Use when adding new error messages to React, or seeing "unknown error code" warnings.
npx skills add tomwangowa/agent-skills --skill "code-reviewer"
Install specific skill from multi-skill repository
# Description
Intelligent code review dispatcher - automatically selects best reviewer based on context and preferences
# SKILL.md
name: code-reviewer
description: Intelligent code review dispatcher - automatically selects best reviewer based on context and preferences
Code Reviewer (Meta-Skill)
Purpose
Automatically routes code review requests to the most appropriate reviewer based on context and user preferences. This meta-skill provides a unified interface for accessing multiple code review implementations.
Available Reviewers
- code-review-claude - Fast, native code review (< 30 seconds)
- Best for: Quick validation checks during development
-
Strength: Speed and immediate feedback
-
code-review-gemini - Deep, external perspective review
- Best for: Thorough reviews with alternative AI viewpoint
- Strength: Different perspective, catches different issues
Trigger Phrases
When the user mentions any of these, activate this skill:
- "review"
- "code review"
- "review my code"
- "review the code"
- "check the code"
- "review changes"
- "review staged files"
Decision Logic
Step 1: Check User Preferences
Read from ~/.claude/CLAUDE.md section 9 (Notice) β Skill Preferences β Code Review:
- Default reviewer preference
- Fallback options
- Any context-specific overrides
Step 2: Analyze Context
Examine the current situation:
| Factor | How to Check | Impact |
|---|---|---|
| File count | git diff --cached --name-only \| wc -l |
Many files β prefer thorough review |
| Line count | git diff --cached --shortstat |
Few lines β quick review acceptable |
| User keywords | Check for "quick", "fast", "thorough", "checklist" | Override preferences |
| Complexity | File types, languages involved | Complex β prefer thorough review |
Step 3: Apply Routing Rules
Execute routing logic in priority order:
| Priority | Condition | Action | Reason |
|---|---|---|---|
| 1 | User says "quick" or "fast" | β code-review-claude |
Speed required |
| 2 | User says "thorough" or "deep" or "gemini" | β code-review-gemini |
Depth required |
| 3 | CLAUDE.md preference set | β Use preferred reviewer | Respect user default |
| 4 | < 50 lines changed AND < 3 files | β code-review-claude |
Small change, quick check |
| 5 | Default (no preference set) | β code-review-gemini |
Best balance |
Step 4: Execute and Report
- Inform user: "Using {skill_name} for this review..."
- Invoke skill: Use the Skill tool to call the selected reviewer
- Pass through results: Return the review output to the user
Implementation
When this skill is triggered:
1. Parse user request for keywords (quick, thorough, checklist)
2. Check for staged changes: git diff --cached --shortstat
3. Read CLAUDE.md preferences (if they exist)
4. Apply routing rules (priority order above)
5. Report decision: "Using code-review-gemini for this review..."
6. Execute: Skill tool with selected reviewer
7. Return results to user
Examples
Example 1: Default Review (with preference)
User: "Review my code"
Assistant: [Checks CLAUDE.md, sees gemini is preferred]
Assistant: "Using code-review-gemini for this review..."
Assistant: [Calls Skill tool with skill="code-review-gemini"]
Example 2: Quick Review Override
User: "Quick review of my changes"
Assistant: [Detects "quick" keyword]
Assistant: "Using code-review-claude for fast feedback..."
Assistant: [Calls Skill tool with skill="code-review-claude"]
Example 3: Small Change Auto-Detection
User: "Review this"
Assistant: [Checks: 15 lines changed in 1 file, no preference set]
Assistant: "Small change detected. Using code-review-claude for quick check..."
Assistant: [Calls Skill tool with skill="code-review-claude"]
Example 4: Large Change Auto-Detection
User: "Review changes"
Assistant: [Checks: 250 lines changed across 8 files]
Assistant: "Substantial changes detected. Using code-review-gemini for thorough review..."
Assistant: [Calls Skill tool with skill="code-review-gemini"]
Configuration
Users can set preferences in ~/.claude/CLAUDE.md:
## 9. Notice
### Skill Preferences
#### Code Review
- **Default reviewer**: code-review-gemini
- **Quick review fallback**: code-review-claude
- **Rationale**: Gemini provides external perspective, Claude for speed
Direct Access
Users can still bypass this meta-skill and call specific reviewers directly:
/skill code-review-gemini # Force gemini
/skill code-review-claude # Force claude
/skill code-review-checklist # Force checklist
This preserves user control while providing intelligent defaults.
Design Principles
- Single Entry Point - Users remember one command: "review my code"
- Transparent Decision - Always tell user which reviewer is being used
- Respect Preferences - Honor user's CLAUDE.md settings
- Context-Aware - Consider file count, line count, complexity
- Override-Friendly - Keywords (quick, thorough) override defaults
- Direct Access Preserved - Users can still call specific skills
Implementation Notes
- This meta-skill does NOT contain review logic
- It only performs routing and delegation
- All actual review work is done by specialized skills
- Decision logic is simple and rule-based (no complex AI needed)
- Routing rules are listed in priority order for clarity
Related Documentation
Pattern: Meta-Skill (Dispatcher)
Version: 1.0
Created: 2026-01-22
# Supported AI Coding Agents
This skill is compatible with the SKILL.md standard and works with all major AI coding agents:
Learn more about the SKILL.md standard and how to use these skills with your preferred AI coding agent.