| name | error-sweep |
| description | Find the highest-signal recurring error in local logs, trace it, and propose a verified fix. |
| triggers | ["/error-sweep","production error sweep","log error triage"] |
| version | 1.0.0 |
| category | loop |
| effort | high |
Production Error Sweep Loop
A convergence loop that picks the single most actionable recurring error each
iteration, traces it to a root cause, and proposes (or applies) a narrow,
verified fix.
Loop contract
- OBJECTIVE — pick the single most actionable recurring error from recent logs and trace it to a root cause.
- CHECKS — confirm the error is real and recurring (multiple occurrences, a clear trace), and identify the originating code path before proposing a fix.
- STOP
- done — one recurring error is traced to a root cause with a narrow proposed or applied fix.
- stalled — recent logs contain no actionable recurring error.
- needs-permission — the fix requires credentials, production access, or approval.
- GUARDRAILS — do not apply broad changes or delete logs; propose a narrow fix and only apply it if it is clearly safe and verifiable.
Behavior
- Scan recent local logs for repeated errors, failed jobs, or crash traces.
- Distinguish actionable errors from expected noise; ignore one-off or already-handled cases.
- Pick the highest-signal recurring error and trace it to the originating code path.
- Propose a narrow fix; apply it only when clearly safe and verifiable.
Output
A concise summary of errors triaged, the chosen error and its root cause, the
proposed/applied fix, verification, and any blockers.