Skill Health

Audit and evaluate installed agent skills for quality, freshness, and structural integrity. This is a meta-skill: it treats other skills as its subject matter.

Boundary Rules

These constraints apply to the entire skill. Read them before any execution steps.

• Do not modify any skill file during audit. This skill is read-only.
• Do not execute any skill during audit. This is static analysis, not dynamic testing.
• If a skill directory contains no SKILL.md and only a .gitkeep, report it as V-S4 (orphaned stub) with anchor 0 for all perspectives.
• If a directory under the skills root contains files but no SKILL.md (and is not just a .gitkeep), flag it as V-S5 (unrecognized layout) with anchor 25 and route to manual review.
• If the skills directory is empty or inaccessible, report anchor 0 overall and stop.
• If auto-routing would create a cycle (e.g., bs-skill-health -> bs-skill-bootstrap -> bs-skill-health), break the cycle by routing to manual review instead.
• Self-audit bias guard: when the audited set includes bs-skill-health itself, tag every finding on it SELF-AUDIT, cap its overall anchor at 75, and add to the report preamble: "bs-skill-health cannot fully audit itself — an auditor shares its own blind spots. Independent review recommended." Never report a clean 100 for yourself.
• Save the report to docs/reviews/bs-skill-health/<YYYY-MM-DD>-health-report.md. If the directory does not exist, create it.

Red Flags — Audit Rationalizations