failure-memory
par leegitw
Notes & PKMRésumé
Stop making the same mistakes — turn failures into patterns that prevent recurrence.
Description
Ce que fait ce skill
Detects failures during agent sessions, records them as observations with recurrence, confirmation, and disconfirmation counters, and promotes patterns to constraints when evidence thresholds are met. All data stays in a local `.learnings/` directory with no external API calls.
Builds constraints from observed evidence rather than manually written rules, so patterns are only promoted when recurrence and confirmation thresholds are actually met.
Quand l'utiliser
- Tracking repeated test failures across a project
- Capturing API timeout patterns during active development
- Searching past observations when a familiar error reappears
- Recording user corrections to agent behavior for pattern analysis
- Promoting recurring lint errors to agent-level constraints
Exemple de workflow
User asks: The payment API keeps timing out, is this a known issue?
- 1
Run /fm detect api with context 'Payment API 504 timeout'
- 2
Check .learnings/observations/ for a matching observation
- 3
Increment R counter on existing observation (payment-api-timeout)
- 4
Run /fm classify to determine evidence tier based on R count
- 5
Notify user the pattern is eligible for constraint generation (R≥3, C≥2)
Observation marked eligible; user prompted to run /ce generate to create a constraint from the pattern
Installation
clawhub install leegitw/failure-memoryExécute la commande CLI, ou colle l'instruction dans Claude Code, Cursor, Codex ou l'agent de code que tu utilises.