/review-prompt

Prompt Review Command

Review Claude Code prompts for quality, structure, and best practices compliance.

Input

Target: $ARGUMENTS

Review Modes

• single (default): Detailed review of a single file
• batch: Summary + individual reports for multiple files
• consistency: Check naming, structure, and style consistency across files

Process

Step 1: Determine Target and Mode

Parse $ARGUMENTS to identify:

• Target type: single file, directory, or glob pattern
• Review mode: single, batch, or consistency (default: single for one file, batch for multiple)

Step 2: Detect Category

For each file/directory, detect category from content:

• Plugin: Target is a directory containing .claude-plugin/plugin.json
• Skill: description starts with "This skill should be used when" or "Use when"; may have version:, context:, effort: in frontmatter
• Agent: Has model:, color:, or tools: in frontmatter, description starts with "Use this agent when", contains <example> blocks; may have memory:, background:, isolation: fields
• Command: Has description: and optionally argument-hint: or allowed-tools:; no <example> blocks

Step 3: Apply Review Criteria

Skill Criteria

• Frontmatter has name (recommended, kebab-case, max 64 chars) and description (recommended)
• description starts with "This skill should be used when the user asks to..." or "Use when..."
• Specific trigger phrases in quotes (3-6 phrases)
• Body uses imperative form (not "You should...")
• Best Practices section uses ✅/❌ emoji format
• SKILL.md under 500 lines (detailed info in separate files)
• Additional Resources section with subsections if references exist
• disable-model-invocation: true set for side-effect workflows
• context: fork skills have explicit instructions (not just guidelines)

Agent Criteria

• Frontmatter has name (required) and description (required)
• description starts with "Use this agent when..."
• 2-4 <example> blocks present
• Each example has Context, user, assistant, commentary
• Body starts with expert persona
• Has Core Responsibilities, Process, Quality Standards, Output Format
• "What NOT to Focus On" section present (recommended)
• Edge Cases considered
• Model selection appropriate (default: inherit)
• color semantically matches purpose (recommended, universally used)
• Plugin agents do NOT use hooks, mcpServers, or permissionMode

Command Criteria

• description is clear, action-oriented
• argument-hint shows usage (if args expected)
• allowed-tools restricts to necessary tools only
• Workflow logically ordered
• $ARGUMENTS handling clear (with conditional for empty case)
• User confirmation at major decision boundaries
• Error handling for external dependencies
• Validation steps included

Plugin Criteria (directory review)

When target is a plugin directory, review the overall plugin structure:

Structure (Critical)

• .claude-plugin/plugin.json exists
• plugin.json has valid name (kebab-case, no spaces)
• Components (commands/, agents/, skills/) are at plugin root, not inside .claude-plugin/
• .claude-plugin/ contains only plugin.json

Configuration (Major)

• version follows semantic versioning
• description clearly describes plugin purpose
• All paths in plugin.json are relative (./ prefix)
• Scripts referenced by hooks use ${CLAUDE_PLUGIN_ROOT}
• Hook scripts have executable permissions

Security (Critical)

• Plugin agents do NOT use hooks, mcpServers, or permissionMode fields
• No secrets or credentials in any file

Quality (Minor)

• LICENSE file exists
• CHANGELOG.md exists
• README.md exists with installation instructions
• claude plugin validate would pass (check structure)

Components (Major)

• Each skill/agent/command inside the plugin also passes its respective criteria
• Naming is consistent across components (plugin namespace)

Step 4: Generate Report

Single Mode Output

# Prompt Review Result

## Summary
- **File**: [path]
- **Category**: [Skill/Agent/Command/Plugin]
- **Rating**: [A/B/C/D]
- **Critical Issues**: [count]

## Strengths
1. [strength 1]
2. [strength 2]

## Issues

### Critical (Must Fix)
1. **[Issue Title]**
   - Current: [current state]
   - Problem: [why it's a problem]
   - Fix: [specific fix]

### Minor (Recommended)
1. **[Issue Title]**
   - Current: [current state]
   - Suggestion: [improvement]

## Corrected Sample
[Show corrected version of problematic sections]

## Checklist Results
| Item | Status | Comment |
|------|--------|---------|
| ... | OK/NG | ... |

Batch Mode Output

# Batch Review Results

## Summary

| File | Category | Rating | Critical | Minor |
|------|----------|--------|----------|-------|
| ... | ... | ... | ... | ... |

## Statistics
- **Total Files**: [count]
- **Average Rating**: [A/B/C/D]
- **Total Critical Issues**: [count]

## Common Patterns
1. **[Pattern]**: [description] - Files: [list]

---

## Individual Reports
[Single mode format for each file]

Consistency Mode Output

# Consistency Review Results

## Target Directory
`[path]`

## Detected Categories
- Plugins: [count] directories
- Skills: [count] files
- Agents: [count] files
- Commands: [count] files

## Consistency Check

### Naming Conventions
| Item | Status | Details |
|------|--------|---------|
| File naming | OK/NG | [details] |
| Frontmatter name | OK/NG | [details] |

### Structure Consistency
| Item | Status | Details |
|------|--------|---------|
| Section structure | OK/NG | [details] |
| Example format | OK/NG | [details] |

### Style Consistency
| Item | Status | Details |
|------|--------|---------|
| description format | OK/NG | [details] |
| Body person | OK/NG | [details] |

## Recommended Actions
1. [action 1]
2. [action 2]

Rating Criteria

Rating	Description
A	Meets all criteria, follows best practices
B	Minor issues only, no functional impact
C	Some important issues, but basically functional
D	Critical issues, needs major revision