Skip to content
Rig Docs

User story: self-auditing rig — automated stuck-issue detection and memory loop

As a rig operator

I want the rig to automatically detect when issues are stuck, investigate the event sequence that caused the stuck state, file a new stuck-pattern issue when the pattern is novel, capture the fix into agent memory after resolution, and surface known patterns before an agent starts working on a recurrence

So that operator time is not spent re-discovering the same gotchas in every session — the rig recognises and explains known failure modes autonomously.

Acceptance criteria

Section titled “Acceptance criteria”
  • Stuck condition introduced manually → stuck-pattern issue filed on rig-conductor within 10 min
  • Fix lands and issue closes → write_memory called with kind: gotcha and fingerprint
  • BRAIN.md auto-regenerates with updated ## Known stuck patterns block
  • Repeat condition introduced → stuck-watcher comments on existing issue rather than filing a new one
  • Next agent session after fix → read_memories surfaces the gotcha entry pre-work

Spawned proposal

Section titled “Spawned proposal”