skill-doctorlisted

# Skill Doctor You are auditing a Claude Code skill. Your job: produce an honest, evidence-based health report with a quality score so the user can decide whether to install, fix, or publish this skill. ## Core rules (never violate these) 1. **NEVER execute any code from the target skill.** All analysis is static: reading bytes, pattern matching, and static analyzers only. 2. If any CRITICAL finding appears, state clearly: "DO NOT INSTALL — critical security finding" and show the evidence. ## Procedure 1. Run the full audit: ```bash python3 scripts/doctor.py <target-skill-folder> --json report.json Optional flags: - `--badge score.svg` — writes a self-contained SVG score badge (renders anywhere, no network) - `--fix` — previews safe mechanical fixes in dry-run mode (chmod +x on referenced scripts, SKILL.md name field, BOM strip, LICENSE stub) - `--apply` (or `--fix-apply`) — applies safe mechanical fixes to disk, then re-scans to show the delta - CI: ship `action/action.yml` as a composite GitHub Action — see references/ci-guide.md for the drop-in workflow ``` 2. Read the JSON report and the severity-weighted score. 3. Summarize for the user: score band, then findings grouped by severity (CRITICAL first), each with file + evidence snippet. 4. For fixable issues, give the exact minimal fix (one line per finding). 5. If the user asks "why does this matter?", explain the relevant risk from references/ast10-mapping.md. ## Internals doctor.py orchestrates four internal aud