Make targeted, surgical fixes to an existing skill based on observed failure modes — tightening vague language, fixing routing misses, removing bloat, or covering missed edge cases. Use this when a skill produces inconsistent output, triggers incorrectly, or real usage revealed problems not covered. Do not use for complete rewrites (use skill-authoring), trigger-only fixes (use skill-trigger-optimization), adapting to a different context (use skill-adaptation), or when the skill is working correctly.
Iteratively improves an existing skill based on observed failure modes, user feedback, or eval results. Focuses on surgical fixes—tightening vague language, fixing routing misses, removing bloat, clarifying edge cases—not rewrites.
Use when:
Do NOT use when:
skill-authoring)skill-description-optimizer)skill-adaptation)## Refinement Summary
**Skill**: [name]
**Failure mode**: [category]
**Evidence**: [what went wrong]
### Change
**Section**: [which section]
**Before**: [original]
**After**: [new]
**Rationale**: [why this fixes it]
### Verification
- [ ] Existing evals still pass
- [ ] Known good cases still work
- [ ] Failure case now handled
skill-anti-patterns checklist for structural issuesskill-authoring for rewrite