feat(kanban): auto-subscribe on create/assign, SIGUSR1 dispatcher wake, claimed event

[jrhouston-trilogy]

Summary

Production audit by EVE (Hermes agent, 2026-05-08) identified two notification gaps:

1. hermes kanban create never auto-subscribed the assignee — callers had to wrap every create with a helper script that duplicated a profile→platform map.
2. The only notification events were terminal (completed/blocked/gave_up/crashed/timed_out). Tasks that ran for >30s gave zero signal that work had started.

Additionally, the dispatcher's 60s tick interval meant up to 60s latency between task creation and worker spawn.

Changes

P1 — CLI auto-subscribe (hermes_cli/kanban.py)

• Added _load_notify_routes() — loads ~/.hermes/kanban/notify-routes.yaml
• Added _auto_subscribe_from_routes() — looks up assignee route and calls kb.add_notify_sub
• Called from _cmd_create and _cmd_assign after successful DB write
• Missing route = silent no-op. Bad route = stderr warning. Subscribe failure = stderr warning, no crash.

P2 — SIGUSR1 dispatcher wake (gateway/run.py)

• Added asyncio event _dispatch_wake wired to a SIGUSR1 handler
• Dispatcher tick now wakes immediately on SIGUSR1 instead of waiting up to dispatch_interval_seconds
• Degrades gracefully on Windows/platforms without signal support

P3 — claimed notification event (kanban_db.py, gateway/run.py)

• claim_task() now appends a claimed event with the claimer ID
• Notifier watches claimed in addition to terminal kinds (via NOTIFY_KINDS)
• Delivers ▶ Kanban <id> started by @<profile> — <title> message
• claimed is NOT terminal — subscriptions are preserved after delivery

P4 — Tests (tests/hermes_cli/test_kanban_notify_routes.py)

• notify-routes.yaml loaded correctly
• Missing route = no subscribe attempt
• Bad route (missing chat_id) = warning, not crash
• Subscribe failure = no exception propagated

Notes

• notify-routes.yaml schema is unchanged from the user-space prototype
• The SIGUSR1 helper in user scripts was already best-effort no-op; this makes it live
• No breaking changes to existing CLI surface

[teknium1]

Thanks @jrhouston-trilogy — closing this one. Status of the three pieces:

• P3 (claimed event) is already landed on main: hermes_cli/kanban_db.py:2091 appends the claimed event in claim_task() and the notifier picks it up.
• P2 (SIGUSR1 dispatcher wake) conflicts with main's existing use of SIGUSR1 for graceful gateway restart. A different signal (or an explicit IPC channel) would be needed; that's a design discussion rather than a salvage.
• P1 (auto-subscribe via notify-routes.yaml) introduces a new YAML schema and a profile→platform map. The kanban subsystem already has add_notify_sub and gateway-create auto-subscribe (feat(kanban): auto-subscribe gateway chat on tool-driven kanban_create #19718/follow-ups landed) covers the chat-id origin path. A YAML-based assignee→route lookup would still be useful for headless CLI flows, but it deserves a focused PR with the schema and merge-with-existing-mechanisms questions answered.

If you want to revive P1 as a fresh, focused PR against current main, happy to take another look. Appreciate the audit and contribution!