name: cm-skill-health description: "Operational health review for CodyMaster skills using current repo signals: validate-skills, suggest, retro logs, memory, and test gates."
cm-skill-health
Use this skill when a CodyMaster skill feels stale, misleading, unreliable, or under-documented.
What it checks
- Discovery drift
- Is the skill present in
skills/, docs indexes, profiles, and README surfaces?
- Is the skill present in
- Invocation friction
- Does
cm suggestpoint users to the skill when the task matches?
- Does
- Operational evidence
- Are there recurring failures or learnings in
.cm/operational-learnings.jsonl?
- Are there recurring failures or learnings in
- Contract health
- Does the skill reference commands, files, or paths that still exist?
- Release safety
- Does the repo still pass
npm run validate:skills,npm run check:skills, and the test gate?
- Does the repo still pass
Workflow
- Confirm the symptom.
- Missing from docs
- Missing from profiles
- Broken references inside
SKILL.md - Repeated runtime pain in retro notes
- Compare the live skill against:
docs/skills/index.mdskills/profiles/full.txtREADME.md- related changelog promises
- Scan evidence sources.
cm advisory handoff --for cm-skill-healthcm suggest --project .cm retro summary --project ..cm/CONTINUITY.mdrgoverskills/,docs/, andsrc/
- Score the issue.
healthy: discoverable, accurate, references validdegraded: present but misleading or inconsistentbroken: missing, invalid, or unusable
- Hand off to:
cm-skill-evolutionto repair or derive the next version
Output
Produce a short health note:
## Skill Health
- Skill: cm-...
- Status: healthy | degraded | broken
- Symptoms: ...
- Evidence: ...
- Recovery path: FIX | DERIVED | CAPTURED
Preferred input contract:
## Advisory Handoff
- Consumer: cm-skill-health
- Skill: cm-...
- Recovery path: FIX | DERIVED | CAPTURED | NONE
- Confidence: 0.xx
- Source analysis: EA-...
- Task: ...
- Status: completed | partial | failed
- Evidence: ...
- Selected skills: ...
- Target skills: ...
- Quality weight: 0.xx
- Next step: ...
Red flags
- Do not claim metric dashboards or automatic scoring unless the repo actually implements them.
- Do not treat README marketing copy as proof that a skill exists.
- Do not evolve the skill before identifying whether the problem is docs drift, packaging drift, or missing implementation.