skill-governance

Use this skill to control multi-skill side effects with a balanced governance model.

When to use

• After installing a new skill (must pass acceptance before becoming ready)
• Daily/heartbeat governance refresh
• Core pool tuning and automatic promotion/demotion
• Routing decisions: core first, then all ready skills, then explore fallback

Policy

• Status model: candidate -> ready -> core -> quarantine -> retired
• Third-party skills default to candidate
• Only ready or core skills are eligible for automatic selection
• Quarantine after 2 consecutive failures
• Demote core skill to ready after 3 days without usage
• Dynamic core cap range: 8 to 14

Security and privacy

• No outbound network calls are required by this skill.
• No credential collection or token upload behavior is implemented.
• Scripts only read/write local workspace files for governance state and evidence.
• Weekly cleanup is non-destructive by design (status transitions only; no file deletion).
• Any action beyond local governance must be explicitly added by the operator.

Source of truth

• skill-registry.json

Scripts

1. Reconcile runtime + filesystem into registry:
2. scripts/reconcile-ready.ps1 -Root <workspace>
3. Audit one skill with 4-step acceptance:
4. scripts/audit-skill.ps1 -Root <workspace> -SkillName <name>
5. Record skill usage outcome:
6. scripts/record-skill-usage.ps1 -Root <workspace> -SkillName <name> -Outcome success|failure|blocked
7. Evaluate core pool promotion/demotion and dynamic cap:
8. scripts/update-core-pool.ps1 -Root <workspace>
9. Route execution chain (core -> ready -> explore):
10. scripts/route-skill.ps1 -Root <workspace> -TaskText "<text>" -Candidates "a,b,c"
11. Weekly cleanup (retire stale or missing skills):
12. scripts/weekly-cleanup.ps1 -Root <workspace>

Required install flow

After installing a new skill:

1. Run reconcile-ready.ps1
2. Run audit-skill.ps1 for that skill
3. Confirm status in skill-registry.json is ready before use in automatic routing