skill-doctorlisted

# skill-doctor — Structural Audit for Claude Code Skills ## Purpose Single-pass structural audit of a Claude Code skill or workflow. Every check is a **binary structural test** (presence/absence, count, pattern match) — no semantic judgments. Same input produces same output on every run. **Grounded in:** - [Claude Code Skills docs](https://code.claude.com/docs/en/skills) — official SKILL.md spec (name, description, allowed-tools, argument-hint, disable-model-invocation) - [Equipping agents with Agent Skills](https://claude.com/blog/equipping-agents-for-the-real-world-with-agent-skills) — 3-layer progressive disclosure architecture - [Agent Patterns — loop prevention](https://www.agentpatterns.tech/en/failures/infinite-loop) - [Anthropic Engineering — Demystifying Evals](https://www.anthropic.com/engineering/demystifying-evals-for-ai-agents) - Skill-creator technical rules (`~/.claude/skills/skill-creator/references/technical-rules.md`) ## Execution Flow ``` $ARGUMENTS -> <skill-directory-path> | v [Phase 1] INTAKE — Locate skill, read SKILL.md, inventory files | v [Phase 2] CHECK — Run 30+ binary checks across 7 categories | v [Phase 3] REPORT — Severity-weighted report + JSON ``` ## Phase 1 — INTAKE Print: `[Phase 1/3] INTAKE` **1a. Locate the skill:** ``` path = $ARGUMENTS ``` If path doesn't end with `/SKILL.md`, look for `SKILL.md` inside the directory. If not found, list available skills in `~/.claude/skills/` an