Audit Skill

Contents

| Reference | When to load | | ------------------------------------ | --------------------------- | | ../../references/rubric.md | The scoring rubric (shared) | | ../../references/write-protocol.md | Write discipline | | templates/scorecard-skeleton.md | Output file shape |

Audit a skill

Score one skill against the 7-dimension rubric. The skill is the input; a scorecard Markdown file is the output.

How it works

1. Resolve {{ skill }} to an absolute path. Accept either a directory path (skills/research) or a bare skill name (research) — glob for the matching SKILL.md.
2. Read the rubric at ${CLAUDE_PLUGIN_ROOT}/references/rubric.md.
3. Read the skill in full: SKILL.md, every file under references/, every file under templates/. The scorecard is grounded in what's in the files, not in the description alone. Reading is the primary work.
4. If a prior /gardener audit exists at plugin-gardener/audit/, read its latest collisions.csv and clusters.csv for cross-reference context. Collisions involving this skill inform Dimension 1 (description quality) and Dimension 5 (proactive-label discipline) — check whether the description carries the right negative-keyword discriminators. For every collision row, open the sibling SKILL.md and verify by reading, not from the cosine alone.
5. Write a scorecard to plugin-gardener/audit/ad-hoc/-.md using templates/scorecard-skeleton.md.
6. Present the total, dimension scores, and top-3 recommended actions inline.

Apply the rubric dimension by dimension. Record a short rationale (1–2 sentences with a file:line citation from the files you read) for every criterion scoring under 3 points. The rationale is the value; the number is the aggregate.

On the role of embedding data: collisions and clusters from a prior /gardener run are binoculars, not the photograph. They point you at the right cross-references to open. Verify every flagged collision by reading both SKILL.md files — a high cosine is an invitation to investigate, not a verdict.

When to invoke

• After authoring a new skill — baseline the score before committing.
• After a monthly /gardener run flags a skill — re-score to confirm the issue.
• After applying fixes — did the score improve?

What this does not do

• Edit the skill.
• Open a PR.
• Change frontmatter.

The scorecard is the deliverable. Changes are the user's to apply.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: theagenticguy
• Source: theagenticguy/agentic-plugins
• License: MIT

Install and usage instructions live in the source repository linked above.