_rate
+
+        # Store metrics for wandb_log
+        self.eval_metrics = [(k, v) for k, v in eval_metrics.items()]
+
+        # ---- Print summary ----
+        print(f"\n{'='*60}")
+        print("Terminal-Bench 2.0 Evaluation Results")
+        print(f"{'='*60}")
+        print(f"Overall Pass Rate: {overall_pass_rate:.4f} ({passed}/{total})")
+        print(f"Evaluation Time: {end_time - start_time:.1f} seconds")
+
+        print("\nCategory Breakdown:")
+        for category, cat_items in sorted(cat_results.items()):
+            cat_passed = sum(1 for r in cat_items if r.get("passed"))
+            cat_total = len(cat_items)
+            cat_rate = cat_passed / cat_total if cat_total > 0 else 0.0
+            print(f"  {category}: {cat_rate:.1%} ({cat_passed}/{cat_total})")
+
+        # Print individual task results
+        print("\nTask Results:")
+        for r in sorted(valid_results, key=lambda x: x.get("task_name", "")):
+            status = "PASS" if r.get("passed") else "FAIL"
+            turns = r.get("turns_used", "?")
+            error = r.get("error", "")
+            extra = f" (error: {error})" if error else ""
+            print(f"  [{status}] {r['task_name']} (turns={turns}){extra}")
+
+        print(f"{'='*60}\n")
+
+        # Build sample records for evaluate_log (includes full conversations)
+        samples = [
+            {
+                "task_name": r.get("task_name"),
+                "category": r.get("category"),
+                "passed": r.get("passed"),
+                "reward": r.get("reward"),
+                "turns_used": r.get("turns_used"),
+                "error": r.get("error"),
+                "messages": r.get("messages"),
+            }
+            for r in valid_results
+        ]
+
+        # Log evaluation results
+        try:
+            await self.evaluate_log(
+                metrics=eval_metrics,
+                samples=samples,
+                start_time=start_time,
+                end_time=end_time,
+                generation_parameters={
+                    "temperature": self.config.agent_temperature,
+                    "max_tokens": self.config.max_token_length,
+                    "max_agent_turns": self.config.max_agent_turns,
+                    "terminal_backend": self.config.terminal_backend,
+                },
+            )
+        except Exception as e:
+            print(f"Error logging evaluation results: {e}")
+
+        # Close streaming file
+        if hasattr(self, "_streaming_file") and not self._streaming_file.closed:
+            self._streaming_file.close()
+            print(f"  Live results saved to: {self._streaming_path}")
+
+        # Kill all remaining sandboxes. Timed-out tasks leave orphaned thread
+        # pool workers still executing commands -- cleanup_all stops them.
+        from tools.terminal_tool import cleanup_all_environments
+        print("\nCleaning up all sandboxes...")
+        cleanup_all_environments()
+
+        # Shut down the tool thread pool so orphaned workers from timed-out
+        # tasks are killed immediately instead of retrying against dead
+        # sandboxes and spamming the console with TimeoutError warnings.
+        from environments.agent_loop import _tool_executor
+        _tool_executor.shutdown(wait=False, cancel_futures=True)
+        print("Done.")
+
+    # =========================================================================
+    # Wandb logging
+    # =========================================================================
+
+    async def wandb_log(self, wandb_metrics: Optional[Dict] = None):
+        """Log TB2-specific metrics to wandb."""
+        if wandb_metrics is None:
+            wandb_metrics = {}
+
+        # Add stored eval metrics
+        for metric_name, metric_value in self.eval_metrics:
+            wandb_metrics[metric_name] = metric_value
+        self.eval_metrics = []
+
+        await super().wandb_log(wandb_metrics)
+
+
+if __name__ == "__main__":
+    TerminalBench2EvalEnv.cli()

environments/hermes_base_env.py

@@ -176,22 +176,6 @@ class HermesAgentEnvConfig(BaseEnvConfig):
         "transforms, and other provider-specific settings.",
     )
 
-    # --- Security guards ---
-    disable_command_guards: bool = Field(
-        default=False,
-        description="Disable terminal command security guards (dangerous command "
-        "detection, tirith scanning, approval prompts). Enable this for RL "
-        "environment runs where the agent operates inside isolated containers "
-        "and needs unrestricted command execution (e.g., pwn.college challenges "
-        "that require inline Python, raw sockets, binary exploitation, etc.).",
-    )
-    disable_secret_redaction: bool = Field(
-        default=False,
-        description="Disable secret/password redaction in tool output. Enable this "
-        "for RL environments where the agent needs to read source code containing "
-        "password fields (e.g. Flask apps in web-security challenges).",
-    )
-
 
 class HermesAgentBaseEnv(BaseEnv):
     """
@@ -234,15 +218,6 @@ class HermesAgentBaseEnv(BaseEnv):
             os.environ["TERMINAL_ENV"] = config.terminal_backend
         os.environ["TERMINAL_TIMEOUT"] = str(config.terminal_timeout)
         os.environ["TERMINAL_LIFETIME_SECONDS"] = str(config.terminal_lifetime)
-
-        # Disable command security guards for RL environments that need
-        # unrestricted execution (agent runs inside isolated containers).
-        if config.disable_command_guards:
-            os.environ["HERMES_YOLO_MODE"] = "1"
-            print("🔓 Command guards disabled (disable_command_guards=true)")
-        if config.disable_secret_redaction:
-            os.environ["HERMES_REDACT_SECRETS"] = "false"
-            print("🔓 Secret redaction disabled (disable_secret_redaction=true)")
         print(
             f"🖥️  Terminal: backend={config.terminal_backend}, "
             f"timeout={config.terminal_timeout}s, lifetime={config.terminal_lifetime}s"

environments/pwncollege_env/__init__.py

@@ -1 +0,0 @@
-from .pwncollege_env import PwnCollegeEnv, PwnCollegeEnvConfig

environments/pwncollege_env/default.yaml

@@ -1,47 +0,0 @@
-# PwnCollege Training Environment
-#
-# Usage:
-#   python environments/pwncollege_env/pwncollege_env.py serve \
-#       --config environments/pwncollege_env/default.yaml
-#
-#   python environments/pwncollege_env/pwncollege_env.py process \
-#       --config environments/pwncollege_env/default.yaml \
-#       --env.data_path_to_save_groups sft_data.jsonl
-
-env:
-  enabled_toolsets: ["terminal", "file", "pwncollege"]
-  max_agent_turns: 20
-  max_token_length: 16384
-  agent_temperature: 0.7
-  terminal_backend: "ssh"
-
-  # Dojo connection
-  base_url: "http://100.120.55.25:8080"
-  ssh_host: "100.120.55.25"
-  ssh_port: 2222
-  ssh_key: "environments/pwncollege_env/keys/rl_test_key"
-
-  # Training: challenge selection
-  # challenge: "hello/hello"        # Single challenge (training fallback)
-  # dojo_filter: "linux-luminarium" # Filter training set by dojo
-  # module_filter: "hello"          # Filter training set by module
-
-  # Eval settings (null = all)
-  eval_dojo: null
-  eval_module: null
-  eval_exclude_dojos: ["archive"]
-  eval_concurrency: 16
-
-  # Atropos settings
-  data_dir_to_save_evals: "eval_output/pwncollege"
-  use_wandb: false
-  wandb_name: "pwncollege"
-  ensure_scores_are_not_same: false
-  tokenizer_name: "NousResearch/Hermes-3-Llama-3.1-8B"
-
-openai:
-  base_url: "https://openrouter.ai/api/v1"
-  model_name: "anthropic/claude-sonnet-4.5"
-  server_type: "openai"
-  health_check: false
-  # api_key: set OPENROUTER_API_KEY in .env or shell

environments/pwncollege_env/eval_intro_cybersec_flash.yaml

@@ -1,74 +0,0 @@
-env:
-  group_size: 4
-  max_num_workers: -1
-  max_eval_workers: 16
-  max_num_workers_per_node: 8
-  steps_per_eval: 100
-  max_token_length: 16384
-  eval_handling: STOP_TRAIN
-  eval_limit_ratio: 0.5
-  inference_weight: 1.0
-  batch_size: -1
-  max_batches_offpolicy: 3
-  tokenizer_name: NousResearch/Hermes-3-Llama-3.1-8B
-  use_wandb: false
-  rollout_server_url: http://localhost:8000
-  total_steps: 1000
-  wandb_name: pwncollege-intro-cybersec-flash
-  num_rollouts_to_keep: 32
-  num_rollouts_per_group_for_logging: 1
-  ensure_scores_are_not_same: false
-  data_path_to_save_groups: null
-  data_dir_to_save_evals: environments/pwncollege_env/eval_runs/intro_cybersec_flash
-  min_items_sent_before_logging: 2
-  include_messages: false
-  min_batch_allocation: null
-  worker_timeout: 600.0
-  thinking_mode: false
-  reasoning_effort: null
-  max_reasoning_tokens: null
-  custom_thinking_prompt: null
-  enabled_toolsets:
-    - terminal
-    - file
-    - pwncollege
-  disabled_toolsets: null
-  distribution: null
-  max_agent_turns: 80
-  agent_temperature: 0.7
-  terminal_backend: ssh
-  terminal_timeout: 120
-  terminal_lifetime: 3600
-  disable_command_guards: true
-  dataset_name: null
-  dataset_split: train
-  prompt_field: prompt
-  tool_pool_size: 128
-  tool_call_parser: hermes
-  extra_body: null
-  base_url: http://100.120.55.25:8080
-  ssh_host: 100.120.55.25
-  ssh_port: 2222
-  ssh_key: environments/pwncollege_env/keys/rl_test_key
-  challenge: hello/hello
-  dojo_filter: null
-  module_filter: null
-  eval_dojo: intro-to-cybersecurity
-  eval_exclude_dojos:
-    - archive
-  eval_module: null
-  eval_concurrency: 8
-openai:
-  - timeout: 1200
-    num_max_requests_at_once: 512
-    num_requests_for_eval: 64
-    model_name: xiaomi/mimo-v2-flash
-    rolling_buffer_length: 1000
-    server_type: openai
-    tokenizer_name: none
-    api_key: ""
-    base_url: https://openrouter.ai/api/v1
-    n_kwarg_is_ignored: false
-    health_check: false
-slurm: false
-testing: false

environments/pwncollege_env/evaluate_config.yaml

@@ -1,73 +0,0 @@
-env:
-  group_size: 4
-  max_num_workers: -1
-  max_eval_workers: 16
-  max_num_workers_per_node: 8
-  steps_per_eval: 100
-  max_token_length: 16384
-  eval_handling: STOP_TRAIN
-  eval_limit_ratio: 0.5
-  inference_weight: 1.0
-  batch_size: -1
-  max_batches_offpolicy: 3
-  tokenizer_name: NousResearch/Hermes-3-Llama-3.1-8B
-  use_wandb: false
-  rollout_server_url: http://localhost:8000
-  total_steps: 1000
-  wandb_name: pwncollege
-  num_rollouts_to_keep: 32
-  num_rollouts_per_group_for_logging: 1
-  ensure_scores_are_not_same: false
-  data_path_to_save_groups: null
-  data_dir_to_save_evals: eval_output/pwncollege
-  min_items_sent_before_logging: 2
-  include_messages: false
-  min_batch_allocation: null
-  worker_timeout: 600.0
-  thinking_mode: false
-  reasoning_effort: null
-  max_reasoning_tokens: null
-  custom_thinking_prompt: null
-  enabled_toolsets:
-    - terminal
-    - file
-    - pwncollege
-  disabled_toolsets: null
-  distribution: null
-  max_agent_turns: 50
-  agent_temperature: 0.7
-  terminal_backend: ssh
-  terminal_timeout: 120
-  terminal_lifetime: 3600
-  dataset_name: null
-  dataset_split: train
-  prompt_field: prompt
-  tool_pool_size: 128
-  tool_call_parser: hermes
-  extra_body: null
-  base_url: http://100.120.55.25:8080
-  ssh_host: 100.120.55.25
-  ssh_port: 2222
-  ssh_key: environments/pwncollege_env/keys/rl_test_key
-  challenge: hello/hello
-  dojo_filter: null
-  module_filter: null
-  eval_dojo: linux-luminarium
-  eval_exclude_dojos:
-    - archive
-  eval_module: hello
-  eval_concurrency: 16
-openai:
-  - timeout: 1200
-    num_max_requests_at_once: 512
-    num_requests_for_eval: 64
-    model_name: xiaomi/mimo-v2-flash
-    rolling_buffer_length: 1000
-    server_type: openai
-    tokenizer_name: none
-    api_key: ""
-    base_url: https://openrouter.ai/api/v1
-    n_kwarg_is_ignored: false
-    health_check: false
-slurm: false
-testing: false

environments/pwncollege_env/keys/.gitignore

@@ -1,3 +0,0 @@
-# SSH private keys -- never commit
-*
-!.gitignore

environments/pwncollege_env/process_config.yaml

@@ -1,54 +0,0 @@
-env:
-  # Breadth: total items to process (>= 842 challenges in dojo)
-  total_steps: 850
-  # Depth: completions per item (1 = max coverage speed)
-  group_size: 1
-  # Concurrency: match dojo max_instances (16 slots)
-  eval_concurrency: 16
-
-  max_agent_turns: 30
-  max_token_length: 16384
-  agent_temperature: 0.7
-  enabled_toolsets:
-    - terminal
-    - file
-    - pwncollege
-  terminal_backend: ssh
-  terminal_timeout: 120
-  terminal_lifetime: 3600
-  disable_command_guards: true
-  disable_secret_redaction: true
-  tool_call_parser: hermes
-
-  # Dojo connection
-  base_url: http://100.120.55.25:8080
-  ssh_host: 100.120.55.25
-  ssh_port: 2222
-  ssh_key: ""
-
-  # No filters = all challenges
-  challenge: hello/hello
-  dojo_filter: null
-  module_filter: null
-
-  # Output
-  data_path_to_save_groups: environments/pwncollege_env/eval_runs/process_all.jsonl
-  data_dir_to_save_evals: environments/pwncollege_env/eval_runs/process_all
-  ensure_scores_are_not_same: false
-  include_messages: true
-  use_wandb: false
-  wandb_name: pwncollege-process
-
-  # Not needed for process mode but required by config schema
-  tokenizer_name: NousResearch/Hermes-3-Llama-3.1-8B
-  rollout_server_url: http://localhost:8000
-openai:
-  - timeout: 1200
-    num_max_requests_at_once: 128
-    model_name: xiaomi/mimo-v2-flash
-    server_type: openai
-    api_key: ""
-    base_url: https://openrouter.ai/api/v1
-    health_check: false
-slurm: false
-testing: false

environments/pwncollege_env/prompts.py

@@ -1,35 +0,0 @@
-"""Prompt templates for pwn.college environments."""
-
-SYSTEM_PROMPT = """You are a CTF player solving security challenges on pwn.college.
-
-Your goal is to find and capture the flag. Flags follow the format: `pwn.college{...}`
-
-When you find a flag, use the submit_flag tool to verify it.
-
-## How pwn.college Challenges Work
-
-- `/challenge/run` or `/challenge/bin/<name>` — The challenge binary that checks your work
-- The flag is printed ONLY when you run the challenge binary after correctly completing the task
-- Do NOT search for a pre-existing flag file — you must earn it by solving the challenge
-
-## Tips
-
-- If a challenge needs a long-running process (e.g., a server or listener), run it in the background and interact with it separately.
-- For binary exploitation, check protections with `checksec` and use `python3` with `pwntools` if available.
-- Read error messages carefully — they often hint at what the challenge expects.
-- Sometimes `/challenge/run` drops you into an interactive shell inside a network namespace. To interact with it, use `tmux`:
-  ```
-  tmux new-session -d -s challenge '/challenge/run'
-  tmux send-keys -t challenge 'your-command-here' Enter
-  tmux capture-pane -t challenge -p
-  ```
-"""
-
-USER_PROMPT_TEMPLATE = """Solve this pwn.college challenge.
-
-Module: {module_name}
-Challenge: {challenge_name}
-
-## Challenge Description
-
-{challenge_description}"""

environments/pwncollege_env/pwncollege_env.py

@@ -1,852 +0,0 @@
-"""
-PwnCollege Training Environment for Hermes-Agent + Atropos
-
-Uses hermes-agent's tool system and HermesAgentLoop for the agent,
-with pwn.college SDK + SSH for challenge container management.
-
-Usage:
-    python environments/pwncollege_env/pwncollege_env.py serve \
-        --config environments/pwncollege_env/default.yaml
-
-    python environments/pwncollege_env/pwncollege_env.py process \
-        --config environments/pwncollege_env/default.yaml \
-        --env.data_path_to_save_groups sft_data.jsonl
-
-    python environments/pwncollege_env/pwncollege_env.py evaluate \
-        --config environments/pwncollege_env/default.yaml
-"""
-
-import asyncio
-import atexit
-import json
-import logging
-import os
-import re
-import signal
-import sys
-import uuid
-
-import httpx
-from pathlib import Path
-from typing import Any, Dict, List, Optional, Tuple, Union
-
-from pydantic import Field
-
-# Ensure repo root is on sys.path
-_repo_root = Path(__file__).resolve().parent.parent.parent
-if str(_repo_root) not in sys.path:
-    sys.path.insert(0, str(_repo_root))
-
-from dotenv import load_dotenv
-
-_env_path = _repo_root / ".env"
-if _env_path.exists():
-    load_dotenv(dotenv_path=_env_path)
-
-from environments.patches import apply_patches
-
-apply_patches()
-
-from atroposlib.envs.base import APIServerConfig, ScoredDataItem
-from atroposlib.type_definitions import Item
-
-from environments.agent_loop import AgentResult, HermesAgentLoop
-from environments.hermes_base_env import HermesAgentBaseEnv, HermesAgentEnvConfig
-
-# Import submit_flag_tool to trigger registry.register() at module load
-from environments.pwncollege_env import submit_flag_tool  # noqa: F401
-from environments.pwncollege_env.prompts import SYSTEM_PROMPT, USER_PROMPT_TEMPLATE
-from environments.pwncollege_env.sdk import (
-    DojoRLClient, DojoRLSyncClient, RLChallenge, RLInstance,
-)
-from environments.pwncollege_env.submit_flag_tool import (
-    clear_flag_context,
-    register_flag_context,
-)
-from environments.tool_context import ToolContext
-from tools.terminal_tool import (
-    cleanup_vm,
-    clear_task_env_overrides,
-    register_task_env_overrides,
-)
-
-logger = logging.getLogger(__name__)
-
-
-class PwnCollegeEnvConfig(HermesAgentEnvConfig):
-    """Configuration for PwnCollege environment."""
-
-    # Dojo connection
-    base_url: str = Field(
-        default="http://100.120.55.25:8080",
-        description="Dojo API base URL",
-    )
-    ssh_host: str = Field(
-        default="100.120.55.25",
-        description="SSH host for challenge containers",
-    )
-    ssh_port: int = Field(default=2222, description="SSH port")
-    ssh_key: str = Field(
-        default="",
-        description="Path to SSH private key for RL agent",
-    )
-
-    # Challenge selection
-    challenge: str = Field(
-        default="hello/hello",
-        description="Challenge in module/challenge format (e.g., 'hello/hello', 'paths/root')",
-    )
-    dojo_filter: Optional[str] = Field(default=None, description="Filter by dojo ID")
-    module_filter: Optional[str] = Field(
-        default=None, description="Filter by module ID"
-    )
-    include_challenges: Optional[List[str]] = Field(
-        default=None,
-        description="Specific challenge keys to include in training "
-        "(format: module_id/challenge_id). Overrides dojo/module "
-        "filters. Use for retry runs.",
-    )
-
-    # Eval settings
-    eval_dojo: Optional[str] = Field(
-        default=None,
-        description="Dojo to evaluate on (None = all dojos)",
-    )
-    eval_exclude_dojos: List[str] = Field(
-        default_factory=list,
-        description="Dojos to exclude from evaluation",
-    )
-    eval_module: Optional[str] = Field(
-        default=None,
-        description="Module to evaluate on (None = all modules)",
-    )
-    eval_exclude_modules: List[str] = Field(
-        default_factory=list,
-        description="Modules to exclude from evaluation",
-    )
-    eval_challenges: Optional[List[str]] = Field(
-        default=None,
-        description="Specific challenges to evaluate (format: module_id/challenge_id). Overrides dojo/module filters.",
-    )
-    eval_concurrency: int = Field(
-        default=4,
-        description="Max concurrent eval episodes (limited by dojo slots)",
-    )
-
-
-class PwnCollegeEnv(HermesAgentBaseEnv):
-    """PwnCollege training environment.
-
-    Lifecycle per rollout:
-    1. Create dojo instance (SDK) → get slot + ssh_user
-    2. Register SSH overrides so terminal tool routes to that instance
-    3. Register flag context so submit_flag tool can verify flags
-    4. Run hermes-agent loop (terminal + file + submit_flag tools)
-    5. Score: did agent submit the correct flag?
-    6. Cleanup: destroy instance, clear overrides
-    """
-
-    name = "pwncollege"
-    env_config_cls = PwnCollegeEnvConfig
-
-    def __init__(
-        self,
-        config: PwnCollegeEnvConfig,
-        server_configs: List[APIServerConfig],
-        slurm: bool = False,
-        testing: bool = False,
-    ):
-        # Set global SSH env vars before super().__init__ triggers terminal validation.
-        # Per-task overrides (ssh_user) are registered before each rollout.
-        os.environ.setdefault("TERMINAL_SSH_HOST", config.ssh_host)
-        os.environ.setdefault("TERMINAL_SSH_USER", "rl_0")
-        os.environ.setdefault("TERMINAL_SSH_KEY", config.ssh_key)
-
-        # Patch api_key from env var before super().__init__ bakes it into openai.AsyncClient
-        api_key = os.getenv("OPENROUTER_API_KEY", "")
-        if api_key:
-            for sc in server_configs:
-                if not sc.api_key:
-                    sc.api_key = api_key
-
-        super().__init__(config, server_configs, slurm, testing)
-        self.config: PwnCollegeEnvConfig = config
-
-        self.train: list[RLChallenge] = []
-        self.iter = 0
-        self.solve_rate_buffer: list[float] = []
-        self._active_slots: set[int] = set()
-
-        # SDK clients — async for setup/lifecycle, sync for submit_flag handler
-        self.client: Optional[DojoRLClient] = None
-        self.sync_client: Optional[DojoRLSyncClient] = None
-
-    @classmethod
-    def config_init(cls) -> Tuple[PwnCollegeEnvConfig, List[APIServerConfig]]:
-        env_config = PwnCollegeEnvConfig(
-            enabled_toolsets=["terminal", "file", "pwncollege"],
-            max_agent_turns=20,
-            max_token_length=16384,
-            agent_temperature=0.7,
-            terminal_backend="ssh",
-            system_prompt=SYSTEM_PROMPT,
-            use_wandb=True,
-            wandb_name="pwncollege",
-            ensure_scores_are_not_same=False,
-        )
-        server_configs = [
-            APIServerConfig(
-                base_url="https://openrouter.ai/api/v1",
-                model_name="anthropic/claude-sonnet-4.5",
-                server_type="openai",
-                api_key=os.getenv("OPENROUTER_API_KEY", ""),
-                health_check=False,
-            ),
-        ]
-        return env_config, server_configs
-
-    def _cleanup_instances(self):
-        """Destroy all running dojo instances. Called on exit/signal."""
-        if not self.sync_client:
-            return
-        try:
-            n = self.sync_client.destroy_all()
-            if n:
-                logger.info("Cleaned up %d dojo instance(s)", n)
-        except Exception as e:
-            logger.warning("Instance cleanup failed: %s", e)
-
-        if hasattr(self, "_auto_ssh_key_dir"):
-            import shutil
-            shutil.rmtree(self._auto_ssh_key_dir, ignore_errors=True)
-
-    def _signal_handler(self, signum, frame):
-        """Handle SIGINT/SIGTERM: clean up instances, then re-raise."""
-        logger.info("Signal %d received, cleaning up dojo instances...", signum)
-        self._cleanup_instances()
-        signal.signal(signum, signal.SIG_DFL)
-        os.kill(os.getpid(), signum)
-
-    async def _ensure_ssh_key(self):
-        """Auto-generate and register an SSH key if none configured."""
-        if self.config.ssh_key and Path(self.config.ssh_key).exists():
-            return
-
-        import subprocess
-        import tempfile
-
-        key_dir = Path(tempfile.mkdtemp(prefix="hermes-ssh-"))
-        key_path = key_dir / "id_ed25519"
-
-        subprocess.run(
-            ["ssh-keygen", "-t", "ed25519", "-f", str(key_path), "-N", "", "-q"],
-            check=True,
-        )
-
-        pub_key = key_path.with_suffix(".pub").read_text().strip()
-        registered = await self.client.register_ssh_key(pub_key)
-        if not registered:
-            raise RuntimeError("Failed to register SSH key with dojo")
-
-        self.config.ssh_key = str(key_path)
-        os.environ["TERMINAL_SSH_KEY"] = str(key_path)
-        self._auto_ssh_key_dir = key_dir
-
-        logger.info("Auto-generated SSH key and registered with dojo")
-
-    async def setup(self):
-        """Load challenges from dojo and initialize SDK clients."""
-        self.client = DojoRLClient(self.config.base_url)
-        self.sync_client = DojoRLSyncClient(self.config.base_url)
-
-        await self._ensure_ssh_key()
-
-        atexit.register(self._cleanup_instances)
-        signal.signal(signal.SIGINT, self._signal_handler)
-        signal.signal(signal.SIGTERM, self._signal_handler)
-
-        # Fetch challenges
-        challenges = await self.client.list_challenges()
-        logger.info("Fetched %d challenges from dojo", len(challenges))
-
-        # Apply filters
-        if self.config.include_challenges:
-            # Explicit include list overrides all other filters
-            include_set = set(self.config.include_challenges)
-            for c in challenges:
-                if c.challenge_key in include_set:
-                    self.train.append(c)
-        else:
-            for c in challenges:
-                if (self.config.dojo_filter
-                        and c.dojo_id != self.config.dojo_filter):
-                    continue
-                if (self.config.module_filter
-                        and c.module_id != self.config.module_filter):
-                    continue
-                self.train.append(c)
-
-        # If a specific challenge is set and no filters matched, use it directly
-        if not self.train and self.config.challenge:
-            parts = self.config.challenge.split("/")
-            self.train.append(
-                RLChallenge(
-                    id=parts[-1],
-                    module_id=parts[0],
-                    dojo_id="unknown",
-                    name=self.config.challenge,
-                    description="",
-                )
-            )
-
-        if not self.train:
-            raise RuntimeError(
-                f"No challenges matched filters (dojo_filter={self.config.dojo_filter}, "
-                f"module_filter={self.config.module_filter}, challenge={self.config.challenge}). "
-                f"Total available: {len(challenges)}"
-            )
-
-        logger.info("Training on %d challenges", len(self.train))
-
-    async def get_next_item(self) -> RLChallenge:
-        """Return next challenge item (round-robin)."""
-        item = self.train[self.iter % len(self.train)]
-        self.iter += 1
-        return item
-
-    def _get_challenge_key(self, item: RLChallenge) -> str:
-        """Extract the challenge key from a challenge."""
-        return item.challenge_key or f"{item.module_id or ''}/{item.id}"
-
-    def format_prompt(self, item: RLChallenge) -> str:
-        """Build user prompt from challenge metadata."""
-        challenge_key = self._get_challenge_key(item)
-        return USER_PROMPT_TEMPLATE.format(
-            module_name=item.module_id or "unknown",
-            challenge_name=item.name or item.id,
-            challenge_description=item.description or f"Solve the challenge: {challenge_key}",
-        )
-
-    async def _acquire_instance(
-        self, challenge_key: str, *, pool_slot: Optional[int] = None,
-    ) -> Optional[RLInstance]:
-        """Acquire a dojo instance for a challenge.
-
-        If *pool_slot* is given (process mode), try to reset the slot.
-        If the slot is dead on the dojo, destroy it and create a fresh
-        one.  The returned instance may have a different slot ID than
-        *pool_slot* — callers must use ``inst.slot`` going forward.
-
-        If *pool_slot* is ``None`` (evaluate / serve modes), create a
-        new instance with transient-error retries.
-        """
-        if pool_slot is not None:
-            # Pool mode: try reset first (fast path)
-            try:
-                return await self.client.reset_instance(
-                    pool_slot, challenge=challenge_key,
-                )
-            except Exception as e:
-                logger.warning(
-                    "reset_instance(%d, %s) failed: %s — "
-                    "destroying and creating fresh slot",
-                    pool_slot, challenge_key, str(e)[:80],
-                )
-                try:
-                    await self.client.destroy_instance(pool_slot)
-                except Exception:
-                    pass
-                # Fall through to create mode
-
-        # Create mode: new instance with transient-error retries
-        max_retries = 10 if pool_slot is not None else 5
-        for attempt in range(max_retries):
-            try:
-                return await self.client.create_instance(
-                    challenge_key,
-                )
-            except Exception as e:
-                err_str = str(e)
-                is_transient = (
-                    isinstance(e, httpx.HTTPStatusError)
-                    and e.response.status_code >= 500
-                    or isinstance(e, (
-                        httpx.ReadTimeout,
-                        httpx.ConnectTimeout,
-                        httpx.ConnectError,
-                    ))
-                    or "No available slots" in err_str
-                )
-                if is_transient and attempt < max_retries - 1:
-                    wait = min(2 ** (attempt + 1), 60)
-                    logger.warning(
-                        "Transient error creating instance "
-                        "for %s (attempt %d/%d): %s, "
-                        "retrying in %ds",
-                        challenge_key, attempt + 1,
-                        max_retries, err_str[:80], wait,
-                    )
-                    await asyncio.sleep(wait)
-                else:
-                    logger.error(
-                        "Failed to create instance for %s "
-                        "after %d attempts: %s",
-                        challenge_key, attempt + 1, e,
-                    )
-                    return None
-        return None
-
-    async def collect_trajectory(
-        self, item: Item, *, pool_instance: Optional[RLInstance] = None,
-    ) -> Tuple[Optional[Union[ScoredDataItem, Any]], List[Item]]:
-        """Run a single rollout with dojo instance lifecycle.
-
-        Wraps the agent loop with:
-        1. Dojo instance creation (SSH-accessible challenge container)
-        2. SSH override registration (routes terminal tool to the instance)
-        3. Flag context registration (enables submit_flag tool)
-        4. Cleanup on completion
-
-        When *pool_instance* is provided (process mode), that
-        pre-acquired instance is used directly and NOT destroyed on
-        completion — the caller manages its lifecycle.
-        """
-        task_id = str(uuid.uuid4())
-        challenge_key = self._get_challenge_key(item)
-        owns_slot = pool_instance is None
-
-        if pool_instance is not None:
-            inst = pool_instance
-        else:
-            inst = await self._acquire_instance(challenge_key)
-            if inst is None:
-                return None, []
-
-        slot = inst.slot
-        self._active_slots.add(slot)
-        register_task_env_overrides(
-            task_id,
-            {
-                "ssh_user": inst.ssh_user,
-                "ssh_host": self.config.ssh_host,
-                "ssh_port": self.config.ssh_port,
-                "ssh_key": self.config.ssh_key,
-            },
-        )
-        register_flag_context(task_id, self.sync_client, slot)
-
-        try:
-            # Resolve tools (includes submit_flag via "pwncollege" toolset)
-            if self._current_group_tools is None:
-                tools, valid_names = self._resolve_tools_for_group()
-            else:
-                tools, valid_names = self._current_group_tools
-
-            messages: List[Dict[str, Any]] = []
-            if self.config.system_prompt:
-                messages.append({"role": "system", "content": self.config.system_prompt})
-            messages.append({"role": "user", "content": self.format_prompt(item)})
-
-            agent = HermesAgentLoop(
-                server=self.server,
-                tool_schemas=tools,
-                valid_tool_names=valid_names,
-                max_turns=self.config.max_agent_turns,
-                task_id=task_id,
-                temperature=self.config.agent_temperature,
-                max_tokens=self.config.max_token_length,
-                extra_body=self.config.extra_body,
-            )
-            result = await agent.run(messages)
-
-            # Skip reward if agent produced no output
-            only_system_and_user = all(
-                msg.get("role") in ("system", "user") for msg in result.messages
-            )
-            if result.turns_used == 0 or only_system_and_user:
-                logger.warning("Agent produced no output for %s", challenge_key)
-                reward = 0.0
-            else:
-                ctx = ToolContext(task_id)
-                try:
-                    reward = await self.compute_reward(item, result, ctx)
-                finally:
-                    ctx.cleanup()
-
-            # Track tool errors
-            if result.tool_errors:
-                for err in result.tool_errors:
-                    self._tool_error_buffer.append({
-                        "turn": err.turn,
-                        "tool": err.tool_name,
-                        "args": err.arguments[:150],
-                        "error": err.error[:300],
-                        "result": err.tool_result[:300],
-                    })
-
-            # Build scored item (Phase 1: placeholder tokens)
-            full_text = "\n".join(
-                msg.get("content", "") for msg in result.messages if msg.get("content")
-            )
-            if self.tokenizer:
-                tokens = self.tokenizer.encode(full_text, add_special_tokens=True)
-            else:
-                tokens = list(range(min(len(full_text) // 4, 128)))
-
-            scored_item = {
-                "tokens": tokens,
-                "masks": [-100] + tokens[1:],
-                "scores": reward,
-                "messages": result.messages,
-            }
-            return scored_item, []
-
-        finally:
-            clear_flag_context(task_id)
-            clear_task_env_overrides(task_id)
-            cleanup_vm(task_id)
-            if owns_slot:
-                # Evaluate/serve mode: we created it, we destroy it
-                try:
-                    await self.client.destroy_instance(slot)
-                except Exception as e:
-                    logger.warning("Failed to destroy instance slot %d: %s", slot, e)
-            # Pool mode: caller is responsible for the slot lifecycle
-            self._active_slots.discard(slot)
-
-    async def compute_reward(
-        self, item: Item, result: AgentResult, ctx: ToolContext
-    ) -> float:
-        """Score the rollout: 1.0 if flag was correctly submitted, 0.0 otherwise.
-
-        Checks two signals:
-        1. Did submit_flag return {"success": true}?
-        2. Fallback: extract pwn.college{...} from terminal output and verify via SDK.
-        """
-        # Check submit_flag tool results in the conversation
-        for msg in result.messages:
-            if msg.get("role") == "tool":
-                try:
-                    data = json.loads(msg.get("content", ""))
-                    if isinstance(data, dict) and data.get("success") is True:
-                        self.solve_rate_buffer.append(1.0)
-                        return 1.0
-                except (json.JSONDecodeError, TypeError):
-                    pass
-
-        # Fallback: scan for flag pattern in all messages
-        for msg in result.messages:
-            content = msg.get("content", "")
-            if not content:
-                continue
-            flag_match = re.search(r"pwn\.college\{[^}]+\}", content)
-            if flag_match:
-                # We can't verify here since instance is being torn down,
-                # but the flag pattern presence suggests partial progress
-                self.solve_rate_buffer.append(0.0)
-                return 0.0
-
-        self.solve_rate_buffer.append(0.0)
-        return 0.0
-
-    async def process_manager(self):
-        """Override: process items concurrently with pre-allocated slot pool.
-
-        Uses a pool of dojo instances (asyncio.Queue) instead of a semaphore.
-        Each task waits for a real dojo slot to become available, resets it
-        to the target challenge, and returns it to the pool on completion.
-        This guarantees zero silent drops from slot contention.
-        """
-        from atroposlib.frontend.jsonl2html import generate_html
-
-        await self.setup()
-
-        if self.config.use_wandb:
-            import random
-            import string
-            from datetime import datetime
-
-            import wandb
-
-            random_id = "".join(random.choices(string.ascii_lowercase, k=6))
-            current_date = datetime.now().strftime("%Y-%m-%d")
-            wandb.init(
-                project=self.wandb_project,
-                name=f"{self.name}-{current_date}-{random_id}",
-                group=self.wandb_group,
-                config=self.config.model_dump(),
-            )
-
-        self.config.group_size = self.group_size_to_process
-        items = self.train[:self.n_groups_to_process]
-
-        total = len(items)
-        concurrency = self.config.eval_concurrency
-        completed = 0
-
-        # --- Pre-allocate slot pool ---
-        # Use the first challenge as a throwaway target; each task will
-        # reset_instance to its own challenge before running.
-        first_key = self._get_challenge_key(items[0]) if items else "hello/hello"
-        slot_pool: asyncio.Queue[int] = asyncio.Queue()
-        pool_size = 0
-
-        logger.info("Pre-allocating %d dojo slots...", concurrency)
-        for i in range(concurrency):
-            try:
-                inst = await self.client.create_instance(first_key)
-                slot_pool.put_nowait(inst.slot)
-                pool_size += 1
-            except Exception as e:
-                # Dojo has a hard slot cap; once full, stop trying
-                logger.info(
-                    "Pre-allocated %d/%d slots (dojo full: %s)",
-                    i, concurrency, e,
-                )
-                break
-
-        if pool_size == 0:
-            raise RuntimeError("Could not allocate any dojo slots")
-
-        logger.info(
-            "Processing %d items (pool_size=%d, group_size=%d)",
-            total, pool_size, self.group_size_to_process,
-        )
-
-        # Resolve tools once before launching concurrent tasks
-        self._current_group_tools = self._resolve_tools_for_group()
-
-        async def process_one(item):
-            nonlocal completed
-            challenge_key = self._get_challenge_key(item)
-
-            # Wait for a real slot (blocks until one is returned)
-            original_slot = await slot_pool.get()
-            # _acquire_instance may create a new slot if the original
-            # died on the dojo, so we track the actual slot to return.
-            actual_slot: int | None = original_slot
-
-            try:
-                # Acquire instance (reset or create)
-                inst = await self._acquire_instance(
-                    challenge_key, pool_slot=original_slot,
-                )
-                if inst is None:
-                    logger.warning(
-                        "Could not acquire instance for %s",
-                        challenge_key,
-                    )
-                    actual_slot = None  # don't poison pool
-                    return
-                actual_slot = inst.slot
-                if actual_slot != original_slot:
-                    logger.info(
-                        "Slot %d replaced with %d for %s",
-                        original_slot, actual_slot,
-                        challenge_key,
-                    )
-
-                # Run the trajectory with the acquired instance
-                scored, _ = await self.collect_trajectory(
-                    item, pool_instance=inst,
-                )
-                if scored is None:
-                    logger.warning(
-                        "No scored data for %s (slot %d)",
-                        challenge_key, actual_slot,
-                    )
-                    return
-
-                # Wrap in ScoredDataGroup for postprocessing
-                to_postprocess = {
-                    "tokens": [scored["tokens"]],
-                    "masks": [scored["masks"]],
-                    "scores": [scored["scores"]],
-                    "advantages": [],
-                    "ref_logprobs": [],
-                    "messages": [scored.get("messages", [])],
-                    "group_overrides": {},
-                    "overrides": [],
-                    "images": [],
-                }
-                processed = await self.postprocess_histories(
-                    to_postprocess,
-                )
-                await self.handle_send_to_api(
-                    processed, item,
-                    do_send_to_api=False,
-                    abort_on_any_max_length_exceeded=False,
-                )
-            except Exception as e:
-                logger.error(
-                    "Failed to process %s: %s", challenge_key, e,
-                )
-            finally:
-                completed += 1
-                logger.info(
-                    "Processed %d/%d (%s)",
-                    completed, total, challenge_key,
-                )
-                # Return the actual slot to pool (may differ from
-                # original_slot if reset failed and a new one was
-                # created). None means acquisition failed entirely.
-                if actual_slot is not None:
-                    slot_pool.put_nowait(actual_slot)
-
-        await asyncio.gather(*[process_one(item) for item in items])
-
-        logger.info("Completed processing %d items", completed)
-
-        # Cleanup: destroy all pooled slots
-        while not slot_pool.empty():
-            slot = slot_pool.get_nowait()
-            try:
-                await self.client.destroy_instance(slot)
-            except Exception as e:
-                logger.warning("Failed to destroy pool slot %d: %s", slot, e)
-
-        if self.jsonl_writer is not None:
-            self.jsonl_writer.close()
-
-        if self.config.data_path_to_save_groups:
-            generate_html(self.config.data_path_to_save_groups)
-
-    async def evaluate(self, *args, **kwargs):
-        """Run evaluation on a dojo/module and report solve rate.
-
-        Fetches challenges matching eval_dojo/eval_module, runs each through
-        the agent loop with concurrency control, and logs results.
-        """
-        import time
-
-        if not self.client:
-            logger.error("SDK client not initialized. Call setup() first.")
-            return
-
-        start_time = time.time()
-
-        # Fetch and filter eval challenges
-        all_challenges = await self.client.list_challenges()
-        if self.config.eval_challenges:
-            challenge_set = set(self.config.eval_challenges)
-            eval_challenges = [c for c in all_challenges if c.challenge_key in challenge_set]
-        else:
-            eval_challenges = [
-                c for c in all_challenges
-                if (self.config.eval_dojo is None or c.dojo_id == self.config.eval_dojo)
-                and (self.config.eval_module is None or c.module_id == self.config.eval_module)
-                and c.dojo_id not in self.config.eval_exclude_dojos
-                and c.module_id not in self.config.eval_exclude_modules
-            ]
-
-        if not eval_challenges:
-            logger.warning(
-                "No challenges found for eval_dojo=%s eval_module=%s",
-                self.config.eval_dojo, self.config.eval_module,
-            )
-            return
-
-        print(
-            f"Evaluating {len(eval_challenges)} challenges from "
-            f"{self.config.eval_dojo or '*'}/{self.config.eval_module or '*'} "
-            f"(concurrency={self.config.eval_concurrency})",
-            flush=True,
-        )
-
-        semaphore = asyncio.Semaphore(self.config.eval_concurrency)
-        completed = 0
-        total = len(eval_challenges)
-
-        async def eval_one(challenge: RLChallenge) -> dict:
-            nonlocal completed
-            challenge_key = self._get_challenge_key(challenge)
-            async with semaphore:
-                try:
-                    scored, _ = await self.collect_trajectory(challenge)
-                    solved = scored is not None and scored.get("scores", 0.0) >= 1.0
-                    completed += 1
-                    status = "PASS" if solved else "FAIL"
-                    reward = scored.get("scores", 0.0) if scored else 0.0
-                    print(
-                        f"  [{completed}/{total}] [{status}] {challenge_key} "
-                        f"(reward={reward:.1f})",
-                        flush=True,
-                    )
-                    result = {
-                        "challenge": challenge_key,
-                        "name": challenge.name,
-                        "solved": solved,
-                        "reward": reward,
-                    }
-                    # Stream-write sample with full conversation for HTML viewer
-                    self.log_eval_sample({
-                        "score": reward,
-                        "challenge": challenge_key,
-                        "solved": solved,
-                        "messages": scored.get("messages", []) if scored else [],
-                    })
-                    return result
-                except Exception as e:
-                    completed += 1
-                    print(
-                        f"  [{completed}/{total}] [ERR ] {challenge_key}: {e}",
-                        flush=True,
-                    )
-                    self.log_eval_sample({
-                        "score": 0.0,
-                        "challenge": challenge_key,
-                        "solved": False,
-                        "messages": [{"role": "system", "content": f"Error: {e}"}],
-                    })
-                    return {
-                        "challenge": challenge_key,
-                        "name": challenge.name,
-                        "solved": False,
-                        "reward": 0.0,
-                        "error": str(e),
-                    }
-
-        tasks = [eval_one(c) for c in eval_challenges]
-        results = await asyncio.gather(*tasks)
-
-        end_time = time.time()
-
-        # Aggregate
-        n = len(results)
-        solved = sum(1 for r in results if r["solved"])
-        solve_rate = solved / n if n else 0.0
-
-        print("=" * 60, flush=True)
-        print(
-            f"Eval: {solved}/{n} solved ({solve_rate * 100:.1f}%) "
-            f"in {end_time - start_time:.1f}s",
-            flush=True,
-        )
-        print("=" * 60, flush=True)
-
-        eval_metrics = {
-            "eval/solve_rate": solve_rate,
-            "eval/solved": solved,
-            "eval/total": n,
-        }
-
-        await self.evaluate_log(
-            metrics=eval_metrics,
-            start_time=start_time,
-            end_time=end_time,
-        )
-
-    async def wandb_log(self, wandb_metrics: Optional[Dict] = None):
-        """Log solve rate metrics to wandb."""
-        if wandb_metrics is None:
-            wandb_metrics = {}
-        if self.solve_rate_buffer:
-            n = len(self.solve_rate_buffer)
-            wandb_metrics["train/solve_rate"] = sum(self.solve_rate_buffer) / n
-            wandb_metrics["train/num_rollouts"] = n
-            self.solve_rate_buffer = []
-        await super().wandb_log(wandb_metrics)
-
-
-if __name__ == "__main__":
-    PwnCollegeEnv.cli()

environments/pwncollege_env/sdk.py

@@ -1,468 +0,0 @@
-"""SDK for pwncollege dojo"""
-
-import asyncio
-import logging
-import re
-from contextlib import asynccontextmanager
-from dataclasses import dataclass, field
-from typing import Any
-
-import httpx
-
-logger = logging.getLogger(__name__)
-
-
-def _extract_csrf_nonce(html: str) -> str | None:
-    match = re.search(r"'csrfNonce': \"([^\"]+)\"", html)
-    return match.group(1) if match else None
-
-
-@dataclass
-class RLInstance:
-    slot: int
-    ssh_user: str
-    challenge_id: str
-    module_id: str
-    dojo_id: str
-    flag: str | None = None
-    created_at: float | None = None
-    status: str | None = None
-
-    @property
-    def challenge_key(self) -> str:
-        return f"{self.module_id}/{self.challenge_id}"
-
-
-@dataclass
-class RLResource:
-    type: str
-    name: str
-    content: str | None = None
-    video: str | None = None
-    slides: str | None = None
-
-
-@dataclass
-class RLChallenge:
-    id: str
-    name: str
-    description: str
-    module_id: str | None = None
-    module_name: str | None = None
-    module_description: str | None = None
-    dojo_id: str | None = None
-    dojo_name: str | None = None
-    dojo_description: str | None = None
-    resources: list[RLResource] = field(default_factory=list)
-
-    @property
-    def challenge_key(self) -> str | None:
-        if self.module_id:
-            return f"{self.module_id}/{self.id}"
-        return None
-
-
-@dataclass
-class RLStatus:
-    enabled: bool
-    max_instances: int
-    running: int
-    instances: list[RLInstance]
-
-
-class DojoRLClient:
-    """Client for the dojo RL API. No auth required."""
-
-    def __init__(self, base_url: str, timeout: float = 120.0):
-        self.base_url = base_url.rstrip("/")
-        self.client = httpx.AsyncClient(
-            base_url=self.base_url,
-            timeout=timeout,
-            follow_redirects=True,
-        )
-
-    async def __aenter__(self):
-        return self
-
-    async def __aexit__(self, *args):
-        await self.close()
-
-    async def close(self):
-        await self.client.aclose()
-
-    def _rl_url(self, path: str) -> str:
-        return f"/pwncollege_api/v1/rl{path}"
-
-    async def _get(self, path: str) -> dict[str, Any]:
-        resp = await self.client.get(self._rl_url(path))
-        resp.raise_for_status()
-        return resp.json()
-
-    async def _post(self, path: str, json: dict | None = None) -> dict[str, Any]:
-        resp = await self.client.post(self._rl_url(path), json=json or {})
-        resp.raise_for_status()
-        return resp.json()
-
-    async def _delete(self, path: str) -> dict[str, Any]:
-        resp = await self.client.delete(self._rl_url(path))
-        resp.raise_for_status()
-        return resp.json()
-
-    # ── Response Parsing ──────────────────────────────────────────────────────
-    # The API uses different field names in create/reset vs get/list responses.
-    # These parsers normalize everything into RLInstance.
-
-    @staticmethod
-    def _parse_create_response(data: dict[str, Any]) -> RLInstance:
-        return RLInstance(
-            slot=data["slot"],
-            ssh_user=data["ssh_user"],
-            challenge_id=data["challenge"],
-            module_id=data["module"],
-            dojo_id=data["dojo"],
-        )
-
-    @staticmethod
-    def _parse_instance_detail(data: dict[str, Any]) -> RLInstance:
-        created_at = data.get("created_at")
-        return RLInstance(
-            slot=data["slot"],
-            ssh_user=data.get("ssh_user", f"rl_{data['slot']}"),
-            challenge_id=data["challenge_id"],
-            module_id=data["module_id"],
-            dojo_id=data["dojo_id"],
-            flag=data.get("flag"),
-            created_at=float(created_at) if created_at else None,
-        )
-
-    @staticmethod
-    def _parse_instance_listing(data: dict[str, Any]) -> RLInstance:
-        created_at = data.get("created_at")
-        return RLInstance(
-            slot=data["slot"],
-            ssh_user=f"rl_{data['slot']}",
-            challenge_id=data["challenge_id"],
-            module_id=data["module_id"],
-            dojo_id=data["dojo_id"],
-            created_at=float(created_at) if created_at else None,
-            status=data.get("status"),
-        )
-
-    @staticmethod
-    def _parse_challenge(data: dict[str, Any]) -> RLChallenge:
-        resources = [
-            RLResource(
-                type=r["type"],
-                name=r["name"],
-                content=r.get("content"),
-                video=r.get("video"),
-                slides=r.get("slides"),
-            )
-            for r in data.get("resources", [])
-        ]
-        return RLChallenge(
-            id=data["id"],
-            name=data["name"],
-            description=data["description"],
-            module_id=data.get("module_id"),
-            module_name=data.get("module_name"),
-            module_description=data.get("module_description"),
-            dojo_id=data.get("dojo_id"),
-            dojo_name=data.get("dojo_name"),
-            dojo_description=data.get("dojo_description"),
-            resources=resources,
-        )
-
-    # ── RL Instance Lifecycle ─────────────────────────────────────────────────
-
-    async def status(self) -> RLStatus:
-        result = await self._get("/status")
-        instances = [
-            self._parse_instance_listing(inst) for inst in result.get("instances", [])
-        ]
-        return RLStatus(
-            enabled=result["enabled"],
-            max_instances=result["max_instances"],
-            running=result["running"],
-            instances=instances,
-        )
-
-    async def create_instance(
-        self, challenge: str, *, variant: int | None = None
-    ) -> RLInstance:
-        data: dict[str, Any] = {"challenge": challenge}
-        if variant is not None:
-            data["variant"] = variant
-        result = await self._post("/instances", json=data)
-        if not result.get("success"):
-            raise RuntimeError(f"Failed to create instance: {result.get('error')}")
-        return self._parse_create_response(result)
-
-    async def get_instance(self, slot: int) -> RLInstance:
-        result = await self._get(f"/instances/{slot}")
-        if not result.get("success"):
-            raise KeyError(f"No instance at slot {slot}")
-        return self._parse_instance_detail(result)
-
-    async def list_instances(self) -> list[RLInstance]:
-        result = await self._get("/instances")
-        return [
-            self._parse_instance_listing(inst) for inst in result.get("instances", [])
-        ]
-
-    async def destroy_instance(self, slot: int) -> None:
-        result = await self._delete(f"/instances/{slot}")
-        if not result.get("success"):
-            raise RuntimeError(f"Failed to destroy instance: {result.get('error')}")
-
-    async def reset_instance(
-        self, slot: int, *, challenge: str | None = None
-    ) -> RLInstance:
-        data: dict[str, Any] = {}
-        if challenge is not None:
-            data["challenge"] = challenge
-        result = await self._post(f"/instances/{slot}/reset", json=data)
-        if not result.get("success"):
-            raise RuntimeError(f"Failed to reset instance: {result.get('error')}")
-        return self._parse_create_response(result)
-
-    async def check_flag(self, slot: int, flag: str) -> bool:
-        result = await self._post(f"/instances/{slot}/check", json={"flag": flag})
-        return result.get("correct", False)
-
-    async def get_flag(self, slot: int) -> str:
-        instance = await self.get_instance(slot)
-        if instance.flag is None:
-            raise RuntimeError(f"No flag available for slot {slot}")
-        return instance.flag
-
-    # ── SSH Key Management ────────────────────────────────────────────────────
-
-    async def register_ssh_key(self, public_key: str) -> bool:
-        result = await self._post("/ssh_key", json={"public_key": public_key})
-        return result.get("success", False)
-
-    async def get_ssh_key(self) -> dict[str, Any]:
-        return await self._get("/ssh_key")
-
-    # ── Challenge Discovery ───────────────────────────────────────────────────
-
-    async def list_challenges(self) -> list[RLChallenge]:
-        result = await self._get("/challenges")
-        return [self._parse_challenge(ch) for ch in result.get("challenges", [])]
-
-    # ── Admin (requires auth) ─────────────────────────────────────────────────
-
-    async def admin_login(
-        self, username: str = "admin", password: str = "admin"
-    ) -> None:
-        resp = await self.client.get("/login")
-        nonce = _extract_csrf_nonce(resp.text)
-        if not nonce:
-            raise RuntimeError("Could not extract CSRF nonce")
-        self._admin_csrf = nonce
-        resp = await self.client.post(
-            "/login",
-            data={"name": username, "password": password, "nonce": nonce},
-        )
-        if resp.status_code not in (200, 302):
-            raise RuntimeError(f"Login failed: {resp.status_code}")
-        resp = await self.client.get("/")
-        self._admin_csrf = _extract_csrf_nonce(resp.text) or self._admin_csrf
-
-    async def load_dojo(self, repository: str) -> str:
-        if not hasattr(self, "_admin_csrf"):
-            raise RuntimeError("Must call admin_login() first")
-        resp = await self.client.post(
-            "/pwncollege_api/v1/dojos/create",
-            json={
-                "repository": repository,
-                "public_key": f"public/{repository}",
-                "private_key": f"private/{repository}",
-            },
-            headers={"CSRF-Token": self._admin_csrf},
-        )
-        resp.raise_for_status()
-        data = resp.json()
-        if not data.get("success", True):
-            raise RuntimeError(f"Failed to load dojo: {data.get('error', data)}")
-        return data.get("dojo", repository)
-
-    async def promote_dojo(self, dojo_id: str) -> None:
-        if not hasattr(self, "_admin_csrf"):
-            raise RuntimeError("Must call admin_login() first")
-        resp = await self.client.post(
-            f"/pwncollege_api/v1/dojos/{dojo_id}/promote",
-            json={},
-            headers={"CSRF-Token": self._admin_csrf},
-        )
-        resp.raise_for_status()
-
-    # ── Bulk Operations ───────────────────────────────────────────────────────
-
-    async def create_batch(self, challenge: str, count: int) -> list[RLInstance]:
-        tasks = [self.create_instance(challenge) for _ in range(count)]
-        return await asyncio.gather(*tasks)
-
-    async def destroy_all(self) -> int:
-        instances = await self.list_instances()
-        for inst in instances:
-            await self.destroy_instance(inst.slot)
-        return len(instances)
-
-
-class DojoRLSyncClient:
-    """Sync wrapper for DojoRLClient.
-
-    Runs all async operations on a dedicated background thread with its own
-    event loop, so it's safe to call from any context — including from inside
-    another running event loop (e.g., Atropos's loop or tool dispatch threads).
-    """
-
-    def __init__(self, base_url: str, timeout: float = 120.0):
-        import threading
-
-        self._async = DojoRLClient(base_url, timeout)
-        self._loop = asyncio.new_event_loop()
-        self._thread = threading.Thread(
-            target=self._loop.run_forever,
-            daemon=True,
-        )
-        self._thread.start()
-
-    def _run(self, coro):
-        return asyncio.run_coroutine_threadsafe(coro, self._loop).result()
-
-    def __enter__(self):
-        return self
-
-    def __exit__(self, *args):
-        self.close()
-
-    def close(self):
-        if not self._loop.is_running():
-            return
-        try:
-            self._run(self._async.close())
-        except Exception:
-            pass
-        self._loop.call_soon_threadsafe(self._loop.stop)
-        self._thread.join(timeout=5)
-
-    def status(self) -> RLStatus:
-        return self._run(self._async.status())
-
-    def create_instance(
-        self, challenge: str, *, variant: int | None = None
-    ) -> RLInstance:
-        return self._run(self._async.create_instance(challenge, variant=variant))
-
-    def get_instance(self, slot: int) -> RLInstance:
-        return self._run(self._async.get_instance(slot))
-
-    def list_instances(self) -> list[RLInstance]:
-        return self._run(self._async.list_instances())
-
-    def destroy_instance(self, slot: int) -> None:
-        return self._run(self._async.destroy_instance(slot))
-
-    def reset_instance(self, slot: int, *, challenge: str | None = None) -> RLInstance:
-        return self._run(self._async.reset_instance(slot, challenge=challenge))
-
-    def check_flag(self, slot: int, flag: str) -> bool:
-        return self._run(self._async.check_flag(slot, flag))
-
-    def get_flag(self, slot: int) -> str:
-        return self._run(self._async.get_flag(slot))
-
-    def list_challenges(self) -> list[RLChallenge]:
-        return self._run(self._async.list_challenges())
-
-    def register_ssh_key(self, public_key: str) -> bool:
-        return self._run(self._async.register_ssh_key(public_key))
-
-    def get_ssh_key(self) -> dict[str, Any]:
-        return self._run(self._async.get_ssh_key())
-
-    def admin_login(self, username: str = "admin", password: str = "admin") -> None:
-        return self._run(self._async.admin_login(username, password))
-
-    def load_dojo(self, repository: str) -> str:
-        return self._run(self._async.load_dojo(repository))
-
-    def promote_dojo(self, dojo_id: str) -> None:
-        return self._run(self._async.promote_dojo(dojo_id))
-
-    def destroy_all(self) -> int:
-        return self._run(self._async.destroy_all())
-
-
-@dataclass
-class EpisodePool:
-    """Manages a pool of RL instances for parallel episode collection."""
-
-    client: DojoRLClient
-    challenge: str
-    pool_size: int = 32
-    acquisition_timeout: float = 300.0
-
-    _available: asyncio.Queue[RLInstance] = field(
-        default_factory=asyncio.Queue, init=False
-    )
-    _all_instances: dict[int, RLInstance] = field(default_factory=dict, init=False)
-    _initialized: bool = field(default=False, init=False)
-
-    async def initialize(self) -> None:
-        if self._initialized:
-            return
-        for _ in range(self.pool_size):
-            instance = await self.client.create_instance(self.challenge)
-            full = await self.client.get_instance(instance.slot)
-            self._all_instances[instance.slot] = full
-            await self._available.put(full)
-        self._initialized = True
-
-    @asynccontextmanager
-    async def acquire(self):
-        if not self._initialized:
-            raise RuntimeError("EpisodePool not initialized")
-        try:
-            instance = await asyncio.wait_for(
-                self._available.get(), timeout=self.acquisition_timeout
-            )
-        except asyncio.TimeoutError:
-            raise RuntimeError(
-                f"No instance available within {self.acquisition_timeout}s"
-            )
-        try:
-            yield instance
-        finally:
-            try:
-                reset = await self.client.reset_instance(
-                    instance.slot, challenge=self.challenge
-                )
-                full = await self.client.get_instance(reset.slot)
-                self._all_instances[reset.slot] = full
-                await self._available.put(full)
-            except Exception as e:
-                logger.error(
-                    "Failed to reset instance slot %d, returning stale instance: %s",
-                    instance.slot,
-                    e,
-                )
-                await self._available.put(instance)
-
-    async def shutdown(self) -> None:
-        errors = []
-        for slot in list(self._all_instances.keys()):
-            try:
-                await self.client.destroy_instance(slot)
-            except Exception as e:
-                errors.append((slot, e))
-                logger.warning("Failed to destroy instance slot %d: %s", slot, e)
-        self._all_instances.clear()
-        self._initialized = False
-        if errors:
-            logger.error(
-                "EpisodePool shutdown: %d instance(s) failed to destroy", len(errors)
-            )

environments/pwncollege_env/smoke_hello.yaml

@@ -1,74 +0,0 @@
-env:
-  group_size: 4
-  max_num_workers: -1
-  max_eval_workers: 16
-  max_num_workers_per_node: 8
-  steps_per_eval: 100
-  max_token_length: 16384
-  eval_handling: STOP_TRAIN
-  eval_limit_ratio: 0.5
-  inference_weight: 1.0
-  batch_size: -1
-  max_batches_offpolicy: 3
-  tokenizer_name: NousResearch/Hermes-3-Llama-3.1-8B
-  use_wandb: false
-  rollout_server_url: http://localhost:8000
-  total_steps: 1000
-  wandb_name: pwncollege-smoke-hello
-  num_rollouts_to_keep: 32
-  num_rollouts_per_group_for_logging: 1
-  ensure_scores_are_not_same: false
-  data_path_to_save_groups: null
-  data_dir_to_save_evals: environments/pwncollege_env/eval_runs/smoke_hello
-  min_items_sent_before_logging: 2
-  include_messages: false
-  min_batch_allocation: null
-  worker_timeout: 600.0
-  thinking_mode: false
-  reasoning_effort: null
-  max_reasoning_tokens: null
-  custom_thinking_prompt: null
-  enabled_toolsets:
-    - terminal
-    - file
-    - pwncollege
-  disabled_toolsets: null
-  distribution: null
-  max_agent_turns: 20
-  agent_temperature: 0.7
-  terminal_backend: ssh
-  terminal_timeout: 120
-  terminal_lifetime: 3600
-  disable_command_guards: true
-  dataset_name: null
-  dataset_split: train
-  prompt_field: prompt
-  tool_pool_size: 128
-  tool_call_parser: hermes
-  extra_body: null
-  base_url: http://100.120.55.25:8080
-  ssh_host: 100.120.55.25
-  ssh_port: 2222
-  ssh_key: environments/pwncollege_env/keys/rl_test_key
-  challenge: hello/hello
-  dojo_filter: null
-  module_filter: null
-  eval_dojo: linux-luminarium
-  eval_exclude_dojos:
-    - archive
-  eval_module: hello
-  eval_concurrency: 3
-openai:
-  - timeout: 1200
-    num_max_requests_at_once: 512
-    num_requests_for_eval: 64
-    model_name: xiaomi/mimo-v2-flash
-    rolling_buffer_length: 1000
-    server_type: openai
-    tokenizer_name: none
-    api_key: ""
-    base_url: https://openrouter.ai/api/v1
-    n_kwarg_is_ignored: false
-    health_check: false
-slurm: false
-testing: false

environments/pwncollege_env/stress_test.py

@@ -1,513 +0,0 @@
-"""
-Capability verification test for pwn-dojo RL infrastructure.
-
-Verifies that RL containers are provisioned with the correct Linux capabilities,
-resource limits, and host configuration for each challenge type.
-
-Usage:
-    python environments/pwncollege_env/stress_test.py -y
-    python environments/pwncollege_env/stress_test.py -y -o report.json --verbose
-"""
-
-import argparse
-import asyncio
-import json
-import sys
-import time
-from dataclasses import asdict, dataclass, field
-from pathlib import Path
-
-_repo_root = Path(__file__).resolve().parent.parent.parent
-if str(_repo_root) not in sys.path:
-    sys.path.insert(0, str(_repo_root))
-
-from environments.pwncollege_env.sdk import DojoRLClient
-
-
-@dataclass
-class SSHConfig:
-    host: str
-    port: int
-    key: str
-
-
-@dataclass
-class CheckResult:
-    name: str
-    passed: bool
-    message: str
-    duration: float = 0.0
-
-
-@dataclass
-class TestResult:
-    name: str
-    challenge: str
-    checks: list[CheckResult] = field(default_factory=list)
-    passed: bool = False
-    skipped: bool = False
-    error: str | None = None
-    duration: float = 0.0
-
-
-@dataclass
-class TestCase:
-    name: str
-    challenge: str
-    checks: list
-
-
-async def ssh_run(
-    cfg: SSHConfig, user: str, command: str, timeout: float = 30.0
-) -> tuple[int, str]:
-    """Run a command over SSH via subprocess. Returns (returncode, output)."""
-    cmd = [
-        "ssh",
-        "-o",
-        "BatchMode=yes",
-        "-o",
-        "StrictHostKeyChecking=accept-new",
-        "-o",
-        "UserKnownHostsFile=/dev/null",
-        "-o",
-        "ConnectTimeout=10",
-        "-o",
-        "LogLevel=ERROR",
-        "-p",
-        str(cfg.port),
-        "-i",
-        cfg.key,
-        f"{user}@{cfg.host}",
-        command,
-    ]
-    proc = await asyncio.create_subprocess_exec(
-        *cmd,
-        stdout=asyncio.subprocess.PIPE,
-        stderr=asyncio.subprocess.STDOUT,
-    )
-    try:
-        stdout, _ = await asyncio.wait_for(proc.communicate(), timeout=timeout)
-        return proc.returncode, stdout.decode(errors="replace")
-    except asyncio.TimeoutError:
-        proc.kill()
-        await proc.wait()
-        return -1, f"[SSH timeout after {timeout}s]"
-
-
-async def wait_ssh_ready(cfg: SSHConfig, user: str, retries: int = 10) -> bool:
-    for i in range(retries):
-        rc, out = await ssh_run(cfg, user, "echo ready", timeout=10)
-        if rc == 0 and "ready" in out:
-            return True
-        await asyncio.sleep(1)
-    return False
-
-
-# ── Check functions ──────────────────────────────────────────────────────────
-
-
-async def check_ssh_echo(cfg: SSHConfig, user: str) -> CheckResult:
-    t0 = time.monotonic()
-    rc, out = await ssh_run(cfg, user, "echo ok")
-    dur = time.monotonic() - t0
-    if rc == 0 and "ok" in out:
-        return CheckResult("ssh_echo", True, "connected", dur)
-    return CheckResult("ssh_echo", False, f"rc={rc}: {out.strip()[:100]}", dur)
-
-
-async def check_unshare_net(cfg: SSHConfig, user: str) -> CheckResult:
-    t0 = time.monotonic()
-    rc, out = await ssh_run(cfg, user, "unshare --net echo ok")
-    dur = time.monotonic() - t0
-    if rc == 0 and "ok" in out:
-        return CheckResult("unshare_net", True, "namespace creation works", dur)
-    return CheckResult("unshare_net", False, f"rc={rc}: {out.strip()[:120]}", dur)
-
-
-async def check_unshare_user(cfg: SSHConfig, user: str) -> CheckResult:
-    t0 = time.monotonic()
-    rc, out = await ssh_run(cfg, user, "unshare --user --map-root-user bash -c 'id'")
-    dur = time.monotonic() - t0
-    if rc == 0 and "uid=0" in out:
-        return CheckResult("unshare_user", True, "user namespace works", dur)
-    return CheckResult("unshare_user", False, f"rc={rc}: {out.strip()[:120]}", dur)
-
-
-async def check_capeff(cfg: SSHConfig, user: str) -> CheckResult:
-    """Check that the container init (PID 1) has SYS_ADMIN capability."""
-    t0 = time.monotonic()
-    rc, out = await ssh_run(cfg, user, "cat /proc/1/status")
-    dur = time.monotonic() - t0
-    if rc != 0:
-        return CheckResult(
-            "capeff", False, f"Cannot read /proc/1/status: {out.strip()[:80]}", dur
-        )
-    for line in out.splitlines():
-        if line.startswith("CapEff:") or line.startswith("CapBnd:"):
-            hex_val = line.split(":")[1].strip()
-            try:
-                val = int(hex_val, 16)
-                has_sysadmin = bool(val & (1 << 21))
-                if has_sysadmin:
-                    label = line.split(":")[0]
-                    return CheckResult(
-                        "capeff", True, f"{label}={hex_val} has SYS_ADMIN", dur
-                    )
-            except ValueError:
-                pass
-    return CheckResult(
-        "capeff", False, "SYS_ADMIN (bit 21) not found in capabilities", dur
-    )
-
-
-async def check_hosts_resolution(cfg: SSHConfig, user: str) -> CheckResult:
-    t0 = time.monotonic()
-    rc, out = await ssh_run(cfg, user, "getent hosts challenge.localhost")
-    dur = time.monotonic() - t0
-    if rc == 0 and out.strip():
-        return CheckResult(
-            "hosts_resolution", True, f"resolves to {out.strip()[:40]}", dur
-        )
-    rc2, out2 = await ssh_run(cfg, user, "grep challenge.localhost /etc/hosts")
-    dur = time.monotonic() - t0
-    if rc2 == 0 and "challenge.localhost" in out2:
-        return CheckResult(
-            "hosts_resolution", True, "/etc/hosts has entry", dur
-        )
-    return CheckResult(
-        "hosts_resolution", False, "challenge.localhost not resolvable", dur
-    )
-
-
-async def check_pids_limit(cfg: SSHConfig, user: str) -> CheckResult:
-    t0 = time.monotonic()
-    rc, out = await ssh_run(
-        cfg,
-        user,
-        "cat /sys/fs/cgroup/pids.max 2>/dev/null || cat /sys/fs/cgroup/pids/pids.max 2>/dev/null",
-    )
-    dur = time.monotonic() - t0
-    val = out.strip()
-    if val == "max":
-        return CheckResult("pids_limit", True, "unlimited", dur)
-    try:
-        limit = int(val)
-        if limit >= 1024:
-            return CheckResult("pids_limit", True, f"pids_limit={limit}", dur)
-        return CheckResult(
-            "pids_limit", False, f"pids_limit={limit} (need >= 1024)", dur
-        )
-    except ValueError:
-        return CheckResult("pids_limit", False, f"Cannot parse: {val[:60]}", dur)
-
-
-async def check_mem_limit(cfg: SSHConfig, user: str) -> CheckResult:
-    t0 = time.monotonic()
-    rc, out = await ssh_run(
-        cfg,
-        user,
-        "cat /sys/fs/cgroup/memory.max 2>/dev/null || cat /sys/fs/cgroup/memory/memory.limit_in_bytes 2>/dev/null",
-    )
-    dur = time.monotonic() - t0
-    val = out.strip()
-    if val == "max":
-        return CheckResult("mem_limit", True, "unlimited", dur)
-    try:
-        limit = int(val)
-        limit_gb = limit / (1024**3)
-        if (
-            limit_gb >= 1.8
-        ):  # 2GB for privileged RL containers (not 4GB to manage memory pressure)
-            return CheckResult("mem_limit", True, f"mem={limit_gb:.1f}GB", dur)
-        return CheckResult(
-            "mem_limit", False, f"mem={limit_gb:.1f}GB (need >= 2GB)", dur
-        )
-    except ValueError:
-        return CheckResult("mem_limit", False, f"Cannot parse: {val[:60]}", dur)
-
-
-async def check_challenge_run(cfg: SSHConfig, user: str) -> CheckResult:
-    """Run /challenge/run and verify no PermissionError."""
-    t0 = time.monotonic()
-    rc, out = await ssh_run(cfg, user, "/challenge/run < /dev/null", timeout=15)
-    dur = time.monotonic() - t0
-    if "PermissionError" in out or "Operation not permitted" in out:
-        snippet = [l for l in out.splitlines() if "Permission" in l or "Operation" in l]
-        return CheckResult(
-            "challenge_run",
-            False,
-            snippet[0][:120] if snippet else "PermissionError",
-            dur,
-        )
-    return CheckResult("challenge_run", True, f"No permission errors (rc={rc})", dur)
-
-
-# ── Test cases ───────────────────────────────────────────────────────────────
-
-TEST_CASES = [
-    TestCase("unprivileged_basic", "hello/hello", [check_ssh_echo]),
-    TestCase(
-        "privileged_caps",
-        "intercepting-communication/udp-1",
-        [check_ssh_echo, check_capeff],
-    ),
-    TestCase(
-        "privileged_challenge_run",
-        "intercepting-communication/udp-1",
-        [check_challenge_run],
-    ),
-    TestCase(
-        "web_challenge_hosts",
-        "web-security/path-traversal-1",
-        [check_ssh_echo, check_hosts_resolution],
-    ),
-    TestCase(
-        "resource_limits",
-        "intercepting-communication/udp-1",
-        [check_pids_limit, check_mem_limit],
-    ),
-]
-
-
-# ── Runner ───────────────────────────────────────────────────────────────────
-
-
-async def run_tests(args) -> dict:
-    cfg = SSHConfig(host=args.ssh_host, port=args.ssh_port, key=args.ssh_key)
-    client = DojoRLClient(args.base_url)
-
-    status = await client.status()
-    print(
-        f"Server: {args.base_url} (RL={'enabled' if status.enabled else 'DISABLED'}, "
-        f"{status.max_instances} max, {status.running} running)"
-    )
-    if status.running > 0:
-        n = await client.destroy_all()
-        print(f"Cleaned up {n} instance(s)")
-    print()
-
-    results: list[TestResult] = []
-    test_num = 0
-    total = len(TEST_CASES) + (0 if args.skip_concurrent else 1)
-    start_time = time.monotonic()
-
-    for tc in TEST_CASES:
-        test_num += 1
-        t0 = time.monotonic()
-        tr = TestResult(name=tc.name, challenge=tc.challenge)
-        print(f"[{test_num}/{total}] {tc.name} ({tc.challenge})")
-
-        try:
-            inst = await client.create_instance(tc.challenge)
-        except Exception as e:
-            err = str(e)
-            if "404" in err or "not found" in err.lower() or "Invalid" in err:
-                tr.skipped = True
-                tr.error = f"Challenge not available: {err[:80]}"
-                print(f"  SKIP  {tr.error}")
-            else:
-                tr.error = f"create_instance failed: {err[:100]}"
-                print(f"  ERR   {tr.error}")
-            tr.duration = time.monotonic() - t0
-            results.append(tr)
-            print(f"  --- {'SKIP' if tr.skipped else 'FAIL'} ({tr.duration:.1f}s)\n")
-            continue
-
-        try:
-            ready = await wait_ssh_ready(cfg, inst.ssh_user)
-            if not ready:
-                tr.error = "SSH not ready after 10 retries"
-                tr.checks.append(
-                    CheckResult("ssh_ready", False, tr.error, time.monotonic() - t0)
-                )
-                print(f"  FAIL  ssh_ready: {tr.error}")
-            else:
-                for check_fn in tc.checks:
-                    cr = await check_fn(cfg, inst.ssh_user)
-                    tr.checks.append(cr)
-                    tag = "PASS" if cr.passed else "FAIL"
-                    extra = f"  ({cr.message})" if args.verbose or not cr.passed else ""
-                    print(f"  {tag}  {cr.name:30s} {cr.duration:.1f}s{extra}")
-                    if not cr.passed:
-                        break
-        finally:
-            try:
-                await client.destroy_instance(inst.slot)
-            except Exception as e:
-                print(f"  WARN  destroy failed: {e}")
-
-        tr.passed = all(c.passed for c in tr.checks) and not tr.error
-        tr.duration = time.monotonic() - t0
-        results.append(tr)
-        print(f"  --- {'PASS' if tr.passed else 'FAIL'} ({tr.duration:.1f}s)\n")
-
-    if not args.skip_concurrent:
-        test_num += 1
-        t0 = time.monotonic()
-        tr = TestResult(name="concurrent_lifecycle", challenge="8x hello/hello")
-        n_concurrent = min(8, status.max_instances)
-        print(
-            f"[{test_num}/{total}] concurrent_lifecycle ({n_concurrent}x hello/hello)"
-        )
-
-        try:
-            ct0 = time.monotonic()
-            tasks = [client.create_instance("hello/hello") for _ in range(n_concurrent)]
-            instances = await asyncio.gather(*tasks, return_exceptions=True)
-            create_dur = time.monotonic() - ct0
-
-            created = [i for i in instances if not isinstance(i, Exception)]
-            errors = [i for i in instances if isinstance(i, Exception)]
-            if errors:
-                tr.checks.append(
-                    CheckResult(
-                        "create_all",
-                        False,
-                        f"{len(errors)}/{n_concurrent} failed: {errors[0]}",
-                        create_dur,
-                    )
-                )
-            else:
-                tr.checks.append(
-                    CheckResult(
-                        "create_all", True, f"{n_concurrent} created", create_dur
-                    )
-                )
-
-            if created:
-                await asyncio.sleep(3)
-                et0 = time.monotonic()
-                echo_tasks = [
-                    ssh_run(cfg, i.ssh_user, "echo ok", timeout=15) for i in created
-                ]
-                echo_results = await asyncio.gather(*echo_tasks, return_exceptions=True)
-                echo_ok = sum(
-                    1
-                    for r in echo_results
-                    if not isinstance(r, Exception) and r[0] == 0
-                )
-                tr.checks.append(
-                    CheckResult(
-                        "ssh_echo_all",
-                        echo_ok == len(created),
-                        f"{echo_ok}/{len(created)} connected",
-                        time.monotonic() - et0,
-                    )
-                )
-
-            dt0 = time.monotonic()
-            destroyed = await client.destroy_all()
-            tr.checks.append(
-                CheckResult(
-                    "destroy_all",
-                    True,
-                    f"destroyed {destroyed}",
-                    time.monotonic() - dt0,
-                )
-            )
-
-            st = await client.status()
-            live = sum(1 for i in st.instances if i.status == "running")
-            tr.checks.append(
-                CheckResult(
-                    "slot_cleanup",
-                    live == 0,
-                    f"running={live} (total listed={st.running})",
-                    0.0,
-                )
-            )
-        except Exception as e:
-            tr.error = str(e)[:200]
-            tr.checks.append(CheckResult("concurrent", False, str(e)[:100], 0.0))
-
-        tr.passed = all(c.passed for c in tr.checks) and not tr.error
-        tr.duration = time.monotonic() - t0
-        results.append(tr)
-        for cr in tr.checks:
-            tag = "PASS" if cr.passed else "FAIL"
-            extra = f"  ({cr.message})" if args.verbose or not cr.passed else ""
-            print(f"  {tag}  {cr.name:30s} {cr.duration:.1f}s{extra}")
-        print(f"  --- {'PASS' if tr.passed else 'FAIL'} ({tr.duration:.1f}s)\n")
-
-    total_dur = time.monotonic() - start_time
-    passed = sum(1 for r in results if r.passed)
-    failed = sum(1 for r in results if not r.passed and not r.skipped)
-    skipped = sum(1 for r in results if r.skipped)
-
-    print("=" * 50)
-    parts = [f"{passed}/{len(results)} passed"]
-    if failed:
-        parts.append(f"{failed} failed")
-    if skipped:
-        parts.append(f"{skipped} skipped")
-    print(f"RESULTS: {', '.join(parts)} in {total_dur:.0f}s")
-    print("=" * 50)
-
-    return {
-        "test": "capability_verification",
-        "timestamp": time.strftime("%Y-%m-%dT%H:%M:%S%z"),
-        "server": args.base_url,
-        "summary": {
-            "total": len(results),
-            "passed": passed,
-            "failed": failed,
-            "skipped": skipped,
-            "duration_seconds": round(total_dur, 1),
-        },
-        "tests": [
-            {
-                "name": r.name,
-                "challenge": r.challenge,
-                "passed": r.passed,
-                "skipped": r.skipped,
-                "error": r.error,
-                "duration": round(r.duration, 1),
-                "checks": [asdict(c) for c in r.checks],
-            }
-            for r in results
-        ],
-    }
-
-
-def main():
-    parser = argparse.ArgumentParser(
-        description="Capability verification test for pwn-dojo RL infrastructure",
-        formatter_class=argparse.ArgumentDefaultsHelpFormatter,
-    )
-    parser.add_argument("--base-url", default="http://100.120.55.25:8080")
-    parser.add_argument("--ssh-host", default="100.120.55.25")
-    parser.add_argument("--ssh-port", type=int, default=2222)
-    parser.add_argument(
-        "--ssh-key", default="environments/pwncollege_env/keys/rl_test_key"
-    )
-    parser.add_argument("--output", "-o", help="Write JSON report")
-    parser.add_argument("--skip-concurrent", action="store_true")
-    parser.add_argument("--verbose", "-v", action="store_true")
-    parser.add_argument("--yes", "-y", action="store_true", help="Skip confirmation")
-    args = parser.parse_args()
-
-    key = Path(args.ssh_key)
-    if not key.exists():
-        key = _repo_root / args.ssh_key
-    if not key.exists():
-        print(f"SSH key not found: {args.ssh_key}")
-        sys.exit(1)
-    args.ssh_key = str(key)
-
-    if not args.yes:
-        print(f"Will test against {args.base_url}")
-        if input("Continue? [y/N] ").lower() != "y":
-            sys.exit(0)
-
-    report = asyncio.run(run_tests(args))
-
-    if args.output:
-        with open(args.output, "w") as f:
-            json.dump(report, f, indent=2)
-        print(f"\nJSON report: {args.output}")
-
-    sys.exit(0 if report["summary"]["failed"] == 0 else 1)
-
-
-if __name__ == "__main__":
-    main()

environments/pwncollege_env/submit_flag_tool.py

@@ -1,102 +0,0 @@
-"""submit_flag tool for pwn.college RL environments.
-
-Registers a `submit_flag` tool in the hermes-agent tool registry under the
-"pwncollege" toolset. The handler checks flags against the dojo RL API using
-per-task context (SDK client + slot) stored in a module-level dict.
-
-Usage in an environment:
-    from environments.pwncollege_env.submit_flag_tool import (
-        register_flag_context, clear_flag_context,
-    )
-
-    # Before agent loop
-    register_flag_context(task_id, sync_client, slot)
-
-    # After agent loop
-    clear_flag_context(task_id)
-"""
-
-import json
-import logging
-from typing import Any, Dict
-
-logger = logging.getLogger(__name__)
-
-# Per-task context: task_id → {"client": DojoRLSyncClient, "slot": int}
-_task_flag_context: Dict[str, Dict[str, Any]] = {}
-
-
-def register_flag_context(task_id: str, sync_client: Any, slot: int) -> None:
-    """Register dojo client + slot for a rollout so submit_flag can verify flags."""
-    _task_flag_context[task_id] = {"client": sync_client, "slot": slot}
-
-
-def clear_flag_context(task_id: str) -> None:
-    """Remove flag context after rollout completes."""
-    _task_flag_context.pop(task_id, None)
-
-
-def _submit_flag_handler(args: dict, **kw) -> str:
-    """Handle submit_flag tool calls by checking the flag against the dojo API."""
-    task_id = kw.get("task_id", "default")
-    flag = args.get("flag", "")
-
-    if not flag:
-        return json.dumps({"success": False, "message": "No flag provided."})
-
-    ctx = _task_flag_context.get(task_id)
-    if not ctx:
-        return json.dumps({
-            "success": False,
-            "message": "No active challenge instance for this task.",
-        })
-
-    try:
-        correct = ctx["client"].check_flag(ctx["slot"], flag)
-    except Exception as e:
-        logger.error("Flag check failed for task %s: %s", task_id, e, exc_info=True)
-        return json.dumps({"success": False, "message": f"Flag check error: {type(e).__name__}"})
-
-    if correct:
-        return json.dumps({"success": True, "message": "Flag accepted! Challenge solved."})
-    return json.dumps({"success": False, "message": "Incorrect flag."})
-
-
-# ---------------------------------------------------------------------------
-# Register in hermes-agent tool registry
-# ---------------------------------------------------------------------------
-
-SUBMIT_FLAG_SCHEMA = {
-    "name": "submit_flag",
-    "description": (
-        "Submit a flag for verification. Use this when you find a flag "
-        "(format: pwn.college{...}) to check if it is correct."
-    ),
-    "parameters": {
-        "type": "object",
-        "properties": {
-            "flag": {
-                "type": "string",
-                "description": "The flag string (format: pwn.college{...}).",
-            }
-        },
-        "required": ["flag"],
-    },
-}
-
-from tools.registry import registry
-from toolsets import create_custom_toolset
-
-registry.register(
-    name="submit_flag",
-    toolset="pwncollege",
-    schema=SUBMIT_FLAG_SCHEMA,
-    handler=_submit_flag_handler,
-    emoji="🚩",
-)
-
-create_custom_toolset(
-    name="pwncollege",
-    description="PwnCollege CTF tools",
-    tools=["submit_flag"],
-)

gateway/builtin_hooks/__init__.py

@@ -0,0 +1 @@
+"""Built-in gateway hooks that are always registered."""

gateway/builtin_hooks/boot_md.py

@@ -0,0 +1,86 @@
+"""Built-in boot-md hook — run ~/.hermes/BOOT.md on gateway startup.
+
+This hook is always registered. It silently skips if no BOOT.md exists.
+To activate, create ``~/.hermes/BOOT.md`` with instructions for the
+agent to execute on every gateway restart.
+
+Example BOOT.md::
+
+    # Startup Checklist
+
+    1. Check if any cron jobs failed overnight
+    2. Send a status update to Discord #general
+    3. If there are errors in /opt/app/deploy.log, summarize them
+
+The agent runs in a background thread so it doesn't block gateway
+startup. If nothing needs attention, it replies with [SILENT] to
+suppress delivery.
+"""
+
+import logging
+import os
+import threading
+from pathlib import Path
+
+logger = logging.getLogger("hooks.boot-md")
+
+HERMES_HOME = Path(os.environ.get("HERMES_HOME", Path.home() / ".hermes"))
+BOOT_FILE = HERMES_HOME / "BOOT.md"
+
+
+def _build_boot_prompt(content: str) -> str:
+    """Wrap BOOT.md content in a system-level instruction."""
+    return (
+        "You are running a startup boot checklist. Follow the BOOT.md "
+        "instructions below exactly.\n\n"
+        "---\n"
+        f"{content}\n"
+        "---\n\n"
+        "Execute each instruction. If you need to send a message to a "
+        "platform, use the send_message tool.\n"
+        "If nothing needs attention and there is nothing to report, "
+        "reply with ONLY: [SILENT]"
+    )
+
+
+def _run_boot_agent(content: str) -> None:
+    """Spawn a one-shot agent session to execute the boot instructions."""
+    try:
+        from run_agent import AIAgent
+
+        prompt = _build_boot_prompt(content)
+        agent = AIAgent(
+            quiet_mode=True,
+            skip_context_files=True,
+            skip_memory=True,
+            max_iterations=20,
+        )
+        result = agent.run_conversation(prompt)
+        response = result.get("final_response", "")
+        if response and "[SILENT]" not in response:
+            logger.info("boot-md completed: %s", response[:200])
+        else:
+            logger.info("boot-md completed (nothing to report)")
+    except Exception as e:
+        logger.error("boot-md agent failed: %s", e)
+
+
+async def handle(event_type: str, context: dict) -> None:
+    """Gateway startup handler — run BOOT.md if it exists."""
+    if not BOOT_FILE.exists():
+        return
+
+    content = BOOT_FILE.read_text(encoding="utf-8").strip()
+    if not content:
+        return
+
+    logger.info("Running BOOT.md (%d chars)", len(content))
+
+    # Run in a background thread so we don't block gateway startup.
+    thread = threading.Thread(
+        target=_run_boot_agent,
+        args=(content,),
+        name="boot-md",
+        daemon=True,
+    )
+    thread.start()

gateway/config.py

@@ -647,14 +647,13 @@ def _apply_env_overrides(config: GatewayConfig) -> None:
             config.platforms[Platform.SLACK] = PlatformConfig()
         config.platforms[Platform.SLACK].enabled = True
         config.platforms[Platform.SLACK].token = slack_token
-        # Home channel
-        slack_home = os.getenv("SLACK_HOME_CHANNEL")
-        if slack_home:
-            config.platforms[Platform.SLACK].home_channel = HomeChannel(
-                platform=Platform.SLACK,
-                chat_id=slack_home,
-                name=os.getenv("SLACK_HOME_CHANNEL_NAME", ""),
-            )
+    slack_home = os.getenv("SLACK_HOME_CHANNEL")
+    if slack_home and Platform.SLACK in config.platforms:
+        config.platforms[Platform.SLACK].home_channel = HomeChannel(
+            platform=Platform.SLACK,
+            chat_id=slack_home,
+            name=os.getenv("SLACK_HOME_CHANNEL_NAME", ""),
+        )
     
     # Signal
     signal_url = os.getenv("SIGNAL_HTTP_URL")
@@ -668,13 +667,13 @@ def _apply_env_overrides(config: GatewayConfig) -> None:
             "account": signal_account,
             "ignore_stories": os.getenv("SIGNAL_IGNORE_STORIES", "true").lower() in ("true", "1", "yes"),
         })
-        signal_home = os.getenv("SIGNAL_HOME_CHANNEL")
-        if signal_home:
-            config.platforms[Platform.SIGNAL].home_channel = HomeChannel(
-                platform=Platform.SIGNAL,
-                chat_id=signal_home,
-                name=os.getenv("SIGNAL_HOME_CHANNEL_NAME", "Home"),
-            )
+    signal_home = os.getenv("SIGNAL_HOME_CHANNEL")
+    if signal_home and Platform.SIGNAL in config.platforms:
+        config.platforms[Platform.SIGNAL].home_channel = HomeChannel(
+            platform=Platform.SIGNAL,
+            chat_id=signal_home,
+            name=os.getenv("SIGNAL_HOME_CHANNEL_NAME", "Home"),
+        )
 
     # Mattermost
     mattermost_token = os.getenv("MATTERMOST_TOKEN")
@@ -687,13 +686,13 @@ def _apply_env_overrides(config: GatewayConfig) -> None:
         config.platforms[Platform.MATTERMOST].enabled = True
         config.platforms[Platform.MATTERMOST].token = mattermost_token
         config.platforms[Platform.MATTERMOST].extra["url"] = mattermost_url
-        mattermost_home = os.getenv("MATTERMOST_HOME_CHANNEL")
-        if mattermost_home:
-            config.platforms[Platform.MATTERMOST].home_channel = HomeChannel(
-                platform=Platform.MATTERMOST,
-                chat_id=mattermost_home,
-                name=os.getenv("MATTERMOST_HOME_CHANNEL_NAME", "Home"),
-            )
+    mattermost_home = os.getenv("MATTERMOST_HOME_CHANNEL")
+    if mattermost_home and Platform.MATTERMOST in config.platforms:
+        config.platforms[Platform.MATTERMOST].home_channel = HomeChannel(
+            platform=Platform.MATTERMOST,
+            chat_id=mattermost_home,
+            name=os.getenv("MATTERMOST_HOME_CHANNEL_NAME", "Home"),
+        )
 
     # Matrix
     matrix_token = os.getenv("MATRIX_ACCESS_TOKEN")
@@ -715,13 +714,13 @@ def _apply_env_overrides(config: GatewayConfig) -> None:
             config.platforms[Platform.MATRIX].extra["password"] = matrix_password
         matrix_e2ee = os.getenv("MATRIX_ENCRYPTION", "").lower() in ("true", "1", "yes")
         config.platforms[Platform.MATRIX].extra["encryption"] = matrix_e2ee
-        matrix_home = os.getenv("MATRIX_HOME_ROOM")
-        if matrix_home:
-            config.platforms[Platform.MATRIX].home_channel = HomeChannel(
-                platform=Platform.MATRIX,
-                chat_id=matrix_home,
-                name=os.getenv("MATRIX_HOME_ROOM_NAME", "Home"),
-            )
+    matrix_home = os.getenv("MATRIX_HOME_ROOM")
+    if matrix_home and Platform.MATRIX in config.platforms:
+        config.platforms[Platform.MATRIX].home_channel = HomeChannel(
+            platform=Platform.MATRIX,
+            chat_id=matrix_home,
+            name=os.getenv("MATRIX_HOME_ROOM_NAME", "Home"),
+        )
 
     # Home Assistant
     hass_token = os.getenv("HASS_TOKEN")
@@ -748,13 +747,13 @@ def _apply_env_overrides(config: GatewayConfig) -> None:
             "imap_host": email_imap,
             "smtp_host": email_smtp,
         })
-        email_home = os.getenv("EMAIL_HOME_ADDRESS")
-        if email_home:
-            config.platforms[Platform.EMAIL].home_channel = HomeChannel(
-                platform=Platform.EMAIL,
-                chat_id=email_home,
-                name=os.getenv("EMAIL_HOME_ADDRESS_NAME", "Home"),
-            )
+    email_home = os.getenv("EMAIL_HOME_ADDRESS")
+    if email_home and Platform.EMAIL in config.platforms:
+        config.platforms[Platform.EMAIL].home_channel = HomeChannel(
+            platform=Platform.EMAIL,
+            chat_id=email_home,
+            name=os.getenv("EMAIL_HOME_ADDRESS_NAME", "Home"),
+        )
 
     # SMS (Twilio)
     twilio_sid = os.getenv("TWILIO_ACCOUNT_SID")
@@ -763,13 +762,13 @@ def _apply_env_overrides(config: GatewayConfig) -> None:
             config.platforms[Platform.SMS] = PlatformConfig()
         config.platforms[Platform.SMS].enabled = True
         config.platforms[Platform.SMS].api_key = os.getenv("TWILIO_AUTH_TOKEN", "")
-        sms_home = os.getenv("SMS_HOME_CHANNEL")
-        if sms_home:
-            config.platforms[Platform.SMS].home_channel = HomeChannel(
-                platform=Platform.SMS,
-                chat_id=sms_home,
-                name=os.getenv("SMS_HOME_CHANNEL_NAME", "Home"),
-            )
+    sms_home = os.getenv("SMS_HOME_CHANNEL")
+    if sms_home and Platform.SMS in config.platforms:
+        config.platforms[Platform.SMS].home_channel = HomeChannel(
+            platform=Platform.SMS,
+            chat_id=sms_home,
+            name=os.getenv("SMS_HOME_CHANNEL_NAME", "Home"),
+        )
 
     # API Server
     api_server_enabled = os.getenv("API_SERVER_ENABLED", "").lower() in ("true", "1", "yes")

gateway/hooks.py

@@ -51,14 +51,33 @@ class HookRegistry:
         """Return metadata about all loaded hooks."""
         return list(self._loaded_hooks)
 
+    def _register_builtin_hooks(self) -> None:
+        """Register built-in hooks that are always active."""
+        try:
+            from gateway.builtin_hooks.boot_md import handle as boot_md_handle
+
+            self._handlers.setdefault("gateway:startup", []).append(boot_md_handle)
+            self._loaded_hooks.append({
+                "name": "boot-md",
+                "description": "Run ~/.hermes/BOOT.md on gateway startup",
+                "events": ["gateway:startup"],
+                "path": "(builtin)",
+            })
+        except Exception as e:
+            print(f"[hooks] Could not load built-in boot-md hook: {e}", flush=True)
+
     def discover_and_load(self) -> None:
         """
         Scan the hooks directory for hook directories and load their handlers.
 
+        Also registers built-in hooks that are always active.
+
         Each hook directory must contain:
           - HOOK.yaml with at least 'name' and 'events' keys
           - handler.py with a top-level 'handle' function (sync or async)
         """
+        self._register_builtin_hooks()
+
         if not HOOKS_DIR.exists():
             return
 

gateway/pairing.py

@@ -25,7 +25,7 @@ import time
 from pathlib import Path
 from typing import Optional
 
-from hermes_cli.config import get_hermes_home
+from hermes_constants import get_hermes_dir
 
 
 # Unambiguous alphabet -- excludes 0/O, 1/I to prevent confusion
@@ -41,7 +41,7 @@ LOCKOUT_SECONDS = 3600              # Lockout duration after too many failures
 MAX_PENDING_PER_PLATFORM = 3        # Max pending codes per platform
 MAX_FAILED_ATTEMPTS = 5             # Failed approvals before lockout
 
-PAIRING_DIR = get_hermes_home() / "pairing"
+PAIRING_DIR = get_hermes_dir("platforms/pairing", "pairing")
 
 
 def _secure_write(path: Path, data: str) -> None:

gateway/platforms/api_server.py

@@ -166,7 +166,7 @@ class ResponseStore:
 
 _CORS_HEADERS = {
     "Access-Control-Allow-Methods": "GET, POST, DELETE, OPTIONS",
-    "Access-Control-Allow-Headers": "Authorization, Content-Type",
+    "Access-Control-Allow-Headers": "Authorization, Content-Type, Idempotency-Key",
 }
 
 
@@ -223,6 +223,23 @@ if AIOHTTP_AVAILABLE:
 else:
     body_limit_middleware = None  # type: ignore[assignment]
 
+_SECURITY_HEADERS = {
+    "X-Content-Type-Options": "nosniff",
+    "Referrer-Policy": "no-referrer",
+}
+
+
+if AIOHTTP_AVAILABLE:
+    @web.middleware
+    async def security_headers_middleware(request, handler):
+        """Add security headers to all responses (including errors)."""
+        response = await handler(request)
+        for k, v in _SECURITY_HEADERS.items():
+            response.headers.setdefault(k, v)
+        return response
+else:
+    security_headers_middleware = None  # type: ignore[assignment]
+
 
 class _IdempotencyCache:
     """In-memory idempotency cache with TTL and basic LRU semantics."""
@@ -307,6 +324,7 @@ class APIServerAdapter(BasePlatformAdapter):
         if "*" in self._cors_origins:
             headers = dict(_CORS_HEADERS)
             headers["Access-Control-Allow-Origin"] = "*"
+            headers["Access-Control-Max-Age"] = "600"
             return headers
 
         if origin not in self._cors_origins:
@@ -315,6 +333,7 @@ class APIServerAdapter(BasePlatformAdapter):
         headers = dict(_CORS_HEADERS)
         headers["Access-Control-Allow-Origin"] = origin
         headers["Vary"] = "Origin"
+        headers["Access-Control-Max-Age"] = "600"
         return headers
 
     def _origin_allowed(self, origin: str) -> bool:
@@ -495,17 +514,21 @@ class APIServerAdapter(BasePlatformAdapter):
                 if delta is not None:
                     _stream_q.put(delta)
 
-            # Start agent in background
+            # Start agent in background.  agent_ref is a mutable container
+            # so the SSE writer can interrupt the agent on client disconnect.
+            agent_ref = [None]
             agent_task = asyncio.ensure_future(self._run_agent(
                 user_message=user_message,
                 conversation_history=history,
                 ephemeral_system_prompt=system_prompt,
                 session_id=session_id,
                 stream_delta_callback=_on_delta,
+                agent_ref=agent_ref,
             ))
 
             return await self._write_sse_chat_completion(
-                request, completion_id, model_name, created, _stream_q, agent_task
+                request, completion_id, model_name, created, _stream_q,
+                agent_task, agent_ref,
             )
 
         # Non-streaming: run the agent (with optional Idempotency-Key)
@@ -568,80 +591,107 @@ class APIServerAdapter(BasePlatformAdapter):
 
     async def _write_sse_chat_completion(
         self, request: "web.Request", completion_id: str, model: str,
-        created: int, stream_q, agent_task,
+        created: int, stream_q, agent_task, agent_ref=None,
     ) -> "web.StreamResponse":
-        """Write real streaming SSE from agent's stream_delta_callback queue."""
+        """Write real streaming SSE from agent's stream_delta_callback queue.
+
+        If the client disconnects mid-stream (network drop, browser tab close),
+        the agent is interrupted via ``agent.interrupt()`` so it stops making
+        LLM API calls, and the asyncio task wrapper is cancelled.
+        """
         import queue as _q
 
-        response = web.StreamResponse(
-            status=200,
-            headers={"Content-Type": "text/event-stream", "Cache-Control": "no-cache"},
-        )
+        sse_headers = {"Content-Type": "text/event-stream", "Cache-Control": "no-cache"}
+        # CORS middleware can't inject headers into StreamResponse after
+        # prepare() flushes them, so resolve CORS headers up front.
+        origin = request.headers.get("Origin", "")
+        cors = self._cors_headers_for_origin(origin) if origin else None
+        if cors:
+            sse_headers.update(cors)
+        response = web.StreamResponse(status=200, headers=sse_headers)
         await response.prepare(request)
 
-        # Role chunk
-        role_chunk = {
-            "id": completion_id, "object": "chat.completion.chunk",
-            "created": created, "model": model,
-            "choices": [{"index": 0, "delta": {"role": "assistant"}, "finish_reason": None}],
-        }
-        await response.write(f"data: {json.dumps(role_chunk)}\n\n".encode())
-
-        # Stream content chunks as they arrive from the agent
-        loop = asyncio.get_event_loop()
-        while True:
-            try:
-                delta = await loop.run_in_executor(None, lambda: stream_q.get(timeout=0.5))
-            except _q.Empty:
-                if agent_task.done():
-                    # Drain any remaining items
-                    while True:
-                        try:
-                            delta = stream_q.get_nowait()
-                            if delta is None:
-                                break
-                            content_chunk = {
-                                "id": completion_id, "object": "chat.completion.chunk",
-                                "created": created, "model": model,
-                                "choices": [{"index": 0, "delta": {"content": delta}, "finish_reason": None}],
-                            }
-                            await response.write(f"data: {json.dumps(content_chunk)}\n\n".encode())
-                        except _q.Empty:
-                            break
-                    break
-                continue
-
-            if delta is None:  # End of stream sentinel
-                break
-
-            content_chunk = {
+        try:
+            # Role chunk
+            role_chunk = {
                 "id": completion_id, "object": "chat.completion.chunk",
                 "created": created, "model": model,
-                "choices": [{"index": 0, "delta": {"content": delta}, "finish_reason": None}],
+                "choices": [{"index": 0, "delta": {"role": "assistant"}, "finish_reason": None}],
             }
-            await response.write(f"data: {json.dumps(content_chunk)}\n\n".encode())
+            await response.write(f"data: {json.dumps(role_chunk)}\n\n".encode())
 
-        # Get usage from completed agent
-        usage = {"input_tokens": 0, "output_tokens": 0, "total_tokens": 0}
-        try:
-            result, agent_usage = await agent_task
-            usage = agent_usage or usage
-        except Exception:
-            pass
+            # Stream content chunks as they arrive from the agent
+            loop = asyncio.get_event_loop()
+            while True:
+                try:
+                    delta = await loop.run_in_executor(None, lambda: stream_q.get(timeout=0.5))
+                except _q.Empty:
+                    if agent_task.done():
+                        # Drain any remaining items
+                        while True:
+                            try:
+                                delta = stream_q.get_nowait()
+                                if delta is None:
+                                    break
+                                content_chunk = {
+                                    "id": completion_id, "object": "chat.completion.chunk",
+                                    "created": created, "model": model,
+                                    "choices": [{"index": 0, "delta": {"content": delta}, "finish_reason": None}],
+                                }
+                                await response.write(f"data: {json.dumps(content_chunk)}\n\n".encode())
+                            except _q.Empty:
+                                break
+                        break
+                    continue
 
-        # Finish chunk
-        finish_chunk = {
-            "id": completion_id, "object": "chat.completion.chunk",
-            "created": created, "model": model,
-            "choices": [{"index": 0, "delta": {}, "finish_reason": "stop"}],
-            "usage": {
-                "prompt_tokens": usage.get("input_tokens", 0),
-                "completion_tokens": usage.get("output_tokens", 0),
-                "total_tokens": usage.get("total_tokens", 0),
-            },
-        }
-        await response.write(f"data: {json.dumps(finish_chunk)}\n\n".encode())
-        await response.write(b"data: [DONE]\n\n")
+                if delta is None:  # End of stream sentinel
+                    break
+
+                content_chunk = {
+                    "id": completion_id, "object": "chat.completion.chunk",
+                    "created": created, "model": model,
+                    "choices": [{"index": 0, "delta": {"content": delta}, "finish_reason": None}],
+                }
+                await response.write(f"data: {json.dumps(content_chunk)}\n\n".encode())
+
+            # Get usage from completed agent
+            usage = {"input_tokens": 0, "output_tokens": 0, "total_tokens": 0}
+            try:
+                result, agent_usage = await agent_task
+                usage = agent_usage or usage
+            except Exception:
+                pass
+
+            # Finish chunk
+            finish_chunk = {
+                "id": completion_id, "object": "chat.completion.chunk",
+                "created": created, "model": model,
+                "choices": [{"index": 0, "delta": {}, "finish_reason": "stop"}],
+                "usage": {
+                    "prompt_tokens": usage.get("input_tokens", 0),
+                    "completion_tokens": usage.get("output_tokens", 0),
+                    "total_tokens": usage.get("total_tokens", 0),
+                },
+            }
+            await response.write(f"data: {json.dumps(finish_chunk)}\n\n".encode())
+            await response.write(b"data: [DONE]\n\n")
+        except (ConnectionResetError, ConnectionAbortedError, BrokenPipeError, OSError):
+            # Client disconnected mid-stream.  Interrupt the agent so it
+            # stops making LLM API calls at the next loop iteration, then
+            # cancel the asyncio task wrapper.
+            agent = agent_ref[0] if agent_ref else None
+            if agent is not None:
+                try:
+                    agent.interrupt("SSE client disconnected")
+                except Exception:
+                    pass
+            if not agent_task.done():
+                agent_task.cancel()
+                try:
+                    await agent_task
+                except (asyncio.CancelledError, Exception):
+                    pass
+            logger.info("SSE client disconnected; interrupted agent task %s", completion_id)
 
         return response
 
@@ -1144,12 +1194,18 @@ class APIServerAdapter(BasePlatformAdapter):
         ephemeral_system_prompt: Optional[str] = None,
         session_id: Optional[str] = None,
         stream_delta_callback=None,
+        agent_ref: Optional[list] = None,
     ) -> tuple:
         """
         Create an agent and run a conversation in a thread executor.
 
         Returns ``(result_dict, usage_dict)`` where *usage_dict* contains
         ``input_tokens``, ``output_tokens`` and ``total_tokens``.
+
+        If *agent_ref* is a one-element list, the AIAgent instance is stored
+        at ``agent_ref[0]`` before ``run_conversation`` begins.  This allows
+        callers (e.g. the SSE writer) to call ``agent.interrupt()`` from
+        another thread to stop in-progress LLM calls.
         """
         loop = asyncio.get_event_loop()
 
@@ -1159,6 +1215,8 @@ class APIServerAdapter(BasePlatformAdapter):
                 session_id=session_id,
                 stream_delta_callback=stream_delta_callback,
             )
+            if agent_ref is not None:
+                agent_ref[0] = agent
             result = agent.run_conversation(
                 user_message=user_message,
                 conversation_history=conversation_history,
@@ -1183,10 +1241,11 @@ class APIServerAdapter(BasePlatformAdapter):
             return False
 
         try:
-            mws = [mw for mw in (cors_middleware, body_limit_middleware) if mw is not None]
+            mws = [mw for mw in (cors_middleware, body_limit_middleware, security_headers_middleware) if mw is not None]
             self._app = web.Application(middlewares=mws)
             self._app["api_server_adapter"] = self
             self._app.router.add_get("/health", self._handle_health)
+            self._app.router.add_get("/v1/health", self._handle_health)
             self._app.router.add_get("/v1/models", self._handle_models)
             self._app.router.add_post("/v1/chat/completions", self._handle_chat_completions)
             self._app.router.add_post("/v1/responses", self._handle_responses)
@@ -1202,6 +1261,17 @@ class APIServerAdapter(BasePlatformAdapter):
             self._app.router.add_post("/api/jobs/{job_id}/resume", self._handle_resume_job)
             self._app.router.add_post("/api/jobs/{job_id}/run", self._handle_run_job)
 
+            # Port conflict detection — fail fast if port is already in use
+            import socket as _socket
+            try:
+                with _socket.socket(_socket.AF_INET, _socket.SOCK_STREAM) as _s:
+                    _s.settimeout(1)
+                    _s.connect(('127.0.0.1', self._port))
+                logger.error('[%s] Port %d already in use. Set a different port in config.yaml: platforms.api_server.port', self.name, self._port)
+                return False
+            except (ConnectionRefusedError, OSError):
+                pass  # port is free
+
             self._runner = web.AppRunner(self._app)
             await self._runner.setup()
             self._site = web.TCPSite(self._runner, self._host, self._port)

gateway/platforms/base.py

@@ -27,6 +27,7 @@ sys.path.insert(0, str(_Path(__file__).resolve().parents[2]))
 from gateway.config import Platform, PlatformConfig
 from gateway.session import SessionSource, build_session_key
 from hermes_cli.config import get_hermes_home
+from hermes_constants import get_hermes_dir
 
 
 GATEWAY_SECRET_CAPTURE_UNSUPPORTED_MESSAGE = (
@@ -44,8 +45,8 @@ GATEWAY_SECRET_CAPTURE_UNSUPPORTED_MESSAGE = (
 # (e.g. Telegram file URLs expire after ~1 hour).
 # ---------------------------------------------------------------------------
 
-# Default location: {HERMES_HOME}/image_cache/
-IMAGE_CACHE_DIR = get_hermes_home() / "image_cache"
+# Default location: {HERMES_HOME}/cache/images/ (legacy: image_cache/)
+IMAGE_CACHE_DIR = get_hermes_dir("cache/images", "image_cache")
 
 
 def get_image_cache_dir() -> Path:
@@ -147,7 +148,7 @@ def cleanup_image_cache(max_age_hours: int = 24) -> int:
 # here so the STT tool (OpenAI Whisper) can transcribe them from local files.
 # ---------------------------------------------------------------------------
 
-AUDIO_CACHE_DIR = get_hermes_home() / "audio_cache"
+AUDIO_CACHE_DIR = get_hermes_dir("cache/audio", "audio_cache")
 
 
 def get_audio_cache_dir() -> Path:
@@ -174,29 +175,51 @@ def cache_audio_from_bytes(data: bytes, ext: str = ".ogg") -> str:
     return str(filepath)
 
 
-async def cache_audio_from_url(url: str, ext: str = ".ogg") -> str:
+async def cache_audio_from_url(url: str, ext: str = ".ogg", retries: int = 2) -> str:
     """
     Download an audio file from a URL and save it to the local cache.
 
+    Retries on transient failures (timeouts, 429, 5xx) with exponential
+    backoff so a single slow CDN response doesn't lose the media.
+
     Args:
         url: The HTTP/HTTPS URL to download from.
         ext: File extension including the dot (e.g. ".ogg", ".mp3").
+        retries: Number of retry attempts on transient failures.
 
     Returns:
         Absolute path to the cached audio file as a string.
     """
+    import asyncio
     import httpx
+    import logging as _logging
+    _log = _logging.getLogger(__name__)
 
+    last_exc = None
     async with httpx.AsyncClient(timeout=30.0, follow_redirects=True) as client:
-        response = await client.get(
-            url,
-            headers={
-                "User-Agent": "Mozilla/5.0 (compatible; HermesAgent/1.0)",
-                "Accept": "audio/*,*/*;q=0.8",
-            },
-        )
-        response.raise_for_status()
-        return cache_audio_from_bytes(response.content, ext)
+        for attempt in range(retries + 1):
+            try:
+                response = await client.get(
+                    url,
+                    headers={
+                        "User-Agent": "Mozilla/5.0 (compatible; HermesAgent/1.0)",
+                        "Accept": "audio/*,*/*;q=0.8",
+                    },
+                )
+                response.raise_for_status()
+                return cache_audio_from_bytes(response.content, ext)
+            except (httpx.TimeoutException, httpx.HTTPStatusError) as exc:
+                last_exc = exc
+                if isinstance(exc, httpx.HTTPStatusError) and exc.response.status_code < 429:
+                    raise
+                if attempt < retries:
+                    wait = 1.5 * (attempt + 1)
+                    _log.debug("Audio cache retry %d/%d for %s (%.1fs): %s",
+                               attempt + 1, retries, url[:80], wait, exc)
+                    await asyncio.sleep(wait)
+                    continue
+                raise
+    raise last_exc
 
 
 # ---------------------------------------------------------------------------
@@ -206,7 +229,7 @@ async def cache_audio_from_url(url: str, ext: str = ".ogg") -> str:
 # here so the agent can reference them by local file path.
 # ---------------------------------------------------------------------------
 
-DOCUMENT_CACHE_DIR = get_hermes_home() / "document_cache"
+DOCUMENT_CACHE_DIR = get_hermes_dir("cache/documents", "document_cache")
 
 SUPPORTED_DOCUMENT_TYPES = {
     ".pdf": "application/pdf",
@@ -333,7 +356,10 @@ class MessageEvent:
             return None
         # Split on space and get first word, strip the /
         parts = self.text.split(maxsplit=1)
-        return parts[0][1:].lower() if parts else None
+        raw = parts[0][1:].lower() if parts else None
+        if raw and "@" in raw:
+            raw = raw.split("@", 1)[0]
+        return raw
     
     def get_command_args(self) -> str:
         """Get the arguments after a command."""
@@ -979,7 +1005,7 @@ class BasePlatformAdapter(ABC):
             # simultaneous messages. Queue them without interrupting the active run,
             # then process them immediately after the current task finishes.
             if event.message_type == MessageType.PHOTO:
-                print(f"[{self.name}] 🖼️ Queuing photo follow-up for session {session_key} without interrupt")
+                logger.debug("[%s] Queuing photo follow-up for session %s without interrupt", self.name, session_key)
                 existing = self._pending_messages.get(session_key)
                 if existing and existing.message_type == MessageType.PHOTO:
                     existing.media_urls.extend(event.media_urls)
@@ -994,7 +1020,7 @@ class BasePlatformAdapter(ABC):
                 return  # Don't interrupt now - will run after current task completes
 
             # Default behavior for non-photo follow-ups: interrupt the running agent
-            print(f"[{self.name}] ⚡ New message while session {session_key} is active - triggering interrupt")
+            logger.debug("[%s] New message while session %s is active — triggering interrupt", self.name, session_key)
             self._pending_messages[session_key] = event
             # Signal the interrupt (the processing task checks this)
             self._active_sessions[session_key].set()
@@ -1180,9 +1206,9 @@ class BasePlatformAdapter(ABC):
                             )
 
                         if not media_result.success:
-                            print(f"[{self.name}] Failed to send media ({ext}): {media_result.error}")
+                            logger.warning("[%s] Failed to send media (%s): %s", self.name, ext, media_result.error)
                     except Exception as media_err:
-                        print(f"[{self.name}] Error sending media: {media_err}")
+                        logger.warning("[%s] Error sending media: %s", self.name, media_err)
 
                 # Send auto-detected local files as native attachments
                 for file_path in local_files:
@@ -1214,7 +1240,7 @@ class BasePlatformAdapter(ABC):
             # Check if there's a pending message that was queued during our processing
             if session_key in self._pending_messages:
                 pending_event = self._pending_messages.pop(session_key)
-                print(f"[{self.name}] 📨 Processing queued message from interrupt")
+                logger.debug("[%s] Processing queued message from interrupt", self.name)
                 # Clean up current session before processing pending
                 if session_key in self._active_sessions:
                     del self._active_sessions[session_key]
@@ -1228,9 +1254,7 @@ class BasePlatformAdapter(ABC):
                 return  # Already cleaned up
                 
         except Exception as e:
-            print(f"[{self.name}] Error handling message: {e}")
-            import traceback
-            traceback.print_exc()
+            logger.error("[%s] Error handling message: %s", self.name, e, exc_info=True)
             # Send the error to the user so they aren't left with radio silence
             try:
                 error_type = type(e).__name__

gateway/platforms/discord.py

@@ -486,6 +486,17 @@ class DiscordAdapter(BasePlatformAdapter):
             return False
         
         try:
+            # Acquire scoped lock to prevent duplicate bot token usage
+            from gateway.status import acquire_scoped_lock
+            self._token_lock_identity = self.config.token
+            acquired, existing = acquire_scoped_lock('discord-bot-token', self._token_lock_identity, metadata={'platform': 'discord'})
+            if not acquired:
+                owner_pid = existing.get('pid') if isinstance(existing, dict) else None
+                message = f'Discord bot token already in use' + (f' (PID {owner_pid})' if owner_pid else '') + '. Stop the other gateway first.'
+                logger.error('[%s] %s', self.name, message)
+                self._set_fatal_error('discord_token_lock', message, retryable=False)
+                return False
+
             # Set up intents -- members intent needed for username-to-ID resolution
             intents = Intents.default()
             intents.message_content = True
@@ -550,6 +561,22 @@ class DiscordAdapter(BasePlatformAdapter):
                             return
                     # "all" falls through to handle_message
                 
+                # If the message @mentions other users but NOT the bot, the
+                # sender is talking to someone else — stay silent.  Only
+                # applies in server channels; in DMs the user is always
+                # talking to the bot (mentions are just references).
+                # Controlled by DISCORD_IGNORE_NO_MENTION (default: true).
+                _ignore_no_mention = os.getenv(
+                    "DISCORD_IGNORE_NO_MENTION", "true"
+                ).lower() in ("true", "1", "yes")
+                if _ignore_no_mention and message.mentions and not isinstance(message.channel, discord.DMChannel):
+                    _bot_mentioned = (
+                        self._client.user is not None
+                        and self._client.user in message.mentions
+                    )
+                    if not _bot_mentioned:
+                        return  # Talking to someone else, don't interrupt
+
                 await self._handle_message(message)
 
             @self._client.event
@@ -622,6 +649,16 @@ class DiscordAdapter(BasePlatformAdapter):
         self._running = False
         self._client = None
         self._ready_event.clear()
+
+        # Release the token lock
+        try:
+            from gateway.status import release_scoped_lock
+            if getattr(self, '_token_lock_identity', None):
+                release_scoped_lock('discord-bot-token', self._token_lock_identity)
+                self._token_lock_identity = None
+        except Exception:
+            pass
+
         logger.info("[%s] Disconnected", self.name)
     
     async def send(
@@ -1413,15 +1450,23 @@ class DiscordAdapter(BasePlatformAdapter):
         command_text: str,
         followup_msg: str | None = None,
     ) -> None:
-        """Common handler for simple slash commands that dispatch a command string."""
+        """Common handler for simple slash commands that dispatch a command string.
+
+        Defers the interaction (shows "thinking..."), dispatches the command,
+        then cleans up the deferred response.  If *followup_msg* is provided
+        the "thinking..." indicator is replaced with that text; otherwise it
+        is deleted so the channel isn't cluttered.
+        """
         await interaction.response.defer(ephemeral=True)
         event = self._build_slash_event(interaction, command_text)
         await self.handle_message(event)
-        if followup_msg:
-            try:
-                await interaction.followup.send(followup_msg, ephemeral=True)
-            except Exception as e:
-                logger.debug("Discord followup failed: %s", e)
+        try:
+            if followup_msg:
+                await interaction.edit_original_response(content=followup_msg)
+            else:
+                await interaction.delete_original_response()
+        except Exception as e:
+            logger.debug("Discord interaction cleanup failed: %s", e)
 
     def _register_slash_commands(self) -> None:
         """Register Discord slash commands on the command tree."""
@@ -1446,9 +1491,7 @@ class DiscordAdapter(BasePlatformAdapter):
         @tree.command(name="reasoning", description="Show or change reasoning effort")
         @discord.app_commands.describe(effort="Reasoning effort: xhigh, high, medium, low, minimal, or none.")
         async def slash_reasoning(interaction: discord.Interaction, effort: str = ""):
-            await interaction.response.defer(ephemeral=True)
-            event = self._build_slash_event(interaction, f"/reasoning {effort}".strip())
-            await self.handle_message(event)
+            await self._run_simple_slash(interaction, f"/reasoning {effort}".strip())
 
         @tree.command(name="personality", description="Set a personality")
         @discord.app_commands.describe(name="Personality name. Leave empty to list available.")
@@ -1521,9 +1564,7 @@ class DiscordAdapter(BasePlatformAdapter):
             discord.app_commands.Choice(name="status — show current mode", value="status"),
         ])
         async def slash_voice(interaction: discord.Interaction, mode: str = ""):
-            await interaction.response.defer(ephemeral=True)
-            event = self._build_slash_event(interaction, f"/voice {mode}".strip())
-            await self.handle_message(event)
+            await self._run_simple_slash(interaction, f"/voice {mode}".strip())
 
         @tree.command(name="update", description="Update Hermes Agent to the latest version")
         async def slash_update(interaction: discord.Interaction):

gateway/platforms/email.py

@@ -43,6 +43,20 @@ from gateway.platforms.base import (
 from gateway.config import Platform, PlatformConfig
 
 logger = logging.getLogger(__name__)
+# Automated sender patterns — emails from these are silently ignored
+_NOREPLY_PATTERNS = (
+    "noreply", "no-reply", "no_reply", "donotreply", "do-not-reply",
+    "mailer-daemon", "postmaster", "bounce", "notifications@",
+    "automated@", "auto-confirm", "auto-reply", "automailer",
+)
+
+# RFC headers that indicate bulk/automated mail
+_AUTOMATED_HEADERS = {
+    "Auto-Submitted": lambda v: v.lower() != "no",
+    "Precedence": lambda v: v.lower() in ("bulk", "list", "junk"),
+    "X-Auto-Response-Suppress": lambda v: bool(v),
+    "List-Unsubscribe": lambda v: bool(v),
+}
 
 # Gmail-safe max length per email body
 MAX_MESSAGE_LENGTH = 50_000
@@ -50,7 +64,17 @@ MAX_MESSAGE_LENGTH = 50_000
 # Supported image extensions for inline detection
 _IMAGE_EXTS = {".jpg", ".jpeg", ".png", ".gif", ".webp"}
 
-
+def _is_automated_sender(address: str, headers: dict) -> bool:
+    """Return True if this email is from an automated/noreply source."""
+    addr = address.lower()
+    if any(pattern in addr for pattern in _NOREPLY_PATTERNS):
+        return True
+    for header, check in _AUTOMATED_HEADERS.items():
+        value = headers.get(header, "")
+        if value and check(value):
+            return True
+    return False
+    
 def check_email_requirements() -> bool:
     """Check if email platform dependencies are available."""
     addr = os.getenv("EMAIL_ADDRESS")
@@ -213,6 +237,7 @@ class EmailAdapter(BasePlatformAdapter):
 
         # Track message IDs we've already processed to avoid duplicates
         self._seen_uids: set = set()
+        self._seen_uids_max: int = 2000   # cap to prevent unbounded memory growth
         self._poll_task: Optional[asyncio.Task] = None
 
         # Map chat_id (sender email) -> last subject + message-id for threading
@@ -220,6 +245,26 @@ class EmailAdapter(BasePlatformAdapter):
 
         logger.info("[Email] Adapter initialized for %s", self._address)
 
+    def _trim_seen_uids(self) -> None:
+        """Keep only the most recent UIDs to prevent unbounded memory growth.
+
+        IMAP UIDs are monotonically increasing integers. When the set grows
+        beyond the cap, we keep only the highest half — old UIDs are safe to
+        drop because new messages always have higher UIDs and IMAP's UNSEEN
+        flag prevents re-delivery regardless.
+        """
+        if len(self._seen_uids) <= self._seen_uids_max:
+            return
+        try:
+            # UIDs are bytes like b'1234' — sort numerically and keep top half
+            sorted_uids = sorted(self._seen_uids, key=lambda u: int(u))
+            keep = self._seen_uids_max // 2
+            self._seen_uids = set(sorted_uids[-keep:])
+            logger.debug("[Email] Trimmed seen UIDs to %d entries", len(self._seen_uids))
+        except (ValueError, TypeError):
+            # Fallback: just clear old entries if sort fails
+            self._seen_uids = set(list(self._seen_uids)[-self._seen_uids_max // 2:])
+
     async def connect(self) -> bool:
         """Connect to the IMAP server and start polling for new messages."""
         try:
@@ -232,6 +277,8 @@ class EmailAdapter(BasePlatformAdapter):
             if status == "OK" and data and data[0]:
                 for uid in data[0].split():
                     self._seen_uids.add(uid)
+            # Keep only the most recent UIDs to prevent unbounded growth
+            self._trim_seen_uids()
             imap.logout()
             logger.info("[Email] IMAP connection test passed. %d existing messages skipped.", len(self._seen_uids))
         except Exception as e:
@@ -290,52 +337,63 @@ class EmailAdapter(BasePlatformAdapter):
         results = []
         try:
             imap = imaplib.IMAP4_SSL(self._imap_host, self._imap_port, timeout=30)
-            imap.login(self._address, self._password)
-            imap.select("INBOX")
+            try:
+                imap.login(self._address, self._password)
+                imap.select("INBOX")
 
-            status, data = imap.uid("search", None, "UNSEEN")
-            if status != "OK" or not data or not data[0]:
-                imap.logout()
-                return results
+                status, data = imap.uid("search", None, "UNSEEN")
+                if status != "OK" or not data or not data[0]:
+                    return results
 
-            for uid in data[0].split():
-                if uid in self._seen_uids:
-                    continue
-                self._seen_uids.add(uid)
+                for uid in data[0].split():
+                    if uid in self._seen_uids:
+                        continue
+                    self._seen_uids.add(uid)
+                    # Trim periodically to prevent unbounded memory growth
+                    if len(self._seen_uids) > self._seen_uids_max:
+                        self._trim_seen_uids()
 
-                status, msg_data = imap.uid("fetch", uid, "(RFC822)")
-                if status != "OK":
-                    continue
+                    status, msg_data = imap.uid("fetch", uid, "(RFC822)")
+                    if status != "OK":
+                        continue
 
-                raw_email = msg_data[0][1]
-                msg = email_lib.message_from_bytes(raw_email)
+                    raw_email = msg_data[0][1]
+                    msg = email_lib.message_from_bytes(raw_email)
 
-                sender_raw = msg.get("From", "")
-                sender_addr = _extract_email_address(sender_raw)
-                sender_name = _decode_header_value(sender_raw)
-                # Remove email from name if present
-                if "<" in sender_name:
-                    sender_name = sender_name.split("<")[0].strip().strip('"')
+                    sender_raw = msg.get("From", "")
+                    sender_addr = _extract_email_address(sender_raw)
+                    sender_name = _decode_header_value(sender_raw)
+                    # Remove email from name if present
+                    if "<" in sender_name:
+                        sender_name = sender_name.split("<")[0].strip().strip('"')
 
-                subject = _decode_header_value(msg.get("Subject", "(no subject)"))
-                message_id = msg.get("Message-ID", "")
-                in_reply_to = msg.get("In-Reply-To", "")
-                body = _extract_text_body(msg)
-                attachments = _extract_attachments(msg, skip_attachments=self._skip_attachments)
+                    subject = _decode_header_value(msg.get("Subject", "(no subject)"))
+                    message_id = msg.get("Message-ID", "")
+                    in_reply_to = msg.get("In-Reply-To", "")
+                    # Skip automated/noreply senders before any processing
+                    msg_headers = dict(msg.items())
+                    if _is_automated_sender(sender_addr, msg_headers):
+                        logger.debug("[Email] Skipping automated sender: %s", sender_addr)
+                        continue
+                    body = _extract_text_body(msg)
+                    attachments = _extract_attachments(msg, skip_attachments=self._skip_attachments)
 
-                results.append({
-                    "uid": uid,
-                    "sender_addr": sender_addr,
-                    "sender_name": sender_name,
-                    "subject": subject,
-                    "message_id": message_id,
-                    "in_reply_to": in_reply_to,
-                    "body": body,
-                    "attachments": attachments,
-                    "date": msg.get("Date", ""),
-                })
-
-            imap.logout()
+                    results.append({
+                        "uid": uid,
+                        "sender_addr": sender_addr,
+                        "sender_name": sender_name,
+                        "subject": subject,
+                        "message_id": message_id,
+                        "in_reply_to": in_reply_to,
+                        "body": body,
+                        "attachments": attachments,
+                        "date": msg.get("Date", ""),
+                    })
+            finally:
+                try:
+                    imap.logout()
+                except Exception:
+                    pass
         except Exception as e:
             logger.error("[Email] IMAP fetch error: %s", e)
         return results
@@ -348,6 +406,11 @@ class EmailAdapter(BasePlatformAdapter):
         if sender_addr == self._address.lower():
             return
 
+        # Never reply to automated senders
+        if _is_automated_sender(sender_addr, {}):
+            logger.debug("[Email] Dropping automated sender at dispatch: %s", sender_addr)
+            return
+
         subject = msg_data["subject"]
         body = msg_data["body"].strip()
         attachments = msg_data["attachments"]
@@ -443,10 +506,15 @@ class EmailAdapter(BasePlatformAdapter):
         msg.attach(MIMEText(body, "plain", "utf-8"))
 
         smtp = smtplib.SMTP(self._smtp_host, self._smtp_port, timeout=30)
-        smtp.starttls(context=ssl.create_default_context())
-        smtp.login(self._address, self._password)
-        smtp.send_message(msg)
-        smtp.quit()
+        try:
+            smtp.starttls(context=ssl.create_default_context())
+            smtp.login(self._address, self._password)
+            smtp.send_message(msg)
+        finally:
+            try:
+                smtp.quit()
+            except Exception:
+                smtp.close()
 
         logger.info("[Email] Sent reply to %s (subject: %s)", to_addr, subject)
         return msg_id
@@ -530,10 +598,15 @@ class EmailAdapter(BasePlatformAdapter):
             msg.attach(part)
 
         smtp = smtplib.SMTP(self._smtp_host, self._smtp_port, timeout=30)
-        smtp.starttls(context=ssl.create_default_context())
-        smtp.login(self._address, self._password)
-        smtp.send_message(msg)
-        smtp.quit()
+        try:
+            smtp.starttls(context=ssl.create_default_context())
+            smtp.login(self._address, self._password)
+            smtp.send_message(msg)
+        finally:
+            try:
+                smtp.quit()
+            except Exception:
+                smtp.close()
 
         return msg_id
 

gateway/platforms/matrix.py

@@ -40,7 +40,9 @@ logger = logging.getLogger(__name__)
 MAX_MESSAGE_LENGTH = 4000
 
 # Store directory for E2EE keys and sync state.
-_STORE_DIR = Path.home() / ".hermes" / "matrix" / "store"
+# Uses get_hermes_home() so each profile gets its own Matrix store.
+from hermes_constants import get_hermes_dir as _get_hermes_dir
+_STORE_DIR = _get_hermes_dir("platforms/matrix/store", "matrix/store")
 
 # Grace period: ignore messages older than this many seconds before startup.
 _STARTUP_GRACE_SECONDS = 5
@@ -161,22 +163,49 @@ class MatrixAdapter(BasePlatformAdapter):
         # Authenticate.
         if self._access_token:
             client.access_token = self._access_token
-            # Resolve user_id if not set.
-            if not self._user_id:
-                resp = await client.whoami()
-                if isinstance(resp, nio.WhoamiResponse):
-                    self._user_id = resp.user_id
-                    client.user_id = resp.user_id
-                    logger.info("Matrix: authenticated as %s", self._user_id)
-                else:
-                    logger.error(
-                        "Matrix: whoami failed — check MATRIX_ACCESS_TOKEN and MATRIX_HOMESERVER"
+
+            # With access-token auth, always resolve whoami so we validate the
+            # token and learn the device_id. The device_id matters for E2EE:
+            # without it, matrix-nio can send plain messages but may fail to
+            # decrypt inbound encrypted events or encrypt outbound room sends.
+            resp = await client.whoami()
+            if isinstance(resp, nio.WhoamiResponse):
+                resolved_user_id = getattr(resp, "user_id", "") or self._user_id
+                resolved_device_id = getattr(resp, "device_id", "")
+                if resolved_user_id:
+                    self._user_id = resolved_user_id
+
+                # restore_login() is the matrix-nio path that binds the access
+                # token to a specific device and loads the crypto store.
+                if resolved_device_id and hasattr(client, "restore_login"):
+                    client.restore_login(
+                        self._user_id or resolved_user_id,
+                        resolved_device_id,
+                        self._access_token,
                     )
-                    await client.close()
-                    return False
+                else:
+                    if self._user_id:
+                        client.user_id = self._user_id
+                    if resolved_device_id:
+                        client.device_id = resolved_device_id
+                    client.access_token = self._access_token
+                    if self._encryption:
+                        logger.warning(
+                            "Matrix: access-token login did not restore E2EE state; "
+                            "encrypted rooms may fail until a device_id is available"
+                        )
+
+                logger.info(
+                    "Matrix: using access token for %s%s",
+                    self._user_id or "(unknown user)",
+                    f" (device {resolved_device_id})" if resolved_device_id else "",
+                )
             else:
-                client.user_id = self._user_id
-                logger.info("Matrix: using access token for %s", self._user_id)
+                logger.error(
+                    "Matrix: whoami failed — check MATRIX_ACCESS_TOKEN and MATRIX_HOMESERVER"
+                )
+                await client.close()
+                return False
         elif self._password and self._user_id:
             resp = await client.login(
                 self._password,
@@ -194,13 +223,18 @@ class MatrixAdapter(BasePlatformAdapter):
             return False
 
         # If E2EE is enabled, load the crypto store.
-        if self._encryption and hasattr(client, "olm"):
+        if self._encryption and getattr(client, "olm", None):
             try:
                 if client.should_upload_keys:
                     await client.keys_upload()
                 logger.info("Matrix: E2EE crypto initialized")
             except Exception as exc:
                 logger.warning("Matrix: crypto init issue: %s", exc)
+        elif self._encryption:
+            logger.warning(
+                "Matrix: E2EE requested but crypto store is not loaded; "
+                "encrypted rooms may fail"
+            )
 
         # Register event callbacks.
         client.add_event_callback(self._on_room_message, nio.RoomMessageText)
@@ -230,6 +264,7 @@ class MatrixAdapter(BasePlatformAdapter):
             )
             # Build DM room cache from m.direct account data.
             await self._refresh_dm_cache()
+            await self._run_e2ee_maintenance()
         else:
             logger.warning("Matrix: initial sync returned %s", type(resp).__name__)
 
@@ -301,13 +336,48 @@ class MatrixAdapter(BasePlatformAdapter):
                     relates_to["m.in_reply_to"] = {"event_id": reply_to}
                 msg_content["m.relates_to"] = relates_to
 
-            resp = await self._client.room_send(
-                chat_id,
-                "m.room.message",
-                msg_content,
-            )
+            async def _room_send_once(*, ignore_unverified_devices: bool = False):
+                return await asyncio.wait_for(
+                    self._client.room_send(
+                        chat_id,
+                        "m.room.message",
+                        msg_content,
+                        ignore_unverified_devices=ignore_unverified_devices,
+                    ),
+                    timeout=45,
+                )
+
+            try:
+                resp = await _room_send_once(ignore_unverified_devices=False)
+            except Exception as exc:
+                retryable = isinstance(exc, asyncio.TimeoutError)
+                olm_unverified = getattr(nio, "OlmUnverifiedDeviceError", None)
+                send_retry = getattr(nio, "SendRetryError", None)
+                if isinstance(olm_unverified, type) and isinstance(exc, olm_unverified):
+                    retryable = True
+                if isinstance(send_retry, type) and isinstance(exc, send_retry):
+                    retryable = True
+
+                if not retryable:
+                    logger.error("Matrix: failed to send to %s: %s", chat_id, exc)
+                    return SendResult(success=False, error=str(exc))
+
+                logger.warning(
+                    "Matrix: initial encrypted send to %s failed (%s); "
+                    "retrying after E2EE maintenance with ignored unverified devices",
+                    chat_id,
+                    exc,
+                )
+                await self._run_e2ee_maintenance()
+                try:
+                    resp = await _room_send_once(ignore_unverified_devices=True)
+                except Exception as retry_exc:
+                    logger.error("Matrix: failed to send to %s after retry: %s", chat_id, retry_exc)
+                    return SendResult(success=False, error=str(retry_exc))
+
             if isinstance(resp, nio.RoomSendResponse):
                 last_event_id = resp.event_id
+                logger.info("Matrix: sent event %s to %s", last_event_id, chat_id)
             else:
                 err = getattr(resp, "message", str(resp))
                 logger.error("Matrix: failed to send to %s: %s", chat_id, err)
@@ -565,6 +635,9 @@ class MatrixAdapter(BasePlatformAdapter):
                         getattr(resp, "message", resp),
                     )
                     await asyncio.sleep(5)
+                    continue
+
+                await self._run_e2ee_maintenance()
             except asyncio.CancelledError:
                 return
             except Exception as exc:
@@ -573,6 +646,38 @@ class MatrixAdapter(BasePlatformAdapter):
                 logger.warning("Matrix: sync error: %s — retrying in 5s", exc)
                 await asyncio.sleep(5)
 
+    async def _run_e2ee_maintenance(self) -> None:
+        """Run matrix-nio E2EE housekeeping between syncs.
+
+        Hermes uses a custom sync loop instead of matrix-nio's sync_forever(),
+        so we need to explicitly drive the key management work that sync_forever()
+        normally handles for encrypted rooms.
+        """
+        client = self._client
+        if not client or not self._encryption or not getattr(client, "olm", None):
+            return
+
+        tasks = [asyncio.create_task(client.send_to_device_messages())]
+
+        if client.should_upload_keys:
+            tasks.append(asyncio.create_task(client.keys_upload()))
+
+        if client.should_query_keys:
+            tasks.append(asyncio.create_task(client.keys_query()))
+
+        if client.should_claim_keys:
+            users = client.get_users_for_key_claiming()
+            if users:
+                tasks.append(asyncio.create_task(client.keys_claim(users)))
+
+        for task in asyncio.as_completed(tasks):
+            try:
+                await task
+            except asyncio.CancelledError:
+                raise
+            except Exception as exc:
+                logger.warning("Matrix: E2EE maintenance task failed: %s", exc)
+
     # ------------------------------------------------------------------
     # Event callbacks
     # ------------------------------------------------------------------

gateway/platforms/mattermost.py

@@ -603,9 +603,19 @@ class MattermostAdapter(BasePlatformAdapter):
         # For DMs, user_id is sufficient.  For channels, check for @mention.
         message_text = post.get("message", "")
 
-        # Mention-only mode: skip channel messages that don't @mention the bot.
-        # DMs (type "D") are always processed.
+        # Mention-gating for non-DM channels.
+        # Config (env vars):
+        #   MATTERMOST_REQUIRE_MENTION: Require @mention in channels (default: true)
+        #   MATTERMOST_FREE_RESPONSE_CHANNELS: Channel IDs where bot responds without mention
         if channel_type_raw != "D":
+            require_mention = os.getenv(
+                "MATTERMOST_REQUIRE_MENTION", "true"
+            ).lower() not in ("false", "0", "no")
+
+            free_channels_raw = os.getenv("MATTERMOST_FREE_RESPONSE_CHANNELS", "")
+            free_channels = {ch.strip() for ch in free_channels_raw.split(",") if ch.strip()}
+            is_free_channel = channel_id in free_channels
+
             mention_patterns = [
                 f"@{self._bot_username}",
                 f"@{self._bot_user_id}",
@@ -614,13 +624,21 @@ class MattermostAdapter(BasePlatformAdapter):
                 pattern.lower() in message_text.lower()
                 for pattern in mention_patterns
             )
-            if not has_mention:
+
+            if require_mention and not is_free_channel and not has_mention:
                 logger.debug(
                     "Mattermost: skipping non-DM message without @mention (channel=%s)",
                     channel_id,
                 )
                 return
 
+            # Strip @mention from the message text so the agent sees clean input.
+            if has_mention:
+                for pattern in mention_patterns:
+                    message_text = re.sub(
+                        re.escape(pattern), "", message_text, flags=re.IGNORECASE
+                    ).strip()
+
         # Resolve sender info.
         sender_id = post.get("user_id", "")
         sender_name = data.get("sender_name", "").lstrip("@") or sender_id

gateway/platforms/signal.py

@@ -22,7 +22,7 @@ import time
 from datetime import datetime, timezone
 from pathlib import Path
 from typing import Dict, List, Optional, Any
-from urllib.parse import unquote
+from urllib.parse import quote, unquote
 
 import httpx
 
@@ -184,6 +184,8 @@ class SignalAdapter(BasePlatformAdapter):
         self._recent_sent_timestamps: set = set()
         self._max_recent_timestamps = 50
 
+        self._phone_lock_identity: Optional[str] = None
+
         logger.info("Signal adapter initialized: url=%s account=%s groups=%s",
                      self.http_url, _redact_phone(self.account),
                      "enabled" if self.group_allow_from else "disabled")
@@ -198,6 +200,29 @@ class SignalAdapter(BasePlatformAdapter):
             logger.error("Signal: SIGNAL_HTTP_URL and SIGNAL_ACCOUNT are required")
             return False
 
+        # Acquire scoped lock to prevent duplicate Signal listeners for the same phone
+        try:
+            from gateway.status import acquire_scoped_lock
+
+            self._phone_lock_identity = self.account
+            acquired, existing = acquire_scoped_lock(
+                "signal-phone",
+                self._phone_lock_identity,
+                metadata={"platform": self.platform.value},
+            )
+            if not acquired:
+                owner_pid = existing.get("pid") if isinstance(existing, dict) else None
+                message = (
+                    "Another local Hermes gateway is already using this Signal account"
+                    + (f" (PID {owner_pid})." if owner_pid else ".")
+                    + " Stop the other gateway before starting a second Signal listener."
+                )
+                logger.error("Signal: %s", message)
+                self._set_fatal_error("signal_phone_lock", message, retryable=False)
+                return False
+        except Exception as e:
+            logger.warning("Signal: Could not acquire phone lock (non-fatal): %s", e)
+
         self.client = httpx.AsyncClient(timeout=30.0)
 
         # Health check — verify signal-cli daemon is reachable
@@ -245,6 +270,14 @@ class SignalAdapter(BasePlatformAdapter):
             await self.client.aclose()
             self.client = None
 
+        if self._phone_lock_identity:
+            try:
+                from gateway.status import release_scoped_lock
+                release_scoped_lock("signal-phone", self._phone_lock_identity)
+            except Exception as e:
+                logger.warning("Signal: Error releasing phone lock: %s", e, exc_info=True)
+            self._phone_lock_identity = None
+
         logger.info("Signal: disconnected")
 
     # ------------------------------------------------------------------
@@ -253,7 +286,7 @@ class SignalAdapter(BasePlatformAdapter):
 
     async def _sse_listener(self) -> None:
         """Listen for SSE events from signal-cli daemon."""
-        url = f"{self.http_url}/api/v1/events?account={self.account}"
+        url = f"{self.http_url}/api/v1/events?account={quote(self.account, safe='')}"
         backoff = SSE_RETRY_DELAY_INITIAL
 
         while self._running:
@@ -521,7 +554,7 @@ class SignalAdapter(BasePlatformAdapter):
         """Fetch an attachment via JSON-RPC and cache it. Returns (path, ext)."""
         result = await self._rpc("getAttachment", {
             "account": self.account,
-            "attachmentId": attachment_id,
+            "id": attachment_id,
         })
 
         if not result:

gateway/platforms/slack.py

@@ -93,6 +93,17 @@ class SlackAdapter(BasePlatformAdapter):
             return False
 
         try:
+            # Acquire scoped lock to prevent duplicate app token usage
+            from gateway.status import acquire_scoped_lock
+            self._token_lock_identity = app_token
+            acquired, existing = acquire_scoped_lock('slack-app-token', app_token, metadata={'platform': 'slack'})
+            if not acquired:
+                owner_pid = existing.get('pid') if isinstance(existing, dict) else None
+                message = f'Slack app token already in use' + (f' (PID {owner_pid})' if owner_pid else '') + '. Stop the other gateway first.'
+                logger.error('[%s] %s', self.name, message)
+                self._set_fatal_error('slack_token_lock', message, retryable=False)
+                return False
+
             self._app = AsyncApp(token=bot_token)
 
             # Get our own bot user ID for mention detection
@@ -138,6 +149,16 @@ class SlackAdapter(BasePlatformAdapter):
             except Exception as e:  # pragma: no cover - defensive logging
                 logger.warning("[Slack] Error while closing Socket Mode handler: %s", e, exc_info=True)
         self._running = False
+
+        # Release the token lock (use stored identity, not re-read env)
+        try:
+            from gateway.status import release_scoped_lock
+            if getattr(self, '_token_lock_identity', None):
+                release_scoped_lock('slack-app-token', self._token_lock_identity)
+                self._token_lock_identity = None
+        except Exception:
+            pass
+
         logger.info("[Slack] Disconnected")
 
     async def send(

gateway/platforms/telegram.py

@@ -345,7 +345,8 @@ class TelegramAdapter(BasePlatformAdapter):
     def _persist_dm_topic_thread_id(self, chat_id: int, topic_name: str, thread_id: int) -> None:
         """Save a newly created thread_id back into config.yaml so it persists across restarts."""
         try:
-            config_path = _Path.home() / ".hermes" / "config.yaml"
+            from hermes_constants import get_hermes_home
+            config_path = get_hermes_home() / "config.yaml"
             if not config_path.exists():
                 logger.warning("[%s] Config file not found at %s, cannot persist thread_id", self.name, config_path)
                 return
@@ -1757,7 +1758,8 @@ class TelegramAdapter(BasePlatformAdapter):
         recognized without a gateway restart.
         """
         try:
-            config_path = _Path.home() / ".hermes" / "config.yaml"
+            from hermes_constants import get_hermes_home
+            config_path = get_hermes_home() / "config.yaml"
             if not config_path.exists():
                 return
 

gateway/platforms/telegram_network.py

@@ -12,6 +12,7 @@ from __future__ import annotations
 import asyncio
 import ipaddress
 import logging
+import os
 import socket
 from typing import Iterable, Optional
 
@@ -43,6 +44,14 @@ _DOH_PROVIDERS: list[dict] = [
 _SEED_FALLBACK_IPS: list[str] = ["149.154.167.220"]
 
 
+def _resolve_proxy_url() -> str | None:
+    for key in ("HTTPS_PROXY", "HTTP_PROXY", "ALL_PROXY", "https_proxy", "http_proxy", "all_proxy"):
+        value = (os.environ.get(key) or "").strip()
+        if value:
+            return value
+    return None
+
+
 class TelegramFallbackTransport(httpx.AsyncBaseTransport):
     """Retry Telegram Bot API requests via fallback IPs while preserving TLS/SNI.
 
@@ -54,6 +63,9 @@ class TelegramFallbackTransport(httpx.AsyncBaseTransport):
 
     def __init__(self, fallback_ips: Iterable[str], **transport_kwargs):
         self._fallback_ips = [ip for ip in dict.fromkeys(_normalize_fallback_ips(fallback_ips))]
+        proxy_url = _resolve_proxy_url()
+        if proxy_url and "proxy" not in transport_kwargs:
+            transport_kwargs["proxy"] = proxy_url
         self._primary = httpx.AsyncHTTPTransport(**transport_kwargs)
         self._fallbacks = {
             ip: httpx.AsyncHTTPTransport(**transport_kwargs) for ip in self._fallback_ips

gateway/platforms/webhook.py

@@ -27,6 +27,7 @@ import hashlib
 import hmac
 import json
 import logging
+import os
 import re
 import subprocess
 import time
@@ -53,6 +54,7 @@ logger = logging.getLogger(__name__)
 DEFAULT_HOST = "0.0.0.0"
 DEFAULT_PORT = 8644
 _INSECURE_NO_AUTH = "INSECURE_NO_AUTH"
+_DYNAMIC_ROUTES_FILENAME = "webhook_subscriptions.json"
 
 
 def check_webhook_requirements() -> bool:
@@ -68,7 +70,10 @@ class WebhookAdapter(BasePlatformAdapter):
         self._host: str = config.extra.get("host", DEFAULT_HOST)
         self._port: int = int(config.extra.get("port", DEFAULT_PORT))
         self._global_secret: str = config.extra.get("secret", "")
-        self._routes: Dict[str, dict] = config.extra.get("routes", {})
+        self._static_routes: Dict[str, dict] = config.extra.get("routes", {})
+        self._dynamic_routes: Dict[str, dict] = {}
+        self._dynamic_routes_mtime: float = 0.0
+        self._routes: Dict[str, dict] = dict(self._static_routes)
         self._runner = None
 
         # Delivery info keyed by session chat_id — consumed by send()
@@ -96,6 +101,9 @@ class WebhookAdapter(BasePlatformAdapter):
     # ------------------------------------------------------------------
 
     async def connect(self) -> bool:
+        # Load agent-created subscriptions before validating
+        self._reload_dynamic_routes()
+
         # Validate routes at startup — secret is required per route
         for name, route in self._routes.items():
             secret = route.get("secret", self._global_secret)
@@ -110,6 +118,17 @@ class WebhookAdapter(BasePlatformAdapter):
         app.router.add_get("/health", self._handle_health)
         app.router.add_post("/webhooks/{route_name}", self._handle_webhook)
 
+        # Port conflict detection — fail fast if port is already in use
+        import socket as _socket
+        try:
+            with _socket.socket(_socket.AF_INET, _socket.SOCK_STREAM) as _s:
+                _s.settimeout(1)
+                _s.connect(('127.0.0.1', self._port))
+            logger.error('[webhook] Port %d already in use. Set a different port in config.yaml: platforms.webhook.port', self._port)
+            return False
+        except (ConnectionRefusedError, OSError):
+            pass  # port is free
+
         self._runner = web.AppRunner(app)
         await self._runner.setup()
         site = web.TCPSite(self._runner, self._host, self._port)
@@ -182,8 +201,46 @@ class WebhookAdapter(BasePlatformAdapter):
         """GET /health — simple health check."""
         return web.json_response({"status": "ok", "platform": "webhook"})
 
+    def _reload_dynamic_routes(self) -> None:
+        """Reload agent-created subscriptions from disk if the file changed."""
+        from pathlib import Path as _Path
+        hermes_home = _Path(
+            os.getenv("HERMES_HOME", str(_Path.home() / ".hermes"))
+        ).expanduser()
+        subs_path = hermes_home / _DYNAMIC_ROUTES_FILENAME
+        if not subs_path.exists():
+            if self._dynamic_routes:
+                self._dynamic_routes = {}
+                self._routes = dict(self._static_routes)
+                logger.debug("[webhook] Dynamic subscriptions file removed, cleared dynamic routes")
+            return
+        try:
+            mtime = subs_path.stat().st_mtime
+            if mtime <= self._dynamic_routes_mtime:
+                return  # No change
+            data = json.loads(subs_path.read_text(encoding="utf-8"))
+            if not isinstance(data, dict):
+                return
+            # Merge: static routes take precedence over dynamic ones
+            self._dynamic_routes = {
+                k: v for k, v in data.items()
+                if k not in self._static_routes
+            }
+            self._routes = {**self._dynamic_routes, **self._static_routes}
+            self._dynamic_routes_mtime = mtime
+            logger.info(
+                "[webhook] Reloaded %d dynamic route(s): %s",
+                len(self._dynamic_routes),
+                ", ".join(self._dynamic_routes.keys()) or "(none)",
+            )
+        except Exception as e:
+            logger.warning("[webhook] Failed to reload dynamic routes: %s", e)
+
     async def _handle_webhook(self, request: "web.Request") -> "web.Response":
         """POST /webhooks/{route_name} — receive and process a webhook event."""
+        # Hot-reload dynamic subscriptions on each request (mtime-gated, cheap)
+        self._reload_dynamic_routes()
+
         route_name = request.match_info.get("route_name", "")
         route_config = self._routes.get(route_name)
 

gateway/platforms/whatsapp.py

@@ -26,6 +26,7 @@ from pathlib import Path
 from typing import Dict, Optional, Any
 
 from hermes_cli.config import get_hermes_home
+from hermes_constants import get_hermes_dir
 
 logger = logging.getLogger(__name__)
 
@@ -134,13 +135,15 @@ class WhatsAppAdapter(BasePlatformAdapter):
         )
         self._session_path: Path = Path(config.extra.get(
             "session_path",
-            get_hermes_home() / "whatsapp" / "session"
+            get_hermes_dir("platforms/whatsapp/session", "whatsapp/session")
         ))
         self._reply_prefix: Optional[str] = config.extra.get("reply_prefix")
         self._message_queue: asyncio.Queue = asyncio.Queue()
         self._bridge_log_fh = None
         self._bridge_log: Optional[Path] = None
         self._poll_task: Optional[asyncio.Task] = None
+        self._http_session: Optional["aiohttp.ClientSession"] = None
+        self._session_lock_identity: Optional[str] = None
     
     async def connect(self) -> bool:
         """
@@ -159,6 +162,29 @@ class WhatsAppAdapter(BasePlatformAdapter):
         
         logger.info("[%s] Bridge found at %s", self.name, bridge_path)
         
+        # Acquire scoped lock to prevent duplicate sessions
+        try:
+            from gateway.status import acquire_scoped_lock
+
+            self._session_lock_identity = str(self._session_path)
+            acquired, existing = acquire_scoped_lock(
+                "whatsapp-session",
+                self._session_lock_identity,
+                metadata={"platform": self.platform.value},
+            )
+            if not acquired:
+                owner_pid = existing.get("pid") if isinstance(existing, dict) else None
+                message = (
+                    "Another local Hermes gateway is already using this WhatsApp session"
+                    + (f" (PID {owner_pid})." if owner_pid else ".")
+                    + " Stop the other gateway before starting a second WhatsApp bridge."
+                )
+                logger.error("[%s] %s", self.name, message)
+                self._set_fatal_error("whatsapp_session_lock", message, retryable=False)
+                return False
+        except Exception as e:
+            logger.warning("[%s] Could not acquire session lock (non-fatal): %s", self.name, e)
+
         # Auto-install npm dependencies if node_modules doesn't exist
         bridge_dir = bridge_path.parent
         if not (bridge_dir / "node_modules").exists():
@@ -199,6 +225,7 @@ class WhatsAppAdapter(BasePlatformAdapter):
                                 print(f"[{self.name}] Using existing bridge (status: {bridge_status})")
                                 self._mark_connected()
                                 self._bridge_process = None  # Not managed by us
+                                self._http_session = aiohttp.ClientSession()
                                 self._poll_task = asyncio.create_task(self._poll_messages())
                                 return True
                             else:
@@ -304,6 +331,9 @@ class WhatsAppAdapter(BasePlatformAdapter):
                     print(f"[{self.name}]   Bridge log: {self._bridge_log}")
                     print(f"[{self.name}]   If session expired, re-pair: hermes whatsapp")
             
+            # Create a persistent HTTP session for all bridge communication
+            self._http_session = aiohttp.ClientSession()
+
             # Start message polling task
             self._poll_task = asyncio.create_task(self._poll_messages())
             
@@ -312,6 +342,12 @@ class WhatsAppAdapter(BasePlatformAdapter):
             return True
             
         except Exception as e:
+            if self._session_lock_identity:
+                try:
+                    from gateway.status import release_scoped_lock
+                    release_scoped_lock("whatsapp-session", self._session_lock_identity)
+                except Exception:
+                    pass
             logger.error("[%s] Failed to start bridge: %s", self.name, e, exc_info=True)
             self._close_bridge_log()
             return False
@@ -369,10 +405,32 @@ class WhatsAppAdapter(BasePlatformAdapter):
         else:
             # Bridge was not started by us, don't kill it
             print(f"[{self.name}] Disconnecting (external bridge left running)")
-        
+
+        # Cancel the poll task explicitly
+        if self._poll_task and not self._poll_task.done():
+            self._poll_task.cancel()
+            try:
+                await self._poll_task
+            except (asyncio.CancelledError, Exception):
+                pass
+        self._poll_task = None
+
+        # Close the persistent HTTP session
+        if self._http_session and not self._http_session.closed:
+            await self._http_session.close()
+        self._http_session = None
+
+        if self._session_lock_identity:
+            try:
+                from gateway.status import release_scoped_lock
+                release_scoped_lock("whatsapp-session", self._session_lock_identity)
+            except Exception as e:
+                logger.warning("[%s] Error releasing WhatsApp session lock: %s", self.name, e, exc_info=True)
+
         self._mark_disconnected()
         self._bridge_process = None
         self._close_bridge_log()
+        self._session_lock_identity = None
         print(f"[{self.name}] Disconnected")
     
     async def send(
@@ -383,7 +441,7 @@ class WhatsAppAdapter(BasePlatformAdapter):
         metadata: Optional[Dict[str, Any]] = None
     ) -> SendResult:
         """Send a message via the WhatsApp bridge."""
-        if not self._running:
+        if not self._running or not self._http_session:
             return SendResult(success=False, error="Not connected")
         bridge_exit = await self._check_managed_bridge_exit()
         if bridge_exit:
@@ -391,36 +449,29 @@ class WhatsAppAdapter(BasePlatformAdapter):
         
         try:
             import aiohttp
+
+            payload = {
+                "chatId": chat_id,
+                "message": content,
+            }
+            if reply_to:
+                payload["replyTo"] = reply_to
             
-            async with aiohttp.ClientSession() as session:
-                payload = {
-                    "chatId": chat_id,
-                    "message": content,
-                }
-                if reply_to:
-                    payload["replyTo"] = reply_to
-                
-                async with session.post(
-                    f"http://127.0.0.1:{self._bridge_port}/send",
-                    json=payload,
-                    timeout=aiohttp.ClientTimeout(total=30)
-                ) as resp:
-                    if resp.status == 200:
-                        data = await resp.json()
-                        return SendResult(
-                            success=True,
-                            message_id=data.get("messageId"),
-                            raw_response=data
-                        )
-                    else:
-                        error = await resp.text()
-                        return SendResult(success=False, error=error)
-                        
-        except ImportError:
-            return SendResult(
-                success=False, 
-                error="aiohttp not installed. Run: pip install aiohttp"
-            )
+            async with self._http_session.post(
+                f"http://127.0.0.1:{self._bridge_port}/send",
+                json=payload,
+                timeout=aiohttp.ClientTimeout(total=30)
+            ) as resp:
+                if resp.status == 200:
+                    data = await resp.json()
+                    return SendResult(
+                        success=True,
+                        message_id=data.get("messageId"),
+                        raw_response=data
+                    )
+                else:
+                    error = await resp.text()
+                    return SendResult(success=False, error=error)
         except Exception as e:
             return SendResult(success=False, error=str(e))
 
@@ -431,28 +482,27 @@ class WhatsAppAdapter(BasePlatformAdapter):
         content: str,
     ) -> SendResult:
         """Edit a previously sent message via the WhatsApp bridge."""
-        if not self._running:
+        if not self._running or not self._http_session:
             return SendResult(success=False, error="Not connected")
         bridge_exit = await self._check_managed_bridge_exit()
         if bridge_exit:
             return SendResult(success=False, error=bridge_exit)
         try:
             import aiohttp
-            async with aiohttp.ClientSession() as session:
-                async with session.post(
-                    f"http://127.0.0.1:{self._bridge_port}/edit",
-                    json={
-                        "chatId": chat_id,
-                        "messageId": message_id,
-                        "message": content,
-                    },
-                    timeout=aiohttp.ClientTimeout(total=15)
-                ) as resp:
-                    if resp.status == 200:
-                        return SendResult(success=True, message_id=message_id)
-                    else:
-                        error = await resp.text()
-                        return SendResult(success=False, error=error)
+            async with self._http_session.post(
+                f"http://127.0.0.1:{self._bridge_port}/edit",
+                json={
+                    "chatId": chat_id,
+                    "messageId": message_id,
+                    "message": content,
+                },
+                timeout=aiohttp.ClientTimeout(total=15)
+            ) as resp:
+                if resp.status == 200:
+                    return SendResult(success=True, message_id=message_id)
+                else:
+                    error = await resp.text()
+                    return SendResult(success=False, error=error)
         except Exception as e:
             return SendResult(success=False, error=str(e))
 
@@ -465,7 +515,7 @@ class WhatsAppAdapter(BasePlatformAdapter):
         file_name: Optional[str] = None,
     ) -> SendResult:
         """Send any media file via bridge /send-media endpoint."""
-        if not self._running:
+        if not self._running or not self._http_session:
             return SendResult(success=False, error="Not connected")
         bridge_exit = await self._check_managed_bridge_exit()
         if bridge_exit:
@@ -486,22 +536,21 @@ class WhatsAppAdapter(BasePlatformAdapter):
             if file_name:
                 payload["fileName"] = file_name
 
-            async with aiohttp.ClientSession() as session:
-                async with session.post(
-                    f"http://127.0.0.1:{self._bridge_port}/send-media",
-                    json=payload,
-                    timeout=aiohttp.ClientTimeout(total=120),
-                ) as resp:
-                    if resp.status == 200:
-                        data = await resp.json()
-                        return SendResult(
-                            success=True,
-                            message_id=data.get("messageId"),
-                            raw_response=data,
-                        )
-                    else:
-                        error = await resp.text()
-                        return SendResult(success=False, error=error)
+            async with self._http_session.post(
+                f"http://127.0.0.1:{self._bridge_port}/send-media",
+                json=payload,
+                timeout=aiohttp.ClientTimeout(total=120),
+            ) as resp:
+                if resp.status == 200:
+                    data = await resp.json()
+                    return SendResult(
+                        success=True,
+                        message_id=data.get("messageId"),
+                        raw_response=data,
+                    )
+                else:
+                    error = await resp.text()
+                    return SendResult(success=False, error=error)
 
         except Exception as e:
             return SendResult(success=False, error=str(e))
@@ -526,6 +575,7 @@ class WhatsAppAdapter(BasePlatformAdapter):
         image_path: str,
         caption: Optional[str] = None,
         reply_to: Optional[str] = None,
+        **kwargs,
     ) -> SendResult:
         """Send a local image file natively via bridge."""
         return await self._send_media_to_bridge(chat_id, image_path, "image", caption)
@@ -536,6 +586,7 @@ class WhatsAppAdapter(BasePlatformAdapter):
         video_path: str,
         caption: Optional[str] = None,
         reply_to: Optional[str] = None,
+        **kwargs,
     ) -> SendResult:
         """Send a video natively via bridge — plays inline in WhatsApp."""
         return await self._send_media_to_bridge(chat_id, video_path, "video", caption)
@@ -547,6 +598,7 @@ class WhatsAppAdapter(BasePlatformAdapter):
         caption: Optional[str] = None,
         file_name: Optional[str] = None,
         reply_to: Optional[str] = None,
+        **kwargs,
     ) -> SendResult:
         """Send a document/file as a downloadable attachment via bridge."""
         return await self._send_media_to_bridge(
@@ -556,45 +608,43 @@ class WhatsAppAdapter(BasePlatformAdapter):
 
     async def send_typing(self, chat_id: str, metadata=None) -> None:
         """Send typing indicator via bridge."""
-        if not self._running:
+        if not self._running or not self._http_session:
             return
         if await self._check_managed_bridge_exit():
             return
         
         try:
             import aiohttp
-            
-            async with aiohttp.ClientSession() as session:
-                await session.post(
-                    f"http://127.0.0.1:{self._bridge_port}/typing",
-                    json={"chatId": chat_id},
-                    timeout=aiohttp.ClientTimeout(total=5)
-                )
+
+            await self._http_session.post(
+                f"http://127.0.0.1:{self._bridge_port}/typing",
+                json={"chatId": chat_id},
+                timeout=aiohttp.ClientTimeout(total=5)
+            )
         except Exception:
             pass  # Ignore typing indicator failures
     
     async def get_chat_info(self, chat_id: str) -> Dict[str, Any]:
         """Get information about a WhatsApp chat."""
-        if not self._running:
+        if not self._running or not self._http_session:
             return {"name": "Unknown", "type": "dm"}
         if await self._check_managed_bridge_exit():
             return {"name": chat_id, "type": "dm"}
         
         try:
             import aiohttp
-            
-            async with aiohttp.ClientSession() as session:
-                async with session.get(
-                    f"http://127.0.0.1:{self._bridge_port}/chat/{chat_id}",
-                    timeout=aiohttp.ClientTimeout(total=10)
-                ) as resp:
-                    if resp.status == 200:
-                        data = await resp.json()
-                        return {
-                            "name": data.get("name", chat_id),
-                            "type": "group" if data.get("isGroup") else "dm",
-                            "participants": data.get("participants", []),
-                        }
+
+            async with self._http_session.get(
+                f"http://127.0.0.1:{self._bridge_port}/chat/{chat_id}",
+                timeout=aiohttp.ClientTimeout(total=10)
+            ) as resp:
+                if resp.status == 200:
+                    data = await resp.json()
+                    return {
+                        "name": data.get("name", chat_id),
+                        "type": "group" if data.get("isGroup") else "dm",
+                        "participants": data.get("participants", []),
+                    }
         except Exception as e:
             logger.debug("Could not get WhatsApp chat info for %s: %s", chat_id, e)
         
@@ -602,29 +652,26 @@ class WhatsAppAdapter(BasePlatformAdapter):
     
     async def _poll_messages(self) -> None:
         """Poll the bridge for incoming messages."""
-        try:
-            import aiohttp
-        except ImportError:
-            print(f"[{self.name}] aiohttp not installed, message polling disabled")
-            return
-        
+        import aiohttp
+
         while self._running:
+            if not self._http_session:
+                break
             bridge_exit = await self._check_managed_bridge_exit()
             if bridge_exit:
                 print(f"[{self.name}] {bridge_exit}")
                 break
             try:
-                async with aiohttp.ClientSession() as session:
-                    async with session.get(
-                        f"http://127.0.0.1:{self._bridge_port}/messages",
-                        timeout=aiohttp.ClientTimeout(total=30)
-                    ) as resp:
-                        if resp.status == 200:
-                            messages = await resp.json()
-                            for msg_data in messages:
-                                event = await self._build_message_event(msg_data)
-                                if event:
-                                    await self.handle_message(event)
+                async with self._http_session.get(
+                    f"http://127.0.0.1:{self._bridge_port}/messages",
+                    timeout=aiohttp.ClientTimeout(total=30)
+                ) as resp:
+                    if resp.status == 200:
+                        messages = await resp.json()
+                        for msg_data in messages:
+                            event = await self._build_message_event(msg_data)
+                            if event:
+                                await self.handle_message(event)
             except asyncio.CancelledError:
                 break
             except Exception as e:

gateway/run.py

@@ -77,6 +77,7 @@ sys.path.insert(0, str(Path(__file__).parent.parent))
 
 # Resolve Hermes home directory (respects HERMES_HOME override)
 from hermes_constants import get_hermes_home
+from utils import atomic_yaml_write
 _hermes_home = get_hermes_home()
 
 # Load environment variables from ~/.hermes/.env first.
@@ -288,7 +289,7 @@ def _resolve_gateway_model(config: dict | None = None) -> str:
     if isinstance(model_cfg, str):
         model = model_cfg
     elif isinstance(model_cfg, dict):
-        model = model_cfg.get("default", model)
+        model = model_cfg.get("default") or model_cfg.get("model") or model
     return model
 
 
@@ -432,7 +433,7 @@ class GatewayRunner:
             from honcho_integration.session import HonchoSessionManager
 
             hcfg = HonchoClientConfig.from_global_config()
-            if not hcfg.enabled or not hcfg.api_key:
+            if not hcfg.enabled or not (hcfg.api_key or hcfg.base_url):
                 return None, hcfg
 
             client = get_honcho_client(hcfg)
@@ -745,10 +746,22 @@ class GatewayRunner:
                 logger.error("No connected messaging platforms remain. Shutting down gateway cleanly.")
             await self.stop()
         elif not self.adapters and self._failed_platforms:
-            logger.warning(
-                "No connected messaging platforms remain, but %d platform(s) queued for reconnection",
-                len(self._failed_platforms),
-            )
+            # All platforms are down and queued for background reconnection.
+            # If the error is retryable, exit with failure so systemd Restart=on-failure
+            # can restart the process. Otherwise stay alive and keep retrying in background.
+            if adapter.fatal_error_retryable:
+                self._exit_reason = adapter.fatal_error_message or "All messaging platforms failed with retryable errors"
+                self._exit_with_failure = True
+                logger.error(
+                    "All messaging platforms failed with retryable errors. "
+                    "Shutting down gateway for service restart (systemd will retry)."
+                )
+                await self.stop()
+            else:
+                logger.warning(
+                    "No connected messaging platforms remain, but %d platform(s) queued for reconnection",
+                    len(self._failed_platforms),
+                )
 
     def _request_clean_exit(self, reason: str) -> None:
         self._exit_cleanly = True
@@ -906,11 +919,12 @@ class GatewayRunner:
         return {}
 
     @staticmethod
-    def _load_fallback_model() -> dict | None:
-        """Load fallback model config from config.yaml.
+    def _load_fallback_model() -> list | dict | None:
+        """Load fallback provider chain from config.yaml.
 
-        Returns a dict with 'provider' and 'model' keys, or None if
-        not configured / both fields empty.
+        Returns a list of provider dicts (``fallback_providers``), a single
+        dict (legacy ``fallback_model``), or None if not configured.
+        AIAgent.__init__ normalizes both formats into a chain.
         """
         try:
             import yaml as _y
@@ -918,8 +932,8 @@ class GatewayRunner:
             if cfg_path.exists():
                 with open(cfg_path, encoding="utf-8") as _f:
                     cfg = _y.safe_load(_f) or {}
-                fb = cfg.get("fallback_model", {}) or {}
-                if fb.get("provider") and fb.get("model"):
+                fb = cfg.get("fallback_providers") or cfg.get("fallback_model") or None
+                if fb:
                     return fb
         except Exception:
             pass
@@ -947,6 +961,13 @@ class GatewayRunner:
         """
         logger.info("Starting Hermes Gateway...")
         logger.info("Session storage: %s", self.config.sessions_dir)
+        try:
+            from hermes_cli.profiles import get_active_profile_name
+            _profile = get_active_profile_name()
+            if _profile and _profile != "default":
+                logger.info("Active profile: %s", _profile)
+        except Exception:
+            pass
         try:
             from gateway.status import write_runtime_status
             write_runtime_status(gateway_state="starting", exit_reason=None)
@@ -2081,7 +2102,7 @@ class GatewayRunner:
                     if isinstance(_model_cfg, str):
                         _hyg_model = _model_cfg
                     elif isinstance(_model_cfg, dict):
-                        _hyg_model = _model_cfg.get("default", _hyg_model)
+                        _hyg_model = _model_cfg.get("default") or _model_cfg.get("model") or _hyg_model
                         # Read explicit context_length override from model config
                         # (same as run_agent.py lines 995-1005)
                         _raw_ctx = _model_cfg.get("context_length")
@@ -2204,6 +2225,15 @@ class GatewayRunner:
                                     ),
                                 )
 
+                                # _compress_context ends the old session and creates
+                                # a new session_id.  Write compressed messages into
+                                # the NEW session so the old transcript stays intact
+                                # and searchable via session_search.
+                                _hyg_new_sid = _hyg_agent.session_id
+                                if _hyg_new_sid != session_entry.session_id:
+                                    session_entry.session_id = _hyg_new_sid
+                                    self.session_store._save()
+
                                 self.session_store.rewrite_transcript(
                                     session_entry.session_id, _compressed
                                 )
@@ -3067,8 +3097,7 @@ class GatewayRunner:
                 if "agent" not in config or not isinstance(config.get("agent"), dict):
                     config["agent"] = {}
                 config["agent"]["system_prompt"] = ""
-                with open(config_path, "w") as f:
-                    yaml.dump(config, f, default_flow_style=False, sort_keys=False)
+                atomic_yaml_write(config_path, config)
             except Exception as e:
                 return f"⚠️ Failed to save personality change: {e}"
             self._ephemeral_system_prompt = ""
@@ -3081,8 +3110,7 @@ class GatewayRunner:
                 if "agent" not in config or not isinstance(config.get("agent"), dict):
                     config["agent"] = {}
                 config["agent"]["system_prompt"] = new_prompt
-                with open(config_path, 'w', encoding="utf-8") as f:
-                    yaml.dump(config, f, default_flow_style=False, sort_keys=False)
+                atomic_yaml_write(config_path, config)
             except Exception as e:
                 return f"⚠️ Failed to save personality change: {e}"
 
@@ -3172,8 +3200,7 @@ class GatewayRunner:
                 with open(config_path, encoding="utf-8") as f:
                     user_config = yaml.safe_load(f) or {}
             user_config[env_key] = chat_id
-            with open(config_path, 'w', encoding="utf-8") as f:
-                yaml.dump(user_config, f, default_flow_style=False)
+            atomic_yaml_write(config_path, user_config)
             # Also set in the current environment so it takes effect immediately
             os.environ[env_key] = str(chat_id)
         except Exception as e:
@@ -3841,8 +3868,7 @@ class GatewayRunner:
                         current[k] = {}
                     current = current[k]
                 current[keys[-1]] = value
-                with open(config_path, "w", encoding="utf-8") as f:
-                    yaml.dump(user_config, f, default_flow_style=False, sort_keys=False)
+                atomic_yaml_write(config_path, user_config)
                 return True
             except Exception as e:
                 logger.error("Failed to save config key %s: %s", key_path, e)
@@ -3950,8 +3976,7 @@ class GatewayRunner:
             if "display" not in user_config or not isinstance(user_config.get("display"), dict):
                 user_config["display"] = {}
             user_config["display"]["tool_progress"] = new_mode
-            with open(config_path, "w", encoding="utf-8") as f:
-                yaml.dump(user_config, f, default_flow_style=False, sort_keys=False)
+            atomic_yaml_write(config_path, user_config)
             return f"{descriptions[new_mode]}\n_(saved to config — takes effect on next message)_"
         except Exception as e:
             logger.warning("Failed to save tool_progress mode: %s", e)
@@ -3998,13 +4023,22 @@ class GatewayRunner:
             loop = asyncio.get_event_loop()
             compressed, _ = await loop.run_in_executor(
                 None,
-                lambda: tmp_agent._compress_context(msgs, "", approx_tokens=approx_tokens),
+                lambda: tmp_agent._compress_context(msgs, "", approx_tokens=approx_tokens)
             )
 
-            self.session_store.rewrite_transcript(session_entry.session_id, compressed)
+            # _compress_context already calls end_session() on the old session
+            # (preserving its full transcript in SQLite) and creates a new
+            # session_id for the continuation.  Write the compressed messages
+            # into the NEW session so the original history stays searchable.
+            new_session_id = tmp_agent.session_id
+            if new_session_id != session_entry.session_id:
+                session_entry.session_id = new_session_id
+                self.session_store._save()
+
+            self.session_store.rewrite_transcript(new_session_id, compressed)
             # Reset stored token count — transcript changed, old value is stale
             self.session_store.update_session(
-                session_entry.session_key, last_prompt_tokens=0,
+                session_entry.session_key, last_prompt_tokens=0
             )
             new_count = len(compressed)
             new_tokens = estimate_messages_tokens_rough(compressed)
@@ -4160,7 +4194,7 @@ class GatewayRunner:
             ]
             ctx = agent.context_compressor
             if ctx.last_prompt_tokens:
-                pct = ctx.last_prompt_tokens / ctx.context_length * 100 if ctx.context_length else 0
+                pct = min(100, ctx.last_prompt_tokens / ctx.context_length * 100) if ctx.context_length else 0
                 lines.append(f"Context: {ctx.last_prompt_tokens:,} / {ctx.context_length:,} ({pct:.0f}%)")
             if ctx.compression_count:
                 lines.append(f"Compressions: {ctx.compression_count}")
@@ -4974,12 +5008,17 @@ class GatewayRunner:
             progress_queue.put(msg)
         
         # Background task to send progress messages
-        # Accumulates tool lines into a single message that gets edited
-        # For DM top-level Slack messages, source.thread_id is None but the
-        # final reply will be threaded under the original message via reply_to.
-        # Use event_message_id as fallback so progress messages land in the
-        # same thread as the final response instead of going to the DM root.
-        _progress_thread_id = source.thread_id or event_message_id
+        # Accumulates tool lines into a single message that gets edited.
+        #
+        # Threading metadata is platform-specific:
+        # - Slack DM threading needs event_message_id fallback (reply thread)
+        # - Telegram uses message_thread_id only for forum topics; passing a
+        #   normal DM/group message id as thread_id causes send failures
+        # - Other platforms should use explicit source.thread_id only
+        if source.platform == Platform.SLACK:
+            _progress_thread_id = source.thread_id or event_message_id
+        else:
+            _progress_thread_id = source.thread_id
         _progress_metadata = {"thread_id": _progress_thread_id} if _progress_thread_id else None
 
         async def send_progress_messages():

hermes_cli/__init__.py

@@ -11,5 +11,5 @@ Provides subcommands for:
 - hermes cron          - Manage cron jobs
 """
 
-__version__ = "0.4.0"
-__release_date__ = "2026.3.23"
+__version__ = "0.5.0"
+__release_date__ = "2026.3.28"

hermes_cli/auth.py

@@ -160,7 +160,7 @@ PROVIDER_REGISTRY: Dict[str, ProviderConfig] = {
         id="alibaba",
         name="Alibaba Cloud (DashScope)",
         auth_type="api_key",
-        inference_base_url="https://dashscope-intl.aliyuncs.com/apps/anthropic",
+        inference_base_url="https://coding-intl.dashscope.aliyuncs.com/v1",
         api_key_env_vars=("DASHSCOPE_API_KEY",),
         base_url_env_var="DASHSCOPE_BASE_URL",
     ),
@@ -212,6 +212,14 @@ PROVIDER_REGISTRY: Dict[str, ProviderConfig] = {
         api_key_env_vars=("KILOCODE_API_KEY",),
         base_url_env_var="KILOCODE_BASE_URL",
     ),
+    "huggingface": ProviderConfig(
+        id="huggingface",
+        name="Hugging Face",
+        auth_type="api_key",
+        inference_base_url="https://router.huggingface.co/v1",
+        api_key_env_vars=("HF_TOKEN",),
+        base_url_env_var="HF_BASE_URL",
+    ),
 }
 
 
@@ -685,6 +693,7 @@ def resolve_provider(
         "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",
+        "hf": "huggingface", "hugging-face": "huggingface", "huggingface-hub": "huggingface",
         "go": "opencode-go", "opencode-go-sub": "opencode-go",
         "kilo": "kilocode", "kilo-code": "kilocode", "kilo-gateway": "kilocode",
     }
@@ -2012,7 +2021,8 @@ def _login_openai_codex(args, pconfig: ProviderConfig) -> None:
     config_path = _update_config_for_provider("openai-codex", creds.get("base_url", DEFAULT_CODEX_BASE_URL))
     print()
     print("Login successful!")
-    print("  Auth state: ~/.hermes/auth.json")
+    from hermes_constants import display_hermes_home as _dhh
+    print(f"  Auth state: {_dhh()}/auth.json")
     print(f"  Config updated: {config_path} (model.provider=openai-codex)")
 
 

hermes_cli/banner.py

@@ -258,7 +258,7 @@ def build_welcome_banner(console: Console, model: str, cwd: str,
         get_toolset_for_tool: Callable to map tool name -> toolset name.
         context_length: Model's context window size in tokens.
     """
-    from model_tools import check_tool_availability
+    from model_tools import check_tool_availability, TOOLSET_REQUIREMENTS
     if get_toolset_for_tool is None:
         from model_tools import get_toolset_for_tool
 
@@ -267,8 +267,18 @@ def build_welcome_banner(console: Console, model: str, cwd: str,
 
     _, unavailable_toolsets = check_tool_availability(quiet=True)
     disabled_tools = set()
+    # Tools whose toolset has a check_fn are lazy-initialized (e.g. honcho,
+    # homeassistant) — they show as unavailable at banner time because the
+    # check hasn't run yet, but they aren't misconfigured.
+    lazy_tools = set()
     for item in unavailable_toolsets:
-        disabled_tools.update(item.get("tools", []))
+        toolset_name = item.get("name", "")
+        ts_req = TOOLSET_REQUIREMENTS.get(toolset_name, {})
+        tools_in_ts = item.get("tools", [])
+        if ts_req.get("check_fn"):
+            lazy_tools.update(tools_in_ts)
+        else:
+            disabled_tools.update(tools_in_ts)
 
     layout_table = Table.grid(padding=(0, 2))
     layout_table.add_column("left", justify="center")
@@ -328,6 +338,8 @@ def build_welcome_banner(console: Console, model: str, cwd: str,
         for name in sorted(tool_names):
             if name in disabled_tools:
                 colored_names.append(f"[red]{name}[/]")
+            elif name in lazy_tools:
+                colored_names.append(f"[yellow]{name}[/]")
             else:
                 colored_names.append(f"[{text}]{name}[/]")
 
@@ -347,6 +359,8 @@ def build_welcome_banner(console: Console, model: str, cwd: str,
                     colored_names.append("[dim]...[/]")
                 elif name in disabled_tools:
                     colored_names.append(f"[red]{name}[/]")
+                elif name in lazy_tools:
+                    colored_names.append(f"[yellow]{name}[/]")
                 else:
                     colored_names.append(f"[{text}]{name}[/]")
             tools_str = ", ".join(colored_names)
@@ -403,6 +417,15 @@ def build_welcome_banner(console: Console, model: str, cwd: str,
     if mcp_connected:
         summary_parts.append(f"{mcp_connected} MCP servers")
     summary_parts.append("/help for commands")
+    # Show active profile name when not 'default'
+    try:
+        from hermes_cli.profiles import get_active_profile_name
+        _profile_name = get_active_profile_name()
+        if _profile_name and _profile_name != "default":
+            right_lines.append(f"[bold {accent}]Profile:[/] [{text}]{_profile_name}[/]")
+    except Exception:
+        pass  # Never break the banner over a profiles.py bug
+
     right_lines.append(f"[dim {dim}]{' · '.join(summary_parts)}[/]")
 
     # Update check — use prefetched result if available

hermes_cli/callbacks.py

@@ -12,6 +12,7 @@ import getpass
 
 from hermes_cli.banner import cprint, _DIM, _RST
 from hermes_cli.config import save_env_value_secure
+from hermes_constants import display_hermes_home
 
 
 def clarify_callback(cli, question, choices):
@@ -131,7 +132,8 @@ def prompt_for_secret(cli, var_name: str, prompt: str, metadata=None) -> dict:
             }
 
         stored = save_env_value_secure(var_name, value)
-        cprint(f"\n{_DIM}  ✓ Stored secret in ~/.hermes/.env as {var_name}{_RST}")
+        _dhh = display_hermes_home()
+        cprint(f"\n{_DIM}  ✓ Stored secret in {_dhh}/.env as {var_name}{_RST}")
         return {
             **stored,
             "skipped": False,
@@ -183,7 +185,8 @@ def prompt_for_secret(cli, var_name: str, prompt: str, metadata=None) -> dict:
                 }
 
             stored = save_env_value_secure(var_name, value)
-            cprint(f"\n{_DIM}  ✓ Stored secret in ~/.hermes/.env as {var_name}{_RST}")
+            _dhh = display_hermes_home()
+            cprint(f"\n{_DIM}  ✓ Stored secret in {_dhh}/.env as {var_name}{_RST}")
             return {
                 **stored,
                 "skipped": False,

hermes_cli/config.py

@@ -135,9 +135,16 @@ def ensure_hermes_home():
 
 DEFAULT_CONFIG = {
     "model": "anthropic/claude-opus-4.6",
+    "fallback_providers": [],
     "toolsets": ["hermes-cli"],
     "agent": {
         "max_turns": 90,
+        # Tool-use enforcement: injects system prompt guidance that tells the
+        # model to actually call tools instead of describing intended actions.
+        # Values: "auto" (default — applies to gpt/codex models), true/false
+        # (force on/off for all models), or a list of model-name substrings
+        # to match (e.g. ["gpt", "codex", "gemini", "qwen"]).
+        "tool_use_enforcement": "auto",
     },
     
     "terminal": {
@@ -221,42 +228,49 @@ DEFAULT_CONFIG = {
             "model": "",
             "base_url": "",
             "api_key": "",
+            "timeout": 30,         # seconds — increase for slow local models
         },
         "compression": {
             "provider": "auto",
             "model": "",
             "base_url": "",
             "api_key": "",
+            "timeout": 120,        # seconds — compression summarises large contexts; increase for local models
         },
         "session_search": {
             "provider": "auto",
             "model": "",
             "base_url": "",
             "api_key": "",
+            "timeout": 30,
         },
         "skills_hub": {
             "provider": "auto",
             "model": "",
             "base_url": "",
             "api_key": "",
+            "timeout": 30,
         },
         "approval": {
             "provider": "auto",
             "model": "",           # fast/cheap model recommended (e.g. gemini-flash, haiku)
             "base_url": "",
             "api_key": "",
+            "timeout": 30,
         },
         "mcp": {
             "provider": "auto",
             "model": "",
             "base_url": "",
             "api_key": "",
+            "timeout": 30,
         },
         "flush_memories": {
             "provider": "auto",
             "model": "",
             "base_url": "",
             "api_key": "",
+            "timeout": 30,
         },
     },
     
@@ -353,6 +367,13 @@ DEFAULT_CONFIG = {
     # Never saved to sessions, logs, or trajectories.
     "prefill_messages_file": "",
     
+    # Skills — external skill directories for sharing skills across tools/agents.
+    # Each path is expanded (~, ${VAR}) and resolved.  Read-only — skill creation
+    # always goes to ~/.hermes/skills/.
+    "skills": {
+        "external_dirs": [],   # e.g. ["~/.agents/skills", "/shared/team-skills"]
+    },
+
     # Honcho AI-native memory -- reads ~/.honcho/config.json as single source of truth.
     # This section is only needed for hermes-specific overrides; everything else
     # (apiKey, workspace, peerName, sessions, enabled) comes from the global config.
@@ -408,6 +429,12 @@ DEFAULT_CONFIG = {
         },
     },
 
+    "cron": {
+        # Wrap delivered cron responses with a header (task name) and footer
+        # ("The agent cannot see this message").  Set to false for clean output.
+        "wrap_response": True,
+    },
+
     # Config schema version - bump this when adding new required fields
     "_config_version": 10,
 }
@@ -547,14 +574,14 @@ OPTIONAL_ENV_VARS = {
         "category": "provider",
     },
     "DASHSCOPE_API_KEY": {
-        "description": "Alibaba Cloud DashScope API key for Qwen models",
+        "description": "Alibaba Cloud DashScope API key (Qwen + multi-provider models)",
         "prompt": "DashScope API Key",
         "url": "https://modelstudio.console.alibabacloud.com/",
         "password": True,
         "category": "provider",
     },
     "DASHSCOPE_BASE_URL": {
-        "description": "Custom DashScope base URL (default: international endpoint)",
+        "description": "Custom DashScope base URL (default: coding-intl OpenAI-compat endpoint)",
         "prompt": "DashScope Base URL",
         "url": "",
         "password": False,
@@ -593,8 +620,31 @@ OPTIONAL_ENV_VARS = {
         "category": "provider",
         "advanced": True,
     },
+    "HF_TOKEN": {
+        "description": "Hugging Face token for Inference Providers (20+ open models via router.huggingface.co)",
+        "prompt": "Hugging Face Token",
+        "url": "https://huggingface.co/settings/tokens",
+        "password": True,
+        "category": "provider",
+    },
+    "HF_BASE_URL": {
+        "description": "Hugging Face Inference Providers base URL override",
+        "prompt": "HF base URL (leave empty for default)",
+        "url": None,
+        "password": False,
+        "category": "provider",
+        "advanced": True,
+    },
 
     # ── Tool API keys ──
+    "EXA_API_KEY": {
+        "description": "Exa API key for AI-native web search and contents",
+        "prompt": "Exa API key",
+        "url": "https://exa.ai/",
+        "tools": ["web_search", "web_extract"],
+        "password": True,
+        "category": "tool",
+    },
     "PARALLEL_API_KEY": {
         "description": "Parallel API key for AI-native web search and extract",
         "prompt": "Parallel API key",
@@ -781,6 +831,20 @@ OPTIONAL_ENV_VARS = {
         "password": False,
         "category": "messaging",
     },
+    "MATTERMOST_REQUIRE_MENTION": {
+        "description": "Require @mention in Mattermost channels (default: true). Set to false to respond to all messages.",
+        "prompt": "Require @mention in channels",
+        "url": None,
+        "password": False,
+        "category": "messaging",
+    },
+    "MATTERMOST_FREE_RESPONSE_CHANNELS": {
+        "description": "Comma-separated Mattermost channel IDs where bot responds without @mention",
+        "prompt": "Free-response channel IDs (comma-separated)",
+        "url": None,
+        "password": False,
+        "category": "messaging",
+    },
     "MATRIX_HOMESERVER": {
         "description": "Matrix homeserver URL (e.g. https://matrix.example.org)",
         "prompt": "Matrix homeserver URL",
@@ -1651,6 +1715,7 @@ def show_config():
     keys = [
         ("OPENROUTER_API_KEY", "OpenRouter"),
         ("VOICE_TOOLS_OPENAI_KEY", "OpenAI (STT/TTS)"),
+        ("EXA_API_KEY", "Exa"),
         ("PARALLEL_API_KEY", "Parallel"),
         ("FIRECRAWL_API_KEY", "Firecrawl"),
         ("TAVILY_API_KEY", "Tavily"),
@@ -1810,7 +1875,7 @@ def set_config_value(key: str, value: str):
     # Check if it's an API key (goes to .env)
     api_keys = [
         'OPENROUTER_API_KEY', 'OPENAI_API_KEY', 'ANTHROPIC_API_KEY', 'VOICE_TOOLS_OPENAI_KEY',
-        'PARALLEL_API_KEY', 'FIRECRAWL_API_KEY', 'FIRECRAWL_API_URL', 'TAVILY_API_KEY',
+        'EXA_API_KEY', 'PARALLEL_API_KEY', 'FIRECRAWL_API_KEY', 'FIRECRAWL_API_URL', 'TAVILY_API_KEY',
         'BROWSERBASE_API_KEY', 'BROWSERBASE_PROJECT_ID', 'BROWSER_USE_API_KEY',
         'FAL_KEY', 'TELEGRAM_BOT_TOKEN', 'DISCORD_BOT_TOKEN',
         'TERMINAL_SSH_HOST', 'TERMINAL_SSH_USER', 'TERMINAL_SSH_KEY',

hermes_cli/curses_ui.py

@@ -4,7 +4,7 @@ Used by `hermes tools` and `hermes skills` for interactive checklists.
 Provides a curses multi-select with keyboard navigation, plus a
 text-based numbered fallback for terminals without curses support.
 """
-from typing import List, Set
+from typing import Callable, List, Optional, Set
 
 from hermes_cli.colors import Colors, color
 
@@ -15,6 +15,7 @@ def curses_checklist(
     selected: Set[int],
     *,
     cancel_returns: Set[int] | None = None,
+    status_fn: Optional[Callable[[Set[int]], str]] = None,
 ) -> Set[int]:
     """Curses multi-select checklist. Returns set of selected indices.
 
@@ -23,6 +24,9 @@ def curses_checklist(
         items: Display labels for each row.
         selected: Indices that start checked (pre-selected).
         cancel_returns: Returned on ESC/q. Defaults to the original *selected*.
+        status_fn: Optional callback ``f(chosen_indices) -> str`` whose return
+            value is rendered on the bottom row of the terminal.  Use this for
+            live aggregate info (e.g. estimated token counts).
     """
     if cancel_returns is None:
         cancel_returns = set(selected)
@@ -47,6 +51,9 @@ def curses_checklist(
                 stdscr.clear()
                 max_y, max_x = stdscr.getmaxyx()
 
+                # Reserve bottom row for status bar when status_fn provided
+                footer_rows = 1 if status_fn else 0
+
                 # Header
                 try:
                     hattr = curses.A_BOLD
@@ -62,7 +69,7 @@ def curses_checklist(
                     pass
 
                 # Scrollable item list
-                visible_rows = max_y - 3
+                visible_rows = max_y - 3 - footer_rows
                 if cursor < scroll_offset:
                     scroll_offset = cursor
                 elif cursor >= scroll_offset + visible_rows:
@@ -72,7 +79,7 @@ def curses_checklist(
                     range(scroll_offset, min(len(items), scroll_offset + visible_rows))
                 ):
                     y = draw_i + 3
-                    if y >= max_y - 1:
+                    if y >= max_y - 1 - footer_rows:
                         break
                     check = "✓" if i in chosen else " "
                     arrow = "→" if i == cursor else " "
@@ -87,6 +94,20 @@ def curses_checklist(
                     except curses.error:
                         pass
 
+                # Status bar (bottom row, right-aligned)
+                if status_fn:
+                    try:
+                        status_text = status_fn(chosen)
+                        if status_text:
+                            # Right-align on the bottom row
+                            sx = max(0, max_x - len(status_text) - 1)
+                            sattr = curses.A_DIM
+                            if curses.has_colors():
+                                sattr |= curses.color_pair(3)
+                            stdscr.addnstr(max_y - 1, sx, status_text, max_x - sx - 1, sattr)
+                    except curses.error:
+                        pass
+
                 stdscr.refresh()
                 key = stdscr.getch()
 
@@ -107,7 +128,7 @@ def curses_checklist(
         return result_holder[0] if result_holder[0] is not None else cancel_returns
 
     except Exception:
-        return _numbered_fallback(title, items, selected, cancel_returns)
+        return _numbered_fallback(title, items, selected, cancel_returns, status_fn)
 
 
 def _numbered_fallback(
@@ -115,6 +136,7 @@ def _numbered_fallback(
     items: List[str],
     selected: Set[int],
     cancel_returns: Set[int],
+    status_fn: Optional[Callable[[Set[int]], str]] = None,
 ) -> Set[int]:
     """Text-based toggle fallback for terminals without curses."""
     chosen = set(selected)
@@ -125,6 +147,10 @@ def _numbered_fallback(
         for i, label in enumerate(items):
             marker = color("[✓]", Colors.GREEN) if i in chosen else "[ ]"
             print(f"  {marker} {i + 1:>2}. {label}")
+        if status_fn:
+            status_text = status_fn(chosen)
+            if status_text:
+                print(color(f"\n  {status_text}", Colors.DIM))
         print()
         try:
             val = input(color("  Toggle # (or Enter to confirm): ", Colors.DIM)).strip()

hermes_cli/doctor.py

@@ -10,9 +10,11 @@ import subprocess
 import shutil
 
 from hermes_cli.config import get_project_root, get_hermes_home, get_env_path
+from hermes_constants import display_hermes_home
 
 PROJECT_ROOT = get_project_root()
 HERMES_HOME = get_hermes_home()
+_DHH = display_hermes_home()  # user-facing display path (e.g. ~/.hermes or ~/.hermes/profiles/coder)
 
 # Load environment variables from ~/.hermes/.env so API key checks work
 from dotenv import load_dotenv
@@ -56,7 +58,7 @@ def _honcho_is_configured_for_doctor() -> bool:
         from honcho_integration.client import HonchoClientConfig
 
         cfg = HonchoClientConfig.from_global_config()
-        return bool(cfg.enabled and cfg.api_key)
+        return bool(cfg.enabled and (cfg.api_key or cfg.base_url))
     except Exception:
         return False
 
@@ -209,14 +211,14 @@ def run_doctor(args):
     # Check ~/.hermes/.env (primary location for user config)
     env_path = HERMES_HOME / '.env'
     if env_path.exists():
-        check_ok("~/.hermes/.env file exists")
+        check_ok(f"{_DHH}/.env file exists")
         
         # Check for common issues
         content = env_path.read_text()
         if _has_provider_env_config(content):
             check_ok("API key or custom endpoint configured")
         else:
-            check_warn("No API key found in ~/.hermes/.env")
+            check_warn(f"No API key found in {_DHH}/.env")
             issues.append("Run 'hermes setup' to configure API keys")
     else:
         # Also check project root as fallback
@@ -224,11 +226,11 @@ def run_doctor(args):
         if fallback_env.exists():
             check_ok(".env file exists (in project directory)")
         else:
-            check_fail("~/.hermes/.env file missing")
+            check_fail(f"{_DHH}/.env file missing")
             if should_fix:
                 env_path.parent.mkdir(parents=True, exist_ok=True)
                 env_path.touch()
-                check_ok("Created empty ~/.hermes/.env")
+                check_ok(f"Created empty {_DHH}/.env")
                 check_info("Run 'hermes setup' to configure API keys")
                 fixed_count += 1
             else:
@@ -238,7 +240,7 @@ def run_doctor(args):
     # Check ~/.hermes/config.yaml (primary) or project cli-config.yaml (fallback)
     config_path = HERMES_HOME / 'config.yaml'
     if config_path.exists():
-        check_ok("~/.hermes/config.yaml exists")
+        check_ok(f"{_DHH}/config.yaml exists")
     else:
         fallback_config = PROJECT_ROOT / 'cli-config.yaml'
         if fallback_config.exists():
@@ -248,11 +250,11 @@ def run_doctor(args):
             if should_fix and example_config.exists():
                 config_path.parent.mkdir(parents=True, exist_ok=True)
                 shutil.copy2(str(example_config), str(config_path))
-                check_ok("Created ~/.hermes/config.yaml from cli-config.yaml.example")
+                check_ok(f"Created {_DHH}/config.yaml from cli-config.yaml.example")
                 fixed_count += 1
             elif should_fix:
                 check_warn("config.yaml not found and no example to copy from")
-                manual_issues.append("Create ~/.hermes/config.yaml manually")
+                manual_issues.append(f"Create {_DHH}/config.yaml manually")
             else:
                 check_warn("config.yaml not found", "(using defaults)")
     
@@ -294,28 +296,28 @@ def run_doctor(args):
     
     hermes_home = HERMES_HOME
     if hermes_home.exists():
-        check_ok("~/.hermes directory exists")
+        check_ok(f"{_DHH} directory exists")
     else:
         if should_fix:
             hermes_home.mkdir(parents=True, exist_ok=True)
-            check_ok("Created ~/.hermes directory")
+            check_ok(f"Created {_DHH} directory")
             fixed_count += 1
         else:
-            check_warn("~/.hermes not found", "(will be created on first use)")
+            check_warn(f"{_DHH} not found", "(will be created on first use)")
     
     # Check expected subdirectories
     expected_subdirs = ["cron", "sessions", "logs", "skills", "memories"]
     for subdir_name in expected_subdirs:
         subdir_path = hermes_home / subdir_name
         if subdir_path.exists():
-            check_ok(f"~/.hermes/{subdir_name}/ exists")
+            check_ok(f"{_DHH}/{subdir_name}/ exists")
         else:
             if should_fix:
                 subdir_path.mkdir(parents=True, exist_ok=True)
-                check_ok(f"Created ~/.hermes/{subdir_name}/")
+                check_ok(f"Created {_DHH}/{subdir_name}/")
                 fixed_count += 1
             else:
-                check_warn(f"~/.hermes/{subdir_name}/ not found", "(will be created on first use)")
+                check_warn(f"{_DHH}/{subdir_name}/ not found", "(will be created on first use)")
     
     # Check for SOUL.md persona file
     soul_path = hermes_home / "SOUL.md"
@@ -324,11 +326,11 @@ def run_doctor(args):
         # Check if it's just the template comments (no real content)
         lines = [l for l in content.splitlines() if l.strip() and not l.strip().startswith(("<!--", "-->", "#"))]
         if lines:
-            check_ok("~/.hermes/SOUL.md exists (persona configured)")
+            check_ok(f"{_DHH}/SOUL.md exists (persona configured)")
         else:
-            check_info("~/.hermes/SOUL.md exists but is empty — edit it to customize personality")
+            check_info(f"{_DHH}/SOUL.md exists but is empty — edit it to customize personality")
     else:
-        check_warn("~/.hermes/SOUL.md not found", "(create it to give Hermes a custom personality)")
+        check_warn(f"{_DHH}/SOUL.md not found", "(create it to give Hermes a custom personality)")
         if should_fix:
             soul_path.parent.mkdir(parents=True, exist_ok=True)
             soul_path.write_text(
@@ -337,13 +339,13 @@ def run_doctor(args):
                 "You are Hermes, a helpful AI assistant.\n",
                 encoding="utf-8",
             )
-            check_ok("Created ~/.hermes/SOUL.md with basic template")
+            check_ok(f"Created {_DHH}/SOUL.md with basic template")
             fixed_count += 1
     
     # Check memory directory
     memories_dir = hermes_home / "memories"
     if memories_dir.exists():
-        check_ok("~/.hermes/memories/ directory exists")
+        check_ok(f"{_DHH}/memories/ directory exists")
         memory_file = memories_dir / "MEMORY.md"
         user_file = memories_dir / "USER.md"
         if memory_file.exists():
@@ -357,10 +359,10 @@ def run_doctor(args):
         else:
             check_info("USER.md not created yet (will be created when the agent first writes a memory)")
     else:
-        check_warn("~/.hermes/memories/ not found", "(will be created on first use)")
+        check_warn(f"{_DHH}/memories/ not found", "(will be created on first use)")
         if should_fix:
             memories_dir.mkdir(parents=True, exist_ok=True)
-            check_ok("Created ~/.hermes/memories/")
+            check_ok(f"Created {_DHH}/memories/")
             fixed_count += 1
     
     # Check SQLite session store
@@ -372,11 +374,11 @@ def run_doctor(args):
             cursor = conn.execute("SELECT COUNT(*) FROM sessions")
             count = cursor.fetchone()[0]
             conn.close()
-            check_ok(f"~/.hermes/state.db exists ({count} sessions)")
+            check_ok(f"{_DHH}/state.db exists ({count} sessions)")
         except Exception as e:
-            check_warn(f"~/.hermes/state.db exists but has issues: {e}")
+            check_warn(f"{_DHH}/state.db exists but has issues: {e}")
     else:
-        check_info("~/.hermes/state.db not created yet (will be created on first session)")
+        check_info(f"{_DHH}/state.db not created yet (will be created on first session)")
 
     _check_gateway_service_linger(issues)
     
@@ -691,7 +693,7 @@ def run_doctor(args):
     if github_token:
         check_ok("GitHub token configured (authenticated API access)")
     else:
-        check_warn("No GITHUB_TOKEN", "(60 req/hr rate limit — set in ~/.hermes/.env for better rates)")
+        check_warn("No GITHUB_TOKEN", f"(60 req/hr rate limit — set in {_DHH}/.env for better rates)")
 
     # =========================================================================
     # Honcho memory
@@ -708,8 +710,8 @@ def run_doctor(args):
             check_warn("Honcho config not found", "run: hermes honcho setup")
         elif not hcfg.enabled:
             check_info(f"Honcho disabled (set enabled: true in {_honcho_cfg_path} to activate)")
-        elif not hcfg.api_key:
-            check_fail("Honcho API key not set", "run: hermes honcho setup")
+        elif not (hcfg.api_key or hcfg.base_url):
+            check_fail("Honcho API key or base URL not set", "run: hermes honcho setup")
             issues.append("No Honcho API key — run 'hermes honcho setup'")
         else:
             from honcho_integration.client import get_honcho_client, reset_honcho_client
@@ -728,6 +730,53 @@ def run_doctor(args):
     except Exception as _e:
         check_warn("Honcho check failed", str(_e))
 
+    # =========================================================================
+    # Profiles
+    # =========================================================================
+    try:
+        from hermes_cli.profiles import list_profiles, _get_wrapper_dir, profile_exists
+        import re as _re
+
+        named_profiles = [p for p in list_profiles() if not p.is_default]
+        if named_profiles:
+            print()
+            print(color("◆ Profiles", Colors.CYAN, Colors.BOLD))
+            check_ok(f"{len(named_profiles)} profile(s) found")
+            wrapper_dir = _get_wrapper_dir()
+            for p in named_profiles:
+                parts = []
+                if p.gateway_running:
+                    parts.append("gateway running")
+                if p.model:
+                    parts.append(p.model[:30])
+                if not (p.path / "config.yaml").exists():
+                    parts.append("⚠ missing config")
+                if not (p.path / ".env").exists():
+                    parts.append("no .env")
+                wrapper = wrapper_dir / p.name
+                if not wrapper.exists():
+                    parts.append("no alias")
+                status = ", ".join(parts) if parts else "configured"
+                check_ok(f"  {p.name}: {status}")
+
+            # Check for orphan wrappers
+            if wrapper_dir.is_dir():
+                for wrapper in wrapper_dir.iterdir():
+                    if not wrapper.is_file():
+                        continue
+                    try:
+                        content = wrapper.read_text()
+                        if "hermes -p" in content:
+                            _m = _re.search(r"hermes -p (\S+)", content)
+                            if _m and not profile_exists(_m.group(1)):
+                                check_warn(f"Orphan alias: {wrapper.name} → profile '{_m.group(1)}' no longer exists")
+                    except Exception:
+                        pass
+    except ImportError:
+        pass
+    except Exception as _e:
+        logger.debug("Profile health check failed: %s", _e)
+
     # =========================================================================
     # Summary
     # =========================================================================

hermes_cli/gateway.py

@@ -15,6 +15,8 @@ from pathlib import Path
 PROJECT_ROOT = Path(__file__).parent.parent.resolve()
 
 from hermes_cli.config import get_env_value, get_hermes_home, save_env_value, is_managed, managed_error
+# display_hermes_home is imported lazily at call sites to avoid ImportError
+# when hermes_constants is cached from a pre-update version during `hermes update`.
 from hermes_cli.setup import (
     print_header, print_info, print_success, print_warning, print_error,
     prompt, prompt_choice, prompt_yes_no,
@@ -125,20 +127,43 @@ _SERVICE_BASE = "hermes-gateway"
 SERVICE_DESCRIPTION = "Hermes Agent Gateway - Messaging Platform Integration"
 
 
+def _profile_suffix() -> str:
+    """Derive a service-name suffix from the current HERMES_HOME.
+
+    Returns ``""`` for the default ``~/.hermes``, the profile name for
+    ``~/.hermes/profiles/<name>``, or a short hash for any other custom
+    HERMES_HOME path.
+    """
+    import hashlib
+    import re
+    from pathlib import Path as _Path
+    home = get_hermes_home().resolve()
+    default = (_Path.home() / ".hermes").resolve()
+    if home == default:
+        return ""
+    # Detect ~/.hermes/profiles/<name> pattern → use the profile name
+    profiles_root = (default / "profiles").resolve()
+    try:
+        rel = home.relative_to(profiles_root)
+        parts = rel.parts
+        if len(parts) == 1 and re.match(r"^[a-z0-9][a-z0-9_-]{0,63}$", parts[0]):
+            return parts[0]
+    except ValueError:
+        pass
+    # Fallback: short hash for arbitrary HERMES_HOME paths
+    return hashlib.sha256(str(home).encode()).hexdigest()[:8]
+
+
 def get_service_name() -> str:
     """Derive a systemd service name scoped to this HERMES_HOME.
 
     Default ``~/.hermes`` returns ``hermes-gateway`` (backward compatible).
-    Any other HERMES_HOME appends a short hash so multiple installations
-    can each have their own systemd service without conflicting.
+    Profile ``~/.hermes/profiles/coder`` returns ``hermes-gateway-coder``.
+    Any other HERMES_HOME appends a short hash for uniqueness.
     """
-    import hashlib
-    from pathlib import Path as _Path  # local import to avoid monkeypatch interference
-    home = get_hermes_home().resolve()
-    default = (_Path.home() / ".hermes").resolve()
-    if home == default:
+    suffix = _profile_suffix()
+    if not suffix:
         return _SERVICE_BASE
-    suffix = hashlib.sha256(str(home).encode()).hexdigest()[:8]
     return f"{_SERVICE_BASE}-{suffix}"
 
 
@@ -369,7 +394,14 @@ def print_systemd_linger_guidance() -> None:
         print("  sudo loginctl enable-linger $USER")
 
 def get_launchd_plist_path() -> Path:
-    return Path.home() / "Library" / "LaunchAgents" / "ai.hermes.gateway.plist"
+    """Return the launchd plist path, scoped per profile.
+
+    Default ``~/.hermes`` → ``ai.hermes.gateway.plist`` (backward compatible).
+    Profile ``~/.hermes/profiles/coder`` → ``ai.hermes.gateway-coder.plist``.
+    """
+    suffix = _profile_suffix()
+    name = f"ai.hermes.gateway-{suffix}" if suffix else "ai.hermes.gateway"
+    return Path.home() / "Library" / "LaunchAgents" / f"{name}.plist"
 
 def _detect_venv_dir() -> Path | None:
     """Detect the active virtualenv directory.
@@ -420,6 +452,17 @@ def get_hermes_cli_path() -> str:
 # Systemd (Linux)
 # =============================================================================
 
+def _build_user_local_paths(home: Path, path_entries: list[str]) -> list[str]:
+    """Return user-local bin dirs that exist and aren't already in *path_entries*."""
+    candidates = [
+        str(home / ".local" / "bin"),       # uv, uvx, pip-installed CLIs
+        str(home / ".cargo" / "bin"),        # Rust/cargo tools
+        str(home / "go" / "bin"),            # Go tools
+        str(home / ".npm-global" / "bin"),   # npm global packages
+    ]
+    return [p for p in candidates if p not in path_entries and Path(p).exists()]
+
+
 def generate_systemd_unit(system: bool = False, run_as_user: str | None = None) -> str:
     python_path = get_python_path()
     working_dir = str(PROJECT_ROOT)
@@ -434,13 +477,16 @@ def generate_systemd_unit(system: bool = False, run_as_user: str | None = None)
         resolved_node_dir = str(Path(resolved_node).resolve().parent)
         if resolved_node_dir not in path_entries:
             path_entries.append(resolved_node_dir)
-    path_entries.extend(["/usr/local/sbin", "/usr/local/bin", "/usr/sbin", "/usr/bin", "/sbin", "/bin"])
-    sane_path = ":".join(path_entries)
 
     hermes_home = str(get_hermes_home().resolve())
 
+    common_bin_paths = ["/usr/local/sbin", "/usr/local/bin", "/usr/sbin", "/usr/bin", "/sbin", "/bin"]
+
     if system:
         username, group_name, home_dir = _system_service_identity(run_as_user)
+        path_entries.extend(_build_user_local_paths(Path(home_dir), path_entries))
+        path_entries.extend(common_bin_paths)
+        sane_path = ":".join(path_entries)
         return f"""[Unit]
 Description={SERVICE_DESCRIPTION}
 After=network-online.target
@@ -472,6 +518,9 @@ StandardError=journal
 WantedBy=multi-user.target
 """
 
+    path_entries.extend(_build_user_local_paths(Path.home(), path_entries))
+    path_entries.extend(common_bin_paths)
+    sane_path = ":".join(path_entries)
     return f"""[Unit]
 Description={SERVICE_DESCRIPTION}
 After=network.target
@@ -752,18 +801,46 @@ def systemd_status(deep: bool = False, system: bool = False):
 # Launchd (macOS)
 # =============================================================================
 
+def get_launchd_label() -> str:
+    """Return the launchd service label, scoped per profile."""
+    suffix = _profile_suffix()
+    return f"ai.hermes.gateway-{suffix}" if suffix else "ai.hermes.gateway"
+
+
 def generate_launchd_plist() -> str:
     python_path = get_python_path()
     working_dir = str(PROJECT_ROOT)
+    hermes_home = str(get_hermes_home().resolve())
     log_dir = get_hermes_home() / "logs"
     log_dir.mkdir(parents=True, exist_ok=True)
-    
+    label = get_launchd_label()
+    # Build a sane PATH for the launchd plist.  launchd provides only a
+    # minimal default (/usr/bin:/bin:/usr/sbin:/sbin) which misses Homebrew,
+    # nvm, cargo, etc.  We prepend venv/bin and node_modules/.bin (matching
+    # the systemd unit), then capture the user's full shell PATH so every
+    # user-installed tool (node, ffmpeg, …) is reachable.
+    detected_venv = _detect_venv_dir()
+    venv_bin = str(detected_venv / "bin") if detected_venv else str(PROJECT_ROOT / "venv" / "bin")
+    venv_dir = str(detected_venv) if detected_venv else str(PROJECT_ROOT / "venv")
+    node_bin = str(PROJECT_ROOT / "node_modules" / ".bin")
+    # Resolve the directory containing the node binary (e.g. Homebrew, nvm)
+    # so it's explicitly in PATH even if the user's shell PATH changes later.
+    priority_dirs = [venv_bin, node_bin]
+    resolved_node = shutil.which("node")
+    if resolved_node:
+        resolved_node_dir = str(Path(resolved_node).resolve().parent)
+        if resolved_node_dir not in priority_dirs:
+            priority_dirs.append(resolved_node_dir)
+    sane_path = ":".join(
+        dict.fromkeys(priority_dirs + [p for p in os.environ.get("PATH", "").split(":") if p])
+    )
+
     return f"""<?xml version="1.0" encoding="UTF-8"?>
 <!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
 <plist version="1.0">
 <dict>
     <key>Label</key>
-    <string>ai.hermes.gateway</string>
+    <string>{label}</string>
     
     <key>ProgramArguments</key>
     <array>
@@ -778,6 +855,16 @@ def generate_launchd_plist() -> str:
     <key>WorkingDirectory</key>
     <string>{working_dir}</string>
     
+    <key>EnvironmentVariables</key>
+    <dict>
+        <key>PATH</key>
+        <string>{sane_path}</string>
+        <key>VIRTUAL_ENV</key>
+        <string>{venv_dir}</string>
+        <key>HERMES_HOME</key>
+        <string>{hermes_home}</string>
+    </dict>
+    
     <key>RunAtLoad</key>
     <true/>
     
@@ -850,7 +937,8 @@ def launchd_install(force: bool = False):
     print()
     print("Next steps:")
     print("  hermes gateway status             # Check status")
-    print("  tail -f ~/.hermes/logs/gateway.log  # View logs")
+    from hermes_constants import display_hermes_home as _dhh
+    print(f"  tail -f {_dhh()}/logs/gateway.log  # View logs")
 
 def launchd_uninstall():
     plist_path = get_launchd_plist_path()
@@ -863,20 +951,33 @@ def launchd_uninstall():
     print("✓ Service uninstalled")
 
 def launchd_start():
-    refresh_launchd_plist_if_needed()
     plist_path = get_launchd_plist_path()
+    label = get_launchd_label()
+
+    # Self-heal if the plist is missing entirely (e.g., manual cleanup, failed upgrade)
+    if not plist_path.exists():
+        print("↻ launchd plist missing; regenerating service definition")
+        plist_path.parent.mkdir(parents=True, exist_ok=True)
+        plist_path.write_text(generate_launchd_plist(), encoding="utf-8")
+        subprocess.run(["launchctl", "load", str(plist_path)], check=True)
+        subprocess.run(["launchctl", "start", label], check=True)
+        print("✓ Service started")
+        return
+
+    refresh_launchd_plist_if_needed()
     try:
-        subprocess.run(["launchctl", "start", "ai.hermes.gateway"], check=True)
+        subprocess.run(["launchctl", "start", label], check=True)
     except subprocess.CalledProcessError as e:
-        if e.returncode != 3 or not plist_path.exists():
+        if e.returncode != 3:
             raise
         print("↻ launchd job was unloaded; reloading service definition")
         subprocess.run(["launchctl", "load", str(plist_path)], check=True)
-        subprocess.run(["launchctl", "start", "ai.hermes.gateway"], check=True)
+        subprocess.run(["launchctl", "start", label], check=True)
     print("✓ Service started")
 
 def launchd_stop():
-    subprocess.run(["launchctl", "stop", "ai.hermes.gateway"], check=True)
+    label = get_launchd_label()
+    subprocess.run(["launchctl", "stop", label], check=True)
     print("✓ Service stopped")
 
 def _wait_for_gateway_exit(timeout: float = 10.0, force_after: float = 5.0):
@@ -931,8 +1032,9 @@ def launchd_restart():
 
 def launchd_status(deep: bool = False):
     plist_path = get_launchd_plist_path()
+    label = get_launchd_label()
     result = subprocess.run(
-        ["launchctl", "list", "ai.hermes.gateway"],
+        ["launchctl", "list", label],
         capture_output=True,
         text=True
     )
@@ -1437,7 +1539,7 @@ def _is_service_running() -> bool:
         return False
     elif is_macos() and get_launchd_plist_path().exists():
         result = subprocess.run(
-            ["launchctl", "list", "ai.hermes.gateway"],
+            ["launchctl", "list", get_launchd_label()],
             capture_output=True, text=True
         )
         return result.returncode == 0

hermes_cli/mcp_config.py

@@ -24,6 +24,7 @@ from hermes_cli.config import (
     get_hermes_home,  # noqa: F401 — used by test mocks
 )
 from hermes_cli.colors import Colors, color
+from hermes_constants import display_hermes_home
 
 logger = logging.getLogger(__name__)
 
@@ -244,7 +245,7 @@ def cmd_mcp_add(args):
                     api_key = _prompt("API key / Bearer token", password=True)
                     if api_key:
                         save_env_value(env_key, api_key)
-                        _success(f"Saved to ~/.hermes/.env as {env_key}")
+                        _success(f"Saved to {display_hermes_home()}/.env as {env_key}")
 
                 # Set header with env var interpolation
                 if api_key or existing_key:
@@ -332,7 +333,7 @@ def cmd_mcp_add(args):
     _save_mcp_server(name, server_config)
 
     print()
-    _success(f"Saved '{name}' to ~/.hermes/config.yaml ({tool_count}/{total} tools enabled)")
+    _success(f"Saved '{name}' to {display_hermes_home()}/config.yaml ({tool_count}/{total} tools enabled)")
     _info("Start a new session to use these tools.")
 
 
@@ -607,6 +608,11 @@ def mcp_command(args):
     """Main dispatcher for ``hermes mcp`` subcommands."""
     action = getattr(args, "mcp_action", None)
 
+    if action == "serve":
+        from mcp_serve import run_mcp_server
+        run_mcp_server(verbose=getattr(args, "verbose", False))
+        return
+
     handlers = {
         "add": cmd_mcp_add,
         "remove": cmd_mcp_remove,
@@ -625,6 +631,7 @@ def mcp_command(args):
         # No subcommand — show list
         cmd_mcp_list()
         print(color("  Commands:", Colors.CYAN))
+        _info("hermes mcp serve                              Run as MCP server")
         _info("hermes mcp add <name> --url <endpoint>        Add an MCP server")
         _info("hermes mcp add <name> --command <cmd>         Add a stdio server")
         _info("hermes mcp remove <name>                      Remove a server")

hermes_cli/models.py

@@ -35,6 +35,8 @@ OPENROUTER_MODELS: list[tuple[str, str]] = [
     ("openai/gpt-5.3-codex",            ""),
     ("google/gemini-3-pro-preview",     ""),
     ("google/gemini-3-flash-preview",   ""),
+    ("google/gemini-3.1-pro-preview",     ""),
+    ("google/gemini-3.1-flash-lite-preview",   ""),
     ("qwen/qwen3.5-plus-02-15",         ""),
     ("qwen/qwen3.5-35b-a3b",            ""),
     ("stepfun/step-3.5-flash",          ""),
@@ -62,6 +64,8 @@ _PROVIDER_MODELS: dict[str, list[str]] = {
         "openai/gpt-5.3-codex",
         "google/gemini-3-pro-preview",
         "google/gemini-3-flash-preview",
+        "google/gemini-3.1-pro-preview",
+        "google/gemini-3.1-flash-lite-preview",
         "qwen/qwen3.5-plus-02-15",
         "qwen/qwen3.5-35b-a3b",
         "stepfun/step-3.5-flash",
@@ -208,14 +212,31 @@ _PROVIDER_MODELS: dict[str, list[str]] = {
         "google/gemini-3-pro-preview",
         "google/gemini-3-flash-preview",
     ],
+    # Alibaba DashScope Coding platform (coding-intl) — default endpoint.
+    # Supports Qwen models + third-party providers (GLM, Kimi, MiniMax).
+    # Users with classic DashScope keys should override DASHSCOPE_BASE_URL
+    # to https://dashscope-intl.aliyuncs.com/compatible-mode/v1 (OpenAI-compat)
+    # or https://dashscope-intl.aliyuncs.com/apps/anthropic (Anthropic-compat).
     "alibaba": [
         "qwen3.5-plus",
-        "qwen3-max",
         "qwen3-coder-plus",
         "qwen3-coder-next",
-        "qwen-plus-latest",
-        "qwen3.5-flash",
-        "qwen-vl-max",
+        # Third-party models available on coding-intl
+        "glm-5",
+        "glm-4.7",
+        "kimi-k2.5",
+        "MiniMax-M2.5",
+    ],
+    # Curated HF model list — only agentic models that map to OpenRouter defaults.
+    "huggingface": [
+        "Qwen/Qwen3.5-397B-A17B",
+        "Qwen/Qwen3.5-35B-A3B",
+        "deepseek-ai/DeepSeek-V3.2",
+        "moonshotai/Kimi-K2.5",
+        "MiniMaxAI/MiniMax-M2.5",
+        "zai-org/GLM-5",
+        "XiaomiMiMo/MiMo-V2-Flash",
+        "moonshotai/Kimi-K2-Thinking",
     ],
 }
 
@@ -236,6 +257,7 @@ _PROVIDER_LABELS = {
     "ai-gateway": "AI Gateway",
     "kilocode": "Kilo Code",
     "alibaba": "Alibaba Cloud (DashScope)",
+    "huggingface": "Hugging Face",
     "custom": "Custom endpoint",
 }
 
@@ -271,6 +293,9 @@ _PROVIDER_ALIASES = {
     "aliyun": "alibaba",
     "qwen": "alibaba",
     "alibaba-cloud": "alibaba",
+    "hf": "huggingface",
+    "hugging-face": "huggingface",
+    "huggingface-hub": "huggingface",
 }
 
 
@@ -304,7 +329,7 @@ def list_available_providers() -> list[dict[str, str]]:
     # Canonical providers in display order
     _PROVIDER_ORDER = [
         "openrouter", "nous", "openai-codex", "copilot", "copilot-acp",
-        "zai", "kimi-coding", "minimax", "minimax-cn", "kilocode", "anthropic", "alibaba",
+        "huggingface", "zai", "kimi-coding", "minimax", "minimax-cn", "kilocode", "anthropic", "alibaba",
         "opencode-zen", "opencode-go",
         "ai-gateway", "deepseek", "custom",
     ]

hermes_cli/plugins.py

@@ -68,6 +68,17 @@ def _env_enabled(name: str) -> bool:
     return os.getenv(name, "").strip().lower() in {"1", "true", "yes", "on"}
 
 
+def _get_disabled_plugins() -> set:
+    """Read the disabled plugins list from config.yaml."""
+    try:
+        from hermes_cli.config import load_config
+        config = load_config()
+        disabled = config.get("plugins", {}).get("disabled", [])
+        return set(disabled) if isinstance(disabled, list) else set()
+    except Exception:
+        return set()
+
+
 # ---------------------------------------------------------------------------
 # Data classes
 # ---------------------------------------------------------------------------
@@ -199,8 +210,15 @@ class PluginManager:
         # 3. Pip / entry-point plugins
         manifests.extend(self._scan_entry_points())
 
-        # Load each manifest
+        # Load each manifest (skip user-disabled plugins)
+        disabled = _get_disabled_plugins()
         for manifest in manifests:
+            if manifest.name in disabled:
+                loaded = LoadedPlugin(manifest=manifest, enabled=False)
+                loaded.error = "disabled via config"
+                self._plugins[manifest.name] = loaded
+                logger.debug("Skipping disabled plugin '%s'", manifest.name)
+                continue
             self._load_plugin(manifest)
 
         if manifests:
@@ -385,16 +403,23 @@ class PluginManager:
     # Hook invocation
     # -----------------------------------------------------------------------
 
-    def invoke_hook(self, hook_name: str, **kwargs: Any) -> None:
+    def invoke_hook(self, hook_name: str, **kwargs: Any) -> List[Any]:
         """Call all registered callbacks for *hook_name*.
 
         Each callback is wrapped in its own try/except so a misbehaving
         plugin cannot break the core agent loop.
+
+        Returns a list of non-``None`` return values from callbacks.
+        This allows hooks like ``pre_llm_call`` to contribute context
+        that the agent core can collect and inject.
         """
         callbacks = self._hooks.get(hook_name, [])
+        results: List[Any] = []
         for cb in callbacks:
             try:
-                cb(**kwargs)
+                ret = cb(**kwargs)
+                if ret is not None:
+                    results.append(ret)
             except Exception as exc:
                 logger.warning(
                     "Hook '%s' callback %s raised: %s",
@@ -402,6 +427,7 @@ class PluginManager:
                     getattr(cb, "__name__", repr(cb)),
                     exc,
                 )
+        return results
 
     # -----------------------------------------------------------------------
     # Introspection
@@ -446,9 +472,12 @@ def discover_plugins() -> None:
     get_plugin_manager().discover_and_load()
 
 
-def invoke_hook(hook_name: str, **kwargs: Any) -> None:
-    """Invoke a lifecycle hook on all loaded plugins."""
-    get_plugin_manager().invoke_hook(hook_name, **kwargs)
+def invoke_hook(hook_name: str, **kwargs: Any) -> List[Any]:
+    """Invoke a lifecycle hook on all loaded plugins.
+
+    Returns a list of non-``None`` return values from plugin callbacks.
+    """
+    return get_plugin_manager().invoke_hook(hook_name, **kwargs)
 
 
 def get_plugin_tool_names() -> Set[str]:

hermes_cli/plugins_cmd.py

@@ -374,6 +374,73 @@ def cmd_remove(name: str) -> None:
     _display_removed(name, plugins_dir)
 
 
+def _get_disabled_set() -> set:
+    """Read the disabled plugins set from config.yaml."""
+    try:
+        from hermes_cli.config import load_config
+        config = load_config()
+        disabled = config.get("plugins", {}).get("disabled", [])
+        return set(disabled) if isinstance(disabled, list) else set()
+    except Exception:
+        return set()
+
+
+def _save_disabled_set(disabled: set) -> None:
+    """Write the disabled plugins list to config.yaml."""
+    from hermes_cli.config import load_config, save_config
+    config = load_config()
+    if "plugins" not in config:
+        config["plugins"] = {}
+    config["plugins"]["disabled"] = sorted(disabled)
+    save_config(config)
+
+
+def cmd_enable(name: str) -> None:
+    """Enable a previously disabled plugin."""
+    from rich.console import Console
+
+    console = Console()
+    plugins_dir = _plugins_dir()
+
+    # Verify the plugin exists
+    target = plugins_dir / name
+    if not target.is_dir():
+        console.print(f"[red]Plugin '{name}' is not installed.[/red]")
+        sys.exit(1)
+
+    disabled = _get_disabled_set()
+    if name not in disabled:
+        console.print(f"[dim]Plugin '{name}' is already enabled.[/dim]")
+        return
+
+    disabled.discard(name)
+    _save_disabled_set(disabled)
+    console.print(f"[green]✓[/green] Plugin [bold]{name}[/bold] enabled. Takes effect on next session.")
+
+
+def cmd_disable(name: str) -> None:
+    """Disable a plugin without removing it."""
+    from rich.console import Console
+
+    console = Console()
+    plugins_dir = _plugins_dir()
+
+    # Verify the plugin exists
+    target = plugins_dir / name
+    if not target.is_dir():
+        console.print(f"[red]Plugin '{name}' is not installed.[/red]")
+        sys.exit(1)
+
+    disabled = _get_disabled_set()
+    if name in disabled:
+        console.print(f"[dim]Plugin '{name}' is already disabled.[/dim]")
+        return
+
+    disabled.add(name)
+    _save_disabled_set(disabled)
+    console.print(f"[yellow]⊘[/yellow] Plugin [bold]{name}[/bold] disabled. Takes effect on next session.")
+
+
 def cmd_list() -> None:
     """List installed plugins."""
     from rich.console import Console
@@ -393,8 +460,11 @@ def cmd_list() -> None:
         console.print("[dim]Install with:[/dim] hermes plugins install owner/repo")
         return
 
+    disabled = _get_disabled_set()
+
     table = Table(title="Installed Plugins", show_lines=False)
     table.add_column("Name", style="bold")
+    table.add_column("Status")
     table.add_column("Version", style="dim")
     table.add_column("Description")
     table.add_column("Source", style="dim")
@@ -420,11 +490,86 @@ def cmd_list() -> None:
         if (d / ".git").exists():
             source = "git"
 
-        table.add_row(name, str(version), description, source)
+        is_disabled = name in disabled or d.name in disabled
+        status = "[red]disabled[/red]" if is_disabled else "[green]enabled[/green]"
+        table.add_row(name, status, str(version), description, source)
 
     console.print()
     console.print(table)
     console.print()
+    console.print("[dim]Interactive toggle:[/dim] hermes plugins")
+    console.print("[dim]Enable/disable:[/dim] hermes plugins enable/disable <name>")
+
+
+def cmd_toggle() -> None:
+    """Interactive curses checklist to enable/disable installed plugins."""
+    from rich.console import Console
+
+    try:
+        import yaml
+    except ImportError:
+        yaml = None
+
+    console = Console()
+    plugins_dir = _plugins_dir()
+
+    dirs = sorted(d for d in plugins_dir.iterdir() if d.is_dir())
+    if not dirs:
+        console.print("[dim]No plugins installed.[/dim]")
+        console.print("[dim]Install with:[/dim] hermes plugins install owner/repo")
+        return
+
+    disabled = _get_disabled_set()
+
+    # Build items list: "name — description" for display
+    names = []
+    labels = []
+    selected = set()
+
+    for i, d in enumerate(dirs):
+        manifest_file = d / "plugin.yaml"
+        name = d.name
+        description = ""
+
+        if manifest_file.exists() and yaml:
+            try:
+                with open(manifest_file) as f:
+                    manifest = yaml.safe_load(f) or {}
+                name = manifest.get("name", d.name)
+                description = manifest.get("description", "")
+            except Exception:
+                pass
+
+        names.append(name)
+        label = f"{name} — {description}" if description else name
+        labels.append(label)
+
+        if name not in disabled and d.name not in disabled:
+            selected.add(i)
+
+    from hermes_cli.curses_ui import curses_checklist
+
+    result = curses_checklist(
+        title="Plugins — toggle enabled/disabled",
+        items=labels,
+        selected=selected,
+    )
+
+    # Compute new disabled set from deselected items
+    new_disabled = set()
+    for i, name in enumerate(names):
+        if i not in result:
+            new_disabled.add(name)
+
+    if new_disabled != disabled:
+        _save_disabled_set(new_disabled)
+        enabled_count = len(names) - len(new_disabled)
+        console.print(
+            f"\n[green]✓[/green] {enabled_count} enabled, {len(new_disabled)} disabled. "
+            f"Takes effect on next session."
+        )
+    else:
+        console.print("\n[dim]No changes.[/dim]")
 
 
 def plugins_command(args) -> None:
@@ -437,8 +582,14 @@ def plugins_command(args) -> None:
         cmd_update(args.name)
     elif action in ("remove", "rm", "uninstall"):
         cmd_remove(args.name)
-    elif action in ("list", "ls") or action is None:
+    elif action == "enable":
+        cmd_enable(args.name)
+    elif action == "disable":
+        cmd_disable(args.name)
+    elif action in ("list", "ls"):
         cmd_list()
+    elif action is None:
+        cmd_toggle()
     else:
         from rich.console import Console
 

hermes_cli/profiles.py

@@ -0,0 +1,906 @@
+"""
+Profile management for multiple isolated Hermes instances.
+
+Each profile is a fully independent HERMES_HOME directory with its own
+config.yaml, .env, memory, sessions, skills, gateway, cron, and logs.
+Profiles live under ``~/.hermes/profiles/<name>/`` by default.
+
+The "default" profile is ``~/.hermes`` itself — backward compatible,
+zero migration needed.
+
+Usage::
+
+    hermes profile create coder          # fresh profile + bundled skills
+    hermes profile create coder --clone  # also copy config, .env, SOUL.md
+    hermes profile create coder --clone-all  # full copy of source profile
+    coder chat                           # use via wrapper alias
+    hermes -p coder chat                 # or via flag
+    hermes profile use coder             # set as sticky default
+    hermes profile delete coder          # remove profile + alias + service
+"""
+
+import json
+import os
+import re
+import shutil
+import stat
+import subprocess
+import sys
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import List, Optional
+
+_PROFILE_ID_RE = re.compile(r"^[a-z0-9][a-z0-9_-]{0,63}$")
+
+# Directories bootstrapped inside every new profile
+_PROFILE_DIRS = [
+    "memories",
+    "sessions",
+    "skills",
+    "skins",
+    "logs",
+    "plans",
+    "workspace",
+    "cron",
+]
+
+# Files copied during --clone (if they exist in the source)
+_CLONE_CONFIG_FILES = [
+    "config.yaml",
+    ".env",
+    "SOUL.md",
+]
+
+# Runtime files stripped after --clone-all (shouldn't carry over)
+_CLONE_ALL_STRIP = [
+    "gateway.pid",
+    "gateway_state.json",
+    "processes.json",
+]
+
+# Names that cannot be used as profile aliases
+_RESERVED_NAMES = frozenset({
+    "hermes", "default", "test", "tmp", "root", "sudo",
+})
+
+# Hermes subcommands that cannot be used as profile names/aliases
+_HERMES_SUBCOMMANDS = frozenset({
+    "chat", "model", "gateway", "setup", "whatsapp", "login", "logout",
+    "status", "cron", "doctor", "config", "pairing", "skills", "tools",
+    "mcp", "sessions", "insights", "version", "update", "uninstall",
+    "profile", "plugins", "honcho", "acp",
+})
+
+
+# ---------------------------------------------------------------------------
+# Path helpers
+# ---------------------------------------------------------------------------
+
+def _get_profiles_root() -> Path:
+    """Return the directory where named profiles are stored.
+
+    Always ``~/.hermes/profiles/`` — anchored to the user's home,
+    NOT to the current HERMES_HOME (which may itself be a profile).
+    This ensures ``coder profile list`` can see all profiles.
+    """
+    return Path.home() / ".hermes" / "profiles"
+
+
+def _get_default_hermes_home() -> Path:
+    """Return the default (pre-profile) HERMES_HOME path."""
+    return Path.home() / ".hermes"
+
+
+def _get_active_profile_path() -> Path:
+    """Return the path to the sticky active_profile file."""
+    return _get_default_hermes_home() / "active_profile"
+
+
+def _get_wrapper_dir() -> Path:
+    """Return the directory for wrapper scripts."""
+    return Path.home() / ".local" / "bin"
+
+
+# ---------------------------------------------------------------------------
+# Validation
+# ---------------------------------------------------------------------------
+
+def validate_profile_name(name: str) -> None:
+    """Raise ``ValueError`` if *name* is not a valid profile identifier."""
+    if name == "default":
+        return  # special alias for ~/.hermes
+    if not _PROFILE_ID_RE.match(name):
+        raise ValueError(
+            f"Invalid profile name {name!r}. Must match "
+            f"[a-z0-9][a-z0-9_-]{{0,63}}"
+        )
+
+
+def get_profile_dir(name: str) -> Path:
+    """Resolve a profile name to its HERMES_HOME directory."""
+    if name == "default":
+        return _get_default_hermes_home()
+    return _get_profiles_root() / name
+
+
+def profile_exists(name: str) -> bool:
+    """Check whether a profile directory exists."""
+    if name == "default":
+        return True
+    return get_profile_dir(name).is_dir()
+
+
+# ---------------------------------------------------------------------------
+# Alias / wrapper script management
+# ---------------------------------------------------------------------------
+
+def check_alias_collision(name: str) -> Optional[str]:
+    """Return a human-readable collision message, or None if the name is safe.
+
+    Checks: reserved names, hermes subcommands, existing binaries in PATH.
+    """
+    if name in _RESERVED_NAMES:
+        return f"'{name}' is a reserved name"
+    if name in _HERMES_SUBCOMMANDS:
+        return f"'{name}' conflicts with a hermes subcommand"
+
+    # Check existing commands in PATH
+    wrapper_dir = _get_wrapper_dir()
+    try:
+        result = subprocess.run(
+            ["which", name], capture_output=True, text=True, timeout=5,
+        )
+        if result.returncode == 0:
+            existing_path = result.stdout.strip()
+            # Allow overwriting our own wrappers
+            if existing_path == str(wrapper_dir / name):
+                try:
+                    content = (wrapper_dir / name).read_text()
+                    if "hermes -p" in content:
+                        return None  # it's our wrapper, safe to overwrite
+                except Exception:
+                    pass
+            return f"'{name}' conflicts with an existing command ({existing_path})"
+    except (FileNotFoundError, subprocess.TimeoutExpired):
+        pass
+
+    return None  # safe
+
+
+def _is_wrapper_dir_in_path() -> bool:
+    """Check if ~/.local/bin is in PATH."""
+    wrapper_dir = str(_get_wrapper_dir())
+    return wrapper_dir in os.environ.get("PATH", "").split(os.pathsep)
+
+
+def create_wrapper_script(name: str) -> Optional[Path]:
+    """Create a shell wrapper script at ~/.local/bin/<name>.
+
+    Returns the path to the created wrapper, or None if creation failed.
+    """
+    wrapper_dir = _get_wrapper_dir()
+    try:
+        wrapper_dir.mkdir(parents=True, exist_ok=True)
+    except OSError as e:
+        print(f"⚠ Could not create {wrapper_dir}: {e}")
+        return None
+
+    wrapper_path = wrapper_dir / name
+    try:
+        wrapper_path.write_text(f'#!/bin/sh\nexec hermes -p {name} "$@"\n')
+        wrapper_path.chmod(wrapper_path.stat().st_mode | stat.S_IEXEC | stat.S_IXGRP | stat.S_IXOTH)
+        return wrapper_path
+    except OSError as e:
+        print(f"⚠ Could not create wrapper at {wrapper_path}: {e}")
+        return None
+
+
+def remove_wrapper_script(name: str) -> bool:
+    """Remove the wrapper script for a profile. Returns True if removed."""
+    wrapper_path = _get_wrapper_dir() / name
+    if wrapper_path.exists():
+        try:
+            # Verify it's our wrapper before removing
+            content = wrapper_path.read_text()
+            if "hermes -p" in content:
+                wrapper_path.unlink()
+                return True
+        except Exception:
+            pass
+    return False
+
+
+# ---------------------------------------------------------------------------
+# ProfileInfo
+# ---------------------------------------------------------------------------
+
+@dataclass
+class ProfileInfo:
+    """Summary information about a profile."""
+    name: str
+    path: Path
+    is_default: bool
+    gateway_running: bool
+    model: Optional[str] = None
+    provider: Optional[str] = None
+    has_env: bool = False
+    skill_count: int = 0
+    alias_path: Optional[Path] = None
+
+
+def _read_config_model(profile_dir: Path) -> tuple:
+    """Read model/provider from a profile's config.yaml. Returns (model, provider)."""
+    config_path = profile_dir / "config.yaml"
+    if not config_path.exists():
+        return None, None
+    try:
+        import yaml
+        with open(config_path, "r") as f:
+            cfg = yaml.safe_load(f) or {}
+        model_cfg = cfg.get("model", {})
+        if isinstance(model_cfg, str):
+            return model_cfg, None
+        if isinstance(model_cfg, dict):
+            return model_cfg.get("model"), model_cfg.get("provider")
+        return None, None
+    except Exception:
+        return None, None
+
+
+def _check_gateway_running(profile_dir: Path) -> bool:
+    """Check if a gateway is running for a given profile directory."""
+    pid_file = profile_dir / "gateway.pid"
+    if not pid_file.exists():
+        return False
+    try:
+        raw = pid_file.read_text().strip()
+        if not raw:
+            return False
+        data = json.loads(raw) if raw.startswith("{") else {"pid": int(raw)}
+        pid = int(data["pid"])
+        os.kill(pid, 0)  # existence check
+        return True
+    except (json.JSONDecodeError, KeyError, ValueError, TypeError,
+            ProcessLookupError, PermissionError, OSError):
+        return False
+
+
+def _count_skills(profile_dir: Path) -> int:
+    """Count installed skills in a profile."""
+    skills_dir = profile_dir / "skills"
+    if not skills_dir.is_dir():
+        return 0
+    count = 0
+    for md in skills_dir.rglob("SKILL.md"):
+        if "/.hub/" not in str(md) and "/.git/" not in str(md):
+            count += 1
+    return count
+
+
+# ---------------------------------------------------------------------------
+# CRUD operations
+# ---------------------------------------------------------------------------
+
+def list_profiles() -> List[ProfileInfo]:
+    """Return info for all profiles, including the default."""
+    profiles = []
+    wrapper_dir = _get_wrapper_dir()
+
+    # Default profile
+    default_home = _get_default_hermes_home()
+    if default_home.is_dir():
+        model, provider = _read_config_model(default_home)
+        profiles.append(ProfileInfo(
+            name="default",
+            path=default_home,
+            is_default=True,
+            gateway_running=_check_gateway_running(default_home),
+            model=model,
+            provider=provider,
+            has_env=(default_home / ".env").exists(),
+            skill_count=_count_skills(default_home),
+        ))
+
+    # Named profiles
+    profiles_root = _get_profiles_root()
+    if profiles_root.is_dir():
+        for entry in sorted(profiles_root.iterdir()):
+            if not entry.is_dir():
+                continue
+            name = entry.name
+            if not _PROFILE_ID_RE.match(name):
+                continue
+            model, provider = _read_config_model(entry)
+            alias_path = wrapper_dir / name
+            profiles.append(ProfileInfo(
+                name=name,
+                path=entry,
+                is_default=False,
+                gateway_running=_check_gateway_running(entry),
+                model=model,
+                provider=provider,
+                has_env=(entry / ".env").exists(),
+                skill_count=_count_skills(entry),
+                alias_path=alias_path if alias_path.exists() else None,
+            ))
+
+    return profiles
+
+
+def create_profile(
+    name: str,
+    clone_from: Optional[str] = None,
+    clone_all: bool = False,
+    clone_config: bool = False,
+    no_alias: bool = False,
+) -> Path:
+    """Create a new profile directory.
+
+    Parameters
+    ----------
+    name:
+        Profile identifier (lowercase, alphanumeric, hyphens, underscores).
+    clone_from:
+        Source profile to clone from. If ``None`` and clone_config/clone_all
+        is True, defaults to the currently active profile.
+    clone_all:
+        If True, do a full copytree of the source (all state).
+    clone_config:
+        If True, copy only config files (config.yaml, .env, SOUL.md).
+    no_alias:
+        If True, skip wrapper script creation.
+
+    Returns
+    -------
+    Path
+        The newly created profile directory.
+    """
+    validate_profile_name(name)
+
+    if name == "default":
+        raise ValueError(
+            "Cannot create a profile named 'default' — it is the built-in profile (~/.hermes)."
+        )
+
+    profile_dir = get_profile_dir(name)
+    if profile_dir.exists():
+        raise FileExistsError(f"Profile '{name}' already exists at {profile_dir}")
+
+    # Resolve clone source
+    source_dir = None
+    if clone_from is not None or clone_all or clone_config:
+        if clone_from is None:
+            # Default: clone from active profile
+            from hermes_constants import get_hermes_home
+            source_dir = get_hermes_home()
+        else:
+            validate_profile_name(clone_from)
+            source_dir = get_profile_dir(clone_from)
+        if not source_dir.is_dir():
+            raise FileNotFoundError(
+                f"Source profile '{clone_from or 'active'}' does not exist at {source_dir}"
+            )
+
+    if clone_all and source_dir:
+        # Full copy of source profile
+        shutil.copytree(source_dir, profile_dir)
+        # Strip runtime files
+        for stale in _CLONE_ALL_STRIP:
+            (profile_dir / stale).unlink(missing_ok=True)
+    else:
+        # Bootstrap directory structure
+        profile_dir.mkdir(parents=True, exist_ok=True)
+        for subdir in _PROFILE_DIRS:
+            (profile_dir / subdir).mkdir(parents=True, exist_ok=True)
+
+        # Clone config files from source
+        if source_dir is not None:
+            for filename in _CLONE_CONFIG_FILES:
+                src = source_dir / filename
+                if src.exists():
+                    shutil.copy2(src, profile_dir / filename)
+
+    return profile_dir
+
+
+def seed_profile_skills(profile_dir: Path, quiet: bool = False) -> Optional[dict]:
+    """Seed bundled skills into a profile via subprocess.
+
+    Uses subprocess because sync_skills() caches HERMES_HOME at module level.
+    Returns the sync result dict, or None on failure.
+    """
+    project_root = Path(__file__).parent.parent.resolve()
+    try:
+        result = subprocess.run(
+            [sys.executable, "-c",
+             "import json; from tools.skills_sync import sync_skills; "
+             "r = sync_skills(quiet=True); print(json.dumps(r))"],
+            env={**os.environ, "HERMES_HOME": str(profile_dir)},
+            cwd=str(project_root),
+            capture_output=True, text=True, timeout=60,
+        )
+        if result.returncode == 0 and result.stdout.strip():
+            return json.loads(result.stdout.strip())
+        if not quiet:
+            print(f"⚠ Skill seeding returned exit code {result.returncode}")
+            if result.stderr.strip():
+                print(f"  {result.stderr.strip()[:200]}")
+        return None
+    except subprocess.TimeoutExpired:
+        if not quiet:
+            print("⚠ Skill seeding timed out (60s)")
+        return None
+    except Exception as e:
+        if not quiet:
+            print(f"⚠ Skill seeding failed: {e}")
+        return None
+
+
+def delete_profile(name: str, yes: bool = False) -> Path:
+    """Delete a profile, its wrapper script, and its gateway service.
+
+    Stops the gateway if running. Disables systemd/launchd service first
+    to prevent auto-restart.
+
+    Returns the path that was removed.
+    """
+    validate_profile_name(name)
+
+    if name == "default":
+        raise ValueError(
+            "Cannot delete the default profile (~/.hermes).\n"
+            "To remove everything, use: hermes uninstall"
+        )
+
+    profile_dir = get_profile_dir(name)
+    if not profile_dir.is_dir():
+        raise FileNotFoundError(f"Profile '{name}' does not exist.")
+
+    # Show what will be deleted
+    model, provider = _read_config_model(profile_dir)
+    gw_running = _check_gateway_running(profile_dir)
+    skill_count = _count_skills(profile_dir)
+
+    print(f"\nProfile: {name}")
+    print(f"Path:    {profile_dir}")
+    if model:
+        print(f"Model:   {model}" + (f" ({provider})" if provider else ""))
+    if skill_count:
+        print(f"Skills:  {skill_count}")
+
+    items = [
+        "All config, API keys, memories, sessions, skills, cron jobs",
+    ]
+
+    # Check for service
+    from hermes_cli.gateway import _profile_suffix, get_service_name
+    wrapper_path = _get_wrapper_dir() / name
+    has_wrapper = wrapper_path.exists()
+    if has_wrapper:
+        items.append(f"Command alias ({wrapper_path})")
+
+    print(f"\nThis will permanently delete:")
+    for item in items:
+        print(f"  • {item}")
+    if gw_running:
+        print(f"  ⚠ Gateway is running — it will be stopped.")
+
+    # Confirmation
+    if not yes:
+        print()
+        try:
+            confirm = input(f"Type '{name}' to confirm: ").strip()
+        except (KeyboardInterrupt, EOFError):
+            print("\nCancelled.")
+            return profile_dir
+        if confirm != name:
+            print("Cancelled.")
+            return profile_dir
+
+    # 1. Disable service (prevents auto-restart)
+    _cleanup_gateway_service(name, profile_dir)
+
+    # 2. Stop running gateway
+    if gw_running:
+        _stop_gateway_process(profile_dir)
+
+    # 3. Remove wrapper script
+    if has_wrapper:
+        if remove_wrapper_script(name):
+            print(f"✓ Removed {wrapper_path}")
+
+    # 4. Remove profile directory
+    try:
+        shutil.rmtree(profile_dir)
+        print(f"✓ Removed {profile_dir}")
+    except Exception as e:
+        print(f"⚠ Could not remove {profile_dir}: {e}")
+
+    # 5. Clear active_profile if it pointed to this profile
+    try:
+        active = get_active_profile()
+        if active == name:
+            set_active_profile("default")
+            print("✓ Active profile reset to default")
+    except Exception:
+        pass
+
+    print(f"\nProfile '{name}' deleted.")
+    return profile_dir
+
+
+def _cleanup_gateway_service(name: str, profile_dir: Path) -> None:
+    """Disable and remove systemd/launchd service for a profile."""
+    import platform as _platform
+
+    # Derive service name for this profile
+    # Temporarily set HERMES_HOME so _profile_suffix resolves correctly
+    old_home = os.environ.get("HERMES_HOME")
+    try:
+        os.environ["HERMES_HOME"] = str(profile_dir)
+        from hermes_cli.gateway import get_service_name, get_launchd_plist_path
+
+        if _platform.system() == "Linux":
+            svc_name = get_service_name()
+            svc_file = Path.home() / ".config" / "systemd" / "user" / f"{svc_name}.service"
+            if svc_file.exists():
+                subprocess.run(
+                    ["systemctl", "--user", "disable", svc_name],
+                    capture_output=True, check=False, timeout=10,
+                )
+                subprocess.run(
+                    ["systemctl", "--user", "stop", svc_name],
+                    capture_output=True, check=False, timeout=10,
+                )
+                svc_file.unlink(missing_ok=True)
+                subprocess.run(
+                    ["systemctl", "--user", "daemon-reload"],
+                    capture_output=True, check=False, timeout=10,
+                )
+                print(f"✓ Service {svc_name} removed")
+
+        elif _platform.system() == "Darwin":
+            plist_path = get_launchd_plist_path()
+            if plist_path.exists():
+                subprocess.run(
+                    ["launchctl", "unload", str(plist_path)],
+                    capture_output=True, check=False, timeout=10,
+                )
+                plist_path.unlink(missing_ok=True)
+                print(f"✓ Launchd service removed")
+    except Exception as e:
+        print(f"⚠ Service cleanup: {e}")
+    finally:
+        if old_home is not None:
+            os.environ["HERMES_HOME"] = old_home
+        elif "HERMES_HOME" in os.environ:
+            del os.environ["HERMES_HOME"]
+
+
+def _stop_gateway_process(profile_dir: Path) -> None:
+    """Stop a running gateway process via its PID file."""
+    import signal as _signal
+    import time as _time
+
+    pid_file = profile_dir / "gateway.pid"
+    if not pid_file.exists():
+        return
+
+    try:
+        raw = pid_file.read_text().strip()
+        data = json.loads(raw) if raw.startswith("{") else {"pid": int(raw)}
+        pid = int(data["pid"])
+        os.kill(pid, _signal.SIGTERM)
+        # Wait up to 10s for graceful shutdown
+        for _ in range(20):
+            _time.sleep(0.5)
+            try:
+                os.kill(pid, 0)
+            except ProcessLookupError:
+                print(f"✓ Gateway stopped (PID {pid})")
+                return
+        # Force kill
+        try:
+            os.kill(pid, _signal.SIGKILL)
+        except ProcessLookupError:
+            pass
+        print(f"✓ Gateway force-stopped (PID {pid})")
+    except (ProcessLookupError, PermissionError):
+        print("✓ Gateway already stopped")
+    except Exception as e:
+        print(f"⚠ Could not stop gateway: {e}")
+
+
+# ---------------------------------------------------------------------------
+# Active profile (sticky default)
+# ---------------------------------------------------------------------------
+
+def get_active_profile() -> str:
+    """Read the sticky active profile name.
+
+    Returns ``"default"`` if no active_profile file exists or it's empty.
+    """
+    path = _get_active_profile_path()
+    try:
+        name = path.read_text().strip()
+        if not name:
+            return "default"
+        return name
+    except (FileNotFoundError, UnicodeDecodeError, OSError):
+        return "default"
+
+
+def set_active_profile(name: str) -> None:
+    """Set the sticky active profile.
+
+    Writes to ``~/.hermes/active_profile``. Use ``"default"`` to clear.
+    """
+    validate_profile_name(name)
+    if name != "default" and not profile_exists(name):
+        raise FileNotFoundError(
+            f"Profile '{name}' does not exist. "
+            f"Create it with: hermes profile create {name}"
+        )
+
+    path = _get_active_profile_path()
+    path.parent.mkdir(parents=True, exist_ok=True)
+    if name == "default":
+        # Remove the file to indicate default
+        path.unlink(missing_ok=True)
+    else:
+        # Atomic write
+        tmp = path.with_suffix(".tmp")
+        tmp.write_text(name + "\n")
+        tmp.replace(path)
+
+
+def get_active_profile_name() -> str:
+    """Infer the current profile name from HERMES_HOME.
+
+    Returns ``"default"`` if HERMES_HOME is not set or points to ``~/.hermes``.
+    Returns the profile name if HERMES_HOME points into ``~/.hermes/profiles/<name>``.
+    Returns ``"custom"`` if HERMES_HOME is set to an unrecognized path.
+    """
+    from hermes_constants import get_hermes_home
+    hermes_home = get_hermes_home()
+    resolved = hermes_home.resolve()
+
+    default_resolved = _get_default_hermes_home().resolve()
+    if resolved == default_resolved:
+        return "default"
+
+    profiles_root = _get_profiles_root().resolve()
+    try:
+        rel = resolved.relative_to(profiles_root)
+        parts = rel.parts
+        if len(parts) == 1 and _PROFILE_ID_RE.match(parts[0]):
+            return parts[0]
+    except ValueError:
+        pass
+
+    return "custom"
+
+
+# ---------------------------------------------------------------------------
+# Export / Import
+# ---------------------------------------------------------------------------
+
+def export_profile(name: str, output_path: str) -> Path:
+    """Export a profile to a tar.gz archive.
+
+    Returns the output file path.
+    """
+    validate_profile_name(name)
+    profile_dir = get_profile_dir(name)
+    if not profile_dir.is_dir():
+        raise FileNotFoundError(f"Profile '{name}' does not exist.")
+
+    output = Path(output_path)
+    # shutil.make_archive wants the base name without extension
+    base = str(output).removesuffix(".tar.gz").removesuffix(".tgz")
+    result = shutil.make_archive(base, "gztar", str(profile_dir.parent), name)
+    return Path(result)
+
+
+def import_profile(archive_path: str, name: Optional[str] = None) -> Path:
+    """Import a profile from a tar.gz archive.
+
+    If *name* is not given, infers it from the archive's top-level directory.
+    Returns the imported profile directory.
+    """
+    import tarfile
+
+    archive = Path(archive_path)
+    if not archive.exists():
+        raise FileNotFoundError(f"Archive not found: {archive}")
+
+    # Peek at the archive to find the top-level directory name
+    with tarfile.open(archive, "r:gz") as tf:
+        top_dirs = {m.name.split("/")[0] for m in tf.getmembers() if "/" in m.name}
+        if not top_dirs:
+            top_dirs = {m.name for m in tf.getmembers() if m.isdir()}
+
+    inferred_name = name or (top_dirs.pop() if len(top_dirs) == 1 else None)
+    if not inferred_name:
+        raise ValueError(
+            "Cannot determine profile name from archive. "
+            "Specify it explicitly: hermes profile import <archive> --name <name>"
+        )
+
+    validate_profile_name(inferred_name)
+    profile_dir = get_profile_dir(inferred_name)
+    if profile_dir.exists():
+        raise FileExistsError(f"Profile '{inferred_name}' already exists at {profile_dir}")
+
+    profiles_root = _get_profiles_root()
+    profiles_root.mkdir(parents=True, exist_ok=True)
+
+    shutil.unpack_archive(str(archive), str(profiles_root))
+
+    # If the archive extracted under a different name, rename
+    extracted = profiles_root / (top_dirs.pop() if top_dirs else inferred_name)
+    if extracted != profile_dir and extracted.exists():
+        extracted.rename(profile_dir)
+
+    return profile_dir
+
+
+# ---------------------------------------------------------------------------
+# Rename
+# ---------------------------------------------------------------------------
+
+def rename_profile(old_name: str, new_name: str) -> Path:
+    """Rename a profile: directory, wrapper script, service, active_profile.
+
+    Returns the new profile directory.
+    """
+    validate_profile_name(old_name)
+    validate_profile_name(new_name)
+
+    if old_name == "default":
+        raise ValueError("Cannot rename the default profile.")
+    if new_name == "default":
+        raise ValueError("Cannot rename to 'default' — it is reserved.")
+
+    old_dir = get_profile_dir(old_name)
+    new_dir = get_profile_dir(new_name)
+
+    if not old_dir.is_dir():
+        raise FileNotFoundError(f"Profile '{old_name}' does not exist.")
+    if new_dir.exists():
+        raise FileExistsError(f"Profile '{new_name}' already exists.")
+
+    # 1. Stop gateway if running
+    if _check_gateway_running(old_dir):
+        _cleanup_gateway_service(old_name, old_dir)
+        _stop_gateway_process(old_dir)
+
+    # 2. Rename directory
+    old_dir.rename(new_dir)
+    print(f"✓ Renamed {old_dir.name} → {new_dir.name}")
+
+    # 3. Update wrapper script
+    remove_wrapper_script(old_name)
+    collision = check_alias_collision(new_name)
+    if not collision:
+        create_wrapper_script(new_name)
+        print(f"✓ Alias updated: {new_name}")
+    else:
+        print(f"⚠ Cannot create alias '{new_name}' — {collision}")
+
+    # 4. Update active_profile if it pointed to old name
+    try:
+        if get_active_profile() == old_name:
+            set_active_profile(new_name)
+            print(f"✓ Active profile updated: {new_name}")
+    except Exception:
+        pass
+
+    return new_dir
+
+
+# ---------------------------------------------------------------------------
+# Tab completion
+# ---------------------------------------------------------------------------
+
+def generate_bash_completion() -> str:
+    """Generate a bash completion script for hermes profile names."""
+    return '''# Hermes Agent profile completion
+# Add to ~/.bashrc: eval "$(hermes completion bash)"
+
+_hermes_profiles() {
+    local profiles_dir="$HOME/.hermes/profiles"
+    local profiles="default"
+    if [ -d "$profiles_dir" ]; then
+        profiles="$profiles $(ls "$profiles_dir" 2>/dev/null)"
+    fi
+    echo "$profiles"
+}
+
+_hermes_completion() {
+    local cur prev
+    cur="${COMP_WORDS[COMP_CWORD]}"
+    prev="${COMP_WORDS[COMP_CWORD-1]}"
+
+    # Complete profile names after -p / --profile
+    if [[ "$prev" == "-p" || "$prev" == "--profile" ]]; then
+        COMPREPLY=($(compgen -W "$(_hermes_profiles)" -- "$cur"))
+        return
+    fi
+
+    # Complete profile subcommands
+    if [[ "${COMP_WORDS[1]}" == "profile" ]]; then
+        case "$prev" in
+            profile)
+                COMPREPLY=($(compgen -W "list use create delete show alias rename export import" -- "$cur"))
+                return
+                ;;
+            use|delete|show|alias|rename|export)
+                COMPREPLY=($(compgen -W "$(_hermes_profiles)" -- "$cur"))
+                return
+                ;;
+        esac
+    fi
+
+    # Top-level subcommands
+    if [[ "$COMP_CWORD" == 1 ]]; then
+        local commands="chat model gateway setup status cron doctor config skills tools mcp sessions profile update version"
+        COMPREPLY=($(compgen -W "$commands" -- "$cur"))
+    fi
+}
+
+complete -F _hermes_completion hermes
+'''
+
+
+def generate_zsh_completion() -> str:
+    """Generate a zsh completion script for hermes profile names."""
+    return '''#compdef hermes
+# Hermes Agent profile completion
+# Add to ~/.zshrc: eval "$(hermes completion zsh)"
+
+_hermes() {
+    local -a profiles
+    profiles=(default)
+    if [[ -d "$HOME/.hermes/profiles" ]]; then
+        profiles+=("${(@f)$(ls $HOME/.hermes/profiles 2>/dev/null)}")
+    fi
+
+    _arguments \\
+        '-p[Profile name]:profile:($profiles)' \\
+        '--profile[Profile name]:profile:($profiles)' \\
+        '1:command:(chat model gateway setup status cron doctor config skills tools mcp sessions profile update version)' \\
+        '*::arg:->args'
+
+    case $words[1] in
+        profile)
+            _arguments '1:action:(list use create delete show alias rename export import)' \\
+                        '2:profile:($profiles)'
+            ;;
+    esac
+}
+
+_hermes "$@"
+'''
+
+
+# ---------------------------------------------------------------------------
+# Profile env resolution (called from _apply_profile_override)
+# ---------------------------------------------------------------------------
+
+def resolve_profile_env(profile_name: str) -> str:
+    """Resolve a profile name to a HERMES_HOME path string.
+
+    Called early in the CLI entry point, before any hermes modules
+    are imported, to set the HERMES_HOME environment variable.
+    """
+    validate_profile_name(profile_name)
+    profile_dir = get_profile_dir(profile_name)
+
+    if profile_name != "default" and not profile_dir.is_dir():
+        raise FileNotFoundError(
+            f"Profile '{profile_name}' does not exist. "
+            f"Create it with: hermes profile create {profile_name}"
+        )
+
+    return str(profile_dir)

hermes_cli/runtime_provider.py

@@ -63,8 +63,11 @@ def _get_model_config() -> Dict[str, Any]:
     model_cfg = config.get("model")
     if isinstance(model_cfg, dict):
         cfg = dict(model_cfg)
-        default = cfg.get("default", "").strip()
-        base_url = cfg.get("base_url", "").strip()
+        # Accept "model" as alias for "default" (users intuitively write model.model)
+        if not cfg.get("default") and cfg.get("model"):
+            cfg["default"] = cfg["model"]
+        default = (cfg.get("default") or "").strip()
+        base_url = (cfg.get("base_url") or "").strip()
         is_local = "localhost" in base_url or "127.0.0.1" in base_url
         is_fallback = not default or default == "anthropic/claude-opus-4.6"
         if is_local and is_fallback and base_url:
@@ -203,7 +206,7 @@ def _resolve_named_custom_runtime(
         or _detect_api_mode_for_url(base_url)
         or "chat_completions",
         "base_url": base_url,
-        "api_key": api_key,
+        "api_key": api_key or "no-key-required",
         "source": f"custom_provider:{custom_provider.get('name', requested_provider)}",
     }
 
@@ -407,12 +410,6 @@ def resolve_runtime_provider(
             # (e.g. https://api.minimax.io/anthropic, https://dashscope.../anthropic)
             elif base_url.rstrip("/").endswith("/anthropic"):
                 api_mode = "anthropic_messages"
-            # MiniMax providers always use Anthropic Messages API.
-            # Auto-correct stale /v1 URLs (from old .env or config) to /anthropic.
-            elif provider in ("minimax", "minimax-cn"):
-                api_mode = "anthropic_messages"
-                if base_url.rstrip("/").endswith("/v1"):
-                    base_url = base_url.rstrip("/")[:-3] + "/anthropic"
         return {
             "provider": provider,
             "api_mode": api_mode,

hermes_cli/setup.py

@@ -80,6 +80,11 @@ _DEFAULT_PROVIDER_MODELS = {
     "minimax-cn": ["MiniMax-M2.7", "MiniMax-M2.7-highspeed", "MiniMax-M2.5", "MiniMax-M2.5-highspeed", "MiniMax-M2.1"],
     "ai-gateway": ["anthropic/claude-opus-4.6", "anthropic/claude-sonnet-4.6", "openai/gpt-5", "google/gemini-3-flash"],
     "kilocode": ["anthropic/claude-opus-4.6", "anthropic/claude-sonnet-4.6", "openai/gpt-5.4", "google/gemini-3-pro-preview", "google/gemini-3-flash-preview"],
+    "huggingface": [
+        "Qwen/Qwen3.5-397B-A17B", "Qwen/Qwen3-235B-A22B-Thinking-2507",
+        "Qwen/Qwen3-Coder-480B-A35B-Instruct", "deepseek-ai/DeepSeek-R1-0528",
+        "deepseek-ai/DeepSeek-V3.2", "moonshotai/Kimi-K2.5",
+    ],
 }
 
 
@@ -284,6 +289,7 @@ from hermes_cli.config import (
     get_env_value,
     ensure_hermes_home,
 )
+# display_hermes_home imported lazily at call sites (stale-module safety during hermes update)
 
 from hermes_cli.colors import Colors, color
 
@@ -580,11 +586,11 @@ def _print_setup_summary(config: dict, hermes_home):
     else:
         tool_status.append(("Mixture of Agents", False, "OPENROUTER_API_KEY"))
 
-    # Web tools (Parallel, Firecrawl, or Tavily)
-    if get_env_value("PARALLEL_API_KEY") or get_env_value("FIRECRAWL_API_KEY") or get_env_value("FIRECRAWL_API_URL") or get_env_value("TAVILY_API_KEY"):
+    # Web tools (Exa, Parallel, Firecrawl, or Tavily)
+    if get_env_value("EXA_API_KEY") or get_env_value("PARALLEL_API_KEY") or get_env_value("FIRECRAWL_API_KEY") or get_env_value("FIRECRAWL_API_URL") or get_env_value("TAVILY_API_KEY"):
         tool_status.append(("Web Search & Extract", True, None))
     else:
-        tool_status.append(("Web Search & Extract", False, "PARALLEL_API_KEY, FIRECRAWL_API_KEY, or TAVILY_API_KEY"))
+        tool_status.append(("Web Search & Extract", False, "EXA_API_KEY, PARALLEL_API_KEY, FIRECRAWL_API_KEY, or TAVILY_API_KEY"))
 
     # Browser tools (local Chromium or Browserbase cloud)
     import shutil
@@ -678,7 +684,8 @@ def _print_setup_summary(config: dict, hermes_home):
         print_warning(
             "Some tools are disabled. Run 'hermes setup tools' to configure them,"
         )
-        print_warning("or edit ~/.hermes/.env directly to add the missing API keys.")
+        from hermes_constants import display_hermes_home as _dhh
+        print_warning(f"or edit {_dhh()}/.env directly to add the missing API keys.")
         print()
 
     # Done banner
@@ -701,7 +708,8 @@ def _print_setup_summary(config: dict, hermes_home):
     print()
 
     # Show file locations prominently
-    print(color("📁 All your files are in ~/.hermes/:", Colors.CYAN, Colors.BOLD))
+    from hermes_constants import display_hermes_home as _dhh
+    print(color(f"📁 All your files are in {_dhh()}/:", Colors.CYAN, Colors.BOLD))
     print()
     print(f"   {color('Settings:', Colors.YELLOW)}  {get_config_path()}")
     print(f"   {color('API Keys:', Colors.YELLOW)}  {get_env_path()}")
@@ -884,6 +892,7 @@ def setup_model_provider(config: dict):
         "OpenCode Go (open models, $10/month subscription)",
         "GitHub Copilot (uses GITHUB_TOKEN or gh auth token)",
         "GitHub Copilot ACP (spawns `copilot --acp --stdio`)",
+        "Hugging Face Inference Providers (20+ open models)",
     ]
     if keep_label:
         provider_choices.append(keep_label)
@@ -1528,7 +1537,26 @@ def setup_model_provider(config: dict):
         _set_model_provider(config, "copilot-acp", pconfig.inference_base_url)
         selected_base_url = pconfig.inference_base_url
 
-    # else: provider_idx == 16 (Keep current) — only shown when a provider already exists
+    elif provider_idx == 16:  # Hugging Face Inference Providers
+        selected_provider = "huggingface"
+        print()
+        print_header("Hugging Face API Token")
+        pconfig = PROVIDER_REGISTRY["huggingface"]
+        print_info(f"Provider: {pconfig.name}")
+        print_info("Get your token at: https://huggingface.co/settings/tokens")
+        print_info("Required permission: 'Make calls to Inference Providers'")
+        print()
+
+        api_key = prompt("  HF Token", password=True)
+        if api_key:
+            save_env_value("HF_TOKEN", api_key)
+            # Clear OpenRouter env vars to prevent routing confusion
+            save_env_value("OPENAI_BASE_URL", "")
+            save_env_value("OPENAI_API_KEY", "")
+        _set_model_provider(config, "huggingface", pconfig.inference_base_url)
+        selected_base_url = pconfig.inference_base_url
+
+    # else: provider_idx == 17 (Keep current) — only shown when a provider already exists
     # Normalize "keep current" to an explicit provider so downstream logic
     # doesn't fall back to the generic OpenRouter/static-model path.
     if selected_provider is None:
@@ -2067,11 +2095,11 @@ def setup_terminal_backend(config: dict):
         print_info("Serverless cloud sandboxes. Each session gets its own container.")
         print_info("Requires a Modal account: https://modal.com")
 
-        # Check if swe-rex[modal] is installed
+        # Check if modal SDK is installed
         try:
-            __import__("swe_rex")
+            __import__("modal")
         except ImportError:
-            print_info("Installing swe-rex[modal]...")
+            print_info("Installing modal SDK...")
             import subprocess
 
             uv_bin = shutil.which("uv")
@@ -2083,22 +2111,22 @@ def setup_terminal_backend(config: dict):
                         "install",
                         "--python",
                         sys.executable,
-                        "swe-rex[modal]",
+                        "modal",
                     ],
                     capture_output=True,
                     text=True,
                 )
             else:
                 result = subprocess.run(
-                    [sys.executable, "-m", "pip", "install", "swe-rex[modal]"],
+                    [sys.executable, "-m", "pip", "install", "modal"],
                     capture_output=True,
                     text=True,
                 )
             if result.returncode == 0:
-                print_success("swe-rex[modal] installed")
+                print_success("modal SDK installed")
             else:
                 print_warning(
-                    "Install failed — run manually: pip install 'swe-rex[modal]'"
+                    "Install failed — run manually: pip install modal"
                 )
 
         # Modal token
@@ -2812,7 +2840,8 @@ def setup_gateway(config: dict):
         save_env_value("WEBHOOK_ENABLED", "true")
         print()
         print_success("Webhooks enabled! Next steps:")
-        print_info("   1. Define webhook routes in ~/.hermes/config.yaml")
+        from hermes_constants import display_hermes_home as _dhh
+        print_info(f"   1. Define webhook routes in {_dhh()}/config.yaml")
         print_info("   2. Point your service (GitHub, GitLab, etc.) at:")
         print_info("      http://your-server:8644/webhooks/<route-name>")
         print()

hermes_cli/skills_config.py

@@ -24,6 +24,10 @@ PLATFORMS = {
     "whatsapp": "📱 WhatsApp",
     "signal":   "📡 Signal",
     "email":    "📧 Email",
+    "homeassistant": "🏠 Home Assistant",
+    "mattermost": "💬 Mattermost",
+    "matrix":   "💬 Matrix",
+    "dingtalk": "💬 DingTalk",
 }
 
 # ─── Config Helpers ───────────────────────────────────────────────────────────

hermes_cli/skills_hub.py

@@ -21,6 +21,7 @@ from rich.table import Table
 
 # Lazy imports to avoid circular dependencies and slow startup.
 # tools.skills_hub and tools.skills_guard are imported inside functions.
+from hermes_constants import display_hermes_home
 
 _console = Console()
 
@@ -304,7 +305,8 @@ def do_browse(page: int = 1, page_size: int = 20, source: str = "all",
 
 
 def do_install(identifier: str, category: str = "", force: bool = False,
-               console: Optional[Console] = None, skip_confirm: bool = False) -> None:
+               console: Optional[Console] = None, skip_confirm: bool = False,
+               invalidate_cache: bool = True) -> None:
     """Fetch, quarantine, scan, confirm, and install a skill."""
     from tools.skills_hub import (
         GitHubAuth, create_source_router, ensure_hub_dirs,
@@ -387,7 +389,7 @@ def do_install(identifier: str, category: str = "", force: bool = False,
                 "[bold bright_cyan]This is an official optional skill maintained by Nous Research.[/]\n\n"
                 "It ships with hermes-agent but is not activated by default.\n"
                 "Installing will copy it to your skills directory where the agent can use it.\n\n"
-                f"Files will be at: [cyan]~/.hermes/skills/{category + '/' if category else ''}{bundle.name}/[/]",
+                f"Files will be at: [cyan]{display_hermes_home()}/skills/{category + '/' if category else ''}{bundle.name}/[/]",
                 title="Official Skill",
                 border_style="bright_cyan",
             ))
@@ -397,7 +399,7 @@ def do_install(identifier: str, category: str = "", force: bool = False,
                 "External skills can contain instructions that influence agent behavior,\n"
                 "shell commands, and scripts. Even after automated scanning, you should\n"
                 "review the installed files before use.\n\n"
-                f"Files will be at: [cyan]~/.hermes/skills/{category + '/' if category else ''}{bundle.name}/[/]",
+                f"Files will be at: [cyan]{display_hermes_home()}/skills/{category + '/' if category else ''}{bundle.name}/[/]",
                 title="Disclaimer",
                 border_style="yellow",
             ))
@@ -417,12 +419,16 @@ def do_install(identifier: str, category: str = "", force: bool = False,
     c.print(f"[bold green]Installed:[/] {install_dir.relative_to(SKILLS_DIR)}")
     c.print(f"[dim]Files: {', '.join(bundle.files.keys())}[/]\n")
 
-    # Invalidate the skills prompt cache so the new skill appears immediately
-    try:
-        from agent.prompt_builder import clear_skills_system_prompt_cache
-        clear_skills_system_prompt_cache(clear_snapshot=True)
-    except Exception:
-        pass
+    if invalidate_cache:
+        # Invalidate the skills prompt cache so the new skill appears immediately
+        try:
+            from agent.prompt_builder import clear_skills_system_prompt_cache
+            clear_skills_system_prompt_cache(clear_snapshot=True)
+        except Exception:
+            pass
+    else:
+        c.print("[dim]Skill will be available in your next session.[/]")
+        c.print("[dim]Use /reset to start a new session now, or --now to activate immediately (invalidates prompt cache).[/]\n")
 
 
 def do_inspect(identifier: str, console: Optional[Console] = None) -> None:
@@ -610,7 +616,8 @@ def do_audit(name: Optional[str] = None, console: Optional[Console] = None) -> N
 
 
 def do_uninstall(name: str, console: Optional[Console] = None,
-                 skip_confirm: bool = False) -> None:
+                 skip_confirm: bool = False,
+                 invalidate_cache: bool = True) -> None:
     """Remove a hub-installed skill with confirmation."""
     from tools.skills_hub import uninstall_skill
 
@@ -630,11 +637,15 @@ def do_uninstall(name: str, console: Optional[Console] = None,
     success, msg = uninstall_skill(name)
     if success:
         c.print(f"[bold green]{msg}[/]\n")
-        try:
-            from agent.prompt_builder import clear_skills_system_prompt_cache
-            clear_skills_system_prompt_cache(clear_snapshot=True)
-        except Exception:
-            pass
+        if invalidate_cache:
+            try:
+                from agent.prompt_builder import clear_skills_system_prompt_cache
+                clear_skills_system_prompt_cache(clear_snapshot=True)
+            except Exception:
+                pass
+        else:
+            c.print("[dim]Change will take effect in your next session.[/]")
+            c.print("[dim]Use /reset to start a new session now, or --now to apply immediately (invalidates prompt cache).[/]\n")
     else:
         c.print(f"[bold red]Error:[/] {msg}\n")
 
@@ -734,7 +745,7 @@ def do_publish(skill_path: str, target: str = "github", repo: str = "",
         auth = GitHubAuth()
         if not auth.is_authenticated():
             c.print("[bold red]Error:[/] GitHub authentication required.\n"
-                    "Set GITHUB_TOKEN in ~/.hermes/.env or run 'gh auth login'.\n")
+                    f"Set GITHUB_TOKEN in {display_hermes_home()}/.env or run 'gh auth login'.\n")
             return
 
         c.print(f"[bold]Publishing '{name}' to {repo}...[/]")
@@ -877,10 +888,15 @@ def do_snapshot_export(output_path: str, console: Optional[Console] = None) -> N
         "taps": tap_list,
     }
 
-    out = Path(output_path)
-    out.write_text(json.dumps(snapshot, indent=2, ensure_ascii=False) + "\n")
-    c.print(f"[bold green]Snapshot exported:[/] {out}")
-    c.print(f"[dim]{len(installed)} skill(s), {len(tap_list)} tap(s)[/]\n")
+    payload = json.dumps(snapshot, indent=2, ensure_ascii=False) + "\n"
+    if output_path == "-":
+        import sys
+        sys.stdout.write(payload)
+    else:
+        out = Path(output_path)
+        out.write_text(payload)
+        c.print(f"[bold green]Snapshot exported:[/] {out}")
+        c.print(f"[dim]{len(installed)} skill(s), {len(tap_list)} tap(s)[/]\n")
 
 
 def do_snapshot_import(input_path: str, force: bool = False,
@@ -1071,19 +1087,23 @@ def handle_skills_slash(cmd: str, console: Optional[Console] = None) -> None:
 
     elif action == "install":
         if not args:
-            c.print("[bold red]Usage:[/] /skills install <identifier> [--category <cat>] [--force|--yes]\n")
+            c.print("[bold red]Usage:[/] /skills install <identifier> [--category <cat>] [--force] [--now]\n")
             return
         identifier = args[0]
         category = ""
-        # --yes / -y bypasses confirmation prompt (needed in TUI mode)
-        # --force handles reinstall override
-        skip_confirm = any(flag in args for flag in ("--yes", "-y"))
+        # Slash commands run inside prompt_toolkit where input() hangs.
+        # Always skip confirmation — the user typing the command is implicit consent.
+        skip_confirm = True
         force = "--force" in args
+        # --now invalidates prompt cache immediately (costs more money).
+        # Default: defer to next session to preserve cache.
+        invalidate_cache = "--now" in args
         for i, a in enumerate(args):
             if a == "--category" and i + 1 < len(args):
                 category = args[i + 1]
         do_install(identifier, category=category, force=force,
-                   skip_confirm=skip_confirm, console=c)
+                   skip_confirm=skip_confirm, invalidate_cache=invalidate_cache,
+                   console=c)
 
     elif action == "inspect":
         if not args:
@@ -1113,10 +1133,13 @@ def handle_skills_slash(cmd: str, console: Optional[Console] = None) -> None:
 
     elif action == "uninstall":
         if not args:
-            c.print("[bold red]Usage:[/] /skills uninstall <name> [--yes]\n")
+            c.print("[bold red]Usage:[/] /skills uninstall <name> [--now]\n")
             return
-        skip_confirm = any(flag in args for flag in ("--yes", "-y"))
-        do_uninstall(args[0], console=c, skip_confirm=skip_confirm)
+        # Slash commands run inside prompt_toolkit where input() hangs.
+        skip_confirm = True
+        invalidate_cache = "--now" in args
+        do_uninstall(args[0], console=c, skip_confirm=skip_confirm,
+                     invalidate_cache=invalidate_cache)
 
     elif action == "publish":
         if not args:

hermes_cli/status.py

@@ -292,8 +292,9 @@ def show_status(args):
         print("  Manager:      systemd (user)")
         
     elif sys.platform == 'darwin':
+        from hermes_cli.gateway import get_launchd_label
         result = subprocess.run(
-            ["launchctl", "list", "ai.hermes.gateway"],
+            ["launchctl", "list", get_launchd_label()],
             capture_output=True,
             text=True
         )

hermes_cli/tools_config.py

@@ -9,6 +9,8 @@ Saves per-platform tool configuration to ~/.hermes/config.yaml under
 the `platform_toolsets` key.
 """
 
+import json as _json
+import logging
 import sys
 from pathlib import Path
 from typing import Dict, List, Optional, Set
@@ -19,6 +21,8 @@ from hermes_cli.config import (
 )
 from hermes_cli.colors import Colors, color
 
+logger = logging.getLogger(__name__)
+
 PROJECT_ROOT = Path(__file__).parent.parent.resolve()
 
 
@@ -108,7 +112,8 @@ def _get_effective_configurable_toolsets():
     """
     result = list(CONFIGURABLE_TOOLSETS)
     try:
-        from hermes_cli.plugins import get_plugin_toolsets
+        from hermes_cli.plugins import discover_plugins, get_plugin_toolsets
+        discover_plugins()  # idempotent — ensures plugins are loaded
         result.extend(get_plugin_toolsets())
     except Exception:
         pass
@@ -118,7 +123,8 @@ def _get_effective_configurable_toolsets():
 def _get_plugin_toolset_keys() -> set:
     """Return the set of toolset keys provided by plugins."""
     try:
-        from hermes_cli.plugins import get_plugin_toolsets
+        from hermes_cli.plugins import discover_plugins, get_plugin_toolsets
+        discover_plugins()  # idempotent — ensures plugins are loaded
         return {ts_key for ts_key, _, _ in get_plugin_toolsets()}
     except Exception:
         return set()
@@ -133,8 +139,10 @@ PLATFORMS = {
     "signal":   {"label": "📡 Signal",     "default_toolset": "hermes-signal"},
     "homeassistant": {"label": "🏠 Home Assistant", "default_toolset": "hermes-homeassistant"},
     "email":    {"label": "📧 Email",      "default_toolset": "hermes-email"},
+    "matrix":   {"label": "💬 Matrix",     "default_toolset": "hermes-matrix"},
     "dingtalk": {"label": "💬 DingTalk",   "default_toolset": "hermes-dingtalk"},
     "api_server": {"label": "🌐 API Server", "default_toolset": "hermes-api-server"},
+    "mattermost": {"label": "💬 Mattermost", "default_toolset": "hermes-mattermost"},
 }
 
 
@@ -186,6 +194,14 @@ TOOL_CATEGORIES = {
                     {"key": "FIRECRAWL_API_KEY", "prompt": "Firecrawl API key", "url": "https://firecrawl.dev"},
                 ],
             },
+            {
+                "name": "Exa",
+                "tag": "AI-native search and contents",
+                "web_backend": "exa",
+                "env_vars": [
+                    {"key": "EXA_API_KEY", "prompt": "Exa API key", "url": "https://exa.ai"},
+                ],
+            },
             {
                 "name": "Parallel",
                 "tag": "AI-native search and extract",
@@ -314,7 +330,8 @@ def _run_post_setup(post_setup_key: str):
             if result.returncode == 0:
                 _print_success("    Node.js dependencies installed")
             else:
-                _print_warning("    npm install failed - run manually: cd ~/.hermes/hermes-agent && npm install")
+                from hermes_constants import display_hermes_home
+                _print_warning(f"    npm install failed - run manually: cd {display_hermes_home()}/hermes-agent && npm install")
         elif not node_modules.exists():
             _print_warning("    Node.js not found - browser tools require: npm install (in hermes-agent directory)")
 
@@ -640,9 +657,61 @@ def _prompt_choice(question: str, choices: list, default: int = 0) -> int:
             return default
 
 
+# ─── Token Estimation ────────────────────────────────────────────────────────
+
+# Module-level cache so discovery + tokenization runs at most once per process.
+_tool_token_cache: Optional[Dict[str, int]] = None
+
+
+def _estimate_tool_tokens() -> Dict[str, int]:
+    """Return estimated token counts per individual tool name.
+
+    Uses tiktoken (cl100k_base) to count tokens in the JSON-serialised
+    OpenAI-format tool schema.  Triggers tool discovery on first call,
+    then caches the result for the rest of the process.
+
+    Returns an empty dict when tiktoken or the registry is unavailable.
+    """
+    global _tool_token_cache
+    if _tool_token_cache is not None:
+        return _tool_token_cache
+
+    try:
+        import tiktoken
+        enc = tiktoken.get_encoding("cl100k_base")
+    except Exception:
+        logger.debug("tiktoken unavailable; skipping tool token estimation")
+        _tool_token_cache = {}
+        return _tool_token_cache
+
+    try:
+        # Trigger full tool discovery (imports all tool modules).
+        import model_tools  # noqa: F401
+        from tools.registry import registry
+    except Exception:
+        logger.debug("Tool registry unavailable; skipping token estimation")
+        _tool_token_cache = {}
+        return _tool_token_cache
+
+    counts: Dict[str, int] = {}
+    for name in registry.get_all_tool_names():
+        schema = registry.get_schema(name)
+        if schema:
+            # Mirror what gets sent to the API:
+            # {"type": "function", "function": <schema>}
+            text = _json.dumps({"type": "function", "function": schema})
+            counts[name] = len(enc.encode(text))
+    _tool_token_cache = counts
+    return _tool_token_cache
+
+
 def _prompt_toolset_checklist(platform_label: str, enabled: Set[str]) -> Set[str]:
     """Multi-select checklist of toolsets. Returns set of selected toolset keys."""
     from hermes_cli.curses_ui import curses_checklist
+    from toolsets import resolve_toolset
+
+    # Pre-compute per-tool token counts (cached after first call).
+    tool_tokens = _estimate_tool_tokens()
 
     effective = _get_effective_configurable_toolsets()
 
@@ -658,11 +727,27 @@ def _prompt_toolset_checklist(platform_label: str, enabled: Set[str]) -> Set[str
         if ts_key in enabled
     }
 
+    # Build a live status function that shows deduplicated total token cost.
+    status_fn = None
+    if tool_tokens:
+        ts_keys = [ts_key for ts_key, _, _ in effective]
+
+        def status_fn(chosen: set) -> str:
+            # Collect unique tool names across all selected toolsets
+            all_tools: set = set()
+            for idx in chosen:
+                all_tools.update(resolve_toolset(ts_keys[idx]))
+            total = sum(tool_tokens.get(name, 0) for name in all_tools)
+            if total >= 1000:
+                return f"Est. tool context: ~{total / 1000:.1f}k tokens"
+            return f"Est. tool context: ~{total} tokens"
+
     chosen = curses_checklist(
         f"Tools for {platform_label}",
         labels,
         pre_selected,
         cancel_returns=pre_selected,
+        status_fn=status_fn,
     )
     return {effective[i][0] for i in chosen}
 
@@ -1252,7 +1337,8 @@ def tools_command(args=None, first_install: bool = False, config: dict = None):
         platform_choices[idx] = f"Configure {pinfo['label']}  ({new_count}/{total} enabled)"
 
     print()
-    print(color("  Tool configuration saved to ~/.hermes/config.yaml", Colors.DIM))
+    from hermes_constants import display_hermes_home
+    print(color(f"  Tool configuration saved to {display_hermes_home()}/config.yaml", Colors.DIM))
     print(color("  Changes take effect on next 'hermes' or gateway restart.", Colors.DIM))
     print()
 

hermes_cli/webhook.py

@@ -0,0 +1,260 @@
+"""hermes webhook — manage dynamic webhook subscriptions from the CLI.
+
+Usage:
+    hermes webhook subscribe <name> [options]
+    hermes webhook list
+    hermes webhook remove <name>
+    hermes webhook test <name> [--payload '{"key": "value"}']
+
+Subscriptions persist to ~/.hermes/webhook_subscriptions.json and are
+hot-reloaded by the webhook adapter without a gateway restart.
+"""
+
+import json
+import os
+import re
+import secrets
+import time
+from pathlib import Path
+from typing import Dict, Optional
+
+from hermes_constants import display_hermes_home
+
+
+_SUBSCRIPTIONS_FILENAME = "webhook_subscriptions.json"
+
+
+def _hermes_home() -> Path:
+    return Path(
+        os.getenv("HERMES_HOME", str(Path.home() / ".hermes"))
+    ).expanduser()
+
+
+def _subscriptions_path() -> Path:
+    return _hermes_home() / _SUBSCRIPTIONS_FILENAME
+
+
+def _load_subscriptions() -> Dict[str, dict]:
+    path = _subscriptions_path()
+    if not path.exists():
+        return {}
+    try:
+        data = json.loads(path.read_text(encoding="utf-8"))
+        return data if isinstance(data, dict) else {}
+    except Exception:
+        return {}
+
+
+def _save_subscriptions(subs: Dict[str, dict]) -> None:
+    path = _subscriptions_path()
+    path.parent.mkdir(parents=True, exist_ok=True)
+    tmp_path = path.with_suffix(".tmp")
+    tmp_path.write_text(
+        json.dumps(subs, indent=2, ensure_ascii=False),
+        encoding="utf-8",
+    )
+    os.replace(str(tmp_path), str(path))
+
+
+def _get_webhook_config() -> dict:
+    """Load webhook platform config. Returns {} if not configured."""
+    try:
+        from hermes_cli.config import load_config
+        cfg = load_config()
+        return cfg.get("platforms", {}).get("webhook", {})
+    except Exception:
+        return {}
+
+
+def _is_webhook_enabled() -> bool:
+    return bool(_get_webhook_config().get("enabled"))
+
+
+def _get_webhook_base_url() -> str:
+    wh = _get_webhook_config().get("extra", {})
+    host = wh.get("host", "0.0.0.0")
+    port = wh.get("port", 8644)
+    display_host = "localhost" if host == "0.0.0.0" else host
+    return f"http://{display_host}:{port}"
+
+
+def _setup_hint() -> str:
+    _dhh = display_hermes_home()
+    return f"""
+  Webhook platform is not enabled. To set it up:
+
+  1. Run the gateway setup wizard:
+     hermes gateway setup
+
+  2. Or manually add to {_dhh}/config.yaml:
+     platforms:
+       webhook:
+         enabled: true
+         extra:
+           host: "0.0.0.0"
+           port: 8644
+           secret: "your-global-hmac-secret"
+
+  3. Or set environment variables in {_dhh}/.env:
+     WEBHOOK_ENABLED=true
+     WEBHOOK_PORT=8644
+     WEBHOOK_SECRET=your-global-secret
+
+  Then start the gateway: hermes gateway run
+"""
+
+
+def _require_webhook_enabled() -> bool:
+    """Check webhook is enabled. Print setup guide and return False if not."""
+    if _is_webhook_enabled():
+        return True
+    print(_setup_hint())
+    return False
+
+
+def webhook_command(args):
+    """Entry point for 'hermes webhook' subcommand."""
+    sub = getattr(args, "webhook_action", None)
+
+    if not sub:
+        print("Usage: hermes webhook {subscribe|list|remove|test}")
+        print("Run 'hermes webhook --help' for details.")
+        return
+
+    if not _require_webhook_enabled():
+        return
+
+    if sub in ("subscribe", "add"):
+        _cmd_subscribe(args)
+    elif sub in ("list", "ls"):
+        _cmd_list(args)
+    elif sub in ("remove", "rm"):
+        _cmd_remove(args)
+    elif sub == "test":
+        _cmd_test(args)
+
+
+def _cmd_subscribe(args):
+    name = args.name.strip().lower().replace(" ", "-")
+    if not re.match(r'^[a-z0-9][a-z0-9_-]*$', name):
+        print(f"Error: Invalid name '{name}'. Use lowercase alphanumeric with hyphens/underscores.")
+        return
+
+    subs = _load_subscriptions()
+    is_update = name in subs
+
+    secret = args.secret or secrets.token_urlsafe(32)
+    events = [e.strip() for e in args.events.split(",")] if args.events else []
+
+    route = {
+        "description": args.description or f"Agent-created subscription: {name}",
+        "events": events,
+        "secret": secret,
+        "prompt": args.prompt or "",
+        "skills": [s.strip() for s in args.skills.split(",")] if args.skills else [],
+        "deliver": args.deliver or "log",
+        "created_at": time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime()),
+    }
+
+    if args.deliver_chat_id:
+        route["deliver_extra"] = {"chat_id": args.deliver_chat_id}
+
+    subs[name] = route
+    _save_subscriptions(subs)
+
+    base_url = _get_webhook_base_url()
+    status = "Updated" if is_update else "Created"
+
+    print(f"\n  {status} webhook subscription: {name}")
+    print(f"  URL:    {base_url}/webhooks/{name}")
+    print(f"  Secret: {secret}")
+    if events:
+        print(f"  Events: {', '.join(events)}")
+    else:
+        print("  Events: (all)")
+    print(f"  Deliver: {route['deliver']}")
+    if route.get("prompt"):
+        prompt_preview = route["prompt"][:80] + ("..." if len(route["prompt"]) > 80 else "")
+        print(f"  Prompt: {prompt_preview}")
+    print(f"\n  Configure your service to POST to the URL above.")
+    print(f"  Use the secret for HMAC-SHA256 signature validation.")
+    print(f"  The gateway must be running to receive events (hermes gateway run).\n")
+
+
+def _cmd_list(args):
+    subs = _load_subscriptions()
+    if not subs:
+        print("  No dynamic webhook subscriptions.")
+        print("  Create one with: hermes webhook subscribe <name>")
+        return
+
+    base_url = _get_webhook_base_url()
+    print(f"\n  {len(subs)} webhook subscription(s):\n")
+    for name, route in subs.items():
+        events = ", ".join(route.get("events", [])) or "(all)"
+        deliver = route.get("deliver", "log")
+        desc = route.get("description", "")
+        print(f"  ◆ {name}")
+        if desc:
+            print(f"    {desc}")
+        print(f"    URL:     {base_url}/webhooks/{name}")
+        print(f"    Events:  {events}")
+        print(f"    Deliver: {deliver}")
+        print()
+
+
+def _cmd_remove(args):
+    name = args.name.strip().lower()
+    subs = _load_subscriptions()
+
+    if name not in subs:
+        print(f"  No subscription named '{name}'.")
+        print("  Note: Static routes from config.yaml cannot be removed here.")
+        return
+
+    del subs[name]
+    _save_subscriptions(subs)
+    print(f"  Removed webhook subscription: {name}")
+
+
+def _cmd_test(args):
+    """Send a test POST to a webhook route."""
+    name = args.name.strip().lower()
+    subs = _load_subscriptions()
+
+    if name not in subs:
+        print(f"  No subscription named '{name}'.")
+        return
+
+    route = subs[name]
+    secret = route.get("secret", "")
+    base_url = _get_webhook_base_url()
+    url = f"{base_url}/webhooks/{name}"
+
+    payload = args.payload or '{"test": true, "event_type": "test", "message": "Hello from hermes webhook test"}'
+
+    import hmac
+    import hashlib
+    sig = "sha256=" + hmac.new(
+        secret.encode(), payload.encode(), hashlib.sha256
+    ).hexdigest()
+
+    print(f"  Sending test POST to {url}")
+    try:
+        import urllib.request
+        req = urllib.request.Request(
+            url,
+            data=payload.encode(),
+            headers={
+                "Content-Type": "application/json",
+                "X-Hub-Signature-256": sig,
+                "X-GitHub-Event": "test",
+            },
+            method="POST",
+        )
+        with urllib.request.urlopen(req, timeout=10) as resp:
+            body = resp.read().decode()
+            print(f"  Response ({resp.status}): {body}")
+    except Exception as e:
+        print(f"  Error: {e}")
+        print("  Is the gateway running? (hermes gateway run)")

hermes_constants.py

@@ -17,6 +17,47 @@ def get_hermes_home() -> Path:
     return Path(os.getenv("HERMES_HOME", Path.home() / ".hermes"))
 
 
+def get_hermes_dir(new_subpath: str, old_name: str) -> Path:
+    """Resolve a Hermes subdirectory with backward compatibility.
+
+    New installs get the consolidated layout (e.g. ``cache/images``).
+    Existing installs that already have the old path (e.g. ``image_cache``)
+    keep using it — no migration required.
+
+    Args:
+        new_subpath: Preferred path relative to HERMES_HOME (e.g. ``"cache/images"``).
+        old_name: Legacy path relative to HERMES_HOME (e.g. ``"image_cache"``).
+
+    Returns:
+        Absolute ``Path`` — old location if it exists on disk, otherwise the new one.
+    """
+    home = get_hermes_home()
+    old_path = home / old_name
+    if old_path.exists():
+        return old_path
+    return home / new_subpath
+
+
+def display_hermes_home() -> str:
+    """Return a user-friendly display string for the current HERMES_HOME.
+
+    Uses ``~/`` shorthand for readability::
+
+        default:  ``~/.hermes``
+        profile:  ``~/.hermes/profiles/coder``
+        custom:   ``/opt/hermes-custom``
+
+    Use this in **user-facing** print/log messages instead of hardcoding
+    ``~/.hermes``.  For code that needs a real ``Path``, use
+    :func:`get_hermes_home` instead.
+    """
+    home = get_hermes_home()
+    try:
+        return "~/" + str(home.relative_to(Path.home()))
+    except ValueError:
+        return str(home)
+
+
 VALID_REASONING_EFFORTS = ("xhigh", "high", "medium", "low", "minimal")
 
 

honcho_integration/cli.py

@@ -270,7 +270,7 @@ def cmd_status(args) -> None:
             print(f"    {peer}: {mode}")
     print(f"  Write freq:     {hcfg.write_frequency}")
 
-    if hcfg.enabled and hcfg.api_key:
+    if hcfg.enabled and (hcfg.api_key or hcfg.base_url):
         print("\n  Connection... ", end="", flush=True)
         try:
             get_honcho_client(hcfg)
@@ -278,7 +278,7 @@ def cmd_status(args) -> None:
         except Exception as e:
             print(f"FAILED ({e})\n")
     else:
-        reason = "disabled" if not hcfg.enabled else "no API key"
+        reason = "disabled" if not hcfg.enabled else "no API key or base URL"
         print(f"\n  Not connected ({reason})\n")
 
 

honcho_integration/client.py

@@ -417,9 +417,18 @@ def get_honcho_client(config: HonchoClientConfig | None = None) -> Honcho:
     else:
         logger.info("Initializing Honcho client (host: %s, workspace: %s)", config.host, config.workspace_id)
 
+    # Local Honcho instances don't require an API key, but the SDK
+    # expects a non-empty string.  Use a placeholder for local URLs.
+    _is_local = resolved_base_url and (
+        "localhost" in resolved_base_url
+        or "127.0.0.1" in resolved_base_url
+        or "::1" in resolved_base_url
+    )
+    effective_api_key = config.api_key or ("local" if _is_local else None)
+
     kwargs: dict = {
         "workspace_id": config.workspace_id,
-        "api_key": config.api_key,
+        "api_key": effective_api_key,
         "environment": config.environment,
     }
     if resolved_base_url:

mcp_serve.py

@@ -0,0 +1,868 @@
+"""
+Hermes MCP Server — expose messaging conversations as MCP tools.
+
+Starts a stdio MCP server that lets any MCP client (Claude Code, Cursor, Codex,
+etc.) list conversations, read message history, send messages, poll for live
+events, and manage approval requests across all connected platforms.
+
+Matches OpenClaw's 9-tool MCP channel bridge surface:
+  conversations_list, conversation_get, messages_read, attachments_fetch,
+  events_poll, events_wait, messages_send, permissions_list_open,
+  permissions_respond
+
+Plus: channels_list (Hermes-specific extra)
+
+Usage:
+    hermes mcp serve
+    hermes mcp serve --verbose
+
+MCP client config (e.g. claude_desktop_config.json):
+    {
+        "mcpServers": {
+            "hermes": {
+                "command": "hermes",
+                "args": ["mcp", "serve"]
+            }
+        }
+    }
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+import os
+import re
+import sys
+import threading
+import time
+from dataclasses import dataclass, field
+from datetime import datetime
+from pathlib import Path
+from typing import Any, Dict, List, Optional
+
+logger = logging.getLogger("hermes.mcp_serve")
+
+# ---------------------------------------------------------------------------
+# Lazy MCP SDK import
+# ---------------------------------------------------------------------------
+
+_MCP_SERVER_AVAILABLE = False
+try:
+    from mcp.server.fastmcp import FastMCP
+
+    _MCP_SERVER_AVAILABLE = True
+except ImportError:
+    FastMCP = None  # type: ignore[assignment,misc]
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+def _get_sessions_dir() -> Path:
+    """Return the sessions directory using HERMES_HOME."""
+    try:
+        from hermes_constants import get_hermes_home
+        return get_hermes_home() / "sessions"
+    except ImportError:
+        return Path(os.environ.get("HERMES_HOME", Path.home() / ".hermes")) / "sessions"
+
+
+def _get_session_db():
+    """Get a SessionDB instance for reading message transcripts."""
+    try:
+        from hermes_state import SessionDB
+        return SessionDB()
+    except Exception as e:
+        logger.debug("SessionDB unavailable: %s", e)
+        return None
+
+
+def _load_sessions_index() -> dict:
+    """Load the gateway sessions.json index directly.
+
+    Returns a dict of session_key -> entry_dict with platform routing info.
+    This avoids importing the full SessionStore which needs GatewayConfig.
+    """
+    sessions_file = _get_sessions_dir() / "sessions.json"
+    if not sessions_file.exists():
+        return {}
+    try:
+        with open(sessions_file, "r", encoding="utf-8") as f:
+            return json.load(f)
+    except Exception as e:
+        logger.debug("Failed to load sessions.json: %s", e)
+        return {}
+
+
+def _load_channel_directory() -> dict:
+    """Load the cached channel directory for available targets."""
+    try:
+        from hermes_constants import get_hermes_home
+        directory_file = get_hermes_home() / "channel_directory.json"
+    except ImportError:
+        directory_file = Path(
+            os.environ.get("HERMES_HOME", Path.home() / ".hermes")
+        ) / "channel_directory.json"
+
+    if not directory_file.exists():
+        return {}
+    try:
+        with open(directory_file, "r", encoding="utf-8") as f:
+            return json.load(f)
+    except Exception as e:
+        logger.debug("Failed to load channel_directory.json: %s", e)
+        return {}
+
+
+def _extract_message_content(msg: dict) -> str:
+    """Extract text content from a message, handling multi-part content."""
+    content = msg.get("content", "")
+    if isinstance(content, list):
+        text_parts = [
+            p.get("text", "") for p in content
+            if isinstance(p, dict) and p.get("type") == "text"
+        ]
+        return "\n".join(text_parts)
+    return str(content) if content else ""
+
+
+def _extract_attachments(msg: dict) -> List[dict]:
+    """Extract non-text attachments from a message.
+
+    Finds: multi-part image/file content blocks, MEDIA: tags in text,
+    image URLs, and file references.
+    """
+    attachments = []
+    content = msg.get("content", "")
+
+    # Multi-part content blocks (image_url, file, etc.)
+    if isinstance(content, list):
+        for part in content:
+            if not isinstance(part, dict):
+                continue
+            ptype = part.get("type", "")
+            if ptype == "image_url":
+                url = part.get("image_url", {}).get("url", "") if isinstance(part.get("image_url"), dict) else ""
+                if url:
+                    attachments.append({"type": "image", "url": url})
+            elif ptype == "image":
+                url = part.get("url", part.get("source", {}).get("url", ""))
+                if url:
+                    attachments.append({"type": "image", "url": url})
+            elif ptype not in ("text",):
+                # Unknown non-text content type
+                attachments.append({"type": ptype, "data": part})
+
+    # MEDIA: tags in text content
+    text = _extract_message_content(msg)
+    if text:
+        media_pattern = re.compile(r'MEDIA:\s*(\S+)')
+        for match in media_pattern.finditer(text):
+            path = match.group(1)
+            attachments.append({"type": "media", "path": path})
+
+    return attachments
+
+
+# ---------------------------------------------------------------------------
+# Event Bridge — polls SessionDB for new messages, maintains event queue
+# ---------------------------------------------------------------------------
+
+QUEUE_LIMIT = 1000
+POLL_INTERVAL = 0.2  # seconds between DB polls (200ms)
+
+
+@dataclass
+class QueueEvent:
+    """An event in the bridge's in-memory queue."""
+    cursor: int
+    type: str  # "message", "approval_requested", "approval_resolved"
+    session_key: str = ""
+    data: dict = field(default_factory=dict)
+
+
+class EventBridge:
+    """Background poller that watches SessionDB for new messages and
+    maintains an in-memory event queue with waiter support.
+
+    This is the Hermes equivalent of OpenClaw's WebSocket gateway bridge.
+    Instead of WebSocket events, we poll the SQLite database for changes.
+    """
+
+    def __init__(self):
+        self._queue: List[QueueEvent] = []
+        self._cursor = 0
+        self._lock = threading.Lock()
+        self._new_event = threading.Event()
+        self._running = False
+        self._thread: Optional[threading.Thread] = None
+        self._last_poll_timestamps: Dict[str, float] = {}  # session_key -> unix timestamp
+        # In-memory approval tracking (populated from events)
+        self._pending_approvals: Dict[str, dict] = {}
+        # mtime cache — skip expensive work when files haven't changed
+        self._sessions_json_mtime: float = 0.0
+        self._state_db_mtime: float = 0.0
+        self._cached_sessions_index: dict = {}
+
+    def start(self):
+        """Start the background polling thread."""
+        if self._running:
+            return
+        self._running = True
+        self._thread = threading.Thread(target=self._poll_loop, daemon=True)
+        self._thread.start()
+        logger.debug("EventBridge started")
+
+    def stop(self):
+        """Stop the background polling thread."""
+        self._running = False
+        self._new_event.set()  # Wake any waiters
+        if self._thread:
+            self._thread.join(timeout=5)
+        logger.debug("EventBridge stopped")
+
+    def poll_events(
+        self,
+        after_cursor: int = 0,
+        session_key: Optional[str] = None,
+        limit: int = 20,
+    ) -> dict:
+        """Return events since after_cursor, optionally filtered by session_key."""
+        with self._lock:
+            events = [
+                e for e in self._queue
+                if e.cursor > after_cursor
+                and (not session_key or e.session_key == session_key)
+            ][:limit]
+
+        next_cursor = events[-1].cursor if events else after_cursor
+        return {
+            "events": [
+                {"cursor": e.cursor, "type": e.type,
+                 "session_key": e.session_key, **e.data}
+                for e in events
+            ],
+            "next_cursor": next_cursor,
+        }
+
+    def wait_for_event(
+        self,
+        after_cursor: int = 0,
+        session_key: Optional[str] = None,
+        timeout_ms: int = 30000,
+    ) -> Optional[dict]:
+        """Block until a matching event arrives or timeout expires."""
+        deadline = time.monotonic() + (timeout_ms / 1000.0)
+
+        while time.monotonic() < deadline:
+            with self._lock:
+                for e in self._queue:
+                    if e.cursor > after_cursor and (
+                        not session_key or e.session_key == session_key
+                    ):
+                        return {
+                            "cursor": e.cursor, "type": e.type,
+                            "session_key": e.session_key, **e.data,
+                        }
+
+            remaining = deadline - time.monotonic()
+            if remaining <= 0:
+                break
+            self._new_event.clear()
+            self._new_event.wait(timeout=min(remaining, POLL_INTERVAL))
+
+        return None
+
+    def list_pending_approvals(self) -> List[dict]:
+        """List approval requests observed during this bridge session."""
+        with self._lock:
+            return sorted(
+                self._pending_approvals.values(),
+                key=lambda a: a.get("created_at", ""),
+            )
+
+    def respond_to_approval(self, approval_id: str, decision: str) -> dict:
+        """Resolve a pending approval (best-effort without gateway IPC)."""
+        with self._lock:
+            approval = self._pending_approvals.pop(approval_id, None)
+
+        if not approval:
+            return {"error": f"Approval not found: {approval_id}"}
+
+        self._enqueue(QueueEvent(
+            cursor=0,  # Will be set by _enqueue
+            type="approval_resolved",
+            session_key=approval.get("session_key", ""),
+            data={"approval_id": approval_id, "decision": decision},
+        ))
+
+        return {"resolved": True, "approval_id": approval_id, "decision": decision}
+
+    def _enqueue(self, event: QueueEvent) -> None:
+        """Add an event to the queue and wake any waiters."""
+        with self._lock:
+            self._cursor += 1
+            event.cursor = self._cursor
+            self._queue.append(event)
+            # Trim queue to limit
+            while len(self._queue) > QUEUE_LIMIT:
+                self._queue.pop(0)
+        self._new_event.set()
+
+    def _poll_loop(self):
+        """Background loop: poll SessionDB for new messages."""
+        db = _get_session_db()
+        if not db:
+            logger.warning("EventBridge: SessionDB unavailable, event polling disabled")
+            return
+
+        while self._running:
+            try:
+                self._poll_once(db)
+            except Exception as e:
+                logger.debug("EventBridge poll error: %s", e)
+            time.sleep(POLL_INTERVAL)
+
+    def _poll_once(self, db):
+        """Check for new messages across all sessions.
+
+        Uses mtime checks on sessions.json and state.db to skip work
+        when nothing has changed — makes 200ms polling essentially free.
+        """
+        # Check if sessions.json has changed (mtime check is ~1μs)
+        sessions_file = _get_sessions_dir() / "sessions.json"
+        try:
+            sj_mtime = sessions_file.stat().st_mtime if sessions_file.exists() else 0.0
+        except OSError:
+            sj_mtime = 0.0
+
+        if sj_mtime != self._sessions_json_mtime:
+            self._sessions_json_mtime = sj_mtime
+            self._cached_sessions_index = _load_sessions_index()
+
+        # Check if state.db has changed
+        try:
+            from hermes_constants import get_hermes_home
+            db_file = get_hermes_home() / "state.db"
+        except ImportError:
+            db_file = Path(os.environ.get("HERMES_HOME", Path.home() / ".hermes")) / "state.db"
+
+        try:
+            db_mtime = db_file.stat().st_mtime if db_file.exists() else 0.0
+        except OSError:
+            db_mtime = 0.0
+
+        if db_mtime == self._state_db_mtime and sj_mtime == self._sessions_json_mtime:
+            return  # Nothing changed since last poll — skip entirely
+
+        self._state_db_mtime = db_mtime
+        entries = self._cached_sessions_index
+
+        for session_key, entry in entries.items():
+            session_id = entry.get("session_id", "")
+            if not session_id:
+                continue
+
+            last_seen = self._last_poll_timestamps.get(session_key, 0.0)
+
+            try:
+                messages = db.get_messages(session_id)
+            except Exception:
+                continue
+
+            if not messages:
+                continue
+
+            # Normalize timestamps to float for comparison
+            def _ts_float(ts) -> float:
+                if isinstance(ts, (int, float)):
+                    return float(ts)
+                if isinstance(ts, str) and ts:
+                    try:
+                        return float(ts)
+                    except ValueError:
+                        # ISO string — parse to epoch
+                        try:
+                            from datetime import datetime
+                            return datetime.fromisoformat(ts).timestamp()
+                        except Exception:
+                            return 0.0
+                return 0.0
+
+            # Find messages newer than our last seen timestamp
+            new_messages = []
+            for msg in messages:
+                ts = _ts_float(msg.get("timestamp", 0))
+                role = msg.get("role", "")
+                if role not in ("user", "assistant"):
+                    continue
+                if ts > last_seen:
+                    new_messages.append(msg)
+
+            for msg in new_messages:
+                content = _extract_message_content(msg)
+                if not content:
+                    continue
+                self._enqueue(QueueEvent(
+                    cursor=0,
+                    type="message",
+                    session_key=session_key,
+                    data={
+                        "role": msg.get("role", ""),
+                        "content": content[:500],
+                        "timestamp": str(msg.get("timestamp", "")),
+                        "message_id": str(msg.get("id", "")),
+                    },
+                ))
+
+            # Update last seen to the most recent message timestamp
+            all_ts = [_ts_float(m.get("timestamp", 0)) for m in messages]
+            if all_ts:
+                latest = max(all_ts)
+                if latest > last_seen:
+                    self._last_poll_timestamps[session_key] = latest
+
+
+# ---------------------------------------------------------------------------
+# MCP Server
+# ---------------------------------------------------------------------------
+
+def create_mcp_server(event_bridge: Optional[EventBridge] = None) -> "FastMCP":
+    """Create and return the Hermes MCP server with all tools registered."""
+    if not _MCP_SERVER_AVAILABLE:
+        raise ImportError(
+            "MCP server requires the 'mcp' package. "
+            "Install with: pip install 'hermes-agent[mcp]'"
+        )
+
+    mcp = FastMCP(
+        "hermes",
+        instructions=(
+            "Hermes Agent messaging bridge. Use these tools to interact with "
+            "conversations across Telegram, Discord, Slack, WhatsApp, Signal, "
+            "Matrix, and other connected platforms."
+        ),
+    )
+
+    bridge = event_bridge or EventBridge()
+
+    # -- conversations_list ------------------------------------------------
+
+    @mcp.tool()
+    def conversations_list(
+        platform: Optional[str] = None,
+        limit: int = 50,
+        search: Optional[str] = None,
+    ) -> str:
+        """List active messaging conversations across connected platforms.
+
+        Returns conversations with their session keys (needed for messages_read),
+        platform, chat type, display name, and last activity time.
+
+        Args:
+            platform: Filter by platform name (telegram, discord, slack, etc.)
+            limit: Maximum number of conversations to return (default 50)
+            search: Optional text to filter conversations by name
+        """
+        entries = _load_sessions_index()
+        conversations = []
+
+        for key, entry in entries.items():
+            origin = entry.get("origin", {})
+            entry_platform = entry.get("platform") or origin.get("platform", "")
+
+            if platform and entry_platform.lower() != platform.lower():
+                continue
+
+            display_name = entry.get("display_name", "")
+            chat_name = origin.get("chat_name", "")
+            if search:
+                search_lower = search.lower()
+                if (search_lower not in display_name.lower()
+                        and search_lower not in chat_name.lower()
+                        and search_lower not in key.lower()):
+                    continue
+
+            conversations.append({
+                "session_key": key,
+                "session_id": entry.get("session_id", ""),
+                "platform": entry_platform,
+                "chat_type": entry.get("chat_type", origin.get("chat_type", "")),
+                "display_name": display_name,
+                "chat_name": chat_name,
+                "user_name": origin.get("user_name", ""),
+                "updated_at": entry.get("updated_at", ""),
+            })
+
+        conversations.sort(key=lambda c: c.get("updated_at", ""), reverse=True)
+        conversations = conversations[:limit]
+
+        return json.dumps({
+            "count": len(conversations),
+            "conversations": conversations,
+        }, indent=2)
+
+    # -- conversation_get --------------------------------------------------
+
+    @mcp.tool()
+    def conversation_get(session_key: str) -> str:
+        """Get detailed info about one conversation by its session key.
+
+        Args:
+            session_key: The session key from conversations_list
+        """
+        entries = _load_sessions_index()
+        entry = entries.get(session_key)
+
+        if not entry:
+            return json.dumps({"error": f"Conversation not found: {session_key}"})
+
+        origin = entry.get("origin", {})
+        return json.dumps({
+            "session_key": session_key,
+            "session_id": entry.get("session_id", ""),
+            "platform": entry.get("platform") or origin.get("platform", ""),
+            "chat_type": entry.get("chat_type", origin.get("chat_type", "")),
+            "display_name": entry.get("display_name", ""),
+            "user_name": origin.get("user_name", ""),
+            "chat_name": origin.get("chat_name", ""),
+            "chat_id": origin.get("chat_id", ""),
+            "thread_id": origin.get("thread_id"),
+            "updated_at": entry.get("updated_at", ""),
+            "created_at": entry.get("created_at", ""),
+            "input_tokens": entry.get("input_tokens", 0),
+            "output_tokens": entry.get("output_tokens", 0),
+            "total_tokens": entry.get("total_tokens", 0),
+        }, indent=2)
+
+    # -- messages_read -----------------------------------------------------
+
+    @mcp.tool()
+    def messages_read(
+        session_key: str,
+        limit: int = 50,
+    ) -> str:
+        """Read recent messages from a conversation.
+
+        Returns the message history in chronological order with role, content,
+        and timestamp for each message.
+
+        Args:
+            session_key: The session key from conversations_list
+            limit: Maximum number of messages to return (default 50, most recent)
+        """
+        entries = _load_sessions_index()
+        entry = entries.get(session_key)
+        if not entry:
+            return json.dumps({"error": f"Conversation not found: {session_key}"})
+
+        session_id = entry.get("session_id", "")
+        if not session_id:
+            return json.dumps({"error": "No session ID for this conversation"})
+
+        db = _get_session_db()
+        if not db:
+            return json.dumps({"error": "Session database unavailable"})
+
+        try:
+            all_messages = db.get_messages(session_id)
+        except Exception as e:
+            return json.dumps({"error": f"Failed to read messages: {e}"})
+
+        filtered = []
+        for msg in all_messages:
+            role = msg.get("role", "")
+            if role in ("user", "assistant"):
+                content = _extract_message_content(msg)
+                if content:
+                    filtered.append({
+                        "id": str(msg.get("id", "")),
+                        "role": role,
+                        "content": content[:2000],
+                        "timestamp": msg.get("timestamp", ""),
+                    })
+
+        messages = filtered[-limit:]
+
+        return json.dumps({
+            "session_key": session_key,
+            "count": len(messages),
+            "total_in_session": len(filtered),
+            "messages": messages,
+        }, indent=2)
+
+    # -- attachments_fetch -------------------------------------------------
+
+    @mcp.tool()
+    def attachments_fetch(
+        session_key: str,
+        message_id: str,
+    ) -> str:
+        """List non-text attachments for a message in a conversation.
+
+        Extracts images, media files, and other non-text content blocks
+        from the specified message.
+
+        Args:
+            session_key: The session key from conversations_list
+            message_id: The message ID from messages_read
+        """
+        entries = _load_sessions_index()
+        entry = entries.get(session_key)
+        if not entry:
+            return json.dumps({"error": f"Conversation not found: {session_key}"})
+
+        session_id = entry.get("session_id", "")
+        if not session_id:
+            return json.dumps({"error": "No session ID for this conversation"})
+
+        db = _get_session_db()
+        if not db:
+            return json.dumps({"error": "Session database unavailable"})
+
+        try:
+            all_messages = db.get_messages(session_id)
+        except Exception as e:
+            return json.dumps({"error": f"Failed to read messages: {e}"})
+
+        # Find the target message
+        target_msg = None
+        for msg in all_messages:
+            if str(msg.get("id", "")) == message_id:
+                target_msg = msg
+                break
+
+        if not target_msg:
+            return json.dumps({"error": f"Message not found: {message_id}"})
+
+        attachments = _extract_attachments(target_msg)
+
+        return json.dumps({
+            "message_id": message_id,
+            "count": len(attachments),
+            "attachments": attachments,
+        }, indent=2)
+
+    # -- events_poll -------------------------------------------------------
+
+    @mcp.tool()
+    def events_poll(
+        after_cursor: int = 0,
+        session_key: Optional[str] = None,
+        limit: int = 20,
+    ) -> str:
+        """Poll for new conversation events since a cursor position.
+
+        Returns events that have occurred since the given cursor. Use the
+        returned next_cursor value for subsequent polls.
+
+        Event types: message, approval_requested, approval_resolved
+
+        Args:
+            after_cursor: Return events after this cursor (0 for all)
+            session_key: Optional filter to one conversation
+            limit: Maximum events to return (default 20)
+        """
+        result = bridge.poll_events(
+            after_cursor=after_cursor,
+            session_key=session_key,
+            limit=limit,
+        )
+        return json.dumps(result, indent=2)
+
+    # -- events_wait -------------------------------------------------------
+
+    @mcp.tool()
+    def events_wait(
+        after_cursor: int = 0,
+        session_key: Optional[str] = None,
+        timeout_ms: int = 30000,
+    ) -> str:
+        """Wait for the next conversation event (long-poll).
+
+        Blocks until a matching event arrives or the timeout expires.
+        Use this for near-real-time event delivery without polling.
+
+        Args:
+            after_cursor: Wait for events after this cursor
+            session_key: Optional filter to one conversation
+            timeout_ms: Maximum wait time in milliseconds (default 30000)
+        """
+        event = bridge.wait_for_event(
+            after_cursor=after_cursor,
+            session_key=session_key,
+            timeout_ms=min(timeout_ms, 300000),  # Cap at 5 minutes
+        )
+        if event:
+            return json.dumps({"event": event}, indent=2)
+        return json.dumps({"event": None, "reason": "timeout"}, indent=2)
+
+    # -- messages_send -----------------------------------------------------
+
+    @mcp.tool()
+    def messages_send(
+        target: str,
+        message: str,
+    ) -> str:
+        """Send a message to a platform conversation.
+
+        The target format is "platform:chat_id" — same format used by the
+        channels_list tool. You can also use human-friendly channel names
+        that will be resolved automatically.
+
+        Examples:
+            target="telegram:6308981865"
+            target="discord:#general"
+            target="slack:#engineering"
+
+        Args:
+            target: Platform target in "platform:identifier" format
+            message: The message text to send
+        """
+        if not target or not message:
+            return json.dumps({"error": "Both target and message are required"})
+
+        try:
+            from tools.send_message_tool import send_message_tool
+            result_str = send_message_tool(
+                {"action": "send", "target": target, "message": message}
+            )
+            return result_str
+        except ImportError:
+            return json.dumps({"error": "Send message tool not available"})
+        except Exception as e:
+            return json.dumps({"error": f"Send failed: {e}"})
+
+    # -- channels_list -----------------------------------------------------
+
+    @mcp.tool()
+    def channels_list(platform: Optional[str] = None) -> str:
+        """List available messaging channels and targets across platforms.
+
+        Returns channels that you can send messages to. The target strings
+        returned here can be used directly with the messages_send tool.
+
+        Args:
+            platform: Filter by platform name (telegram, discord, slack, etc.)
+        """
+        directory = _load_channel_directory()
+        if not directory:
+            entries = _load_sessions_index()
+            targets = []
+            seen = set()
+            for key, entry in entries.items():
+                origin = entry.get("origin", {})
+                p = entry.get("platform") or origin.get("platform", "")
+                chat_id = origin.get("chat_id", "")
+                if not p or not chat_id:
+                    continue
+                if platform and p.lower() != platform.lower():
+                    continue
+                target_str = f"{p}:{chat_id}"
+                if target_str in seen:
+                    continue
+                seen.add(target_str)
+                targets.append({
+                    "target": target_str,
+                    "platform": p,
+                    "name": entry.get("display_name") or origin.get("chat_name", ""),
+                    "chat_type": entry.get("chat_type", origin.get("chat_type", "")),
+                })
+            return json.dumps({"count": len(targets), "channels": targets}, indent=2)
+
+        channels = []
+        for plat, entries_list in directory.items():
+            if platform and plat.lower() != platform.lower():
+                continue
+            if isinstance(entries_list, list):
+                for ch in entries_list:
+                    if isinstance(ch, dict):
+                        chat_id = ch.get("id", ch.get("chat_id", ""))
+                        channels.append({
+                            "target": f"{plat}:{chat_id}" if chat_id else plat,
+                            "platform": plat,
+                            "name": ch.get("name", ch.get("display_name", "")),
+                            "chat_type": ch.get("type", ""),
+                        })
+
+        return json.dumps({"count": len(channels), "channels": channels}, indent=2)
+
+    # -- permissions_list_open ---------------------------------------------
+
+    @mcp.tool()
+    def permissions_list_open() -> str:
+        """List pending approval requests observed during this bridge session.
+
+        Returns exec and plugin approval requests that the bridge has seen
+        since it started. Approvals are live-session only — older approvals
+        from before the bridge connected are not included.
+        """
+        approvals = bridge.list_pending_approvals()
+        return json.dumps({
+            "count": len(approvals),
+            "approvals": approvals,
+        }, indent=2)
+
+    # -- permissions_respond -----------------------------------------------
+
+    @mcp.tool()
+    def permissions_respond(
+        id: str,
+        decision: str,
+    ) -> str:
+        """Respond to a pending approval request.
+
+        Args:
+            id: The approval ID from permissions_list_open
+            decision: One of "allow-once", "allow-always", or "deny"
+        """
+        if decision not in ("allow-once", "allow-always", "deny"):
+            return json.dumps({
+                "error": f"Invalid decision: {decision}. "
+                         f"Must be allow-once, allow-always, or deny"
+            })
+
+        result = bridge.respond_to_approval(id, decision)
+        return json.dumps(result, indent=2)
+
+    return mcp
+
+
+# ---------------------------------------------------------------------------
+# Entry point
+# ---------------------------------------------------------------------------
+
+def run_mcp_server(verbose: bool = False) -> None:
+    """Start the Hermes MCP server on stdio."""
+    if not _MCP_SERVER_AVAILABLE:
+        print(
+            "Error: MCP server requires the 'mcp' package.\n"
+            "Install with: pip install 'hermes-agent[mcp]'",
+            file=sys.stderr,
+        )
+        sys.exit(1)
+
+    if verbose:
+        logging.basicConfig(level=logging.DEBUG, stream=sys.stderr)
+    else:
+        logging.basicConfig(level=logging.WARNING, stream=sys.stderr)
+
+    bridge = EventBridge()
+    bridge.start()
+
+    server = create_mcp_server(event_bridge=bridge)
+
+    import asyncio
+
+    async def _run():
+        try:
+            await server.run_stdio_async()
+        finally:
+            bridge.stop()
+
+    try:
+        asyncio.run(_run())
+    except KeyboardInterrupt:
+        bridge.stop()

nix/nixosModules.nix

@@ -111,6 +111,7 @@
       fi
       mkdir -p "$TARGET_HOME"
       chown "$HERMES_UID:$HERMES_GID" "$TARGET_HOME"
+      chmod 0750 "$TARGET_HOME"
 
       # Ensure HERMES_HOME is owned by the target user
       if [ -n "''${HERMES_HOME:-}" ] && [ -d "$HERMES_HOME" ]; then
@@ -551,8 +552,8 @@
       # ── Directories ───────────────────────────────────────────────────
       {
         systemd.tmpfiles.rules = [
-          "d ${cfg.stateDir}                0755 ${cfg.user} ${cfg.group} - -"
-          "d ${cfg.stateDir}/.hermes        0755 ${cfg.user} ${cfg.group} - -"
+          "d ${cfg.stateDir}                0750 ${cfg.user} ${cfg.group} - -"
+          "d ${cfg.stateDir}/.hermes        0750 ${cfg.user} ${cfg.group} - -"
           "d ${cfg.stateDir}/home           0750 ${cfg.user} ${cfg.group} - -"
           "d ${cfg.workingDirectory}         0750 ${cfg.user} ${cfg.group} - -"
         ];
@@ -566,21 +567,23 @@
           mkdir -p ${cfg.stateDir}/home
           mkdir -p ${cfg.workingDirectory}
           chown ${cfg.user}:${cfg.group} ${cfg.stateDir} ${cfg.stateDir}/.hermes ${cfg.stateDir}/home ${cfg.workingDirectory}
+          chmod 0750 ${cfg.stateDir} ${cfg.stateDir}/.hermes ${cfg.stateDir}/home ${cfg.workingDirectory}
 
           # Merge Nix settings into existing config.yaml.
           # Preserves user-added keys (skills, streaming, etc.); Nix keys win.
           # If configFile is user-provided (not generated), overwrite instead of merge.
           ${if cfg.configFile != null then ''
-            install -o ${cfg.user} -g ${cfg.group} -m 0644 -D ${configFile} ${cfg.stateDir}/.hermes/config.yaml
+            install -o ${cfg.user} -g ${cfg.group} -m 0640 -D ${configFile} ${cfg.stateDir}/.hermes/config.yaml
           '' else ''
             ${configMergeScript} ${generatedConfigFile} ${cfg.stateDir}/.hermes/config.yaml
             chown ${cfg.user}:${cfg.group} ${cfg.stateDir}/.hermes/config.yaml
-            chmod 0644 ${cfg.stateDir}/.hermes/config.yaml
+            chmod 0640 ${cfg.stateDir}/.hermes/config.yaml
           ''}
 
           # Managed mode marker (so interactive shells also detect NixOS management)
           touch ${cfg.stateDir}/.hermes/.managed
           chown ${cfg.user}:${cfg.group} ${cfg.stateDir}/.hermes/.managed
+          chmod 0644 ${cfg.stateDir}/.hermes/.managed
 
           # Seed auth file if provided
           ${lib.optionalString (cfg.authFile != null) ''
@@ -612,7 +615,7 @@ HERMES_NIX_ENV_EOF
 
           # Link documents into workspace
           ${lib.concatStringsSep "\n" (lib.mapAttrsToList (name: _value: ''
-            install -o ${cfg.user} -g ${cfg.group} -m 0644 ${documentDerivation}/${name} ${cfg.workingDirectory}/${name}
+            install -o ${cfg.user} -g ${cfg.group} -m 0640 ${documentDerivation}/${name} ${cfg.workingDirectory}/${name}
           '') cfg.documents)}
         '';
       }

optional-skills/communication/DESCRIPTION.md

@@ -0,0 +1 @@
+Communication and decision-making frameworks — structured response formats for proposals, trade-off analysis, and stakeholder-ready recommendations.

optional-skills/communication/one-three-one-rule/SKILL.md

@@ -0,0 +1,103 @@
+---
+name: one-three-one-rule
+description: >
+  Structured decision-making framework for technical proposals and trade-off analysis.
+  When the user faces a choice between multiple approaches (architecture decisions,
+  tool selection, refactoring strategies, migration paths), this skill produces a
+  1-3-1 format: one clear problem statement, three distinct options with pros/cons,
+  and one concrete recommendation with definition of done and implementation plan.
+  Use when the user asks for a "1-3-1", says "give me options", or needs help
+  choosing between competing approaches.
+version: 1.0.0
+author: Willard Moore
+license: MIT
+category: communication
+metadata:
+  hermes:
+    tags: [communication, decision-making, proposals, trade-offs]
+---
+
+# 1-3-1 Communication Rule
+
+Structured decision-making format for when a task has multiple viable approaches and the user needs a clear recommendation. Produces a concise problem framing, three options with trade-offs, and an actionable plan for the recommended path.
+
+## When to Use
+
+- The user explicitly asks for a "1-3-1" response.
+- The user says "give me options" or "what are my choices" for a technical decision.
+- A task has multiple viable approaches with meaningful trade-offs (architecture, tooling, migration strategy).
+- The user needs a proposal they can forward to a team or stakeholder.
+
+Do NOT use for simple questions with one obvious answer, debugging sessions, or tasks where the user has already decided on an approach.
+
+## Procedure
+
+1. **Problem** (one sentence)
+   - State the core decision or desired outcome in a single concise sentence.
+   - Focus on the *what*, not the *how* — no implementation details, tool names, or specific technologies.
+   - Keep it tight. If you need "and", you're describing two problems.
+
+2. **Options** (exactly three)
+   - Present three distinct, viable approaches labeled A, B, C.
+   - Each option gets a brief description, pros, and cons.
+   - Options should represent genuinely different strategies, not minor variations of the same approach.
+
+3. **Recommendation** (one option)
+   - State which option you recommend and why, based on the user's context and priorities.
+   - Be direct — this is your professional judgment, not a hedge.
+
+4. **Definition of Done**
+   - List the specific success criteria for the recommended option.
+   - These are concrete, verifiable outcomes — not vague aspirations.
+   - If the user picks a different option, revise this section to match.
+
+5. **Implementation Plan**
+   - Concrete steps to execute the recommended option.
+   - Include specific commands, tools, or actions where applicable.
+   - If the user picks a different option, revise this section to match.
+
+## Verification
+
+- Response has exactly one Problem sentence.
+- Response has exactly three Options (A, B, C) with pros and cons for each.
+- Response has a single Recommendation that picks one option with reasoning.
+- Definition of Done and Implementation Plan align with the recommended option.
+- If the user selects a different option, Recommendation, DoD, and Implementation Plan update accordingly.
+
+## Example
+
+User: "Give me a 1-3-1 for adding retry logic to our API client."
+
+Agent:
+
+**Problem:** API calls to external services fail intermittently under load and need graceful retry handling.
+
+**Options:**
+
+- **Option A: Exponential backoff with jitter in a shared utility.**
+  Pros: Simple, well-understood pattern; single implementation shared across all callers.
+  Cons: No circuit-breaking; callers still block during retries.
+
+- **Option B: Retry library (tenacity) as a decorator on each endpoint call.**
+  Pros: Configurable per-endpoint; built-in support for circuit breakers and rate limits.
+  Cons: Adds a dependency; decorator pattern can obscure control flow.
+
+- **Option C: Queue-based async retry with dead letter handling.**
+  Pros: Non-blocking; failed calls are tracked and replayable; scales well.
+  Cons: Significant complexity increase; requires a queue backend; overkill for low-volume services.
+
+**Recommendation:** Option A. The call volume doesn't justify queue infrastructure, and a shared utility with exponential backoff + jitter covers the failure mode cleanly without new dependencies.
+
+**Definition of Done:**
+- `api_retry()` utility handles transient HTTP errors (429, 502, 503, 504).
+- Retries up to 3 times with exponential backoff and jitter.
+- All external API calls route through the retry utility.
+- Logs each retry attempt with endpoint and attempt number.
+- Tests cover retry exhaustion, immediate success, and transient-then-success scenarios.
+
+**Implementation Plan:**
+1. Create `utils/api_retry.py` with configurable max retries, base delay, and retryable status codes.
+2. Add jitter using `random.uniform(0, base_delay)` to prevent thundering herd.
+3. Wrap existing API calls in `api_client.py` with the retry utility.
+4. Add unit tests mocking HTTP responses for each retry scenario.
+5. Verify under load with a simple stress test against a flaky endpoint mock.

optional-skills/productivity/canvas/SKILL.md

@@ -0,0 +1,97 @@
+---
+name: canvas
+description: Canvas LMS integration — fetch enrolled courses and assignments using API token authentication.
+version: 1.0.0
+author: community
+license: MIT
+prerequisites:
+  env_vars: [CANVAS_API_TOKEN, CANVAS_BASE_URL]
+metadata:
+  hermes:
+    tags: [Canvas, LMS, Education, Courses, Assignments]
+---
+
+# Canvas LMS — Course & Assignment Access
+
+Read-only access to Canvas LMS for listing courses and assignments.
+
+## Scripts
+
+- `scripts/canvas_api.py` — Python CLI for Canvas API calls
+
+## Setup
+
+1. Log in to your Canvas instance in a browser
+2. Go to **Account → Settings** (click your profile icon, then Settings)
+3. Scroll to **Approved Integrations** and click **+ New Access Token**
+4. Name the token (e.g., "Hermes Agent"), set an optional expiry, and click **Generate Token**
+5. Copy the token and add to `~/.hermes/.env`:
+
+```
+CANVAS_API_TOKEN=your_token_here
+CANVAS_BASE_URL=https://yourschool.instructure.com
+```
+
+The base URL is whatever appears in your browser when you're logged into Canvas (no trailing slash).
+
+## Usage
+
+```bash
+CANVAS="python $HERMES_HOME/skills/productivity/canvas/scripts/canvas_api.py"
+
+# List all active courses
+$CANVAS list_courses --enrollment-state active
+
+# List all courses (any state)
+$CANVAS list_courses
+
+# List assignments for a specific course
+$CANVAS list_assignments 12345
+
+# List assignments ordered by due date
+$CANVAS list_assignments 12345 --order-by due_at
+```
+
+## Output Format
+
+**list_courses** returns:
+```json
+[{"id": 12345, "name": "Intro to CS", "course_code": "CS101", "workflow_state": "available", "start_at": "...", "end_at": "..."}]
+```
+
+**list_assignments** returns:
+```json
+[{"id": 67890, "name": "Homework 1", "due_at": "2025-02-15T23:59:00Z", "points_possible": 100, "submission_types": ["online_upload"], "html_url": "...", "description": "...", "course_id": 12345}]
+```
+
+Note: Assignment descriptions are truncated to 500 characters. The `html_url` field links to the full assignment page in Canvas.
+
+## API Reference (curl)
+
+```bash
+# List courses
+curl -s -H "Authorization: Bearer $CANVAS_API_TOKEN" \
+  "$CANVAS_BASE_URL/api/v1/courses?enrollment_state=active&per_page=10"
+
+# List assignments for a course
+curl -s -H "Authorization: Bearer $CANVAS_API_TOKEN" \
+  "$CANVAS_BASE_URL/api/v1/courses/COURSE_ID/assignments?per_page=10&order_by=due_at"
+```
+
+Canvas uses `Link` headers for pagination. The Python script handles pagination automatically.
+
+## Rules
+
+- This skill is **read-only** — it only fetches data, never modifies courses or assignments
+- On first use, verify auth by running `$CANVAS list_courses` — if it fails with 401, guide the user through setup
+- Canvas rate-limits to ~700 requests per 10 minutes; check `X-Rate-Limit-Remaining` header if hitting limits
+
+## Troubleshooting
+
+| Problem | Fix |
+|---------|-----|
+| 401 Unauthorized | Token invalid or expired — regenerate in Canvas Settings |
+| 403 Forbidden | Token lacks permission for this course |
+| Empty course list | Try `--enrollment-state active` or omit the flag to see all states |
+| Wrong institution | Verify `CANVAS_BASE_URL` matches the URL in your browser |
+| Timeout errors | Check network connectivity to your Canvas instance |

optional-skills/productivity/canvas/scripts/canvas_api.py

@@ -0,0 +1,157 @@
+#!/usr/bin/env python3
+"""Canvas LMS API CLI for Hermes Agent.
+
+A thin CLI wrapper around the Canvas REST API.
+Authenticates using a personal access token from environment variables.
+
+Usage:
+  python canvas_api.py list_courses [--per-page N] [--enrollment-state STATE]
+  python canvas_api.py list_assignments COURSE_ID [--per-page N] [--order-by FIELD]
+"""
+
+import argparse
+import json
+import os
+import sys
+
+import requests
+
+CANVAS_API_TOKEN = os.environ.get("CANVAS_API_TOKEN", "")
+CANVAS_BASE_URL = os.environ.get("CANVAS_BASE_URL", "").rstrip("/")
+
+
+def _check_config():
+    """Validate required environment variables are set."""
+    missing = []
+    if not CANVAS_API_TOKEN:
+        missing.append("CANVAS_API_TOKEN")
+    if not CANVAS_BASE_URL:
+        missing.append("CANVAS_BASE_URL")
+    if missing:
+        print(
+            f"Missing required environment variables: {', '.join(missing)}\n"
+            "Set them in ~/.hermes/.env or export them in your shell.\n"
+            "See the canvas skill SKILL.md for setup instructions.",
+            file=sys.stderr,
+        )
+        sys.exit(1)
+
+
+def _headers():
+    return {"Authorization": f"Bearer {CANVAS_API_TOKEN}"}
+
+
+def _paginated_get(url, params=None, max_items=200):
+    """Fetch all pages up to max_items, following Canvas Link headers."""
+    results = []
+    while url and len(results) < max_items:
+        resp = requests.get(url, headers=_headers(), params=params, timeout=30)
+        resp.raise_for_status()
+        results.extend(resp.json())
+        params = None  # params are included in the Link URL for subsequent pages
+        url = None
+        link = resp.headers.get("Link", "")
+        for part in link.split(","):
+            if 'rel="next"' in part:
+                url = part.split(";")[0].strip().strip("<>")
+    return results[:max_items]
+
+
+# =========================================================================
+# Commands
+# =========================================================================
+
+
+def list_courses(args):
+    """List enrolled courses."""
+    _check_config()
+    url = f"{CANVAS_BASE_URL}/api/v1/courses"
+    params = {"per_page": args.per_page}
+    if args.enrollment_state:
+        params["enrollment_state"] = args.enrollment_state
+    try:
+        courses = _paginated_get(url, params)
+    except requests.HTTPError as e:
+        print(f"API error: {e.response.status_code} {e.response.text}", file=sys.stderr)
+        sys.exit(1)
+    output = [
+        {
+            "id": c["id"],
+            "name": c.get("name", ""),
+            "course_code": c.get("course_code", ""),
+            "enrollment_term_id": c.get("enrollment_term_id"),
+            "start_at": c.get("start_at"),
+            "end_at": c.get("end_at"),
+            "workflow_state": c.get("workflow_state", ""),
+        }
+        for c in courses
+    ]
+    print(json.dumps(output, indent=2))
+
+
+def list_assignments(args):
+    """List assignments for a course."""
+    _check_config()
+    url = f"{CANVAS_BASE_URL}/api/v1/courses/{args.course_id}/assignments"
+    params = {"per_page": args.per_page}
+    if args.order_by:
+        params["order_by"] = args.order_by
+    try:
+        assignments = _paginated_get(url, params)
+    except requests.HTTPError as e:
+        print(f"API error: {e.response.status_code} {e.response.text}", file=sys.stderr)
+        sys.exit(1)
+    output = [
+        {
+            "id": a["id"],
+            "name": a.get("name", ""),
+            "description": (a.get("description") or "")[:500],
+            "due_at": a.get("due_at"),
+            "points_possible": a.get("points_possible"),
+            "submission_types": a.get("submission_types", []),
+            "html_url": a.get("html_url", ""),
+            "course_id": a.get("course_id"),
+        }
+        for a in assignments
+    ]
+    print(json.dumps(output, indent=2))
+
+
+# =========================================================================
+# CLI parser
+# =========================================================================
+
+
+def main():
+    parser = argparse.ArgumentParser(
+        description="Canvas LMS API CLI for Hermes Agent"
+    )
+    sub = parser.add_subparsers(dest="command", required=True)
+
+    # --- list_courses ---
+    p = sub.add_parser("list_courses", help="List enrolled courses")
+    p.add_argument("--per-page", type=int, default=50, help="Results per page (default 50)")
+    p.add_argument(
+        "--enrollment-state",
+        default="",
+        help="Filter by enrollment state (active, invited_or_pending, completed)",
+    )
+    p.set_defaults(func=list_courses)
+
+    # --- list_assignments ---
+    p = sub.add_parser("list_assignments", help="List assignments for a course")
+    p.add_argument("course_id", help="Canvas course ID")
+    p.add_argument("--per-page", type=int, default=50, help="Results per page (default 50)")
+    p.add_argument(
+        "--order-by",
+        default="",
+        help="Order by field (due_at, name, position)",
+    )
+    p.set_defaults(func=list_assignments)
+
+    args = parser.parse_args()
+    args.func(args)
+
+
+if __name__ == "__main__":
+    main()

optional-skills/productivity/memento-flashcards/SKILL.md

@@ -0,0 +1,324 @@
+---
+name: memento-flashcards
+description: >-
+  Spaced-repetition flashcard system. Create cards from facts or text,
+  chat with flashcards using free-text answers graded by the agent,
+  generate quizzes from YouTube transcripts, review due cards with
+  adaptive scheduling, and export/import decks as CSV.
+version: 1.0.0
+author: Memento AI
+license: MIT
+platforms: [macos, linux]
+metadata:
+  hermes:
+    tags: [Education, Flashcards, Spaced Repetition, Learning, Quiz, YouTube]
+    requires_toolsets: [terminal]
+    category: productivity
+---
+
+# Memento Flashcards — Spaced-Repetition Flashcard Skill
+
+## Overview
+
+Memento gives you a local, file-based flashcard system with spaced-repetition scheduling.
+Users can chat with their flashcards by answering in free text and having the agent grade the response before scheduling the next review.
+Use it whenever the user wants to:
+
+- **Remember a fact** — turn any statement into a Q/A flashcard
+- **Study with spaced repetition** — review due cards with adaptive intervals and agent-graded free-text answers
+- **Quiz from a YouTube video** — fetch a transcript and generate a 5-question quiz
+- **Manage decks** — organise cards into collections, export/import CSV
+
+All card data lives in a single JSON file. No external API keys are required — you (the agent) generate flashcard content and quiz questions directly.
+
+User-facing response style for Memento Flashcards:
+- Use plain text only. Do not use Markdown formatting in replies to the user.
+- Keep review and quiz feedback brief and neutral. Avoid extra praise, pep, or long explanations.
+
+## When to Use
+
+Use this skill when the user wants to:
+- Save facts as flashcards for later review
+- Review due cards with spaced repetition
+- Generate a quiz from a YouTube video transcript
+- Import, export, inspect, or delete flashcard data
+
+Do not use this skill for general Q&A, coding help, or non-memory tasks.
+
+## Quick Reference
+
+| User intent | Action |
+|---|---|
+| "Remember that X" / "save this as a flashcard" | Generate a Q/A card, call `memento_cards.py add` |
+| Sends a fact without mentioning flashcards | Ask "Want me to save this as a Memento flashcard?" — only create if confirmed |
+| "Create a flashcard" | Ask for Q, A, collection; call `memento_cards.py add` |
+| "Review my cards" | Call `memento_cards.py due`, present cards one-by-one |
+| "Quiz me on [YouTube URL]" | Call `youtube_quiz.py fetch VIDEO_ID`, generate 5 questions, call `memento_cards.py add-quiz` |
+| "Export my cards" | Call `memento_cards.py export --output PATH` |
+| "Import cards from CSV" | Call `memento_cards.py import --file PATH --collection NAME` |
+| "Show my stats" | Call `memento_cards.py stats` |
+| "Delete a card" | Call `memento_cards.py delete --id ID` |
+| "Delete a collection" | Call `memento_cards.py delete-collection --collection NAME` |
+
+## Card Storage
+
+Cards are stored in a JSON file at:
+
+```
+~/.hermes/skills/productivity/memento-flashcards/data/cards.json
+```
+
+**Never edit this file directly.** Always use `memento_cards.py` subcommands. The script handles atomic writes (write to temp file, then rename) to prevent corruption.
+
+The file is created automatically on first use.
+
+## Procedure
+
+### Creating Cards from Facts
+
+### Activation Rules
+
+Not every factual statement should become a flashcard. Use this three-tier check:
+
+1. **Explicit intent** — the user mentions "memento", "flashcard", "remember this", "save this card", "add a card", or similar phrasing that clearly requests a flashcard → **create the card directly**, no confirmation needed.
+2. **Implicit intent** — the user sends a factual statement without mentioning flashcards (e.g. "The speed of light is 299,792 km/s") → **ask first**: "Want me to save this as a Memento flashcard?" Only create the card if the user confirms.
+3. **No intent** — the message is a coding task, a question, instructions, normal conversation, or anything that is clearly not a fact to memorize → **do NOT activate this skill at all**. Let other skills or default behavior handle it.
+
+When activation is confirmed (tier 1 directly, tier 2 after confirmation), generate a flashcard:
+
+**Step 1:** Turn the statement into a Q/A pair. Use this format internally:
+
+```
+Turn the factual statement into a front-back pair.
+Return exactly two lines:
+Q: <question text>
+A: <answer text>
+
+Statement: "{statement}"
+```
+
+Rules:
+- The question should test recall of the key fact
+- The answer should be concise and direct
+
+**Step 2:** Call the script to store the card:
+
+```bash
+python3 ~/.hermes/skills/productivity/memento-flashcards/scripts/memento_cards.py add \
+  --question "What year did World War 2 end?" \
+  --answer "1945" \
+  --collection "History"
+```
+
+If the user doesn't specify a collection, use `"General"` as the default.
+
+The script outputs JSON confirming the created card.
+
+### Manual Card Creation
+
+When the user explicitly asks to create a flashcard, ask them for:
+1. The question (front of card)
+2. The answer (back of card)
+3. The collection name (optional — default to `"General"`)
+
+Then call `memento_cards.py add` as above.
+
+### Reviewing Due Cards
+
+When the user wants to review, fetch all due cards:
+
+```bash
+python3 ~/.hermes/skills/productivity/memento-flashcards/scripts/memento_cards.py due
+```
+
+This returns a JSON array of cards where `next_review_at <= now`. If a collection filter is needed:
+
+```bash
+python3 ~/.hermes/skills/productivity/memento-flashcards/scripts/memento_cards.py due --collection "History"
+```
+
+**Review flow (free-text grading):**
+
+Here is an example of the EXACT interaction pattern you must follow. The user answers, you grade them, tell them the correct answer, then rate the card.
+
+**Example interaction:**
+
+> **Agent:** What year did the Berlin Wall fall?
+>
+> **User:** 1991
+>
+> **Agent:** Not quite. The Berlin Wall fell in 1989. Next review is tomorrow.
+> *(agent calls: memento_cards.py rate --id ABC --rating hard --user-answer "1991")*
+>
+> Next question: Who was the first person to walk on the moon?
+
+**The rules:**
+
+1. Show only the question. Wait for the user to answer.
+2. After receiving their answer, compare it to the expected answer and grade it:
+   - **correct** → user got the key fact right (even if worded differently)
+   - **partial** → right track but missing the core detail
+   - **incorrect** → wrong or off-topic
+3. **You MUST tell the user the correct answer and how they did.** Keep it short and plain-text. Use this format:
+   - correct: "Correct. Answer: {answer}. Next review in 7 days."
+   - partial: "Close. Answer: {answer}. {what they missed}. Next review in 3 days."
+   - incorrect: "Not quite. Answer: {answer}. Next review tomorrow."
+4. Then call the rate command: correct→easy, partial→good, incorrect→hard.
+5. Then show the next question.
+
+```bash
+python3 ~/.hermes/skills/productivity/memento-flashcards/scripts/memento_cards.py rate \
+  --id CARD_ID --rating easy --user-answer "what the user said"
+```
+
+**Never skip step 3.** The user must always see the correct answer and feedback before you move on.
+
+If no cards are due, tell the user: "No cards due for review right now. Check back later!"
+
+**Retire override:** At any point the user can say "retire this card" to permanently remove it from reviews. Use `--rating retire` for this.
+
+### Spaced Repetition Algorithm
+
+The rating determines the next review interval:
+
+| Rating | Interval | ease_streak | Status change |
+|---|---|---|---|
+| **hard** | +1 day | reset to 0 | stays learning |
+| **good** | +3 days | reset to 0 | stays learning |
+| **easy** | +7 days | +1 | if ease_streak >= 3 → retired |
+| **retire** | permanent | reset to 0 | → retired |
+
+- **learning**: card is actively in rotation
+- **retired**: card won't appear in reviews (user has mastered it or manually retired it)
+- Three consecutive "easy" ratings automatically retire a card
+
+### YouTube Quiz Generation
+
+When the user sends a YouTube URL and wants a quiz:
+
+**Step 1:** Extract the video ID from the URL (e.g. `dQw4w9WgXcQ` from `https://www.youtube.com/watch?v=dQw4w9WgXcQ`).
+
+**Step 2:** Fetch the transcript:
+
+```bash
+python3 ~/.hermes/skills/productivity/memento-flashcards/scripts/youtube_quiz.py fetch VIDEO_ID
+```
+
+This returns `{"title": "...", "transcript": "..."}` or an error.
+
+If the script reports `missing_dependency`, tell the user to install it:
+```bash
+pip install youtube-transcript-api
+```
+
+**Step 3:** Generate 5 quiz questions from the transcript. Use these rules:
+
+```
+You are creating a 5-question quiz for a podcast episode.
+Return ONLY a JSON array with exactly 5 objects.
+Each object must contain keys 'question' and 'answer'.
+
+Selection criteria:
+- Prioritize important, surprising, or foundational facts.
+- Skip filler, obvious details, and facts that require heavy context.
+- Never return true/false questions.
+- Never ask only for a date.
+
+Question rules:
+- Each question must test exactly one discrete fact.
+- Use clear, unambiguous wording.
+- Prefer What, Who, How many, Which.
+- Avoid open-ended Describe or Explain prompts.
+
+Answer rules:
+- Each answer must be under 240 characters.
+- Lead with the answer itself, not preamble.
+- Add only minimal clarifying detail if needed.
+```
+
+Use the first 15,000 characters of the transcript as context. Generate the questions yourself (you are the LLM).
+
+**Step 4:** Validate the output is valid JSON with exactly 5 items, each having non-empty `question` and `answer` strings. If validation fails, retry once.
+
+**Step 5:** Store quiz cards:
+
+```bash
+python3 ~/.hermes/skills/productivity/memento-flashcards/scripts/memento_cards.py add-quiz \
+  --video-id "VIDEO_ID" \
+  --questions '[{"question":"...","answer":"..."},...]' \
+  --collection "Quiz - Episode Title"
+```
+
+The script deduplicates by `video_id` — if cards for that video already exist, it skips creation and reports the existing cards.
+
+**Step 6:** Present questions one-by-one using the same free-text grading flow:
+1. Show "Question 1/5: ..." and wait for the user's answer. Never include the answer or any hint about revealing it.
+2. Wait for the user to answer in their own words
+3. Grade their answer using the grading prompt (see "Reviewing Due Cards" section)
+4. **IMPORTANT: You MUST reply to the user with feedback before doing anything else.** Show the grade, the correct answer, and when the card is next due. Do NOT silently skip to the next question. Keep it short and plain-text. Example: "Not quite. Answer: {answer}. Next review tomorrow."
+5. **After showing feedback**, call the rate command and then show the next question in the same message:
+```bash
+python3 ~/.hermes/skills/productivity/memento-flashcards/scripts/memento_cards.py rate \
+  --id CARD_ID --rating easy --user-answer "what the user said"
+```
+6. Repeat. Every answer MUST receive visible feedback before the next question.
+
+### Export/Import CSV
+
+**Export:**
+```bash
+python3 ~/.hermes/skills/productivity/memento-flashcards/scripts/memento_cards.py export \
+  --output ~/flashcards.csv
+```
+
+Produces a 3-column CSV: `question,answer,collection` (no header row).
+
+**Import:**
+```bash
+python3 ~/.hermes/skills/productivity/memento-flashcards/scripts/memento_cards.py import \
+  --file ~/flashcards.csv \
+  --collection "Imported"
+```
+
+Reads a CSV with columns: question, answer, and optionally collection (column 3). If the collection column is missing, uses the `--collection` argument.
+
+### Statistics
+
+```bash
+python3 ~/.hermes/skills/productivity/memento-flashcards/scripts/memento_cards.py stats
+```
+
+Returns JSON with:
+- `total`: total card count
+- `learning`: cards in active rotation
+- `retired`: mastered cards
+- `due_now`: cards due for review right now
+- `collections`: breakdown by collection name
+
+## Pitfalls
+
+- **Never edit `cards.json` directly** — always use the script subcommands to avoid corruption
+- **Transcript failures** — some YouTube videos have no English transcript or have transcripts disabled; inform the user and suggest another video
+- **Optional dependency** — `youtube_quiz.py` needs `youtube-transcript-api`; if missing, tell the user to run `pip install youtube-transcript-api`
+- **Large imports** — CSV imports with thousands of rows work fine but the JSON output may be verbose; summarize the result for the user
+- **Video ID extraction** — support both `youtube.com/watch?v=ID` and `youtu.be/ID` URL formats
+
+## Verification
+
+Verify the helper scripts directly:
+
+```bash
+python3 ~/.hermes/skills/productivity/memento-flashcards/scripts/memento_cards.py stats
+python3 ~/.hermes/skills/productivity/memento-flashcards/scripts/memento_cards.py add --question "Capital of France?" --answer "Paris" --collection "General"
+python3 ~/.hermes/skills/productivity/memento-flashcards/scripts/memento_cards.py due
+```
+
+If you are testing from the repo checkout, run:
+
+```bash
+pytest tests/skills/test_memento_cards.py tests/skills/test_youtube_quiz.py -q
+```
+
+Agent-level verification:
+- Start a review and confirm feedback is plain text, brief, and always includes the correct answer before the next card
+- Run a YouTube quiz flow and confirm each answer receives visible feedback before the next question

optional-skills/productivity/memento-flashcards/scripts/memento_cards.py

@@ -0,0 +1,353 @@
+#!/usr/bin/env python3
+"""Memento card storage, spaced-repetition engine, and CSV I/O.
+
+Stdlib-only. All output is JSON for agent parsing.
+Data file: $HERMES_HOME/skills/productivity/memento-flashcards/data/cards.json
+"""
+
+import argparse
+import csv
+import json
+import os
+import sys
+import tempfile
+import uuid
+from datetime import datetime, timedelta, timezone
+from pathlib import Path
+
+_HERMES_HOME = Path(os.environ.get("HERMES_HOME", Path.home() / ".hermes"))
+DATA_DIR = _HERMES_HOME / "skills" / "productivity" / "memento-flashcards" / "data"
+CARDS_FILE = DATA_DIR / "cards.json"
+
+RETIRED_SENTINEL = "9999-12-31T23:59:59+00:00"
+
+
+def _now() -> datetime:
+    return datetime.now(timezone.utc)
+
+
+def _iso(dt: datetime) -> str:
+    return dt.isoformat()
+
+
+def _parse_iso(s: str) -> datetime:
+    return datetime.fromisoformat(s)
+
+
+def _empty_store() -> dict:
+    return {"cards": [], "version": 1}
+
+
+def _load() -> dict:
+    if not CARDS_FILE.exists():
+        return _empty_store()
+    try:
+        with open(CARDS_FILE, "r", encoding="utf-8") as f:
+            data = json.load(f)
+        if not isinstance(data, dict) or "cards" not in data:
+            return _empty_store()
+        return data
+    except (json.JSONDecodeError, OSError):
+        return _empty_store()
+
+
+def _save(data: dict) -> None:
+    DATA_DIR.mkdir(parents=True, exist_ok=True)
+    fd, tmp = tempfile.mkstemp(dir=DATA_DIR, suffix=".tmp")
+    try:
+        with os.fdopen(fd, "w", encoding="utf-8") as f:
+            json.dump(data, f, indent=2, ensure_ascii=False)
+            f.write("\n")
+        os.replace(tmp, CARDS_FILE)
+    except BaseException:
+        try:
+            os.unlink(tmp)
+        except OSError:
+            pass
+        raise
+
+
+def _out(obj: object) -> None:
+    json.dump(obj, sys.stdout, indent=2, ensure_ascii=False)
+    sys.stdout.write("\n")
+
+
+# ── Subcommands ──────────────────────────────────────────────────────────────
+
+def cmd_add(args: argparse.Namespace) -> None:
+    data = _load()
+    now = _now()
+    card = {
+        "id": str(uuid.uuid4()),
+        "question": args.question,
+        "answer": args.answer,
+        "collection": args.collection or "General",
+        "status": "learning",
+        "ease_streak": 0,
+        "next_review_at": _iso(now),
+        "created_at": _iso(now),
+        "video_id": None,
+        "last_user_answer": None,
+    }
+    data["cards"].append(card)
+    _save(data)
+    _out({"ok": True, "card": card})
+
+
+def cmd_add_quiz(args: argparse.Namespace) -> None:
+    data = _load()
+    now = _now()
+
+    try:
+        questions = json.loads(args.questions)
+    except json.JSONDecodeError as exc:
+        _out({"ok": False, "error": f"Invalid JSON for --questions: {exc}"})
+        sys.exit(1)
+
+    # Dedup: skip if cards with this video_id already exist
+    existing_ids = {c["video_id"] for c in data["cards"] if c.get("video_id")}
+    if args.video_id in existing_ids:
+        existing = [c for c in data["cards"] if c.get("video_id") == args.video_id]
+        _out({"ok": True, "skipped": True, "reason": "duplicate_video_id", "existing_count": len(existing), "cards": existing})
+        return
+
+    created = []
+    for qa in questions:
+        card = {
+            "id": str(uuid.uuid4()),
+            "question": qa["question"],
+            "answer": qa["answer"],
+            "collection": args.collection or "Quiz",
+            "status": "learning",
+            "ease_streak": 0,
+            "next_review_at": _iso(now),
+            "created_at": _iso(now),
+            "video_id": args.video_id,
+            "last_user_answer": None,
+        }
+        data["cards"].append(card)
+        created.append(card)
+
+    _save(data)
+    _out({"ok": True, "created_count": len(created), "cards": created})
+
+
+def cmd_due(args: argparse.Namespace) -> None:
+    data = _load()
+    now = _now()
+    due = []
+    for card in data["cards"]:
+        if card["status"] == "retired":
+            continue
+        review_at = _parse_iso(card["next_review_at"])
+        if review_at <= now:
+            if args.collection and card["collection"] != args.collection:
+                continue
+            due.append(card)
+    _out({"ok": True, "count": len(due), "cards": due})
+
+
+def cmd_rate(args: argparse.Namespace) -> None:
+    data = _load()
+    now = _now()
+    card = None
+    for c in data["cards"]:
+        if c["id"] == args.id:
+            card = c
+            break
+    if not card:
+        _out({"ok": False, "error": f"Card not found: {args.id}"})
+        sys.exit(1)
+
+    rating = args.rating
+    user_answer = getattr(args, "user_answer", None)
+    if user_answer is not None:
+        card["last_user_answer"] = user_answer
+
+    if rating == "retire":
+        card["status"] = "retired"
+        card["next_review_at"] = RETIRED_SENTINEL
+        card["ease_streak"] = 0
+    elif rating == "hard":
+        card["next_review_at"] = _iso(now + timedelta(days=1))
+        card["ease_streak"] = 0
+    elif rating == "good":
+        card["next_review_at"] = _iso(now + timedelta(days=3))
+        card["ease_streak"] = 0
+    elif rating == "easy":
+        card["next_review_at"] = _iso(now + timedelta(days=7))
+        card["ease_streak"] = card.get("ease_streak", 0) + 1
+        if card["ease_streak"] >= 3:
+            card["status"] = "retired"
+
+    _save(data)
+    _out({"ok": True, "card": card})
+
+
+def cmd_list(args: argparse.Namespace) -> None:
+    data = _load()
+    cards = data["cards"]
+    if args.collection:
+        cards = [c for c in cards if c["collection"] == args.collection]
+    if args.status:
+        cards = [c for c in cards if c["status"] == args.status]
+    _out({"ok": True, "count": len(cards), "cards": cards})
+
+
+def cmd_stats(args: argparse.Namespace) -> None:
+    data = _load()
+    now = _now()
+    total = len(data["cards"])
+    learning = sum(1 for c in data["cards"] if c["status"] == "learning")
+    retired = sum(1 for c in data["cards"] if c["status"] == "retired")
+    due_now = 0
+    for c in data["cards"]:
+        if c["status"] != "retired" and _parse_iso(c["next_review_at"]) <= now:
+            due_now += 1
+
+    collections: dict[str, int] = {}
+    for c in data["cards"]:
+        name = c["collection"]
+        collections[name] = collections.get(name, 0) + 1
+
+    _out({
+        "ok": True,
+        "total": total,
+        "learning": learning,
+        "retired": retired,
+        "due_now": due_now,
+        "collections": collections,
+    })
+
+
+def cmd_export(args: argparse.Namespace) -> None:
+    data = _load()
+    output_path = Path(args.output).expanduser()
+    with open(output_path, "w", newline="", encoding="utf-8") as f:
+        writer = csv.writer(f, lineterminator="\n")
+        for card in data["cards"]:
+            writer.writerow([card["question"], card["answer"], card["collection"]])
+    _out({"ok": True, "exported": len(data["cards"]), "path": str(output_path)})
+
+
+def cmd_import(args: argparse.Namespace) -> None:
+    data = _load()
+    now = _now()
+    file_path = Path(args.file).expanduser()
+
+    if not file_path.exists():
+        _out({"ok": False, "error": f"File not found: {file_path}"})
+        sys.exit(1)
+
+    created = 0
+    with open(file_path, "r", encoding="utf-8") as f:
+        reader = csv.reader(f)
+        for row in reader:
+            if len(row) < 2:
+                continue
+            question = row[0].strip()
+            answer = row[1].strip()
+            collection = row[2].strip() if len(row) >= 3 and row[2].strip() else (args.collection or "Imported")
+            if not question or not answer:
+                continue
+            card = {
+                "id": str(uuid.uuid4()),
+                "question": question,
+                "answer": answer,
+                "collection": collection,
+                "status": "learning",
+                "ease_streak": 0,
+                "next_review_at": _iso(now),
+                "created_at": _iso(now),
+                "video_id": None,
+                "last_user_answer": None,
+            }
+            data["cards"].append(card)
+            created += 1
+
+    _save(data)
+    _out({"ok": True, "imported": created})
+
+
+def cmd_delete(args: argparse.Namespace) -> None:
+    data = _load()
+    original = len(data["cards"])
+    data["cards"] = [c for c in data["cards"] if c["id"] != args.id]
+    removed = original - len(data["cards"])
+    if removed == 0:
+        _out({"ok": False, "error": f"Card not found: {args.id}"})
+        sys.exit(1)
+    _save(data)
+    _out({"ok": True, "deleted": args.id})
+
+
+def cmd_delete_collection(args: argparse.Namespace) -> None:
+    data = _load()
+    original = len(data["cards"])
+    data["cards"] = [c for c in data["cards"] if c["collection"] != args.collection]
+    removed = original - len(data["cards"])
+    _save(data)
+    _out({"ok": True, "deleted_count": removed, "collection": args.collection})
+
+
+# ── CLI ──────────────────────────────────────────────────────────────────────
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description="Memento flashcard manager")
+    sub = parser.add_subparsers(dest="command", required=True)
+
+    p_add = sub.add_parser("add", help="Create one card")
+    p_add.add_argument("--question", required=True)
+    p_add.add_argument("--answer", required=True)
+    p_add.add_argument("--collection", default="General")
+
+    p_quiz = sub.add_parser("add-quiz", help="Batch-add quiz cards")
+    p_quiz.add_argument("--video-id", required=True)
+    p_quiz.add_argument("--questions", required=True, help="JSON array of {question, answer}")
+    p_quiz.add_argument("--collection", default="Quiz")
+
+    p_due = sub.add_parser("due", help="List due cards")
+    p_due.add_argument("--collection", default=None)
+
+    p_rate = sub.add_parser("rate", help="Rate a card")
+    p_rate.add_argument("--id", required=True)
+    p_rate.add_argument("--rating", required=True, choices=["easy", "good", "hard", "retire"])
+    p_rate.add_argument("--user-answer", default=None)
+
+    p_list = sub.add_parser("list", help="List cards")
+    p_list.add_argument("--collection", default=None)
+    p_list.add_argument("--status", default=None, choices=["learning", "retired"])
+
+    sub.add_parser("stats", help="Show statistics")
+
+    p_export = sub.add_parser("export", help="Export cards to CSV")
+    p_export.add_argument("--output", required=True)
+
+    p_import = sub.add_parser("import", help="Import cards from CSV")
+    p_import.add_argument("--file", required=True)
+    p_import.add_argument("--collection", default="Imported")
+
+    p_del = sub.add_parser("delete", help="Delete one card")
+    p_del.add_argument("--id", required=True)
+
+    p_delcol = sub.add_parser("delete-collection", help="Delete all cards in a collection")
+    p_delcol.add_argument("--collection", required=True)
+
+    args = parser.parse_args()
+    cmd_map = {
+        "add": cmd_add,
+        "add-quiz": cmd_add_quiz,
+        "due": cmd_due,
+        "rate": cmd_rate,
+        "list": cmd_list,
+        "stats": cmd_stats,
+        "export": cmd_export,
+        "import": cmd_import,
+        "delete": cmd_delete,
+        "delete-collection": cmd_delete_collection,
+    }
+    cmd_map[args.command](args)
+
+
+if __name__ == "__main__":
+    main()

optional-skills/productivity/memento-flashcards/scripts/youtube_quiz.py

@@ -0,0 +1,88 @@
+#!/usr/bin/env python3
+"""Fetch YouTube transcripts for Memento quiz generation.
+
+Requires: pip install youtube-transcript-api
+The quiz question *generation* is done by the agent's LLM — this script only fetches transcripts.
+"""
+
+import argparse
+import json
+import re
+import sys
+
+
+def _out(obj: object) -> None:
+    json.dump(obj, sys.stdout, indent=2, ensure_ascii=False)
+    sys.stdout.write("\n")
+
+
+def _normalize_segments(segments: list) -> str:
+    parts = []
+    for seg in segments:
+        text = str(seg.get("text", "")).strip()
+        if text:
+            parts.append(text)
+    return re.sub(r"\s+", " ", " ".join(parts)).strip()
+
+
+def cmd_fetch(args: argparse.Namespace) -> None:
+    try:
+        import youtube_transcript_api  # noqa: F811
+    except ImportError:
+        _out({
+            "ok": False,
+            "error": "missing_dependency",
+            "message": "Run: pip install youtube-transcript-api",
+        })
+        sys.exit(1)
+
+    video_id = args.video_id
+    languages = ["en", "en-US", "en-GB", "en-CA", "en-AU"]
+
+    api = youtube_transcript_api.YouTubeTranscriptApi()
+    try:
+        raw = api.fetch(video_id, languages=languages)
+    except Exception as exc:
+        error_type = type(exc).__name__
+        _out({
+            "ok": False,
+            "error": "transcript_unavailable",
+            "error_type": error_type,
+            "message": f"Could not fetch transcript for {video_id}: {exc}",
+        })
+        sys.exit(1)
+
+    segments = raw
+    if hasattr(raw, "to_raw_data"):
+        segments = raw.to_raw_data()
+
+    text = _normalize_segments(segments)
+    if not text:
+        _out({
+            "ok": False,
+            "error": "empty_transcript",
+            "message": f"Transcript for {video_id} contained no usable text.",
+        })
+        sys.exit(1)
+
+    _out({
+        "ok": True,
+        "video_id": video_id,
+        "transcript": text,
+    })
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description="Memento YouTube transcript fetcher")
+    sub = parser.add_subparsers(dest="command", required=True)
+
+    p_fetch = sub.add_parser("fetch", help="Fetch transcript for a video")
+    p_fetch.add_argument("video_id", help="YouTube video ID")
+
+    args = parser.parse_args()
+    if args.command == "fetch":
+        cmd_fetch(args)
+
+
+if __name__ == "__main__":
+    main()

optional-skills/productivity/siyuan/SKILL.md

@@ -0,0 +1,297 @@
+---
+name: siyuan
+description: SiYuan Note API for searching, reading, creating, and managing blocks and documents in a self-hosted knowledge base via curl.
+version: 1.0.0
+author: FEUAZUR
+license: MIT
+metadata:
+  hermes:
+    tags: [SiYuan, Notes, Knowledge Base, PKM, API]
+    related_skills: [obsidian, notion]
+    homepage: https://github.com/siyuan-note/siyuan
+prerequisites:
+  env_vars: [SIYUAN_TOKEN]
+  commands: [curl, jq]
+required_environment_variables:
+  - name: SIYUAN_TOKEN
+    prompt: SiYuan API token
+    help: "Settings > About in SiYuan desktop app"
+  - name: SIYUAN_URL
+    prompt: SiYuan instance URL (default http://127.0.0.1:6806)
+    required_for: remote instances
+---
+
+# SiYuan Note API
+
+Use the [SiYuan](https://github.com/siyuan-note/siyuan) kernel API via curl to search, read, create, update, and delete blocks and documents in a self-hosted knowledge base. No extra tools needed -- just curl and an API token.
+
+## Prerequisites
+
+1. Install and run SiYuan (desktop or Docker)
+2. Get your API token: **Settings > About > API token**
+3. Store it in `~/.hermes/.env`:
+   ```
+   SIYUAN_TOKEN=your_token_here
+   SIYUAN_URL=http://127.0.0.1:6806
+   ```
+   `SIYUAN_URL` defaults to `http://127.0.0.1:6806` if not set.
+
+## API Basics
+
+All SiYuan API calls are **POST with JSON body**. Every request follows this pattern:
+
+```bash
+curl -s -X POST "${SIYUAN_URL:-http://127.0.0.1:6806}/api/..." \
+  -H "Authorization: Token $SIYUAN_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{"param": "value"}'
+```
+
+Responses are JSON with this structure:
+```json
+{"code": 0, "msg": "", "data": { ... }}
+```
+`code: 0` means success. Any other value is an error -- check `msg` for details.
+
+**ID format:** SiYuan IDs look like `20210808180117-6v0mkxr` (14-digit timestamp + 7 alphanumeric chars).
+
+## Quick Reference
+
+| Operation | Endpoint |
+|-----------|----------|
+| Full-text search | `/api/search/fullTextSearchBlock` |
+| SQL query | `/api/query/sql` |
+| Read block | `/api/block/getBlockKramdown` |
+| Read children | `/api/block/getChildBlocks` |
+| Get path | `/api/filetree/getHPathByID` |
+| Get attributes | `/api/attr/getBlockAttrs` |
+| List notebooks | `/api/notebook/lsNotebooks` |
+| List documents | `/api/filetree/listDocsByPath` |
+| Create notebook | `/api/notebook/createNotebook` |
+| Create document | `/api/filetree/createDocWithMd` |
+| Append block | `/api/block/appendBlock` |
+| Update block | `/api/block/updateBlock` |
+| Rename document | `/api/filetree/renameDocByID` |
+| Set attributes | `/api/attr/setBlockAttrs` |
+| Delete block | `/api/block/deleteBlock` |
+| Delete document | `/api/filetree/removeDocByID` |
+| Export as Markdown | `/api/export/exportMdContent` |
+
+## Common Operations
+
+### Search (Full-Text)
+
+```bash
+curl -s -X POST "${SIYUAN_URL:-http://127.0.0.1:6806}/api/search/fullTextSearchBlock" \
+  -H "Authorization: Token $SIYUAN_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{"query": "meeting notes", "page": 0}' | jq '.data.blocks[:5]'
+```
+
+### Search (SQL)
+
+Query the blocks database directly. Only SELECT statements are safe.
+
+```bash
+curl -s -X POST "${SIYUAN_URL:-http://127.0.0.1:6806}/api/query/sql" \
+  -H "Authorization: Token $SIYUAN_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{"stmt": "SELECT id, content, type, box FROM blocks WHERE content LIKE '\''%keyword%'\'' AND type='\''p'\'' LIMIT 20"}' | jq '.data'
+```
+
+Useful columns: `id`, `parent_id`, `root_id`, `box` (notebook ID), `path`, `content`, `type`, `subtype`, `created`, `updated`.
+
+### Read Block Content
+
+Returns block content in Kramdown (Markdown-like) format.
+
+```bash
+curl -s -X POST "${SIYUAN_URL:-http://127.0.0.1:6806}/api/block/getBlockKramdown" \
+  -H "Authorization: Token $SIYUAN_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{"id": "20210808180117-6v0mkxr"}' | jq '.data.kramdown'
+```
+
+### Read Child Blocks
+
+```bash
+curl -s -X POST "${SIYUAN_URL:-http://127.0.0.1:6806}/api/block/getChildBlocks" \
+  -H "Authorization: Token $SIYUAN_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{"id": "20210808180117-6v0mkxr"}' | jq '.data'
+```
+
+### Get Human-Readable Path
+
+```bash
+curl -s -X POST "${SIYUAN_URL:-http://127.0.0.1:6806}/api/filetree/getHPathByID" \
+  -H "Authorization: Token $SIYUAN_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{"id": "20210808180117-6v0mkxr"}' | jq '.data'
+```
+
+### Get Block Attributes
+
+```bash
+curl -s -X POST "${SIYUAN_URL:-http://127.0.0.1:6806}/api/attr/getBlockAttrs" \
+  -H "Authorization: Token $SIYUAN_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{"id": "20210808180117-6v0mkxr"}' | jq '.data'
+```
+
+### List Notebooks
+
+```bash
+curl -s -X POST "${SIYUAN_URL:-http://127.0.0.1:6806}/api/notebook/lsNotebooks" \
+  -H "Authorization: Token $SIYUAN_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{}' | jq '.data.notebooks[] | {id, name, closed}'
+```
+
+### List Documents in a Notebook
+
+```bash
+curl -s -X POST "${SIYUAN_URL:-http://127.0.0.1:6806}/api/filetree/listDocsByPath" \
+  -H "Authorization: Token $SIYUAN_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{"notebook": "NOTEBOOK_ID", "path": "/"}' | jq '.data.files[] | {id, name}'
+```
+
+### Create a Document
+
+```bash
+curl -s -X POST "${SIYUAN_URL:-http://127.0.0.1:6806}/api/filetree/createDocWithMd" \
+  -H "Authorization: Token $SIYUAN_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{
+    "notebook": "NOTEBOOK_ID",
+    "path": "/Meeting Notes/2026-03-22",
+    "markdown": "# Meeting Notes\n\n- Discussed project timeline\n- Assigned tasks"
+  }' | jq '.data'
+```
+
+### Create a Notebook
+
+```bash
+curl -s -X POST "${SIYUAN_URL:-http://127.0.0.1:6806}/api/notebook/createNotebook" \
+  -H "Authorization: Token $SIYUAN_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{"name": "My New Notebook"}' | jq '.data.notebook.id'
+```
+
+### Append Block to Document
+
+```bash
+curl -s -X POST "${SIYUAN_URL:-http://127.0.0.1:6806}/api/block/appendBlock" \
+  -H "Authorization: Token $SIYUAN_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{
+    "parentID": "DOCUMENT_OR_BLOCK_ID",
+    "data": "New paragraph added at the end.",
+    "dataType": "markdown"
+  }' | jq '.data'
+```
+
+Also available: `/api/block/prependBlock` (same params, inserts at the beginning) and `/api/block/insertBlock` (uses `previousID` instead of `parentID` to insert after a specific block).
+
+### Update Block Content
+
+```bash
+curl -s -X POST "${SIYUAN_URL:-http://127.0.0.1:6806}/api/block/updateBlock" \
+  -H "Authorization: Token $SIYUAN_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{
+    "id": "BLOCK_ID",
+    "data": "Updated content here.",
+    "dataType": "markdown"
+  }' | jq '.data'
+```
+
+### Rename a Document
+
+```bash
+curl -s -X POST "${SIYUAN_URL:-http://127.0.0.1:6806}/api/filetree/renameDocByID" \
+  -H "Authorization: Token $SIYUAN_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{"id": "DOCUMENT_ID", "title": "New Title"}'
+```
+
+### Set Block Attributes
+
+Custom attributes must be prefixed with `custom-`:
+
+```bash
+curl -s -X POST "${SIYUAN_URL:-http://127.0.0.1:6806}/api/attr/setBlockAttrs" \
+  -H "Authorization: Token $SIYUAN_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{
+    "id": "BLOCK_ID",
+    "attrs": {
+      "custom-status": "reviewed",
+      "custom-priority": "high"
+    }
+  }'
+```
+
+### Delete a Block
+
+```bash
+curl -s -X POST "${SIYUAN_URL:-http://127.0.0.1:6806}/api/block/deleteBlock" \
+  -H "Authorization: Token $SIYUAN_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{"id": "BLOCK_ID"}'
+```
+
+To delete a whole document: use `/api/filetree/removeDocByID` with `{"id": "DOC_ID"}`.
+To delete a notebook: use `/api/notebook/removeNotebook` with `{"notebook": "NOTEBOOK_ID"}`.
+
+### Export Document as Markdown
+
+```bash
+curl -s -X POST "${SIYUAN_URL:-http://127.0.0.1:6806}/api/export/exportMdContent" \
+  -H "Authorization: Token $SIYUAN_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{"id": "DOCUMENT_ID"}' | jq -r '.data.content'
+```
+
+## Block Types
+
+Common `type` values in SQL queries:
+
+| Type | Description |
+|------|-------------|
+| `d` | Document (root block) |
+| `p` | Paragraph |
+| `h` | Heading |
+| `l` | List |
+| `i` | List item |
+| `c` | Code block |
+| `m` | Math block |
+| `t` | Table |
+| `b` | Blockquote |
+| `s` | Super block |
+| `html` | HTML block |
+
+## Pitfalls
+
+- **All endpoints are POST** -- even read-only operations. Do not use GET.
+- **SQL safety**: only use SELECT queries. INSERT/UPDATE/DELETE/DROP are dangerous and should never be sent.
+- **ID validation**: IDs match the pattern `YYYYMMDDHHmmss-xxxxxxx`. Reject anything else.
+- **Error responses**: always check `code != 0` in responses before processing `data`.
+- **Large documents**: block content and export results can be very large. Use `LIMIT` in SQL and pipe through `jq` to extract only what you need.
+- **Notebook IDs**: when working with a specific notebook, get its ID first via `lsNotebooks`.
+
+## Alternative: MCP Server
+
+If you prefer a native integration instead of curl, install the SiYuan MCP server:
+
+```yaml
+# In ~/.hermes/config.yaml under mcp_servers:
+mcp_servers:
+  siyuan:
+    command: npx
+    args: ["-y", "@porkll/siyuan-mcp"]
+    env:
+      SIYUAN_TOKEN: "your_token"
+      SIYUAN_URL: "http://127.0.0.1:6806"
+```

optional-skills/research/scrapling/SKILL.md

@@ -0,0 +1,335 @@
+---
+name: scrapling
+description: Web scraping with Scrapling - HTTP fetching, stealth browser automation, Cloudflare bypass, and spider crawling via CLI and Python.
+version: 1.0.0
+author: FEUAZUR
+license: MIT
+metadata:
+  hermes:
+    tags: [Web Scraping, Browser, Cloudflare, Stealth, Crawling, Spider]
+    related_skills: [duckduckgo-search, domain-intel]
+    homepage: https://github.com/D4Vinci/Scrapling
+prerequisites:
+  commands: [scrapling, python]
+---
+
+# Scrapling
+
+[Scrapling](https://github.com/D4Vinci/Scrapling) is a web scraping framework with anti-bot bypass, stealth browser automation, and a spider framework. It provides three fetching strategies (HTTP, dynamic JS, stealth/Cloudflare) and a full CLI.
+
+**This skill is for educational and research purposes only.** Users must comply with local/international data scraping laws and respect website Terms of Service.
+
+## When to Use
+
+- Scraping static HTML pages (faster than browser tools)
+- Scraping JS-rendered pages that need a real browser
+- Bypassing Cloudflare Turnstile or bot detection
+- Crawling multiple pages with a spider
+- When the built-in `web_extract` tool does not return the data you need
+
+## Installation
+
+```bash
+pip install "scrapling[all]"
+scrapling install
+```
+
+Minimal install (HTTP only, no browser):
+```bash
+pip install scrapling
+```
+
+With browser automation only:
+```bash
+pip install "scrapling[fetchers]"
+scrapling install
+```
+
+## Quick Reference
+
+| Approach | Class | Use When |
+|----------|-------|----------|
+| HTTP | `Fetcher` / `FetcherSession` | Static pages, APIs, fast bulk requests |
+| Dynamic | `DynamicFetcher` / `DynamicSession` | JS-rendered content, SPAs |
+| Stealth | `StealthyFetcher` / `StealthySession` | Cloudflare, anti-bot protected sites |
+| Spider | `Spider` | Multi-page crawling with link following |
+
+## CLI Usage
+
+### Extract Static Page
+
+```bash
+scrapling extract get 'https://example.com' output.md
+```
+
+With CSS selector and browser impersonation:
+
+```bash
+scrapling extract get 'https://example.com' output.md \
+  --css-selector '.content' \
+  --impersonate 'chrome'
+```
+
+### Extract JS-Rendered Page
+
+```bash
+scrapling extract fetch 'https://example.com' output.md \
+  --css-selector '.dynamic-content' \
+  --disable-resources \
+  --network-idle
+```
+
+### Extract Cloudflare-Protected Page
+
+```bash
+scrapling extract stealthy-fetch 'https://protected-site.com' output.html \
+  --solve-cloudflare \
+  --block-webrtc \
+  --hide-canvas
+```
+
+### POST Request
+
+```bash
+scrapling extract post 'https://example.com/api' output.json \
+  --json '{"query": "search term"}'
+```
+
+### Output Formats
+
+The output format is determined by the file extension:
+- `.html` -- raw HTML
+- `.md` -- converted to Markdown
+- `.txt` -- plain text
+- `.json` / `.jsonl` -- JSON
+
+## Python: HTTP Scraping
+
+### Single Request
+
+```python
+from scrapling.fetchers import Fetcher
+
+page = Fetcher.get('https://quotes.toscrape.com/')
+quotes = page.css('.quote .text::text').getall()
+for q in quotes:
+    print(q)
+```
+
+### Session (Persistent Cookies)
+
+```python
+from scrapling.fetchers import FetcherSession
+
+with FetcherSession(impersonate='chrome') as session:
+    page = session.get('https://example.com/', stealthy_headers=True)
+    links = page.css('a::attr(href)').getall()
+    for link in links[:5]:
+        sub = session.get(link)
+        print(sub.css('h1::text').get())
+```
+
+### POST / PUT / DELETE
+
+```python
+page = Fetcher.post('https://api.example.com/data', json={"key": "value"})
+page = Fetcher.put('https://api.example.com/item/1', data={"name": "updated"})
+page = Fetcher.delete('https://api.example.com/item/1')
+```
+
+### With Proxy
+
+```python
+page = Fetcher.get('https://example.com', proxy='http://user:pass@proxy:8080')
+```
+
+## Python: Dynamic Pages (JS-Rendered)
+
+For pages that require JavaScript execution (SPAs, lazy-loaded content):
+
+```python
+from scrapling.fetchers import DynamicFetcher
+
+page = DynamicFetcher.fetch('https://example.com', headless=True)
+data = page.css('.js-loaded-content::text').getall()
+```
+
+### Wait for Specific Element
+
+```python
+page = DynamicFetcher.fetch(
+    'https://example.com',
+    wait_selector=('.results', 'visible'),
+    network_idle=True,
+)
+```
+
+### Disable Resources for Speed
+
+Blocks fonts, images, media, stylesheets (~25% faster):
+
+```python
+from scrapling.fetchers import DynamicSession
+
+with DynamicSession(headless=True, disable_resources=True, network_idle=True) as session:
+    page = session.fetch('https://example.com')
+    items = page.css('.item::text').getall()
+```
+
+### Custom Page Automation
+
+```python
+from playwright.sync_api import Page
+from scrapling.fetchers import DynamicFetcher
+
+def scroll_and_click(page: Page):
+    page.mouse.wheel(0, 3000)
+    page.wait_for_timeout(1000)
+    page.click('button.load-more')
+    page.wait_for_selector('.extra-results')
+
+page = DynamicFetcher.fetch('https://example.com', page_action=scroll_and_click)
+results = page.css('.extra-results .item::text').getall()
+```
+
+## Python: Stealth Mode (Anti-Bot Bypass)
+
+For Cloudflare-protected or heavily fingerprinted sites:
+
+```python
+from scrapling.fetchers import StealthyFetcher
+
+page = StealthyFetcher.fetch(
+    'https://protected-site.com',
+    headless=True,
+    solve_cloudflare=True,
+    block_webrtc=True,
+    hide_canvas=True,
+)
+content = page.css('.protected-content::text').getall()
+```
+
+### Stealth Session
+
+```python
+from scrapling.fetchers import StealthySession
+
+with StealthySession(headless=True, solve_cloudflare=True) as session:
+    page1 = session.fetch('https://protected-site.com/page1')
+    page2 = session.fetch('https://protected-site.com/page2')
+```
+
+## Element Selection
+
+All fetchers return a `Selector` object with these methods:
+
+### CSS Selectors
+
+```python
+page.css('h1::text').get()              # First h1 text
+page.css('a::attr(href)').getall()      # All link hrefs
+page.css('.quote .text::text').getall() # Nested selection
+```
+
+### XPath
+
+```python
+page.xpath('//div[@class="content"]/text()').getall()
+page.xpath('//a/@href').getall()
+```
+
+### Find Methods
+
+```python
+page.find_all('div', class_='quote')       # By tag + attribute
+page.find_by_text('Read more', tag='a')    # By text content
+page.find_by_regex(r'\$\d+\.\d{2}')       # By regex pattern
+```
+
+### Similar Elements
+
+Find elements with similar structure (useful for product listings, etc.):
+
+```python
+first_product = page.css('.product')[0]
+all_similar = first_product.find_similar()
+```
+
+### Navigation
+
+```python
+el = page.css('.target')[0]
+el.parent                # Parent element
+el.children              # Child elements
+el.next_sibling          # Next sibling
+el.prev_sibling          # Previous sibling
+```
+
+## Python: Spider Framework
+
+For multi-page crawling with link following:
+
+```python
+from scrapling.spiders import Spider, Request, Response
+
+class QuotesSpider(Spider):
+    name = "quotes"
+    start_urls = ["https://quotes.toscrape.com/"]
+    concurrent_requests = 10
+    download_delay = 1
+
+    async def parse(self, response: Response):
+        for quote in response.css('.quote'):
+            yield {
+                "text": quote.css('.text::text').get(),
+                "author": quote.css('.author::text').get(),
+                "tags": quote.css('.tag::text').getall(),
+            }
+
+        next_page = response.css('.next a::attr(href)').get()
+        if next_page:
+            yield response.follow(next_page)
+
+result = QuotesSpider().start()
+print(f"Scraped {len(result.items)} quotes")
+result.items.to_json("quotes.json")
+```
+
+### Multi-Session Spider
+
+Route requests to different fetcher types:
+
+```python
+from scrapling.fetchers import FetcherSession, AsyncStealthySession
+
+class SmartSpider(Spider):
+    name = "smart"
+    start_urls = ["https://example.com/"]
+
+    def configure_sessions(self, manager):
+        manager.add("fast", FetcherSession(impersonate="chrome"))
+        manager.add("stealth", AsyncStealthySession(headless=True), lazy=True)
+
+    async def parse(self, response: Response):
+        for link in response.css('a::attr(href)').getall():
+            if "protected" in link:
+                yield Request(link, sid="stealth")
+            else:
+                yield Request(link, sid="fast", callback=self.parse)
+```
+
+### Pause/Resume Crawling
+
+```python
+spider = QuotesSpider(crawldir="./crawl_checkpoint")
+spider.start()  # Ctrl+C to pause, re-run to resume from checkpoint
+```
+
+## Pitfalls
+
+- **Browser install required**: run `scrapling install` after pip install -- without it, `DynamicFetcher` and `StealthyFetcher` will fail
+- **Timeouts**: DynamicFetcher/StealthyFetcher timeout is in **milliseconds** (default 30000), Fetcher timeout is in **seconds**
+- **Cloudflare bypass**: `solve_cloudflare=True` adds 5-15 seconds to fetch time -- only enable when needed
+- **Resource usage**: StealthyFetcher runs a real browser -- limit concurrent usage
+- **Legal**: always check robots.txt and website ToS before scraping. This library is for educational and research purposes
+- **Python version**: requires Python 3.10+

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "hermes-agent"
-version = "0.4.0"
+version = "0.5.0"
 description = "The self-improving AI agent — creates skills from experience, improves them during use, and runs anywhere"
 readme = "README.md"
 requires-python = ">=3.11"
@@ -26,6 +26,7 @@ dependencies = [
   # Interactive CLI (prompt_toolkit is used directly by cli.py)
   "prompt_toolkit>=3.0.52,<4",
   # Tools
+  "exa-py>=2.9.0,<3",
   "firecrawl-py>=4.16.0,<5",
   "parallel-web>=0.4.2,<1",
   "fal-client>=0.13.1,<1",
@@ -37,7 +38,7 @@ dependencies = [
 ]
 
 [project.optional-dependencies]
-modal = ["swe-rex[modal]>=1.4.0,<2"]
+modal = ["modal>=1.0.0,<2"]
 daytona = ["daytona>=0.148.0,<1"]
 dev = ["pytest>=9.0.2,<10", "pytest-asyncio>=1.3.0,<2", "pytest-xdist>=3.0,<4", "mcp>=1.2.0,<2"]
 messaging = ["python-telegram-bot>=22.6,<23", "discord.py[voice]>=2.7.1,<3", "aiohttp>=3.13.3,<4", "slack-bolt>=1.18.0,<2", "slack-sdk>=3.27.0,<4"]
@@ -58,7 +59,7 @@ sms = ["aiohttp>=3.9.0,<4"]
 acp = ["agent-client-protocol>=0.8.1,<0.9"]
 dingtalk = ["dingtalk-stream>=0.1.0,<1"]
 rl = [
-  "atroposlib @ git+https://github.com/NousResearch/atropos.git@main",
+  "atroposlib @ git+https://github.com/NousResearch/atropos.git",
   "tinker @ git+https://github.com/thinking-machines-lab/tinker.git",
   "fastapi>=0.104.0,<1",
   "uvicorn[standard]>=0.24.0,<1",

run_agent.py

@@ -88,7 +88,7 @@ from agent.model_metadata import (
 )
 from agent.context_compressor import ContextCompressor
 from agent.prompt_caching import apply_anthropic_cache_control
-from agent.prompt_builder import build_skills_system_prompt, build_context_files_prompt, load_soul_md
+from agent.prompt_builder import build_skills_system_prompt, build_context_files_prompt, load_soul_md, TOOL_USE_ENFORCEMENT_GUIDANCE, TOOL_USE_ENFORCEMENT_MODELS
 from agent.usage_pricing import estimate_usage_cost, normalize_usage
 from agent.display import (
     KawaiiSpinner, build_tool_preview as _build_tool_preview,
@@ -361,6 +361,85 @@ def _inject_honcho_turn_context(content, turn_context: str):
     return f"{text}\n\n{note}"
 
 
+# Budget warning text patterns injected by _get_budget_warning().
+_BUDGET_WARNING_RE = re.compile(
+    r"\[BUDGET(?:\s+WARNING)?:\s+Iteration\s+\d+/\d+\..*?\]",
+    re.DOTALL,
+)
+
+
+# Regex to match lone surrogate code points (U+D800..U+DFFF).
+# These are invalid in UTF-8 and cause UnicodeEncodeError when the OpenAI SDK
+# serialises messages to JSON.  Common source: clipboard paste from Google Docs
+# or other rich-text editors on some platforms.
+_SURROGATE_RE = re.compile(r'[\ud800-\udfff]')
+
+
+def _sanitize_surrogates(text: str) -> str:
+    """Replace lone surrogate code points with U+FFFD (replacement character).
+
+    Surrogates are invalid in UTF-8 and will crash ``json.dumps()`` inside the
+    OpenAI SDK.  This is a fast no-op when the text contains no surrogates.
+    """
+    if _SURROGATE_RE.search(text):
+        return _SURROGATE_RE.sub('\ufffd', text)
+    return text
+
+
+def _sanitize_messages_surrogates(messages: list) -> bool:
+    """Sanitize surrogate characters from all string content in a messages list.
+
+    Walks message dicts in-place.  Returns True if any surrogates were found
+    and replaced, False otherwise.
+    """
+    found = False
+    for msg in messages:
+        if not isinstance(msg, dict):
+            continue
+        content = msg.get("content")
+        if isinstance(content, str) and _SURROGATE_RE.search(content):
+            msg["content"] = _SURROGATE_RE.sub('\ufffd', content)
+            found = True
+        elif isinstance(content, list):
+            for part in content:
+                if isinstance(part, dict):
+                    text = part.get("text")
+                    if isinstance(text, str) and _SURROGATE_RE.search(text):
+                        part["text"] = _SURROGATE_RE.sub('\ufffd', text)
+                        found = True
+    return found
+
+
+def _strip_budget_warnings_from_history(messages: list) -> None:
+    """Remove budget pressure warnings from tool-result messages in-place.
+
+    Budget warnings are turn-scoped signals that must not leak into replayed
+    history.  They live in tool-result ``content`` either as a JSON key
+    (``_budget_warning``) or appended plain text.
+    """
+    for msg in messages:
+        if not isinstance(msg, dict) or msg.get("role") != "tool":
+            continue
+        content = msg.get("content")
+        if not isinstance(content, str) or "_budget_warning" not in content and "[BUDGET" not in content:
+            continue
+
+        # Try JSON first (the common case: _budget_warning key in a dict)
+        try:
+            parsed = json.loads(content)
+            if isinstance(parsed, dict) and "_budget_warning" in parsed:
+                del parsed["_budget_warning"]
+                msg["content"] = json.dumps(parsed, ensure_ascii=False)
+                continue
+        except (json.JSONDecodeError, TypeError):
+            pass
+
+        # Fallback: strip the text pattern from plain-text tool results
+        cleaned = _BUDGET_WARNING_RE.sub("", content).strip()
+        if cleaned != content:
+            msg["content"] = cleaned
+
+
 class AIAgent:
     """
     AI Agent with tool calling capabilities.
@@ -782,6 +861,25 @@ class AIAgent:
                     }
             
             self._client_kwargs = client_kwargs  # stored for rebuilding after interrupt
+
+            # Enable fine-grained tool streaming for Claude on OpenRouter.
+            # Without this, Anthropic buffers the entire tool call and goes
+            # silent for minutes while thinking — OpenRouter's upstream proxy
+            # times out during the silence.  The beta header makes Anthropic
+            # stream tool call arguments token-by-token, keeping the
+            # connection alive.
+            _effective_base = str(client_kwargs.get("base_url", "")).lower()
+            if "openrouter" in _effective_base and "claude" in (self.model or "").lower():
+                headers = client_kwargs.get("default_headers") or {}
+                existing_beta = headers.get("x-anthropic-beta", "")
+                _FINE_GRAINED = "fine-grained-tool-streaming-2025-05-14"
+                if _FINE_GRAINED not in existing_beta:
+                    if existing_beta:
+                        headers["x-anthropic-beta"] = f"{existing_beta},{_FINE_GRAINED}"
+                    else:
+                        headers["x-anthropic-beta"] = _FINE_GRAINED
+                    client_kwargs["default_headers"] = headers
+
             self.api_key = client_kwargs.get("api_key", "")
             try:
                 self.client = self._create_openai_client(client_kwargs, reason="agent_init", shared=True)
@@ -798,16 +896,30 @@ class AIAgent:
             except Exception as e:
                 raise RuntimeError(f"Failed to initialize OpenAI client: {e}")
         
-        # Provider fallback — a single backup model/provider tried when the
-        # primary is exhausted (rate-limit, overload, connection failure).
-        # Config shape: {"provider": "openrouter", "model": "anthropic/claude-sonnet-4"}
-        self._fallback_model = fallback_model if isinstance(fallback_model, dict) else None
+        # Provider fallback chain — ordered list of backup providers tried
+        # when the primary is exhausted (rate-limit, overload, connection
+        # failure).  Supports both legacy single-dict ``fallback_model`` and
+        # new list ``fallback_providers`` format.
+        if isinstance(fallback_model, list):
+            self._fallback_chain = [
+                f for f in fallback_model
+                if isinstance(f, dict) and f.get("provider") and f.get("model")
+            ]
+        elif isinstance(fallback_model, dict) and fallback_model.get("provider") and fallback_model.get("model"):
+            self._fallback_chain = [fallback_model]
+        else:
+            self._fallback_chain = []
+        self._fallback_index = 0
         self._fallback_activated = False
-        if self._fallback_model:
-            fb_p = self._fallback_model.get("provider", "")
-            fb_m = self._fallback_model.get("model", "")
-            if fb_p and fb_m and not self.quiet_mode:
-                print(f"🔄 Fallback model: {fb_m} ({fb_p})")
+        # Legacy attribute kept for backward compat (tests, external callers)
+        self._fallback_model = self._fallback_chain[0] if self._fallback_chain else None
+        if self._fallback_chain and not self.quiet_mode:
+            if len(self._fallback_chain) == 1:
+                fb = self._fallback_chain[0]
+                print(f"🔄 Fallback model: {fb['model']} ({fb['provider']})")
+            else:
+                print(f"🔄 Fallback chain ({len(self._fallback_chain)} providers): " +
+                      " → ".join(f"{f['model']} ({f['provider']})" for f in self._fallback_chain))
 
         # Get available tools with filtering
         self.tools = get_tool_definitions(
@@ -986,8 +1098,8 @@ class AIAgent:
                     else:
                         if not hcfg.enabled:
                             logger.debug("Honcho disabled in global config")
-                        elif not hcfg.api_key:
-                            logger.debug("Honcho enabled but no API key configured")
+                        elif not (hcfg.api_key or hcfg.base_url):
+                            logger.debug("Honcho enabled but no API key or base URL configured")
                         else:
                             logger.debug("Honcho enabled but missing API key or disabled in config")
             except Exception as e:
@@ -1024,6 +1136,13 @@ class AIAgent:
         except Exception:
             pass
 
+        # Tool-use enforcement config: "auto" (default — matches hardcoded
+        # model list), true (always), false (never), or list of substrings.
+        _agent_section = _agent_cfg.get("agent", {})
+        if not isinstance(_agent_section, dict):
+            _agent_section = {}
+        self._tool_use_enforcement = _agent_section.get("tool_use_enforcement", "auto")
+
         # Initialize context compressor for automatic context management
         # Compresses conversation when approaching model's context limit
         # Configuration via config.yaml (compression section)
@@ -2187,8 +2306,14 @@ class AIAgent:
     # ── Honcho integration helpers ──
 
     def _honcho_should_activate(self, hcfg) -> bool:
-        """Return True when remote Honcho should be active."""
-        if not hcfg or not hcfg.enabled or not hcfg.api_key:
+        """Return True when Honcho should be active.
+
+        Self-hosted Honcho may be configured with a base_url and no API key,
+        so activation should accept either credential style.
+        """
+        if not hcfg or not hcfg.enabled:
+            return False
+        if not (hcfg.api_key or hcfg.base_url):
             return False
         return True
 
@@ -2454,6 +2579,30 @@ class AIAgent:
         if tool_guidance:
             prompt_parts.append(" ".join(tool_guidance))
 
+        # Tool-use enforcement: tells the model to actually call tools instead
+        # of describing intended actions.  Controlled by config.yaml
+        # agent.tool_use_enforcement:
+        #   "auto" (default) — matches TOOL_USE_ENFORCEMENT_MODELS
+        #   true  — always inject (all models)
+        #   false — never inject
+        #   list  — custom model-name substrings to match
+        if self.valid_tool_names:
+            _enforce = self._tool_use_enforcement
+            _inject = False
+            if _enforce is True or (isinstance(_enforce, str) and _enforce.lower() in ("true", "always", "yes", "on")):
+                _inject = True
+            elif _enforce is False or (isinstance(_enforce, str) and _enforce.lower() in ("false", "never", "no", "off")):
+                _inject = False
+            elif isinstance(_enforce, list):
+                model_lower = (self.model or "").lower()
+                _inject = any(p.lower() in model_lower for p in _enforce if isinstance(p, str))
+            else:
+                # "auto" or any unrecognised value — use hardcoded defaults
+                model_lower = (self.model or "").lower()
+                _inject = any(p in model_lower for p in TOOL_USE_ENFORCEMENT_MODELS)
+            if _inject:
+                prompt_parts.append(TOOL_USE_ENFORCEMENT_GUIDANCE)
+
         # Honcho CLI awareness: tell Hermes about its own management commands
         # so it can refer the user to them rather than reinventing answers.
         if self._honcho and self._honcho_session_key:
@@ -3772,7 +3921,7 @@ class AIAgent:
         def _call_chat_completions():
             """Stream a chat completions response."""
             import httpx as _httpx
-            _base_timeout = float(os.getenv("HERMES_API_TIMEOUT", 900.0))
+            _base_timeout = float(os.getenv("HERMES_API_TIMEOUT", 1800.0))
             _stream_read_timeout = float(os.getenv("HERMES_STREAM_READ_TIMEOUT", 60.0))
             stream_kwargs = {
                 **api_kwargs,
@@ -3796,6 +3945,12 @@ class AIAgent:
             content_parts: list = []
             tool_calls_acc: dict = {}
             tool_gen_notified: set = set()
+            # Ollama-compatible endpoints reuse index 0 for every tool call
+            # in a parallel batch, distinguishing them only by id.  Track
+            # the last seen id per raw index so we can detect a new tool
+            # call starting at the same index and redirect it to a fresh slot.
+            _last_id_at_idx: dict = {}      # raw_index -> last seen non-empty id
+            _active_slot_by_idx: dict = {}  # raw_index -> current slot in tool_calls_acc
             finish_reason = None
             model_name = None
             role = "assistant"
@@ -3837,11 +3992,45 @@ class AIAgent:
                         _fire_first_delta()
                         self._fire_stream_delta(delta.content)
                         deltas_were_sent["yes"] = True
+                    else:
+                        # Tool calls suppress regular content streaming (avoids
+                        # displaying chatty "I'll use the tool..." text alongside
+                        # tool calls).  But reasoning tags embedded in suppressed
+                        # content should still reach the display — otherwise the
+                        # reasoning box only appears as a post-response fallback,
+                        # rendering it confusingly after the already-streamed
+                        # response.  Route suppressed content through the stream
+                        # delta callback so its tag extraction can fire the
+                        # reasoning display.  Non-reasoning text is harmlessly
+                        # suppressed by the CLI's _stream_delta when the stream
+                        # box is already closed (tool boundary flush).
+                        if self.stream_delta_callback:
+                            try:
+                                self.stream_delta_callback(delta.content)
+                            except Exception:
+                                pass
 
                 # Accumulate tool call deltas — notify display on first name
                 if delta and delta.tool_calls:
                     for tc_delta in delta.tool_calls:
-                        idx = tc_delta.index if tc_delta.index is not None else 0
+                        raw_idx = tc_delta.index if tc_delta.index is not None else 0
+                        delta_id = tc_delta.id or ""
+
+                        # Ollama fix: detect a new tool call reusing the same
+                        # raw index (different id) and redirect to a fresh slot.
+                        if raw_idx not in _active_slot_by_idx:
+                            _active_slot_by_idx[raw_idx] = raw_idx
+                        if (
+                            delta_id
+                            and raw_idx in _last_id_at_idx
+                            and delta_id != _last_id_at_idx[raw_idx]
+                        ):
+                            new_slot = max(tool_calls_acc, default=-1) + 1
+                            _active_slot_by_idx[raw_idx] = new_slot
+                        if delta_id:
+                            _last_id_at_idx[raw_idx] = delta_id
+                        idx = _active_slot_by_idx[raw_idx]
+
                         if idx not in tool_calls_acc:
                             tool_calls_acc[idx] = {
                                 "id": tc_delta.id or "",
@@ -3994,7 +4183,37 @@ class AIAgent:
                             e, (_httpx.ConnectError, _httpx.RemoteProtocolError, ConnectionError)
                         )
 
-                        if _is_timeout or _is_conn_err:
+                        # SSE error events from proxies (e.g. OpenRouter sends
+                        # {"error":{"message":"Network connection lost."}}) are
+                        # raised as APIError by the OpenAI SDK.  These are
+                        # semantically identical to httpx connection drops —
+                        # the upstream stream died — and should be retried with
+                        # a fresh connection.  Distinguish from HTTP errors:
+                        # APIError from SSE has no status_code, while
+                        # APIStatusError (4xx/5xx) always has one.
+                        _is_sse_conn_err = False
+                        if not _is_timeout and not _is_conn_err:
+                            from openai import APIError as _APIError
+                            if isinstance(e, _APIError) and not getattr(e, "status_code", None):
+                                _err_lower_sse = str(e).lower()
+                                _SSE_CONN_PHRASES = (
+                                    "connection lost",
+                                    "connection reset",
+                                    "connection closed",
+                                    "connection terminated",
+                                    "network error",
+                                    "network connection",
+                                    "terminated",
+                                    "peer closed",
+                                    "broken pipe",
+                                    "upstream connect error",
+                                )
+                                _is_sse_conn_err = any(
+                                    phrase in _err_lower_sse
+                                    for phrase in _SSE_CONN_PHRASES
+                                )
+
+                        if _is_timeout or _is_conn_err or _is_sse_conn_err:
                             # Transient network / timeout error. Retry the
                             # streaming request with a fresh connection first.
                             if _stream_attempt < _max_stream_retries:
@@ -4113,25 +4332,26 @@ class AIAgent:
     # ── Provider fallback ──────────────────────────────────────────────────
 
     def _try_activate_fallback(self) -> bool:
-        """Switch to the configured fallback model/provider.
+        """Switch to the next fallback model/provider in the chain.
 
-        Called when the primary model is failing after retries.  Swaps the
+        Called when the current model is failing after retries.  Swaps the
         OpenAI client, model slug, and provider in-place so the retry loop
-        can continue with the new backend.  One-shot: returns False if
-        already activated or not configured.
+        can continue with the new backend.  Advances through the chain on
+        each call; returns False when exhausted.
 
         Uses the centralized provider router (resolve_provider_client) for
         auth resolution and client construction — no duplicated provider→key
         mappings.
         """
-        if self._fallback_activated or not self._fallback_model:
+        if self._fallback_index >= len(self._fallback_chain):
             return False
 
-        fb = self._fallback_model
+        fb = self._fallback_chain[self._fallback_index]
+        self._fallback_index += 1
         fb_provider = (fb.get("provider") or "").strip().lower()
         fb_model = (fb.get("model") or "").strip()
         if not fb_provider or not fb_model:
-            return False
+            return self._try_activate_fallback()  # skip invalid, try next
 
         # Use centralized router for client construction.
         # raw_codex=True because the main agent needs direct responses.stream()
@@ -4144,7 +4364,7 @@ class AIAgent:
                 logging.warning(
                     "Fallback to %s failed: provider not configured",
                     fb_provider)
-                return False
+                return self._try_activate_fallback()  # try next in chain
 
             # Determine api_mode from provider / base URL
             fb_api_mode = "chat_completions"
@@ -4219,8 +4439,8 @@ class AIAgent:
             )
             return True
         except Exception as e:
-            logging.error("Failed to activate fallback model: %s", e)
-            return False
+            logging.error("Failed to activate fallback %s: %s", fb_model, e)
+            return self._try_activate_fallback()  # try next in chain
 
     # ── End provider fallback ──────────────────────────────────────────────
 
@@ -4378,6 +4598,10 @@ class AIAgent:
         if self.api_mode == "anthropic_messages":
             from agent.anthropic_adapter import build_anthropic_kwargs
             anthropic_messages = self._prepare_anthropic_messages_for_api(api_messages)
+            # Pass context_length so the adapter can clamp max_tokens if the
+            # user configured a smaller context window than the model's output limit.
+            ctx_len = getattr(self, "context_compressor", None)
+            ctx_len = ctx_len.context_length if ctx_len else None
             return build_anthropic_kwargs(
                 model=self.model,
                 messages=anthropic_messages,
@@ -4386,6 +4610,7 @@ class AIAgent:
                 reasoning_config=self.reasoning_config,
                 is_oauth=self._is_anthropic_oauth,
                 preserve_dots=self._anthropic_preserve_dots(),
+                context_length=ctx_len,
             )
 
         if self.api_mode == "codex_responses":
@@ -4496,12 +4721,27 @@ class AIAgent:
         api_kwargs = {
             "model": self.model,
             "messages": sanitized_messages,
-            "tools": self.tools if self.tools else None,
-            "timeout": float(os.getenv("HERMES_API_TIMEOUT", 900.0)),
+            "timeout": float(os.getenv("HERMES_API_TIMEOUT", 1800.0)),
         }
+        if self.tools:
+            api_kwargs["tools"] = self.tools
 
         if self.max_tokens is not None:
             api_kwargs.update(self._max_tokens_param(self.max_tokens))
+        elif self._is_openrouter_url() and "claude" in (self.model or "").lower():
+            # OpenRouter translates requests to Anthropic's Messages API,
+            # which requires max_tokens as a mandatory field.  When we omit
+            # it, OpenRouter picks a default that can be too low — the model
+            # spends its output budget on thinking and has almost nothing
+            # left for the actual response (especially large tool calls like
+            # write_file).  Sending the model's real output limit ensures
+            # full capacity.  Other providers handle the default fine.
+            try:
+                from agent.anthropic_adapter import _get_anthropic_max_output
+                _model_output_limit = _get_anthropic_max_output(self.model)
+                api_kwargs["max_tokens"] = _model_output_limit
+            except Exception:
+                pass  # fail open — let OpenRouter pick its default
 
         extra_body = {}
 
@@ -5783,6 +6023,14 @@ class AIAgent:
         # Installed once, transparent when streams are healthy, prevents crash on write.
         _install_safe_stdio()
 
+        # Sanitize surrogate characters from user input.  Clipboard paste from
+        # rich-text editors (Google Docs, Word, etc.) can inject lone surrogates
+        # that are invalid UTF-8 and crash JSON serialization in the OpenAI SDK.
+        if isinstance(user_message, str):
+            user_message = _sanitize_surrogates(user_message)
+        if isinstance(persist_user_message, str):
+            persist_user_message = _sanitize_surrogates(persist_user_message)
+
         # Store stream callback for _interruptible_api_call to pick up
         self._stream_callback = stream_callback
         self._persist_user_message_idx = None
@@ -5799,6 +6047,7 @@ class AIAgent:
         self._codex_incomplete_retries = 0
         self._last_content_with_tools = None
         self._mute_post_response = False
+        self._surrogate_sanitized = False
         # NOTE: _turns_since_memory and _iters_since_skill are NOT reset here.
         # They are initialized in __init__ and must persist across run_conversation
         # calls so that nudge logic accumulates correctly in CLI mode.
@@ -5806,6 +6055,14 @@ class AIAgent:
         
         # Initialize conversation (copy to avoid mutating the caller's list)
         messages = list(conversation_history) if conversation_history else []
+
+        # Strip budget pressure warnings from previous turns.  These are
+        # turn-scoped signals injected by _get_budget_warning() into tool
+        # result content.  If left in the replayed history, models (especially
+        # GPT-family) interpret them as still-active instructions and avoid
+        # making tool calls in ALL subsequent turns.
+        if messages:
+            _strip_budget_warnings_from_history(messages)
         
         # Hydrate todo store from conversation history (gateway creates a fresh
         # AIAgent per message, so the in-memory store is empty -- we need to
@@ -5901,6 +6158,22 @@ class AIAgent:
                     self._cached_system_prompt = (
                         self._cached_system_prompt + "\n\n" + self._honcho_context
                     ).strip()
+
+                # Plugin hook: on_session_start
+                # Fired once when a brand-new session is created (not on
+                # continuation).  Plugins can use this to initialise
+                # session-scoped state (e.g. warm a memory cache).
+                try:
+                    from hermes_cli.plugins import invoke_hook as _invoke_hook
+                    _invoke_hook(
+                        "on_session_start",
+                        session_id=self.session_id,
+                        model=self.model,
+                        platform=getattr(self, "platform", None) or "",
+                    )
+                except Exception as exc:
+                    logger.warning("on_session_start hook failed: %s", exc)
+
                 # Store the system prompt snapshot in SQLite
                 if self._session_db:
                     try:
@@ -5962,6 +6235,34 @@ class AIAgent:
                     if _preflight_tokens < self.context_compressor.threshold_tokens:
                         break  # Under threshold
 
+        # Plugin hook: pre_llm_call
+        # Fired once per turn before the tool-calling loop.  Plugins can
+        # return a dict with a ``context`` key whose value is a string
+        # that will be appended to the ephemeral system prompt for every
+        # API call in this turn (not persisted to session DB or cache).
+        _plugin_turn_context = ""
+        try:
+            from hermes_cli.plugins import invoke_hook as _invoke_hook
+            _pre_results = _invoke_hook(
+                "pre_llm_call",
+                session_id=self.session_id,
+                user_message=original_user_message,
+                conversation_history=list(messages),
+                is_first_turn=(not bool(conversation_history)),
+                model=self.model,
+                platform=getattr(self, "platform", None) or "",
+            )
+            _ctx_parts = []
+            for r in _pre_results:
+                if isinstance(r, dict) and r.get("context"):
+                    _ctx_parts.append(str(r["context"]))
+                elif isinstance(r, str) and r.strip():
+                    _ctx_parts.append(r)
+            if _ctx_parts:
+                _plugin_turn_context = "\n\n".join(_ctx_parts)
+        except Exception as exc:
+            logger.warning("pre_llm_call hook failed: %s", exc)
+
         # Main conversation loop
         api_call_count = 0
         final_response = None
@@ -6059,6 +6360,9 @@ class AIAgent:
             effective_system = active_system_prompt or ""
             if self.ephemeral_system_prompt:
                 effective_system = (effective_system + "\n\n" + self.ephemeral_system_prompt).strip()
+            # Plugin context from pre_llm_call hooks — ephemeral, not cached.
+            if _plugin_turn_context:
+                effective_system = (effective_system + "\n\n" + _plugin_turn_context).strip()
             if effective_system:
                 api_messages = [{"role": "system", "content": effective_system}] + api_messages
 
@@ -6240,9 +6544,9 @@ class AIAgent:
                         # Eager fallback: empty/malformed responses are a common
                         # rate-limit symptom.  Switch to fallback immediately
                         # rather than retrying with extended backoff.
-                        if not self._fallback_activated:
+                        if self._fallback_index < len(self._fallback_chain):
                             self._emit_status("⚠️ Empty/malformed response — switching to fallback...")
-                        if not self._fallback_activated and self._try_activate_fallback():
+                        if self._try_activate_fallback():
                             retry_count = 0
                             continue
 
@@ -6335,6 +6639,62 @@ class AIAgent:
                     if finish_reason == "length":
                         self._vprint(f"{self.log_prefix}⚠️  Response truncated (finish_reason='length') - model hit max output tokens", force=True)
 
+                        # ── Detect thinking-budget exhaustion ──────────────
+                        # When the model spends ALL output tokens on reasoning
+                        # and has none left for the response, continuation
+                        # retries are pointless.  Detect this early and give a
+                        # targeted error instead of wasting 3 API calls.
+                        _trunc_content = None
+                        if self.api_mode == "chat_completions":
+                            _trunc_msg = response.choices[0].message if (hasattr(response, "choices") and response.choices) else None
+                            _trunc_content = getattr(_trunc_msg, "content", None) if _trunc_msg else None
+                        elif self.api_mode == "anthropic_messages":
+                            # Anthropic response.content is a list of blocks
+                            _text_parts = []
+                            for _blk in getattr(response, "content", []):
+                                if getattr(_blk, "type", None) == "text":
+                                    _text_parts.append(getattr(_blk, "text", ""))
+                            _trunc_content = "\n".join(_text_parts) if _text_parts else None
+
+                        _thinking_exhausted = (
+                            _trunc_content is not None
+                            and not self._has_content_after_think_block(_trunc_content)
+                        ) or _trunc_content is None
+
+                        if _thinking_exhausted:
+                            _exhaust_error = (
+                                "Model used all output tokens on reasoning with none left "
+                                "for the response. Try lowering reasoning effort or "
+                                "increasing max_tokens."
+                            )
+                            self._vprint(
+                                f"{self.log_prefix}💭 Reasoning exhausted the output token budget — "
+                                f"no visible response was produced.",
+                                force=True,
+                            )
+                            # Return a user-friendly message as the response so
+                            # CLI (response box) and gateway (chat message) both
+                            # display it naturally instead of a suppressed error.
+                            _exhaust_response = (
+                                "⚠️ **Thinking Budget Exhausted**\n\n"
+                                "The model used all its output tokens on reasoning "
+                                "and had none left for the actual response.\n\n"
+                                "To fix this:\n"
+                                "→ Lower reasoning effort: `/thinkon low` or `/thinkon minimal`\n"
+                                "→ Increase the output token limit: "
+                                "set `model.max_tokens` in config.yaml"
+                            )
+                            self._cleanup_task_resources(effective_task_id)
+                            self._persist_session(messages, conversation_history)
+                            return {
+                                "final_response": _exhaust_response,
+                                "messages": messages,
+                                "api_calls": api_call_count,
+                                "completed": False,
+                                "partial": True,
+                                "error": _exhaust_error,
+                            }
+
                         if self.api_mode == "chat_completions":
                             assistant_message = response.choices[0].message
                             if not assistant_message.tool_calls:
@@ -6523,6 +6883,24 @@ class AIAgent:
                     if self.thinking_callback:
                         self.thinking_callback("")
 
+                    # -----------------------------------------------------------
+                    # Surrogate character recovery.  UnicodeEncodeError happens
+                    # when the messages contain lone surrogates (U+D800..U+DFFF)
+                    # that are invalid UTF-8.  Common source: clipboard paste
+                    # from Google Docs or similar rich-text editors.  We sanitize
+                    # the entire messages list in-place and retry once.
+                    # -----------------------------------------------------------
+                    if isinstance(api_error, UnicodeEncodeError) and not getattr(self, '_surrogate_sanitized', False):
+                        self._surrogate_sanitized = True
+                        if _sanitize_messages_surrogates(messages):
+                            self._vprint(
+                                f"{self.log_prefix}⚠️  Stripped invalid surrogate characters from messages. Retrying...",
+                                force=True,
+                            )
+                            continue
+                        # Surrogates weren't in messages — might be in system
+                        # prompt or prefill.  Fall through to normal error path.
+
                     status_code = getattr(api_error, "status_code", None)
                     if (
                         self.api_mode == "codex_responses"
@@ -6562,8 +6940,10 @@ class AIAgent:
                         print(f"{self.log_prefix}   Auth method: {auth_method}")
                         print(f"{self.log_prefix}   Token prefix: {key[:12]}..." if key and len(key) > 12 else f"{self.log_prefix}   Token: (empty or short)")
                         print(f"{self.log_prefix}   Troubleshooting:")
-                        print(f"{self.log_prefix}     • Check ANTHROPIC_TOKEN in ~/.hermes/.env for Hermes-managed OAuth/setup tokens")
-                        print(f"{self.log_prefix}     • Check ANTHROPIC_API_KEY in ~/.hermes/.env for API keys or legacy token values")
+                        from hermes_constants import display_hermes_home as _dhh_fn
+                        _dhh = _dhh_fn()
+                        print(f"{self.log_prefix}     • Check ANTHROPIC_TOKEN in {_dhh}/.env for Hermes-managed OAuth/setup tokens")
+                        print(f"{self.log_prefix}     • Check ANTHROPIC_API_KEY in {_dhh}/.env for API keys or legacy token values")
                         print(f"{self.log_prefix}     • For API keys: verify at https://console.anthropic.com/settings/keys")
                         print(f"{self.log_prefix}     • For Claude Code: run 'claude /login' to refresh, then retry")
                         print(f"{self.log_prefix}     • Clear stale keys: hermes config set ANTHROPIC_TOKEN \"\"")
@@ -6629,7 +7009,7 @@ class AIAgent:
                         or "usage limit" in error_msg
                         or "quota" in error_msg
                     )
-                    if is_rate_limited and not self._fallback_activated:
+                    if is_rate_limited and self._fallback_index < len(self._fallback_chain):
                         self._emit_status("⚠️ Rate limited — switching to fallback provider...")
                         if self._try_activate_fallback():
                             retry_count = 0
@@ -6791,8 +7171,13 @@ class AIAgent:
                     # 529 (Anthropic overloaded) is also transient.
                     # Also catch local validation errors (ValueError, TypeError) — these
                     # are programming bugs, not transient failures.
+                    # Exclude UnicodeEncodeError — it's a ValueError subclass but is
+                    # handled separately by the surrogate sanitization path above.
                     _RETRYABLE_STATUS_CODES = {413, 429, 529}
-                    is_local_validation_error = isinstance(api_error, (ValueError, TypeError))
+                    is_local_validation_error = (
+                        isinstance(api_error, (ValueError, TypeError))
+                        and not isinstance(api_error, UnicodeEncodeError)
+                    )
                     # Detect generic 400s from Anthropic OAuth (transient server-side failures).
                     # Real invalid_request_error responses include a descriptive message;
                     # transient ones contain only "Error" or are empty. (ref: issue #1608)
@@ -6860,8 +7245,41 @@ class AIAgent:
                             retry_count = 0
                             continue
                         _final_summary = self._summarize_api_error(api_error)
-                        self._vprint(f"{self.log_prefix}❌ Max retries ({max_retries}) exceeded. Giving up.", force=True)
+                        if is_rate_limited:
+                            self._vprint(f"{self.log_prefix}❌ Rate limit persisted after {max_retries} retries. Please try again later.", force=True)
+                        else:
+                            self._vprint(f"{self.log_prefix}❌ Max retries ({max_retries}) exceeded. Giving up.", force=True)
                         self._vprint(f"{self.log_prefix}   💀 Final error: {_final_summary}", force=True)
+
+                        # Detect SSE stream-drop pattern (e.g. "Network
+                        # connection lost") and surface actionable guidance.
+                        # This typically happens when the model generates a
+                        # very large tool call (write_file with huge content)
+                        # and the proxy/CDN drops the stream mid-response.
+                        _is_stream_drop = (
+                            not getattr(api_error, "status_code", None)
+                            and any(p in error_msg for p in (
+                                "connection lost", "connection reset",
+                                "connection closed", "network connection",
+                                "network error", "terminated",
+                            ))
+                        )
+                        if _is_stream_drop:
+                            self._vprint(
+                                f"{self.log_prefix}   💡 The provider's stream "
+                                f"connection keeps dropping. This often happens "
+                                f"when the model tries to write a very large "
+                                f"file in a single tool call.",
+                                force=True,
+                            )
+                            self._vprint(
+                                f"{self.log_prefix}      Try asking the model "
+                                f"to use execute_code with Python's open() for "
+                                f"large files, or to write the file in smaller "
+                                f"sections.",
+                                force=True,
+                            )
+
                         logging.error(
                             "%sAPI call failed after %s retries. %s | provider=%s model=%s msgs=%s tokens=~%s",
                             self.log_prefix, max_retries, _final_summary,
@@ -6871,8 +7289,18 @@ class AIAgent:
                             api_kwargs, reason="max_retries_exhausted", error=api_error,
                         )
                         self._persist_session(messages, conversation_history)
+                        _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 "
+                                "dropping — this often happens when generating "
+                                "very large tool call responses (e.g. write_file "
+                                "with long content). Try asking me to use "
+                                "execute_code with Python's open() for large "
+                                "files, or to write in smaller sections."
+                            )
                         return {
-                            "final_response": f"API call failed after {max_retries} retries: {_final_summary}",
+                            "final_response": _final_response,
                             "messages": messages,
                             "api_calls": api_call_count,
                             "completed": False,
@@ -6880,8 +7308,22 @@ class AIAgent:
                             "error": _final_summary,
                         }
 
-                    wait_time = min(2 ** retry_count, 60)  # Exponential backoff: 2s, 4s, 8s, 16s, 32s, 60s, 60s
-                    self._emit_status(f"⏳ Retrying in {wait_time}s (attempt {retry_count}/{max_retries})...")
+                    # For rate limits, respect the Retry-After header if present
+                    _retry_after = None
+                    if is_rate_limited:
+                        _resp_headers = getattr(getattr(api_error, "response", None), "headers", None)
+                        if _resp_headers and hasattr(_resp_headers, "get"):
+                            _ra_raw = _resp_headers.get("retry-after") or _resp_headers.get("Retry-After")
+                            if _ra_raw:
+                                try:
+                                    _retry_after = min(int(_ra_raw), 120)  # Cap at 2 minutes
+                                except (TypeError, ValueError):
+                                    pass
+                    wait_time = _retry_after if _retry_after else min(2 ** retry_count, 60)
+                    if is_rate_limited:
+                        self._emit_status(f"⏱️ Rate limit reached. Waiting {wait_time}s before retry (attempt {retry_count + 1}/{max_retries})...")
+                    else:
+                        self._emit_status(f"⏳ Retrying in {wait_time}s (attempt {retry_count}/{max_retries})...")
                     logger.warning(
                         "Retrying API call in %ss (attempt %s/%s) %s error=%s",
                         wait_time,
@@ -7250,7 +7692,6 @@ class AIAgent:
                         except Exception:
                             pass
 
-                    _msg_count_before_tools = len(messages)
                     self._execute_tool_calls(assistant_message, messages, effective_task_id, api_call_count)
 
                     # Signal that a paragraph break is needed before the next
@@ -7268,18 +7709,18 @@ class AIAgent:
                     if _tc_names == {"execute_code"}:
                         self.iteration_budget.refund()
                     
-                    # Estimate next prompt size using real token counts from the
-                    # last API response + rough estimate of newly appended tool
-                    # results.  This catches cases where tool results push the
-                    # context past the limit that last_prompt_tokens alone misses
-                    # (e.g. large file reads, web extractions).
+                    # Use real token counts from the API response to decide
+                    # compression.  prompt_tokens + completion_tokens is the
+                    # actual context size the provider reported plus the
+                    # assistant turn — a tight lower bound for the next prompt.
+                    # Tool results appended above aren't counted yet, but the
+                    # threshold (default 50%) leaves ample headroom; if tool
+                    # results push past it, the next API call will report the
+                    # real total and trigger compression then.
                     _compressor = self.context_compressor
-                    _new_tool_msgs = messages[_msg_count_before_tools:]
-                    _new_chars = sum(len(str(m.get("content", "") or "")) for m in _new_tool_msgs)
-                    _estimated_next_prompt = (
+                    _real_tokens = (
                         _compressor.last_prompt_tokens
                         + _compressor.last_completion_tokens
-                        + _new_chars // 3  # conservative: JSON-heavy tool results ≈ 3 chars/token
                     )
 
                     # ── Context pressure warnings (user-facing only) ──────────
@@ -7289,12 +7730,12 @@ class AIAgent:
                     # Does not inject into messages — just prints to CLI output
                     # and fires status_callback for gateway platforms.
                     if _compressor.threshold_tokens > 0:
-                        _compaction_progress = _estimated_next_prompt / _compressor.threshold_tokens
+                        _compaction_progress = _real_tokens / _compressor.threshold_tokens
                         if _compaction_progress >= 0.85 and not self._context_pressure_warned:
                             self._context_pressure_warned = True
                             self._emit_context_pressure(_compaction_progress, _compressor)
 
-                    if self.compression_enabled and _compressor.should_compress(_estimated_next_prompt):
+                    if self.compression_enabled and _compressor.should_compress(_real_tokens):
                         messages, active_system_prompt = self._compress_context(
                             messages, system_message,
                             approx_tokens=self.context_compressor.last_prompt_tokens,
@@ -7541,6 +7982,25 @@ class AIAgent:
             self._honcho_sync(original_user_message, final_response)
             self._queue_honcho_prefetch(original_user_message)
 
+        # Plugin hook: post_llm_call
+        # Fired once per turn after the tool-calling loop completes.
+        # Plugins can use this to persist conversation data (e.g. sync
+        # to an external memory system).
+        if final_response and not interrupted:
+            try:
+                from hermes_cli.plugins import invoke_hook as _invoke_hook
+                _invoke_hook(
+                    "post_llm_call",
+                    session_id=self.session_id,
+                    user_message=original_user_message,
+                    assistant_response=final_response,
+                    conversation_history=list(messages),
+                    model=self.model,
+                    platform=getattr(self, "platform", None) or "",
+                )
+            except Exception as exc:
+                logger.warning("post_llm_call hook failed: %s", exc)
+
         # Extract reasoning from the last assistant message (if any)
         last_reasoning = None
         for msg in reversed(messages):
@@ -7606,6 +8066,22 @@ class AIAgent:
             except Exception:
                 pass  # Background review is best-effort
 
+        # Plugin hook: on_session_end
+        # Fired at the very end of every run_conversation call.
+        # Plugins can use this for cleanup, flushing buffers, etc.
+        try:
+            from hermes_cli.plugins import invoke_hook as _invoke_hook
+            _invoke_hook(
+                "on_session_end",
+                session_id=self.session_id,
+                completed=completed,
+                interrupted=interrupted,
+                model=self.model,
+                platform=getattr(self, "platform", None) or "",
+            )
+        except Exception as exc:
+            logger.warning("on_session_end hook failed: %s", exc)
+
         return result
 
     def chat(self, message: str, stream_callback: Optional[callable] = None) -> str:

scripts/kill_modal.sh

@@ -2,7 +2,7 @@
 # Kill all running Modal apps (sandboxes, deployments, etc.)
 #
 # Usage:
-#   bash scripts/kill_modal.sh          # Stop swe-rex (the sandbox app)
+#   bash scripts/kill_modal.sh          # Stop hermes-agent sandboxes
 #   bash scripts/kill_modal.sh --all    # Stop ALL Modal apps
 
 set -uo pipefail
@@ -17,10 +17,10 @@ if [[ "${1:-}" == "--all" ]]; then
         modal app stop "$app_id" 2>/dev/null || true
     done
 else
-    echo "Stopping swe-rex sandboxes..."
-    APPS=$(echo "$APP_LIST" | grep 'swe-rex' | grep -oE 'ap-[A-Za-z0-9]+' || true)
+    echo "Stopping hermes-agent sandboxes..."
+    APPS=$(echo "$APP_LIST" | grep 'hermes-agent' | grep -oE 'ap-[A-Za-z0-9]+' || true)
     if [[ -z "$APPS" ]]; then
-        echo "  No swe-rex apps found."
+        echo "  No hermes-agent apps found."
     else
         echo "$APPS" | while read app_id; do
             echo "  Stopping $app_id"
@@ -30,5 +30,5 @@ else
 fi
 
 echo ""
-echo "Current swe-rex status:"
-modal app list 2>/dev/null | grep -E 'State|swe-rex' || echo "  (none)"
+echo "Current hermes-agent status:"
+modal app list 2>/dev/null | grep -E 'State|hermes-agent' || echo "  (none)"

skills/devops/webhook-subscriptions/SKILL.md

@@ -0,0 +1,180 @@
+---
+name: webhook-subscriptions
+description: Create and manage webhook subscriptions for event-driven agent activation. Use when the user wants external services to trigger agent runs automatically.
+version: 1.0.0
+metadata:
+  hermes:
+    tags: [webhook, events, automation, integrations]
+---
+
+# Webhook Subscriptions
+
+Create dynamic webhook subscriptions so external services (GitHub, GitLab, Stripe, CI/CD, IoT sensors, monitoring tools) can trigger Hermes agent runs by POSTing events to a URL.
+
+## Setup (Required First)
+
+The webhook platform must be enabled before subscriptions can be created. Check with:
+```bash
+hermes webhook list
+```
+
+If it says "Webhook platform is not enabled", set it up:
+
+### Option 1: Setup wizard
+```bash
+hermes gateway setup
+```
+Follow the prompts to enable webhooks, set the port, and set a global HMAC secret.
+
+### Option 2: Manual config
+Add to `~/.hermes/config.yaml`:
+```yaml
+platforms:
+  webhook:
+    enabled: true
+    extra:
+      host: "0.0.0.0"
+      port: 8644
+      secret: "generate-a-strong-secret-here"
+```
+
+### Option 3: Environment variables
+Add to `~/.hermes/.env`:
+```bash
+WEBHOOK_ENABLED=true
+WEBHOOK_PORT=8644
+WEBHOOK_SECRET=generate-a-strong-secret-here
+```
+
+After configuration, start (or restart) the gateway:
+```bash
+hermes gateway run
+# Or if using systemd:
+systemctl --user restart hermes-gateway
+```
+
+Verify it's running:
+```bash
+curl http://localhost:8644/health
+```
+
+## Commands
+
+All management is via the `hermes webhook` CLI command:
+
+### Create a subscription
+```bash
+hermes webhook subscribe <name> \
+  --prompt "Prompt template with {payload.fields}" \
+  --events "event1,event2" \
+  --description "What this does" \
+  --skills "skill1,skill2" \
+  --deliver telegram \
+  --deliver-chat-id "12345" \
+  --secret "optional-custom-secret"
+```
+
+Returns the webhook URL and HMAC secret. The user configures their service to POST to that URL.
+
+### List subscriptions
+```bash
+hermes webhook list
+```
+
+### Remove a subscription
+```bash
+hermes webhook remove <name>
+```
+
+### Test a subscription
+```bash
+hermes webhook test <name>
+hermes webhook test <name> --payload '{"key": "value"}'
+```
+
+## Prompt Templates
+
+Prompts support `{dot.notation}` for accessing nested payload fields:
+
+- `{issue.title}` — GitHub issue title
+- `{pull_request.user.login}` — PR author
+- `{data.object.amount}` — Stripe payment amount
+- `{sensor.temperature}` — IoT sensor reading
+
+If no prompt is specified, the full JSON payload is dumped into the agent prompt.
+
+## Common Patterns
+
+### GitHub: new issues
+```bash
+hermes webhook subscribe github-issues \
+  --events "issues" \
+  --prompt "New GitHub issue #{issue.number}: {issue.title}\n\nAction: {action}\nAuthor: {issue.user.login}\nBody:\n{issue.body}\n\nPlease triage this issue." \
+  --deliver telegram \
+  --deliver-chat-id "-100123456789"
+```
+
+Then in GitHub repo Settings → Webhooks → Add webhook:
+- Payload URL: the returned webhook_url
+- Content type: application/json
+- Secret: the returned secret
+- Events: "Issues"
+
+### GitHub: PR reviews
+```bash
+hermes webhook subscribe github-prs \
+  --events "pull_request" \
+  --prompt "PR #{pull_request.number} {action}: {pull_request.title}\nBy: {pull_request.user.login}\nBranch: {pull_request.head.ref}\n\n{pull_request.body}" \
+  --skills "github-code-review" \
+  --deliver github_comment
+```
+
+### Stripe: payment events
+```bash
+hermes webhook subscribe stripe-payments \
+  --events "payment_intent.succeeded,payment_intent.payment_failed" \
+  --prompt "Payment {data.object.status}: {data.object.amount} cents from {data.object.receipt_email}" \
+  --deliver telegram \
+  --deliver-chat-id "-100123456789"
+```
+
+### CI/CD: build notifications
+```bash
+hermes webhook subscribe ci-builds \
+  --events "pipeline" \
+  --prompt "Build {object_attributes.status} on {project.name} branch {object_attributes.ref}\nCommit: {commit.message}" \
+  --deliver discord \
+  --deliver-chat-id "1234567890"
+```
+
+### Generic monitoring alert
+```bash
+hermes webhook subscribe alerts \
+  --prompt "Alert: {alert.name}\nSeverity: {alert.severity}\nMessage: {alert.message}\n\nPlease investigate and suggest remediation." \
+  --deliver origin
+```
+
+## Security
+
+- Each subscription gets an auto-generated HMAC-SHA256 secret (or provide your own with `--secret`)
+- The webhook adapter validates signatures on every incoming POST
+- Static routes from config.yaml cannot be overwritten by dynamic subscriptions
+- Subscriptions persist to `~/.hermes/webhook_subscriptions.json`
+
+## How It Works
+
+1. `hermes webhook subscribe` writes to `~/.hermes/webhook_subscriptions.json`
+2. The webhook adapter hot-reloads this file on each incoming request (mtime-gated, negligible overhead)
+3. When a POST arrives matching a route, the adapter formats the prompt and triggers an agent run
+4. The agent's response is delivered to the configured target (Telegram, Discord, GitHub comment, etc.)
+
+## Troubleshooting
+
+If webhooks aren't working:
+
+1. **Is the gateway running?** Check with `systemctl --user status hermes-gateway` or `ps aux | grep gateway`
+2. **Is the webhook server listening?** `curl http://localhost:8644/health` should return `{"status": "ok"}`
+3. **Check gateway logs:** `grep webhook ~/.hermes/logs/gateway.log | tail -20`
+4. **Signature mismatch?** Verify the secret in your service matches the one from `hermes webhook list`. GitHub sends `X-Hub-Signature-256`, GitLab sends `X-Gitlab-Token`.
+5. **Firewall/NAT?** The webhook URL must be reachable from the service. For local development, use a tunnel (ngrok, cloudflared).
+6. **Wrong event type?** Check `--events` filter matches what the service sends. Use `hermes webhook test <name>` to verify the route works.

skills/github/github-auth/SKILL.md

@@ -219,6 +219,9 @@ if command -v gh &>/dev/null && gh auth status &>/dev/null; then
   echo "AUTH_METHOD=gh"
 elif [ -n "$GITHUB_TOKEN" ]; then
   echo "AUTH_METHOD=curl"
+elif [ -f ~/.hermes/.env ] && grep -q "^GITHUB_TOKEN=" ~/.hermes/.env; then
+  export GITHUB_TOKEN=$(grep "^GITHUB_TOKEN=" ~/.hermes/.env | head -1 | cut -d= -f2 | tr -d '\n\r')
+  echo "AUTH_METHOD=curl"
 elif grep -q "github.com" ~/.git-credentials 2>/dev/null; then
   export GITHUB_TOKEN=$(grep "github.com" ~/.git-credentials | head -1 | sed 's|https://[^:]*:\([^@]*\)@.*|\1|')
   echo "AUTH_METHOD=curl"