$npx -y skills add tody-agent/codymaster --skill cm-skill-evolutionRepair or extend CodyMaster skills with a three-mode loop: FIX, DERIVED, and CAPTURED, grounded in current repo tooling.
| 1 | # cm-skill-evolution |
| 2 | |
| 3 | Use this skill after `cm-skill-health` identifies a degraded or broken skill, or when `cm advisory handoff --for cm-skill-evolution` produces a structured recovery note. |
| 4 | |
| 5 | ## Modes |
| 6 | |
| 7 | ### FIX |
| 8 | Use when the skill should exist already but is inaccurate, broken, or partially missing. |
| 9 | |
| 10 | Checklist: |
| 11 | - repair broken references |
| 12 | - restore missing support files |
| 13 | - align docs and profiles |
| 14 | - re-run skill validation and test gate |
| 15 | |
| 16 | ### DERIVED |
| 17 | Use when the original promise was too ambitious, but the repo has enough primitives to ship a truthful MVP. |
| 18 | |
| 19 | Checklist: |
| 20 | - keep the same user problem |
| 21 | - reduce claims to what the code can support today |
| 22 | - reuse existing repo building blocks instead of inventing a new subsystem |
| 23 | |
| 24 | ### CAPTURED |
| 25 | Use when the main value is operational learning rather than a new code path. |
| 26 | |
| 27 | Checklist: |
| 28 | - append the lesson with `cm retro --project . --tool skill --note "..."` |
| 29 | - record durable context in `.cm/CONTINUITY.md` |
| 30 | - update the relevant skill so future sessions do not repeat the same failure |
| 31 | |
| 32 | ## Evolution Loop |
| 33 | |
| 34 | 1. Start from the health note. |
| 35 | - Preferred source: `cm advisory handoff --for cm-skill-evolution` |
| 36 | 2. Pick one mode only. |
| 37 | 3. Define the smallest truthful recovery. |
| 38 | 4. Patch the skill and its discovery surfaces. |
| 39 | 5. Verify: |
| 40 | - `npm run validate:skills` |
| 41 | - `npm run check:skills` |
| 42 | - repo test gate if code or docs wiring changed materially |
| 43 | 6. Capture the lesson in retro and continuity. |
| 44 | |
| 45 | ## Decision guide |
| 46 | |
| 47 | - The feature existed and drifted: `FIX` |
| 48 | - The changelog promised more than the repo ever shipped: `DERIVED` |
| 49 | - The issue is mainly process and should inform future work: `CAPTURED` |
| 50 | |
| 51 | ## Output |
| 52 | |
| 53 | ```md |
| 54 | ## Skill Evolution |
| 55 | - Skill: cm-... |
| 56 | - Mode: FIX | DERIVED | CAPTURED |
| 57 | - Change: ... |
| 58 | - Verification: ... |
| 59 | - Learning captured: yes | no |
| 60 | ``` |
| 61 | |
| 62 | Preferred advisory input: |
| 63 | |
| 64 | ```md |
| 65 | ## Advisory Handoff |
| 66 | - Consumer: cm-skill-evolution |
| 67 | - Skill: cm-... |
| 68 | - Recovery path: FIX | DERIVED | CAPTURED | NONE |
| 69 | - Confidence: 0.xx |
| 70 | - Source analysis: EA-... |
| 71 | - Task: ... |
| 72 | - Status: completed | partial | failed |
| 73 | - Evidence: ... |
| 74 | - Selected skills: ... |
| 75 | - Target skills: ... |
| 76 | - Quality weight: 0.xx |
| 77 | - Next step: ... |
| 78 | ``` |