Second integration audit covering surfaces the first pass didn't hit.
Found eight issues spanning kernel, dashboard frontend, notifier, and CLI.
All behavioral / UX fixes; no schema change.
Kernel
- complete_task on a never-claimed task (ready/blocked → done with no
run in flight) was silently dropping the summary/metadata/result
onto a non-existent run. Now synthesizes a zero-duration run
(started_at == ended_at) so attempt history is complete. Only
fires when there's actually handoff data to persist — bare
complete_task(tid) remains a no-op for run creation.
- block_task on a never-claimed task had the same bug for --reason.
Same fix: synthesize a zero-duration run when a reason is passed.
- Event dataclass gained a `run_id: Optional[int] = None` field.
list_events, unseen_events_for_sub, and the dashboard _event_dict
were all SELECTing the column but dropping it on the way out,
so downstream consumers couldn't group events by attempt. Every
read path now surfaces run_id.
- claim_task got a defensive invariant-recovery step: if somehow
`current_run_id` is non-NULL on a task in 'ready' status (invariant
violation from an unknown code path), close the leaked run as
'reclaimed' inside the same txn as the new claim. No-op in the
common case; belt-and-suspenders in case a future code path forgets
to clear the pointer.
Dashboard
- GET /tasks/:id events array now carries run_id per event (via
_event_dict).
- WebSocket /events SELECT now includes run_id in the pushed event
payload.
- TaskDrawer reloads itself on live events for its own task id. New
`taskEventTick[taskId]` state in the Board, incremented on every
WS event, passed down as `eventTick` prop; drawer's useEffect
depends on it. Previously, background workers completing a task
the user was viewing left the drawer showing stale data until
manual close/reopen.
- CSS: added `.hermes-kanban-run--ended` rule for the fallback class
the JS emits when outcome is unset. Harmless before; just
inconsistent.
CLI
- `hermes kanban watch --kinds` help text listed the legacy event
name `spawn_auto_blocked`. The kernel migration renames it to
`gave_up`, so users typing the documented name got zero matches.
Now shows the current lexicon (`completed,blocked,gave_up,
crashed,timed_out`).
Tests (+6 in core functionality, +1 in dashboard plugin)
- complete_never_claimed_task_synthesizes_run
- block_never_claimed_task_synthesizes_run
- complete_never_claimed_without_handoff_skips_synthesis
- event_dataclass_carries_run_id (created.run_id None, completed.run_id matches)
- unseen_events_for_sub_includes_run_id (notifier path)
- claim_task_recovers_from_invariant_leak (engineer the leak, verify recovery)
- event_dict_includes_run_id (dashboard API shape)
171/171 kanban suite pass under scripts/run_tests.sh. Live-smoke (isolated
HERMES_HOME via execute_code) exercised all six fixed paths plus the
claim-after-leak recovery sequence.
Docs
- Runs section: new 'Synthetic runs for never-claimed completions'
and 'Live drawer refresh' paragraphs explaining the invariants.
- Event reference: `created` / `promoted` / `unblocked` entries now
explicitly note `run_id` is `NULL`; `completed` / `blocked`
describe synthetic-run fallback.
Integration audit of the runs-as-first-class work (0146cb2bd) found five
bugs where structured runs got orphaned or dashboard parity was missing.
All behavioral fixes; no schema change needed.
Kernel
- archive_task: when called on a running task, now closes the
in-flight run with outcome='reclaimed' and clears current_run_id.
Previously, dashboard bulk-archive or CLI `kanban archive <running>`
would leave the task_runs row open with ended_at=NULL forever and
strand the pointer. Adds the claim_lock / claim_expires / worker_pid
clearing to the UPDATE so the task row is clean too.
- complete_task: embeds the first-line handoff summary in the
`completed` event payload (capped at 400 chars). Notifier can now
render `✔ task done — <title>\n<summary>` without a second SQL hit,
and the full summary still lives on the run row.
Dashboard plugin
- _set_status_direct: drag-drop OFF 'running' (to 'ready', 'todo',
'triage', 'done' — anywhere except back to 'running') now closes
the active run with outcome='reclaimed'. Clears worker_pid too.
Snapshots previous status + current_run_id before the UPDATE so
the decision has the right before-state. status event rows now
carry run_id when closing a run, NULL otherwise.
- UpdateTaskBody: adds `summary` and `metadata` fields. PATCH
/tasks/:id with status='done' now forwards them to complete_task,
giving the dashboard parity with `hermes kanban complete --summary
... --metadata ...`. Previously these fields only existed on the
CLI.
CLI
- `hermes kanban complete a b c --summary X` or `--metadata Y`:
refused with a clear stderr message instead of silently applying
the same handoff to every task. Bulk-close without handoff flags
still works. (Note: hermes_cli.main discards subcommand exit
codes via `args.func(args)` without propagating; tracked
separately. Side-effect check is the real guard.)
Gateway notifier
- Completion message prefers run.summary (carried in event payload)
over task.result. task.result remains the fallback for legacy rows
written before runs shipped.
- Docstring: renamed stale `spawn_auto_blocked` reference to
`gave_up` / `timed_out` — matches the actual TERMINAL_KINDS
tuple, which was already correct in code.
Tests (+8 in core functionality, +3 in dashboard plugin)
- archive_of_running_task_closes_run
- archive_of_ready_task_does_not_create_spurious_run
- dashboard_direct_status_change_off_running_closes_run
- dashboard_direct_status_change_within_same_state_is_noop_for_runs
- cli_bulk_complete_with_summary_rejects (side-effect assertion)
- cli_bulk_complete_without_summary_still_works
- completed_event_payload_carries_summary
- completed_event_payload_summary_none_when_missing
- patch_status_done_with_summary_and_metadata
- patch_status_done_without_summary_still_works (legacy path)
- patch_status_archive_closes_running_run (E2E through FastAPI TestClient)
164/164 kanban suite pass under scripts/run_tests.sh. Live smoke
(execute_code with isolated HERMES_HOME) covered all five fixed paths
plus a re-claim-after-drag-drop to confirm the fresh run is tracked
correctly after the orphan close.
Addresses vulcan-artivus's RFC review on issue #16102. Picks up the
structural changes that are expensive to retrofit later and zero-cost
to land now; defers workflow-template routing + per-stage lanes to v2
(kept forward-compat hooks in the schema).
Kernel
- New `task_runs` table. Each claim opens a run (pid, claim_lock,
heartbeat, max_runtime, started_at), each terminal transition
closes it with an outcome (completed / blocked / crashed /
timed_out / spawn_failed / gave_up / reclaimed). Multiple rows per
task when retries happen, preserving full attempt history.
- `tasks.current_run_id` points at the active run (NULL when idle);
denormalised for cheap reads.
- `task_events.run_id` carries the run a given event belongs to so
UIs group events by attempt. claim/spawned/complete/block/crash/
timeout/spawn_fail/gave_up/heartbeat events are all run-scoped;
created/promoted/assigned/edited stay task-scoped (run_id=NULL).
- Legacy DBs: migration adds the columns + indexes + synthesizes a
run row for any task that's 'running' before the runs table
existed, so subsequent complete/heartbeat/reclaim calls have a
target. Idempotent.
Structured handoff
- `complete_task(summary=, metadata=)` persists both on the closing
run. `summary` falls back to `result` when omitted so single-run
callers don't duplicate. `metadata` is a free-form dict
({changed_files, tests_run, findings, ...}).
- `build_worker_context` rewrites: now reads "Prior attempts on this
task" (closed runs: outcome, summary, error, metadata) and
"Parent task results" pulls run.summary + run.metadata of the
most-recent completed run per parent, falling back to task.result
for legacy rows without runs. Retrying workers see why earlier
attempts failed; downstream workers see parent handoffs
structurally, not as loose `result` strings.
CLI
- `hermes kanban complete <id> --summary "..." --metadata '{"files":1}'`.
JSON is parsed and rejected with exit-2 if malformed.
- New `hermes kanban runs <id> [--json]` verb. Shows per-run rows:
outcome, profile, elapsed, summary, error. JSON mode serializes
the full run dataclass for scripting.
Dashboard plugin
- GET /tasks/:id now carries a runs[] array alongside task / events /
comments / links. Each run serialised with outcome, summary,
metadata, worker_pid, elapsed fields.
- New Run History section in the drawer. Outcome-coloured left
border (green=active, blue=completed, amber=reclaimed,
red=crashed/timed_out/gave_up/blocked). Collapsed when >3 runs
with a '+N earlier' toggle. Shows summary + error + metadata
inline.
Forward-compat for v2 (vulcan's workflow templates + stages)
- `tasks.workflow_template_id` and `tasks.current_step_key` added as
nullable columns. v1 kernel ignores them for routing; v2 will add
workflow_templates + workflow_steps tables and wire the dispatcher
to consult them. task_runs has a matching `step_key` column. Lets
a v2 release land additively without another schema migration.
Tests (+22 in test_kanban_core_functionality.py, +2 in dashboard)
- run_created_on_claim / run_closed_on_complete_with_summary
- run_summary_falls_back_to_result
- multiple_attempts_preserved_as_runs (3 attempts: reclaimed →
crashed → completed, all visible in list_runs)
- run_on_block_with_reason / run_on_spawn_failure_records_failed_runs
(5 spawn_failed runs + 1 gave_up run)
- event_rows_carry_run_id (task-scoped vs run-scoped split)
- build_worker_context_includes_prior_attempts
- build_worker_context_uses_parent_run_summary (metadata JSON in context)
- migration_backfills_inflight_run_for_legacy_db (simulates a
pre-migration running task, re-runs init_db, asserts backfill)
- forward_compat_columns_writable
- cli_runs_verb + cli_runs_json
- cli_complete_with_summary_and_metadata (JSON round-trip through
shlex + argparse)
- cli_complete_bad_metadata_exits_nonzero
- task_detail_includes_runs / task_detail_runs_empty_before_claim
269/269 kanban suite pass under scripts/run_tests.sh. Live-smoke
covered: single-attempt complete → run closed + summary persisted;
retry scenario → two runs visible (blocked + completed); parent run
summary + metadata surfaced to child via build_worker_context;
forward-compat columns writable via UPDATE; GET /tasks/:id returns
runs[].
Docs
- New 'Runs — one row per attempt' section in kanban.md: the
why (full attempt history, structured metadata), the two-table
model (task is logical, run is execution), the structured handoff
shape (--summary / --metadata), example CLI + dashboard output,
forward-compat note for v2.
- Event reference updated to mention task_events.run_id.
- CLI reference gains 'hermes kanban runs <id>'.
Not in v1 (deferred to v2):
- Workflow templates (workflow_templates + workflow_steps tables,
stage-based routing, success/failure step links).
- 'stage' as a distinct axis from status in the UI.
- Shared-by-default workspace binding across stages of the same
workflow run.
- Pipeline replacement for the kanban-orchestrator skill (the
orchestrator's 'decompose, don't execute' guidance is still
correct; it becomes partly redundant once workflows land).
Ports four items from the Multica audit (https://github.com/multica-ai/multica).
Dropped their cross-host server/daemon architecture and their Postgres+pgvector
skill search — both the wrong shape for our single-host SQLite kernel.
1. Per-task max-runtime (`max_runtime_seconds` column)
- New kernel function `enforce_max_runtime(conn)` runs in every dispatch
tick. When a running task's elapsed time exceeds the cap, we SIGTERM
the worker, wait a 5 s grace (polling _pid_alive), then SIGKILL. The
task goes back to 'ready' with a `timed_out` event and re-queues
on the next tick (unless the spawn-failure circuit breaker has
already parked it).
- Host-local only: lock prefix must match this host's claimer_id so we
never signal a PID on another machine.
- CLI: `hermes kanban create --max-runtime 30m | 2h | 1d | <seconds>`.
New `_parse_duration` helper accepts s/m/h/d suffixes or bare
integers.
- Dashboard POST body + the card's `max_runtime_seconds` field.
2. Worker heartbeat (`last_heartbeat_at` column, `heartbeat` event)
- `heartbeat_worker(conn, task_id, note=None)` emits the event and
touches last_heartbeat_at. Refused when the task isn't running.
- CLI: `hermes kanban heartbeat <id> [--note "..."]`.
- kanban-worker skill instructs workers to heartbeat during long
loops (training runs, encodes, crawls, batch uploads).
- Separate signal from PID crash detection: a worker's Python can
still be alive while the actual work process is stuck. Heartbeat
absence is diagnostic; future work can auto-block on stale
heartbeats but v1 just surfaces the signal.
3. Assignee enumeration (`known_assignees`, `list_profiles_on_disk`)
- Scans ~/.hermes/profiles/ for dirs containing config.yaml + unions
with current assignees on the board. Each entry returns
{name, on_disk, counts: {status: n}}.
- CLI: `hermes kanban assignees [--json]`. Also hooked into
`hermes kanban init` which now prints discovered profiles so new
installs see 'these are the assignees you can target' immediately.
- Dashboard: GET /api/plugins/kanban/assignees for the picker.
4. Event vocab cleanup (three renames + three new kinds)
- `ready` → `promoted` (fires when deps clear; clearer semantic).
- `priority` → `reprioritized` (past-tense verb, matches others).
- `spawn_auto_blocked` → `gave_up` (short, memorable; the circuit
breaker gave up on this task).
- New: `spawned` (emitted with {pid} on successful spawn),
`heartbeat` ({note?}), `timed_out`
({pid, elapsed_seconds, limit_seconds, sigkill}).
- One-shot migration in `_migrate_add_optional_columns` renames
legacy rows in-place on init_db(), so existing DBs upgrade cleanly.
- Gateway notifier's TERMINAL_KINDS set updated; timed_out gets its
own ⏱ message template, gave_up renamed from 'auto-blocked'.
- Plugin_api.py's two 'priority' emit sites renamed to
'reprioritized'.
- Documented in a new 'Event reference' section in kanban.md,
grouped into three clusters (lifecycle / edits / worker
telemetry) with payload shapes.
Tests (+18 in tests/hermes_cli/test_kanban_core_functionality.py,
136/136 pass):
- max_runtime_terminates_overrun_worker: real SIGTERM flow with
_pid_alive stub, verifies event payload + state reset.
- max_runtime_none_means_no_cap: unbounded tasks aren't timed out.
- create_task_persists_max_runtime.
- enforce_max_runtime_integrates_with_dispatch: kernel-level +
dispatch_once chaining.
- heartbeat_on_running_task + heartbeat_refused_when_not_running.
- cli_heartbeat_verb with --note round-trip.
- recompute_ready_emits_promoted_not_ready.
- spawn_failure_circuit_breaker_emits_gave_up.
- spawned_event_emitted_with_pid.
- migration_renames_legacy_event_kinds (injects old rows, re-runs
init_db, asserts rename).
- list_profiles_on_disk (tmp_path + config.yaml filter).
- known_assignees_merges_disk_and_board (profiles on disk + board
assignees + per-status counts).
- cli_assignees_json.
- parse_duration_accepts_formats (s/m/h/d/float).
- parse_duration_rejects_garbage.
- cli_create_max_runtime_via_duration (2h → 7200).
- cli_create_max_runtime_bad_format_exits_nonzero.
Live smoke: POST /tasks with max_runtime_seconds round-trips;
/assignees returns the union of on-disk + board-assigned names;
PATCH priority produces 'reprioritized' events (not 'priority');
board cards expose max_runtime_seconds + last_heartbeat_at.
Docs (website/docs/user-guide/features/kanban.md):
- New 'Event reference' section with three-cluster table
(lifecycle / edits / worker telemetry) + payload shapes.
- CLI reference updated for --max-runtime, heartbeat, assignees.
- Gateway notifications section updated for the new TERMINAL_KINDS.
Not ported from Multica (deliberate, documented in the out-of-scope
section already): Postgres+pgvector skill search (heavy deps conflict
with SQLite kernel), server+daemon cross-host model (we're
single-host on purpose), first-class agent identity with threaded
comments (we keep the board profile-agnostic).
Eliminates every 'known broken on day one' item in the core functionality
audit. The board is now self-driving (daemon, not cron), self-healing
(crash detection, spawn-failure circuit breaker), and self-reporting
(logs, stats, gateway notifications).
Dispatcher
- New `hermes kanban daemon` long-lived loop with --interval, --max,
--failure-limit, --pidfile, --verbose, signal-clean shutdown
(SIGINT/SIGTERM via threading.Event). A kb.run_daemon() entry point
lets tests drive it inline without subprocess.
- `hermes kanban init` now prints the dispatcher setup hint so users
don't leave the board off-by-default. Ships a systemd user unit at
plugins/kanban/systemd/hermes-kanban-dispatcher.service.
- Removed the old 'add this to cron' doc path. Cron runs agent
prompts (LLM cost per tick) — unacceptable for a per-minute
coordination loop.
Worker aliveness / safety
- Spawn returns the child's PID; dispatcher stores it on the task row
and calls detect_crashed_workers() every tick. If the PID is gone
but the claim TTL hasn't expired, the task drops back to ready with
a 'crashed' event. Host-local only — cross-host PIDs are ignored
per the single-host design.
- Spawn-failure circuit breaker: after N consecutive spawn_failed
events on the same task (default 5), the dispatcher auto-blocks
with the last error as the reason. Success resets the counter.
Workspace-resolution failures count against the same budget.
- Log rotation: _rotate_worker_log trims at 2 MiB, keeps one
generation (.log.1), bounds per-task disk usage at ~4 MiB.
Idempotency / dedup
- create_task(idempotency_key=...) returns the existing non-archived
task id for retried webhooks. --idempotency-key on the CLI, json
body field on the dashboard plugin. Archived tasks don't block a
fresh create with the same key.
CLI surface
- Bulk verbs: complete, unblock, archive accept multiple ids;
block accepts --ids for sibling blocks with the same reason.
- New verbs: daemon, watch (live event tail filtered by
assignee/tenant/kinds), stats, log, notify-subscribe,
notify-list, notify-unsubscribe.
- dispatch gains --failure-limit + crashed/auto_blocked columns in
JSON output and human-readable output.
- gc accepts --event-retention-days / --log-retention-days; prunes
task_events for terminal tasks and old log files.
Gateway integration
- New GatewayRunner._kanban_notifier_watcher: polls
kanban_notify_subs every 5s, pushes ✔/⏸/✖ messages to subscribed
chats for completed/blocked/spawn_auto_blocked/crashed events.
Cursor-advanced per-sub; auto-removed when the task reaches
done/archived. Runs alongside the session expiry and platform
reconnect watchers — SQLite work in asyncio.to_thread so the
event loop never blocks.
- /kanban create in the gateway auto-subscribes the originating
chat (platform + chat_id + thread_id). Users see
'(subscribed — you'll be notified when t_abcd completes or
blocks)' appended to the response.
Dashboard plugin
- GET /stats returns board_stats (by_status, by_assignee,
oldest_ready_age_seconds).
- GET /tasks/:id/log returns the worker log with optional ?tail=N
cap. 404 on unknown task, exists=false when the task has never
spawned.
- POST /tasks accepts idempotency_key; both Pydantic body and the
create_task kwarg now round-trip.
- /board attaches task.age (created/started/time_to_complete in
seconds) so the UI can colour stale cards without recomputing.
- Card CSS: amber border after N minutes, red border when clearly
stuck (tier per status: running 10m/60m, ready 1h/24h, todo
7d/30d, blocked 1h/24h).
- Drawer: new Worker log section, auto-loads on mount, last 100 KB
cap with on-disk path surfaced when truncated.
Kernel
- Schema additions: tasks.idempotency_key, tasks.spawn_failures,
tasks.worker_pid, tasks.last_spawn_error; new
kanban_notify_subs table. All gated by _migrate_add_optional_columns
so legacy DBs upgrade cleanly.
- release_stale_claims / complete_task / block_task now all clear
worker_pid so crash detection doesn't false-positive on reclaimed
tasks.
- read_worker_log fixed: tail-skip no longer eats one-giant-line
logs (common with child processes that don't flush newlines
before dying).
Tests (tests/hermes_cli/test_kanban_core_functionality.py, 28 new)
- Idempotency: same key returns existing, archived doesn't block,
no key never collides
- Circuit breaker: auto-blocks after limit, success resets counter,
workspace-resolution failure counts against budget
- Aliveness: _pid_alive helper, detect_crashed_workers reclaims
exited child
- Daemon: runs and stops cleanly via stop_event, survives a tick
exception
- Stats + task_age helpers
- Notify subs: CRUD, cursor advances, distinct-thread is a separate row
- GC: events-only-for-terminal-tasks, old worker logs deleted
- Log: rotation keeps one generation, read_worker_log tail
- CLI: bulk complete/archive/unblock/block, create with
--idempotency-key, stats --json, notify-subscribe+list, log
missing task, gc reports counts
- run_slash parity: smoke-tests every registered verb (23
invocations); none may raise or return empty string
Full kanban test suite: 234/234 pass under scripts/run_tests.sh
(60 original + 30 dashboard plugin + 28 new core + 116 command
registry). Live smoke covers /stats, idempotency, age, log endpoint
with and without content, log?tail= truncation signal, 404 on unknown
task.
Docs (website/docs/user-guide/features/kanban.md)
- 'Core concepts' rewritten: new statuses (triage), idempotency key,
dispatcher-as-daemon-not-cron with circuit breaker behaviour
documented.
- Quick start swapped to daemon. New systemd section covers user
service install.
- New sections: idempotent create, bulk verbs, gateway
notifications, out-of-scope single-host note (kanban.db is local;
don't expect multi-host).
- CLI reference updated for every new verb, every new flag.
Follows up on the initial dashboard plugin with the items called out
during self-review — ships the GUI-reality claims the PR body made,
closes the WebSocket auth gap, and lands the 'Triage' status the design
spec's Fusion-style screenshot leads with.
Kernel changes
- kanban_db.VALID_STATUSES gains 'triage'. status is TEXT without a
CHECK constraint so no schema migration is needed.
- create_task(triage=True) forces the initial status to 'triage'
regardless of parents, and parent ids are still validated so the
eventual link rows don't dangle. recompute_ready() only promotes
'todo' -> 'ready', so triage tasks are naturally isolated from the
dispatcher pipeline.
- hermes kanban create gains --triage.
Patterns table (docs) gains P9 'Triage specifier'.
Plugin backend (plugins/kanban/dashboard/plugin_api.py)
- GET /board now auto-init's kanban.db on first read (idempotent).
A fresh install shows an empty board instead of 'failed to load'.
- GET /board returns a new 'progress' field per task — {done, total}
of child-task completion, or None if the task has no children.
- BOARD_COLUMNS prepends 'triage'.
- POST /tasks accepts {triage: bool}; PATCH /tasks/:id accepts
{status: 'triage'}.
- WebSocket /events now requires ?token=<session_token> as a query
param — browsers can't set Authorization on a WS upgrade, so this
matches the pattern the in-browser PTY bridge uses. Constant-time
compare against hermes_cli.web_server._SESSION_TOKEN. In bare-test
contexts (no dashboard module) the check no-ops so the tail loop
stays testable. Security boundary documented in the module header
and in website/docs/user-guide/features/kanban.md.
Plugin UI (plugins/kanban/dashboard/dist/index.js + style.css)
- Adds the Triage column (lilac dot) with helper text
'Raw ideas — a specifier will flesh out the spec'. Inline-create
from the Triage column parks new tasks in triage.
- Status action row in the drawer gains '→ triage'.
- Progress pill (N/M) on cards that have children. Full-complete
state tints the pill green.
- 'Lanes by profile' toolbar toggle — sub-groups the Running column
by assignee so you see at a glance which specialist is busy on
what.
- Destructive status moves (done / archived / blocked) via drag-drop
OR via the drawer action row now prompt for confirmation.
- Escape closes the drawer.
- Live-update reloads are debounced (250ms) so a burst of
task_events triggers one refetch, not N.
- WebSocket includes ?token= built from window.__HERMES_SESSION_TOKEN__.
- WebSocket reconnect uses exponential backoff capped at 30s, not
a fixed 1.5s spin loop, and surfaces a user-visible error on
code-1008 (auth rejected) instead of reconnecting forever.
- ErrorBoundary wraps the page — a bad card render shows a
'rendering error, reload view' card instead of crashing the tab.
Tests (tests/plugins/test_kanban_dashboard_plugin.py, +5 tests = 21)
- empty-board shape now asserts all 6 columns including 'triage'
- create_triage_lands_in_triage_column
- triage_task_not_promoted_to_ready (dispatcher bypasses triage)
- patch_status_triage_works (both into triage and out of it)
- board_progress_rollup (0/2 -> 1/2 -> childless cards = None)
- board_auto_initializes_missing_db
- ws_events_rejects_when_token_required (three sub-assertions:
missing → 1008, wrong → 1008, correct → handshake accepted)
All 82 kanban tests pass under scripts/run_tests.sh.
Docs
- kanban.md 'What the plugin gives you' fully rewritten to match
shipped reality (triage, progress pill, assignee lanes,
destructive-confirm, Escape-close, debounce).
- New 'Security model' subsection documents the explicit-plugin-
route-bypass, the WS token requirement, and the --host 0.0.0.0
warning; also notes that kanban.db is profile-agnostic on purpose
(the coordination primitive) so cross-profile visibility is
expected.
- CLI command reference shows --triage.
- Collaboration patterns table adds P9 'Triage specifier'.
New `hermes kanban` CLI subcommand + `/kanban` slash command + skills for
worker and orchestrator profiles. SQLite-backed task board
(~/.hermes/kanban.db) shared across all profiles on the host. Zero
changes to run_agent.py, no new core tools, no tool-schema bloat.
Motivation: delegate_task is a function call — sync fork/join, anonymous
subagent, no resumability, no human-in-the-loop. Kanban is the durable
shape needed for research triage, scheduled ops, digital twins,
engineering pipelines, and fleet work. They coexist (workers may call
delegate_task internally).
What this adds
- hermes_cli/kanban_db.py — schema, CAS claim, dependency resolution,
dispatcher, workspace resolution, worker-context builder.
- hermes_cli/kanban.py — 15-verb CLI surface and shared run_slash()
entry point used by both CLI and gateway.
- skills/devops/kanban-worker — how a profile should work a claimed task.
- skills/devops/kanban-orchestrator — "you are a dispatcher, not a
worker" template with anti-temptation rules.
- /kanban slash command wired into cli.py and gateway/run.py. Bypasses
the running-agent guard (board writes don't touch agent state), so
/kanban unblock can free a stuck worker mid-conversation.
- Design spec at docs/hermes-kanban-v1-spec.pdf — comparative analysis
vs Cline Kanban, Paperclip, NanoClaw, Gemini Enterprise; 8 patterns;
4 user stories; implementation plan; concurrency correctness.
- Docs: website/docs/user-guide/features/kanban.md, CLI reference
updated, sidebar entry added.
Architecture highlights
- Three planes: control (user + gateway), state (board + dispatcher),
execution (pool of profile processes).
- Every worker is a full OS process, spawned as `hermes -p <profile>`.
No in-process subagent swarms — solves NanoClaw's SDK-lifecycle
failure class.
- Atomic claim via SQLite CAS in a BEGIN IMMEDIATE transaction; stale
claims reclaimed 15 min after their TTL expires.
- Tenant namespacing via one nullable column — one specialist fleet
can serve many businesses with data isolation by workspace path.
Tests: 60 targeted tests (schema, CAS atomicity, dependency resolution,
dispatcher, workspace kinds, tenancy, CLI + slash surface). All pass
hermetic via scripts/run_tests.sh.
New `hermes kanban` CLI subcommand + `/kanban` slash command + skills for
worker and orchestrator profiles. SQLite-backed task board
(~/.hermes/kanban.db) shared across all profiles on the host. Zero
changes to run_agent.py, no new core tools, no tool-schema bloat.
Motivation: delegate_task is a function call — sync fork/join, anonymous
subagent, no resumability, no human-in-the-loop. Kanban is the durable
shape needed for research triage, scheduled ops, digital twins,
engineering pipelines, and fleet work. They coexist (workers may call
delegate_task internally).
What this adds
- hermes_cli/kanban_db.py — schema, CAS claim, dependency resolution,
dispatcher, workspace resolution, worker-context builder.
- hermes_cli/kanban.py — 15-verb CLI surface and shared run_slash()
entry point used by both CLI and gateway.
- skills/devops/kanban-worker — how a profile should work a claimed task.
- skills/devops/kanban-orchestrator — "you are a dispatcher, not a
worker" template with anti-temptation rules.
- /kanban slash command wired into cli.py and gateway/run.py. Bypasses
the running-agent guard (board writes don't touch agent state), so
/kanban unblock can free a stuck worker mid-conversation.
- Design spec at docs/hermes-kanban-v1-spec.pdf — comparative analysis
vs Cline Kanban, Paperclip, NanoClaw, Gemini Enterprise; 8 patterns;
4 user stories; implementation plan; concurrency correctness.
- Docs: website/docs/user-guide/features/kanban.md, CLI reference
updated, sidebar entry added.
Architecture highlights
- Three planes: control (user + gateway), state (board + dispatcher),
execution (pool of profile processes).
- Every worker is a full OS process, spawned as `hermes -p <profile>`.
No in-process subagent swarms — solves NanoClaw's SDK-lifecycle
failure class.
- Atomic claim via SQLite CAS in a BEGIN IMMEDIATE transaction; stale
claims reclaimed 15 min after their TTL expires.
- Tenant namespacing via one nullable column — one specialist fleet
can serve many businesses with data isolation by workspace path.
Tests: 60 targeted tests (schema, CAS atomicity, dependency resolution,
dispatcher, workspace kinds, tenancy, CLI + slash surface). All pass
hermetic via scripts/run_tests.sh.
The ephemeral no-tools side-question variant of /btw confused users who
expected 'by-the-way' to mean 'run this off to the side with tools' —
they'd type /btw and get a toolless agent that couldn't do the work.
/bg worked because it was /background with full tools.
Collapse the two: /btw and /bg both alias to /background. One command,
one behavior, no more gotchas about which variant has tools.
Removed:
- _handle_btw_command in cli.py and gateway/run.py
- _run_btw_task + _active_btw_tasks state in gateway/run.py
- prompt.btw JSON-RPC method + btw.complete event in tui_gateway
- BtwStartResponse type + btw.complete case in ui-tui
- Standalone /btw slash tree registration in Discord
- Standalone btw CommandDef in hermes_cli/commands.py
Updated:
- background CommandDef aliases: (bg,) -> (bg, btw)
- TUI session.ts: local btw handler merged into background
- Docs and tips updated to describe /btw as a /background alias
Manage the fallback_providers chain from the CLI instead of hand-editing
config.yaml. The picker reuses select_provider_and_model() from 'hermes
model' — same provider list, same credential prompts, same model picker.
hermes fallback [list] Show the current chain (primary + fallbacks)
hermes fallback add Run the model picker, append selection to chain
hermes fallback remove Pick an entry to delete (arrow-key menu)
hermes fallback clear Remove all entries (with confirmation)
'add' snapshots config['model'] before calling the picker, extracts the
user's selection from the post-picker state, then restores the primary
and appends {provider, model, base_url?, api_mode?} to fallback_providers.
Auth store's active_provider is snapshot/restored too so OAuth-provider
fallbacks don't silently deactivate the user's primary. Duplicates and
self-as-fallback are rejected. Legacy single-dict 'fallback_model' entries
are auto-migrated to the list format on first write.
Instead of a blocking first-run questionnaire, show a one-time hint the first
time the user hits each behavior fork:
1. First message while the agent is working — appends a hint to the busy-ack
explaining the /busy queue vs /busy interrupt knob, phrased to match the
mode that was just applied (don't tell a queue-mode user to switch to
queue).
2. First tool that runs for >= 30s in the noisiest progress mode
(tool_progress: all) — prints a hint about /verbose to cycle display
modes (all -> new -> off -> verbose). Gated on /verbose actually being
usable on the surface: always shown on CLI; on gateway only shown when
display.tool_progress_command is enabled.
Each hint is latched in config.yaml under onboarding.seen.<flag>, so it
fires exactly once per install across CLI, gateway, and cron, then never
again. Users can wipe the section to re-see hints.
New:
- agent/onboarding.py — is_seen / mark_seen / hint strings, shared by
both CLI and gateway.
- onboarding.seen in DEFAULT_CONFIG (hermes_cli/config.py) and in
load_cli_config defaults (cli.py). No _config_version bump — deep
merge handles new keys.
Wired:
- gateway/run.py: _handle_active_session_busy_message appends the hint
after building the ack. progress_callback tracks tool.completed
duration and queues the tool-progress hint into the progress bubble.
- cli.py: CLI input loop appends the busy-input hint on the first busy
Enter; _on_tool_progress appends the tool-progress hint on the first
>=30s tool completion. In-memory CLI_CONFIG is also updated so
subsequent fires in the same process are suppressed immediately.
All writes go through atomic_yaml_write and are wrapped in try/except
so onboarding can never break the input/busy-ack paths.
OpenRouter and Nous Portal curated picker lists now resolve via a JSON
manifest served by the docs site, falling back to the in-repo snapshot
when unreachable. Lets us update model lists without shipping a release.
Live URL: https://hermes-agent.nousresearch.com/docs/api/model-catalog.json
(source at website/static/api/model-catalog.json; auto-deploys via the
existing deploy-site.yml GitHub Pages pipeline on every merge to main).
Schema (v1) carries id + optional description + free-form metadata at
manifest, provider, and model levels. Pricing and context length stay
live-fetched via existing machinery (/v1/models endpoints, models.dev).
Config (new model_catalog section, default enabled):
model_catalog.url master manifest URL
model_catalog.ttl_hours disk cache TTL (default 24h)
model_catalog.providers.<name>.url optional per-provider override
Fetch pipeline: in-process cache -> disk cache (fresh < TTL) -> HTTP
fetch -> disk-cache-on-failure fallback -> in-repo snapshot as last
resort. Never raises to callers; at worst returns the bundled list.
Changes:
- website/static/api/model-catalog.json initial manifest (35 OR + 31 Nous)
- scripts/build_model_catalog.py regenerator from in-repo lists
- hermes_cli/model_catalog.py fetch + validate + cache module
- hermes_cli/models.py fetch_openrouter_models() +
new get_curated_nous_model_ids()
- hermes_cli/main.py, hermes_cli/auth.py Nous flows use the helper
- hermes_cli/config.py model_catalog defaults
- website/docs/reference/model-catalog.md + sidebars.ts
- tests/hermes_cli/test_model_catalog.py 21 tests (validation, fetch
success/failure, accessors,
disabled, overrides, integration)
Plugin hooks fired after a tool dispatch now receive an integer
duration_ms kwarg measuring how long the tool's registry.dispatch()
call took (time.monotonic() before/after). Inspired by Claude Code
2.1.119 which added the same field to PostToolUse hook inputs.
Wire points:
- model_tools.py: measure dispatch latency, pass duration_ms to
invoke_hook("post_tool_call", ...) and invoke_hook("transform_tool_result", ...)
- hermes_cli/hooks.py: include duration_ms in the synthetic payload
used by 'hermes hooks test' and 'hermes hooks doctor' so shell-hook
authors see the same shape at development time as runtime
- shell hooks (agent/shell_hooks.py): no code change needed;
_serialize_payload already surfaces non-top-level kwargs under
payload['extra'], so duration_ms lands at extra.duration_ms for
shell-hook scripts
Plugin authors can now build latency dashboards, per-tool SLO alerts,
and regression canaries without having to wrap every tool manually.
Test: tests/test_model_tools.py::test_post_tool_call_receives_non_negative_integer_duration_ms
E2E: real PluginManager + dispatch monkey-patched with a 50ms sleep,
hook callback observes duration_ms=50 (int).
Refs: https://code.claude.com/docs/en/changelog (2.1.119, Apr 23 2026)
Bare `hermes setup` on a returning user now drops straight into the
full reconfigure wizard — every prompt shows the current value as its
default, press Enter to keep or type a new value to change it. The
returning-user menu is gone.
Behavior:
- First-time user: first-time wizard (unchanged)
- Returning user, bare command: full reconfigure wizard (new default)
- Returning user, `--quick`: only prompt for missing/unset items
- Returning user, one section: `hermes setup model|terminal|gateway|tools|agent`
- `--reconfigure`: preserved as backwards-compat alias (no-op since it's now default)
The section functions already used current values as prompt defaults —
this change just removes the extra click to get to them.
The 'Quick Setup - configure missing items only' menu option is now
exposed as the explicit `--quick` flag; it's the narrow case of
filling in missing config (e.g. after a partial OpenClaw migration or
when a required API key got cleared).
Inspired by Mercury Agent's `mercury doctor` UX.
Also removes:
- RETURNING_USER_MENU_SECTION_KEYS (orphaned constant)
- Two returning-user menu tests in test_setup_noninteractive.py
(guarding behavior that no longer exists — covered by
test_setup_reconfigure.py instead)
The azure-foundry wizard now probes the endpoint before asking the user
to pick anything by hand:
1. URL path sniff — endpoints ending in /anthropic are Azure Foundry
Claude routes and skip to anthropic_messages.
2. GET <base>/models probe — if the endpoint returns an OpenAI-shaped
model list, we switch to chat_completions and prefill the picker
with the returned deployment/model IDs.
3. Anthropic Messages probe — fallback for endpoints that don't expose
/models but do speak the Anthropic Messages shape.
4. Manual fallback — private endpoints / custom routes still work;
the user picks API mode + types a deployment name.
Context length for the selected model is resolved through the existing
agent.model_metadata.get_model_context_length chain (models.dev,
provider metadata, hardcoded family fallbacks) and stored in
model.context_length when a non-default value is found.
Also refactors runtime_provider so Azure Foundry resolution is reused
between the explicit-credentials path and the default top-level path —
previously the /v1 strip for Anthropic-style Azure only ran when the
caller passed explicit_* args, which meant config-driven sessions
hit a double-/v1 URL.
New module hermes_cli/azure_detect.py with 19 unit tests covering:
- path sniff, model ID extraction, probe fallbacks
- HTTP error handling (URLError, HTTPError)
- context-length lookup passthrough
- DEFAULT_FALLBACK_CONTEXT rejection
New runtime tests cover:
- OpenAI-style Azure Foundry
- Anthropic-style Azure Foundry with /v1 stripping
- Missing base_url / API key raising AuthError
Rationale: Microsoft confirms there's no pure-API-key endpoint to list
Azure deployments (that requires ARM management auth). The v1 Azure
OpenAI endpoint does expose /models with the resource's available
model catalog, which is good enough for picker prefill in the common
case. Users on private/gated endpoints fall through to manual entry.
Add support for Azure Foundry as a new inference provider. Azure Foundry
endpoints can use either OpenAI-style (/v1/chat/completions) or
Anthropic-style (/v1/messages) API formats.
Changes:
- Add azure-foundry to PROVIDER_REGISTRY (auth.py)
- Add azure-foundry overlay in HERMES_OVERLAYS (providers.py)
- Add empty model list for azure-foundry (models.py)
- Add _model_flow_azure_foundry() interactive setup (main.py)
- Add azure-foundry runtime resolution with api_mode support (runtime_provider.py)
- Add AZURE_FOUNDRY_API_KEY and AZURE_FOUNDRY_BASE_URL env vars (config.py)
Usage:
hermes model -> More providers -> Azure Foundry
The setup wizard prompts for:
- Endpoint URL
- API format (OpenAI or Anthropic-style)
- API key
- Model name
Configuration is saved to config.yaml (model.provider, model.base_url,
model.api_mode, model.default) and ~/.hermes/.env (AZURE_FOUNDRY_API_KEY).
Fixes#15779. Custom-provider per-model context_length (`custom_providers[].models.<id>.context_length`) is now honored across every resolution path, not just agent startup. Also adds 256K as the top probe tier and default fallback.
## What changed
New helper `hermes_cli.config.get_custom_provider_context_length()` — single source of truth for the per-model override lookup, with trailing-slash-insensitive base-url matching.
`agent.model_metadata.get_model_context_length()` gains an optional `custom_providers=` kwarg (step 0b — runs after explicit `config_context_length` but before every other probe).
Wired through five call sites that previously either duplicated the lookup or ignored it entirely:
- `run_agent.py` startup — refactored to use the new helper (dedups legacy inline loop, keeps invalid-value warning)
- `AIAgent.switch_model()` — re-reads custom_providers from live config on every /model switch
- `hermes_cli.model_switch.resolve_display_context_length()` — new `custom_providers=` kwarg
- `gateway/run.py` /model confirmation (picker callback + text path)
- `gateway/run.py` `_format_session_info` (/info)
## Context probe tiers
`CONTEXT_PROBE_TIERS = [256_000, 128_000, 64_000, 32_000, 16_000, 8_000]` — was `[128_000, ...]`. `DEFAULT_FALLBACK_CONTEXT` follows tier[0], so unknown models now default to 256K. The stale `128000` literal in the OpenRouter metadata-miss path is replaced with `DEFAULT_FALLBACK_CONTEXT` for consistency.
## Repro (from #15779)
```yaml
custom_providers:
- name: my-custom-endpoint
base_url: https://example.invalid/v1
model: gpt-5.5
models:
gpt-5.5:
context_length: 1050000
```
`/model gpt-5.5 --provider custom:my-custom-endpoint` → previously "Context: 128,000", now "Context: 1,050,000".
## Tests
- `tests/hermes_cli/test_custom_provider_context_length.py` — new file, 19 tests covering the helper, step-0b integration, and the 256K tier invariants
- `tests/hermes_cli/test_model_switch_context_display.py` — added regression tests for #15779 through the display resolver
- `tests/gateway/test_session_info.py` — updated default-fallback assertion (128K → 256K)
- `tests/agent/test_model_metadata.py` — updated tier assertions for the new top tier
The raw-template lookup added in PR #15817 went through
`get_compatible_custom_providers(read_raw_config())`, which calls
`_normalize_custom_provider_entry` → `urlparse(base_url)`. Any
entry whose `base_url` is itself an env-ref (`${NEURALWATT_API_BASE}`)
was dropped as 'not a valid URL', so `api_key_ref` stayed empty and the
resolved secret was still written to `model.api_key` — the exact case
the original Discord report described.
Replace the normalizer-gated lookup with a direct read of
`raw['custom_providers']` and `raw['providers']`, indexed by name
(case-insensitive, optionally qualified by model) so the loaded
(expanded) entry can be matched regardless of how `base_url` is
written.
Add an integration regression test driving the real
`select_provider_and_model` entry point with the Discord-reported
NeuralWatt config (`${VAR}` in both `base_url` and `api_key`).
This test fails on the PR-only fix and passes with the broadened
lookup.
When switching models on a custom endpoint (ollama-launch):
- Same-provider switches no longer re-resolve credentials (fixes base_url
being lost for 'custom' provider on subsequent switches)
- Named providers (ollama-launch) are resolved via user_providers so
switch_model can find their base_url from config
- Models not in the /v1/models probe but present in the user's saved
provider config are accepted with a warning instead of rejected
- CLI /model and TUI /model both pass user_providers/custom_providers
to switch_model so the config model list is available for validation
Closes#15088
- expand short model aliases like sonnet/opus via static catalogs during startup runtime resolution
- keep startup alias resolution network-free and add regression tests in models and tui gateway suites
The post-graceful-drain is-active poll used a fixed 10s timeout, but
systemd's hermes-gateway.service has RestartSec=30 — so systemd won't
respawn the unit for 30s after exit-75, and our poll gives up during
the cooldown. Result: every 'hermes update' printed
⚠ hermes-gateway drained but didn't relaunch — forcing restart
followed by a redundant 'systemctl restart' that kicked the newly-
respawning gateway again (and re-started WhatsApp / Discord a second
time in the process).
Fix: read RestartUSec from the unit via 'systemctl show' and set the
poll budget to max(10s, RestartSec + 10s slack). Units without
RestartSec set (or value=infinity) fall back to the original 10s.
Observed timeline from journalctl before fix:
08:56:22.262 old PID exits 75
08:56:32.707 systemd logs Stopped -> Started (10.4s gap, > 10s budget)
After fix the poll covers 40s — comfortably inside RestartSec + slack.
Validation:
- RestartUSec parser tested against '30s', '100ms', '1min 30s',
'infinity', '', 'garbage', '500us', '2min' — all correct.
- Against the live hermes-gateway.service: parses to 30.0s.
- tests/hermes_cli/test_update_gateway_restart.py: 41/41 pass.
Makes hermes -z usable by sweeper without mutating user config.
- Top-level -m/--model and --provider flags that apply to -z/--oneshot
(mirrors hermes chat's plumbing).
- HERMES_INFERENCE_MODEL env var as the parallel to HERMES_INFERENCE_PROVIDER
for CI / scripted invocations.
- resolve_runtime_provider() gets the requested provider; when --model is
given without --provider, detect_provider_for_model() auto-selects the
provider that serves it (same semantic as /model in an interactive session).
- --provider without --model errors out with exit 2 — carrying a config
model across to a different provider is usually wrong, and silently
picking the provider's catalog default hides the mismatch.
Config defaults still used when both flags are omitted (existing behavior).
Validation (all live against OpenRouter):
-z 'x' ....................... uses config default (opus-4.7)
-z 'x' --model haiku-4.5 ..... haiku-4.5 via auto-detected openrouter
-z 'x' --model ... --provider pair as given
HERMES_INFERENCE_MODEL=... -z haiku-4.5 via env var
-z 'x' --provider anthropic .. exits 2 with error to stderr
* feat: add `hermes -z <prompt>` one-shot mode
Top-level flag that runs a single prompt and prints ONLY the final
response text to stdout. No banner, no spinner, no tool previews, no
session_id line — stdout is machine-readable, stderr is silent.
Tools, memory, rules, and AGENTS.md in the CWD are loaded as normal.
Approvals are auto-bypassed (sets HERMES_YOLO_MODE=1 for the call).
Bypasses cli.py entirely — goes straight to AIAgent.chat().
* feat(oneshot): handle interactive-callback gaps explicitly
Document (and where needed, patch) the interactive surfaces that have
no user to answer in oneshot mode:
- clarify — inject a callback that tells the agent to pick the
best default and continue (previously returned a
generic 'not available in this execution context'
error that wastes a tool call)
- sudo password — terminal_tool already gates on HERMES_INTERACTIVE
(we don't set it); sudo fails gracefully
- shell hooks — HERMES_ACCEPT_HOOKS=1 auto-approves; also falls
back to deny on non-tty stdin
- dangerous cmd — HERMES_YOLO_MODE=1 short-circuits before input()
- secret capture— tool returns gracefully when no callback wired
Live-tested: agent asked clarify(['red','blue']) and got 'red' back,
replied with only 'red'.
The AIAgent.flush_memories pre-compression save, the gateway
_flush_memories_for_session, and everything feeding them are
obsolete now that the background memory/skill review handles
persistent memory extraction.
Problems with flush_memories:
- Pre-dates the background review loop. It was the only memory-save
path when introduced; the background review now fires every 10 user
turns on CLI and gateway alike, which is far more frequent than
compression or session reset ever triggered flush.
- Blocking and synchronous. Pre-compression flush ran on the live agent
before compression, blocking the user-visible response.
- Cache-breaking. Flush built a temporary conversation prefix
(system prompt + memory-only tool list) that diverged from the live
conversation's cached prefix, invalidating prompt caching. The
gateway variant spawned a fresh AIAgent with its own clean prompt
for each finalized session — still cache-breaking, just in a
different process.
- Redundant. Background review runs in the live conversation's
session context, gets the same content, writes to the same memory
store, and doesn't break the cache. Everything flush_memories
claimed to preserve is already covered.
What this removes:
- AIAgent.flush_memories() method (~248 LOC in run_agent.py)
- Pre-compression flush call in _compress_context
- flush_memories call sites in cli.py (/new + exit)
- GatewayRunner._flush_memories_for_session + _async_flush_memories
(and the 3 call sites: session expiry watcher, /new, /resume)
- 'flush_memories' entry from DEFAULT_CONFIG auxiliary tasks,
hermes tools UI task list, auxiliary_client docstrings
- _memory_flush_min_turns config + init
- #15631's headroom-deduction math in
_check_compression_model_feasibility (headroom was only needed
because flush dragged the full main-agent system prompt along;
the compression summariser sends a single user-role prompt so
new_threshold = aux_context is safe again)
- The dedicated test files and assertions that exercised
flush-specific paths
What this renames (with read-time backcompat on sessions.json):
- SessionEntry.memory_flushed -> SessionEntry.expiry_finalized.
The session-expiry watcher still uses the flag to avoid re-running
finalize/eviction on the same expired session; the new name
reflects what it now actually gates. from_dict() reads
'expiry_finalized' first, falls back to the legacy 'memory_flushed'
key so existing sessions.json files upgrade seamlessly.
Supersedes #15631 and #15638.
Tested: 383 targeted tests pass across run_agent/, agent/, cli/,
and gateway/ session-boundary suites. No behavior regressions —
background memory review continues to handle persistent memory
extraction on both CLI and gateway.
The auto-restart path in `hermes update` verifies systemd unit health with
`time.sleep(3)` + a single `systemctl is-active` call. The unit's
Stopped -> Started transition after a graceful SIGUSR1 exit (or a hard
restart) is not always complete inside that 3s window, so the verify
races and reports 'drained but didn't relaunch' even though systemd is
about to bring the unit back up a fraction of a second later. Users
then see a spurious warning, a redundant fallback `systemctl restart`
fires, and adapters (Discord, WhatsApp) get restarted twice.
Replace the three sleep+oneshot sites with a small `_wait_for_service_active()`
closure that polls `is-active` every 0.5s for up to 10s. Behaviour
is unchanged when the unit is healthy or truly dead — only the race
window around a clean restart is now handled correctly.
Tests: tests/hermes_cli/test_update_gateway_restart.py (41/41).
`hermes tools` → "reconfigure existing" listed Spotify twice because
the Apr 24 refactor that moved Spotify into plugins/spotify/ (PR #15174)
left the entry in CONFIGURABLE_TOOLSETS. _get_effective_configurable_toolsets()
unconditionally appended get_plugin_toolsets() on top, so the same
'spotify' key showed up from both sources.
Dedupe by key — built-in CONFIGURABLE_TOOLSETS entry wins (it has the
nicer label and description). Also guards against future bundled plugins
that share a toolset key with a built-in.
Both discord (read/participate) and discord_admin (server admin) are now
configurable via `hermes tools` with default-OFF. Previously the core
discord tool (fetch_messages, search_members, create_thread) auto-loaded
on every Discord install with DISCORD_BOT_TOKEN set — 19 tools the user
never opted into.
Adds a platform-scoping mechanism (_TOOLSET_PLATFORM_RESTRICTIONS) so
the discord toolsets only show up in the Discord platform's checklist,
not on CLI/Telegram/Slack/etc. Applied at four gates:
- _prompt_toolset_checklist: checklist filter
- _get_platform_tools: resolution filter (both branches)
- _save_platform_tools: save-time filter (covers 'Configure all
platforms' and hand-edited config.yaml)
- tools_disable_enable_command: rejects `hermes tools enable discord`
on non-Discord platforms with a clear error
build_session_context_prompt now injects the Discord IDs block only
when both conditions hold: the discord/discord_admin toolset is
enabled AND DISCORD_BOT_TOKEN is set. Toolset alone isn't enough —
the tool's check_fn gates on the token at registry time, so opting
in without a token yields no tools and the IDs block would lie.
Otherwise keep the stale-API disclaimer.
Split the monolithic discord_server tool (14 actions) into two:
- discord: core actions (fetch_messages, search_members, create_thread)
that are useful for the agent's normal operation. Auto-enabled on
the discord platform via the pipeline fix.
- discord_admin: server management actions (list channels/roles, pins,
role assignment) that require explicit opt-in via hermes tools.
Added to CONFIGURABLE_TOOLSETS and _DEFAULT_OFF_TOOLSETS.
The reverse-mapping loop in _get_platform_tools only checked
CONFIGURABLE_TOOLSETS, silently dropping platform-specific toolsets
like discord and feishu_doc whose tools were in the composite but
had no configurable key. Add a second pass over TOOLSETS that picks
up unclaimed toolsets whose tools are present in the resolved
composite.
YAML parses bare numeric toolset names (e.g. 12306:) as int, causing
TypeError in sorted() since the read path normalizes to str but the
save path did not.
The no_mcp sentinel was preserved in existing entries even when the
user re-enabled MCP servers, causing MCP to stay silently disabled.
Subagents run inside a ThreadPoolExecutor. The CLI's interactive approval
callback lives in tools/terminal_tool.py's threading.local(), which worker
threads do not inherit. When a subagent hits a dangerous-command guard,
prompt_dangerous_approval() falls back to input() from the worker thread,
deadlocking against the parent's prompt_toolkit TUI that owns stdin.
Fix: install a non-interactive callback into every subagent worker thread
via ThreadPoolExecutor(initializer=set_approval_callback, initargs=(cb,)).
The callback is config-gated by delegation.subagent_auto_approve:
false (default) -> _subagent_auto_deny (safe; matches leaf tool blocklist)
true -> _subagent_auto_approve (opt-in YOLO for cron/batch)
Both emit a logger.warning audit line. Gateway sessions are unaffected
because they resolve approvals via tools/approval.py's per-session queue,
not through these TLS callbacks. Diagnosis credit: @MorAlekss (#14685).
- hermes_cli/config.py: DEFAULT_CONFIG.delegation.subagent_auto_approve: False
- cli-config.yaml.example: documented, commented (default)
- tools/delegate_tool.py: _subagent_auto_deny, _subagent_auto_approve,
_get_subagent_approval_callback, wired into the child timeout executor
- tests/tools/test_delegate.py: 7 tests covering defaults, truthy coercion,
and TLS scoping in the worker thread
/model gpt-5.5 on openai-codex showed 'Context: 1,050,000 tokens' because
the display block used ModelInfo.context_window directly from models.dev.
Codex OAuth actually enforces 272K for the same slug, and the agent's
compressor already runs at 272K via get_model_context_length() — so the
banner + real context budget said 272K while /model lied with 1M.
Route the display context through a new resolve_display_context_length()
helper that always prefers agent.model_metadata.get_model_context_length
(which knows about Codex OAuth, Copilot, Nous caps) and only falls back
to models.dev when that returns nothing.
Fix applied to all 3 /model display sites:
cli.py _handle_model_switch
gateway/run.py picker on_model_selected callback
gateway/run.py text-fallback confirmation
Reported by @emilstridell (Telegram, April 2026).
_load_auth_store() caught all parse/read exceptions and silently
returned an empty store, making corruption look like a logout with
no diagnostic information and no way to recover the original file.
Now copies the corrupt file to auth.json.corrupt before resetting,
and logs a warning with the exception and backup path.
_submit_anthropic_pkce() retrieved sess under _oauth_sessions_lock but
wrote back to sess["status"] and sess["error_message"] outside the lock.
A concurrent session GC or cancel could race with these writes, producing
inconsistent session state.
Wrap all 4 sess write sites in _oauth_sessions_lock:
- network exception path (Token exchange failed)
- missing access_token path
- credential save failure path
- success path (approved)
The web UI schema advertised 'block' as a busy_input_mode choice, but
no implementation ever existed — the gateway and CLI both silently
collapsed 'block' (and anything other than 'queue') to 'interrupt'.
Users who picked 'block' in the dashboard got interrupts anyway.
Drop 'block' from the select options. The two supported modes are
'interrupt' (default) and 'queue'.
Replaces gpt-5.4 / gpt-5.4-pro entries in the OpenRouter fallback snapshot
and the Nous Portal curated list. Other aggregators (Vercel AI Gateway)
and provider-native lists are unchanged.
Teknium