| name | error-remediation |
| description | This skill should be used whenever an error, exception, failed build, failing test, stack trace, compiler/linter error, CI failure, or panic appears, or when the user asks "why is this failing" or "how do I fix this". Covers troubleshoot_diagnose, remediation_search, remediation_record, and remediation_feedback to match errors to past fixes and record new ones. |
| version | 0.5.0 |
Error Remediation
Overview
contextd tracks error → fix patterns. When something breaks, check whether this exact failure was fixed before, diagnose it, fix it, then record the fix so the next occurrence is instant.
The flow
1. Diagnose
troubleshoot_diagnose(error)
Get AI-powered analysis of the error first — it categorizes the failure and suggests likely causes.
2. Search for a known fix
remediation_search(query)
Pass the salient part of the error message. If contextd has seen it, it returns the fix that worked before — do not re-derive it.
3. Record the fix
After resolving it:
remediation_record(error, fix, ...)
Capture:
- The error signature (the stable part of the message, not volatile paths/timestamps).
- The fix that actually worked.
- Any root cause worth noting.
4. Feedback
remediation_feedback — rate whether a suggested fix actually helped, so its confidence stays accurate.
Good vs. weak remediations
| Good | Weak |
|---|
ErrMissingTenant on Search → wrap ctx with ContextWithTenant before calling the store; fail-closed is intentional. | "It works now." |
| Stable error signature + concrete fix + cause | Logging the full transient stack trace as the signature |
When NOT to record
- One-off typos with no reusable pattern.
- Environment-specific noise unlikely to recur.
