feat(KB-206): add stuck task detection and auto-recovery

- Add taskStuckTimeoutMs setting to detect tasks with no activity
- Create StuckTaskDetector to poll in-progress tasks every 30s
- Integrate heartbeat tracking into executor via step callbacks
- Implement recovery flow: abort stuck sessions, retry with preserved progress
- Export detector and wire into dashboard for real-time monitoring
- Add comprehensive tests and documentation to AGENTS.md
This commit is contained in:
gsxdsm
2026-03-30 17:16:35 -07:00
parent f8622daee3
commit b7349a3f1f
8 changed files with 754 additions and 4 deletions

View File

@@ -0,0 +1,5 @@
---
"@dustinbyrne/kb": minor
---
Add stuck task detection and recovery. When `taskStuckTimeoutMs` is configured, tasks with stagnant agent sessions (no text, tool, or progress activity) are automatically terminated and retried from their current step.