What problem does it solve? Agent Skills collections (SKILL.md files and slash commands) degrade over time: descriptions stop triggering correctly, references to tools and docs break after renames, skills overlap in responsibility, and some become dead weight. This Skill standardizes a repeatable audit so these problems are detected with the same criteria and output format every time instead of ad-hoc manual reviews. ## Core Features & Use Cases - Mechanical reference verification: Runs tools/check_skill_references.py to validate that every path mentioned in SKILL.md, commands, and rule files actually exists in the repository, with line-level exclusion markers and a self-test mode. - Multi-perspective detection: Checks for trigger conflicts, dead skills (unreferenced and untouched for 90+ days), stale content contradicted by current lessons, bloat over 500 lines, responsibility duplication, and missing skills. - Description scoring and Issue filing: Scores each skill description out of 10 for trigger precision, then files deduplicated GitHub Issues (max 5 per run) with one root cause per Issue. - Use Case: Run a quarterly inventory of .claude/skills and .claude/commands to find two skills whose descriptions both claim the same trigger phrase, one skill referencing a deleted script, and one oversized SKILL.md, then get three ready-to-triage Issues. ## Quick Start Ask the AI to audit the Agent Skills in this repository by saying "Skill γζ£εΈγγγ¦" or invoking /skill-audit.
