Skill Reviewer
Reviews skills against official best practices from Anthropic's skill authoring guide.
Review Workflow
1. Create TodoWrite Checklist
Create systematic checklist from official quality checklist (see REFERENCE.md):
TodoWrite([
// Core Quality (10 items from official checklist)
{content: "Description specific with key terms", status: "pending", activeForm: "Checking description specificity"},
{content: "Description includes what + when", status: "pending", activeForm: "Verifying what and when clauses"},
{content: "SKILL.md body under 500 lines", status: "pending", activeForm: "Counting SKILL.md lines"},
{content: "Additional details in separate files", status: "pending", activeForm: "Checking file organization"},
{content: "No time-sensitive information", status: "pending", activeForm: "Scanning for dates/versions"},
{content: "Consistent terminology throughout", status: "pending", activeForm: "Verifying terminology"},
{content: "Examples concrete not abstract", status: "pending", activeForm: "Checking example quality"},
{content: "File references one level deep", status: "pending", activeForm: "Checking reference depth"},
{content: "Progressive disclosure used appropriately", status: "pending", activeForm: "Verifying progressive disclosure"},
{content: "Workflows have clear steps", status: "pending", activeForm: "Checking workflow clarity"},
// Code and Scripts (8 items if scripts present)
{content: "Scripts solve problems not punt to Claude", status: "pending", activeForm: "Checking script purpose"},
{content: "Error handling explicit and helpful", status: "pending", activeForm: "Reviewing error handling"},
{content: "No voodoo constants (all justified)", status: "pending", activeForm: "Checking for magic numbers"},
{content: "Required packages listed and verified", status: "pending", activeForm: "Verifying dependencies"},
{content: "Scripts have clear documentation", status: "pending", activeForm: "Checking script docs"},
{content: "Forward slashes only in paths", status: "pending", activeForm: "Checking path format"},
{content: "Validation steps for critical operations", status: "pending", activeForm: "Checking validation"},
{content: "Feedback loops for quality tasks", status: "pending", activeForm: "Checking feedback loops"},
// Testing (4 items)
{content: "At least three evaluation scenarios", status: "pending", activeForm: "Checking evaluations"},
{content: "Tested with Haiku Sonnet Opus", status: "pending", activeForm: "Checking model testing"},
{content: "Real usage scenarios tested", status: "pending", activeForm: "Checking real-world tests"},
{content: "Team feedback incorporated", status: "pending", activeForm: "Checking feedback integration"}
])
Mark items as they are verified.
2. Validate YAML Frontmatter
Verify name, description, and allowed-tools fields meet requirements. See REFERENCE.md for detailed specifications.
3. Count SKILL.md Lines
Run wc -l SKILL.md to count lines. Target: Under 500 lines. See REFERENCE.md for remediation if over limit.
4. Evaluate Description Quality
Verify description includes specific capabilities, file types, when clause, and uses third person voice. See REFERENCE.md for good/bad examples.
5. Check Progressive Disclosure
Verify file references are one level deep from SKILL.md. See REFERENCE.md for structure patterns.
6. Verify Content Quality
Check for concise writing, consistent terminology, concrete examples, and no time-sensitive information. See REFERENCE.md for detailed standards.
7. Check Workflows
Verify workflows have clear sequential steps with checklists and feedback loops. See REFERENCE.md for examples.
8. Review Scripts (if present)
Verify scripts solve problems (don't punt), have no magic numbers, clear execution intent, and documented dependencies. See REFERENCE.md for detailed standards.
9. Verify File Paths
Check all paths use forward slashes (not Windows backslashes). See REFERENCE.md.
10. Check Naming Conventions
Verify skill name uses gerund form (verb + -ing) when possible. See REFERENCE.md for conventions.
11. Assess Degree of Freedom
Evaluate if specificity matches task fragility. See REFERENCE.md.
12. Test Discovery
Based on description, suggest 3-5 test queries:
Example: If description mentions "PDF", "extract", "forms":
- "Extract text from this PDF"
- "Fill out this PDF form"
- "Merge these PDF documents"
Skill should activate for these queries.
Provide Structured Feedback
## Skill Review: [skill-name]
### Line Count
SKILL.md: XXX lines (target: <500)
### ✓ Passes Official Checklist
- [Items that meet official requirements]
### ✗ Issues Found
**Critical**:
- SKILL.md body: 650 lines (must be under 500)
- Description missing when clause
**Major**:
- References nested too deep (SKILL.md → advanced.md → details.md)
- Windows-style paths in scripts section
- Time-sensitive information: "Before 2025, use v1 API"
**Minor**:
- Name not in gerund form: `commit-message` → `generating-commit-messages`
- Description uses second person: "you" → third person
### 🔧 Recommended Fixes
1. **Reduce SKILL.md length** (650 → <500 lines):
- Move examples to `examples.md`
- Move API reference to `references/api.md`
- Keep only core workflow in SKILL.md
2. **Fix description**:
- Current: "Helps create commit messages"
- Should be: "Generates descriptive commit messages by analyzing git diffs. Use when writing commit messages or reviewing staged changes."
3. **Flatten file references**:
- Move content from `advanced.md → details.md` directly into `details.md`
- Reference from SKILL.md: `See [details.md](details.md)`
### Test Queries
Based on description, test activation with:
- "[query using description keywords]"
- "[another test query]"
### Checklist: X/22 items passed
[Show TodoWrite status]
### Official References
- Best Practices: https://docs.claude.com/en/docs/agents-and-tools/agent-skills/best-practices
- Complete checklist: [REFERENCE.md](REFERENCE.md#official-quality-checklist)
Common Issues and Anti-Patterns
See REFERENCE.md for common issues and REFERENCE.md for anti-patterns to avoid.
MCP Tool References
If skill uses MCP tools, verify fully qualified names: ServerName:tool_name. See REFERENCE.md.
Version History
- v2.0.0 (2025-11-15): Renamed from reviewing-skills to skill-reviewer, created REFERENCE.md with detailed specifications, removed duplications from SKILL.md
- v1.0.0: Initial version