Claude Plugin Audit

Validates plugin structure, components, and quality against best practices.

Steps

1. Load the outfitter:claude-plugins skill for plugin structure knowledge
2. Analyze plugin at target path (default: current directory)
3. Check each component type against standards
4. Generate findings with severity and fix recommendations

Audit Scope

Severity Levels

Output Format

# Plugin Audit: {PLUGIN_NAME}

**Path**: {PATH}
**Status**: {PASS|WARNINGS|FAIL}
**Issues**: {CRITICAL} critical, {WARNINGS} warnings, {INFO} info

## Critical Issues

- `◆◆` {component}: {issue}
  - **Fix**: {specific remediation}

## Warnings

- `◆` {component}: {issue}
  - **Fix**: {specific remediation}

## Suggestions

- `◇` {component}: {suggestion}

## Summary

{1-2 sentence overall assessment}

Checks by Component

plugin.json

• File exists at .claude-plugin/plugin.json
• Valid JSON syntax
• name present and valid (lowercase, hyphens, 2-64 chars)
• version present and semver format
• description present and meaningful
• No unknown top-level fields

Commands

• Frontmatter has description
• Description is action-oriented
• argument-hint uses <required> / [optional] syntax
• No broken file references (@path)
• Bash commands in backticks are valid

Agents

• name matches filename (without .md)
• description has trigger conditions and examples
• tools field uses correct syntax (comma-separated)
• model is valid if specified

Skills

• SKILL.md exists in skill directory
• Frontmatter has name and description
• Name matches directory name
• Description includes trigger keywords
• Under 500 lines (progressive disclosure)
• Referenced files exist

Hooks

• Valid hook types (PreToolUse, PostToolUse, etc.)
• Matchers use valid glob/tool patterns
• Scripts have execute permissions
• Timeouts are reasonable (< 30s default)

Auto-Fixable Issues

These can be fixed automatically:

Flag auto-fixable issues in output:

- `◆` commands/deploy.md: Missing description [auto-fixable]
  - **Fix**: Add `description: "Deploy to environment"`

Rules

Always:

• Check every component type present
• Provide specific file paths in findings
• Include concrete fix instructions
• Flag auto-fixable issues

Never:

• Modify files (audit only)
• Skip components due to quantity
• Give vague recommendations