skill-doctorlisted

# Skill-Doctor Diagnose a skill library. Find skills whose descriptions **compete for the same prompts**, then fix the routing. ## When to run - You installed many skills and "the wrong one keeps firing" - Before publishing a skill pack (catch collisions reviewers will hit) - After adding a skill (does it collide with an existing one?) - Periodic health check on a growing library ## Input Point the engine at EITHER: - A **directory** of skills (`<root>/<name>/SKILL.md`), or - A Claude Code **manifest.json** (`skills-plugin/.../manifest.json`) ## How to run The engine lives in `scripts/` next to this file. Zero dependencies (Python 3.8+ stdlib). ```bash # From the directory containing this SKILL.md: python3 -m scripts <path-to-skills-or-manifest> # Markdown report python3 -m scripts <path> --format json --out report.json # machine output python3 -m scripts <path> --fail-on critical # CI gate (exit 1) ``` Common options: - `--merge-at 0.55` — collision score that triggers a MERGE recommendation - `--disambig-at 0.30` — score that triggers a DISAMBIGUATE recommendation - `--vague-threshold 0.5` — flag descriptions vaguer than this - `--top 12` — how many top collisions to list - `--format md|json|both` ## What it detects (issue codes) | Code | Severity | Meaning | |------|----------|---------| | `DUPLICATE_TRIGGER` | 🔴 | Same trigger phrase claimed by 2+ skills (top cause of mis-routing) | | `DUPLICATE_NAME` | 🔴 | Two skills share a name | |