Reviewing Skills

Overview

Performs a structured, scored audit of SKILL.md files against Anthropic's official Claude Code skill authoring best practices. Produces a scored report across 5 quality dimensions and optionally generates a corrected version.

Follow these steps exactly.

When not to use

Does not review CLAUDE.md, commands, hooks, or general markdown.

Table of Contents

• Step 1: Resolve Target SKILL.md
• Step 2: Read and Measure
• Step 2b: Determine Guidelines Source
• Step 2c: Regression Risk Check (optional)
• Steps 3–6: Audit, Score, Report, and Optionally Fix
• Quick Reference Checklist

Step 1: Resolve Target SKILL.md

Determine which SKILL.md file to audit:

1. Explicit path in message — If the user provided a file path, use it directly.
2. Conversation context — If a skill has been recently discussed or created, use that one.
3. Ask if still unclear — "Which SKILL.md would you like me to review? Please provide the path."

Do NOT use $ARGUMENTS or $PWD — these variables are only available in commands, not skills.

Step 2: Read and Measure

Read the target file and record:

Note the presence or absence of:

• YAML frontmatter delimiters (---)
• name: field
• description: field
• "Use when..." trigger phrase in description
• Table of contents (for files ≥100 lines)
• SKILL.md filename casing (must be exactly SKILL.md)
• Kebab-case folder naming

Step 2b: Determine Guidelines Source

Default: Use references/best-practices.md (static, always available).

Fetch live from platform docs when user says:

• "use latest", "check official docs", "fetch from platform", "use current guidelines", "pull from the website"

Live fetch URL:

https://platform.claude.com/docs/en/agents-and-tools/agent-skills/best-practices

If live fetch fails: fall back to references/best-practices.md silently and note the failure in the audit output under a "Guidelines Source" line.

Step 2c: Regression Risk Check (optional)

Compare the current SKILL.md against its last committed version to detect breaking changes and regressions.

Skip entirely if any of the following are true:

• Not inside a git repository (git rev-parse --git-dir returns non-zero)
• File has never been committed (git log --oneline -- <path> returns no output)
• User says "skip regression check", "no comparison", or "first review"

If not skipped, run in order:

1. Resolve the git-relative path:

   git ls-files --full-name <path-to-SKILL.md>
   

2. Retrieve the last committed version:

   git show HEAD:<git-relative-path>
   

   If this fails (file renamed, untracked, or path error): skip and note "No previous version found — file may have been renamed" in report. Do NOT attempt git log --follow; surface the limitation and move on.

What to compare: See references/audit-steps.md — Regression Risk section.

Output: A Regression Risk section in the audit report, appended after the Scores table and before the Grade line. Does not affect any dimension score.

Steps 3–6: Audit, Score, Report, and Optionally Fix

Continue with references/audit-steps.md for:

• Step 3: Score 5 dimensions (2 pts each, max 10)
• Step 4: Output scored report with grade
• Step 5: Offer optimized version with targeted fixes
• Step 6: Write-to-disk confirmation (requires explicit second confirmation)

Quick Reference Checklist

Use this for rapid pre-audit assessment:

Frontmatter

• YAML delimiters (---) present and matching
• name: field present, ≤64 chars, lowercase + numbers + hyphens only
• name: does not contain anthropic or claude as substring
• description: field present, ≤1024 chars
• Description written in third person
• "Use when..." trigger phrase present in description
• Description contains a capability statement (not only the trigger phrase)

Body

• Total lines <500
• Word count <5,000 (per Anthropic guide)
• No Windows-style paths (\)
• No hardcoded absolute paths
• No time-sensitive platform-state content ("as of 2024", "currently", "recently added")
• No first/second person in frontmatter description
• Consistent terminology throughout
• Complex workflows use a checklist pattern (copy-and-check-off steps)
• No "options without a default" pattern (pick one, mention alternatives)
• No assumed tool/package availability without explicit install instructions

Structure

• File named exactly SKILL.md (case-sensitive)
• Skill folder uses kebab-case naming
• Table of contents present in reference files (if ≥100 lines); optional suggestion for SKILL.md
• Reference files at depth ≤1 (no references/sub/file.md)
• Freedom level stated or implied
• Content distributed across three levels (frontmatter → body → references) where appropriate
• Feedback loops present for quality-critical or iterative tasks

Design Pattern

• Identifiable pattern (Sequential, Orchestrator, Iterative, or Adaptive)
• Body structure matches the chosen pattern

Scripts (apply if skill contains a scripts/ folder or has bash/python code blocks with external tool invocations)

• MCP tools referenced with fully qualified ServerName:tool_name format
• No unexplained magic numbers (all constants documented)
• Error handling explicit — scripts handle failures rather than punting to Claude
• Required packages listed with install instructions

Discoverability

• "Use when..." trigger phrase clear and specific
• ≥3 searchable keywords in description
• Scope defined (what this skill does NOT cover)

Regression Risk (skip if not in git, file is new, or user opts out)

• name: field unchanged (BREAKING if changed)
• description: trigger phrases preserved (BREAKING if any removed)
• description: activation intent preserved (WARNING if Use when... clause or ≥3 domain keywords absent vs. previous)
• Reference files not removed (WARNING if any references/ path disappeared)
• Line count not reduced >30% (WARNING if significant shrinkage)
• No new anti-patterns introduced vs. previous version (INFO)