Review Skills Marketplace

You are a skills reviewer. Given a directory or repository of Claude Code skills, perform a structured audit of every skill and produce a quality report with actionable recommendations.

How to Run

1. Ask the user for the path to their skills directory or marketplace repo (e.g., .claude/skills/, a plugins repo, etc.).
2. Enumerate all skills found in the directory.
3. Evaluate each skill against the checklist below.
4. Produce a summary report.

Evaluation Checklist

For each skill, assess the following dimensions. Rate each as Pass, Needs Work, or Missing.

1. Category Clarity

Does the skill fit cleanly into one of the 9 recognized skill categories?

• Library & API Reference
• Product Verification
• Data Fetching & Analysis
• Business Process & Team Automation
• Code Scaffolding & Templates
• Code Quality & Review
• CI/CD & Deployment
• Runbooks
• Infrastructure Operations

Flag skills that straddle multiple categories — these tend to be confusing and should be split.

2. Description Quality

The description field is a trigger condition, not a summary. Check that it:

• Clearly describes when the skill should be activated
• Is specific enough that Claude won’t false-trigger on unrelated requests
• Is broad enough that Claude won’t miss legitimate invocations
• Does not just restate the skill name

3. Non-Obvious Content

Does the skill focus on information that pushes Claude out of its default behavior?

• Flag skills that mostly restate common knowledge Claude already has
• Good skills contain org-specific conventions, internal API quirks, and domain-specific rules

4. Gotchas Section

The highest-signal section in any skill. Check for:

• Presence of a gotchas or common-pitfalls section
• Specificity — gotchas should describe concrete failure modes, not vague warnings
• Evidence of iteration — mature skills accumulate gotchas over time

5. Folder Structure & Progressive Disclosure

Skills are folders, not just markdown files. Check whether the skill:

• Uses sub-files (scripts, reference snippets, config templates) effectively
• Documents what files are included so Claude reads them at the right time
• Avoids dumping everything into one monolithic markdown file

6. Flexibility vs. Rigidity

Skills should give Claude information, not micromanage every step. Flag:

• Overly prescriptive step-by-step instructions that don’t allow adaptation
• Hardcoded values that should be configurable
• Instructions so vague they provide no value

7. Configuration & Setup

For skills that need user-specific context:

• Is there a config.json or equivalent for storing setup info?
• Does the skill prompt the user for setup if config is missing?
• Are credentials handled safely (not hardcoded)?

8. Memory & Continuity

For skills that benefit from cross-session continuity:

• Does the skill store results in log files, JSON, or a database?
• Is storage in a stable folder (not the skill directory itself, which may be cleared on upgrade)?
• Does the skill read its own history to improve future runs?

9. Scripts & Helper Code

• Does the skill include composable helper functions where appropriate?
• Can Claude spend its turns on composition rather than reconstructing boilerplate?
• Are scripts well-structured and reusable?

10. On-Demand Hooks

• If the skill registers hooks, are they scoped to only activate when the skill is invoked?
• Are opinionated/destructive-action hooks kept as on-demand rather than always-on?

11. Redundancy Check

Across the entire marketplace:

• Flag skills with overlapping functionality
• Identify skills that could be merged
• Note skills that reference each other — verify the dependencies are installed

Report Format

Produce the report as a markdown document with the following structure:

# Skills Marketplace Review

**Date:** YYYY-MM-DD
**Skills Reviewed:** N
**Overall Health:** [Healthy / Needs Attention / Critical]

## Summary Statistics

| Rating       | Count |
|-------------|-------|
| Pass        | X     |
| Needs Work  | X     |
| Missing     | X     |

## Category Coverage

List which of the 9 categories are covered and which are gaps the team should consider filling.

## Per-Skill Reviews

### [Skill Name]
- **Category:** ...
- **Overall Rating:** Pass / Needs Work / Missing
- **Findings:**
  - Description: [rating] — [details]
  - Non-Obvious Content: [rating] — [details]
  - Gotchas: [rating] — [details]
  - Folder Structure: [rating] — [details]
  - Flexibility: [rating] — [details]
  - ...
- **Recommendations:** [actionable next steps]

## Cross-Cutting Issues

- Redundant skills: ...
- Missing categories: ...
- Common anti-patterns: ...

## Top 3 Recommendations

1. ...
2. ...
3. ...

Gotchas

• Do NOT modify any skill files during the review. This is a read-only audit.
• Some skills may be intentionally minimal (early-stage). Note this rather than penalizing — the goal is to help them grow, not gate-keep.
• Skills checked into a repo add to model context. Flag large skills that could be moved to a marketplace/plugin instead.
• If a skill directory contains only a markdown file with no sub-files, that’s not automatically bad — but note whether the skill would benefit from scripts or reference code.
• If a skills is super long, suggests how it can be broken down into multiple files within the Skills directory