PRD Example 2: Code Review Skill

Difficulty: Intermediate | Project Type: Custom Skill | Time: 4-6 hours

---

Code Review Skill

Overview

What	An AI skill that reviews code changes and provides structured feedback
Who	Solo developers who want quick feedback before committing
Why	Catch bugs, style issues, and security problems before they reach production

Features

1. Review Staged Changes: /review → Analyzes git diff --staged and provides feedback
2. Structured Output: Returns findings in categories: Bugs, Security, Style, Suggestions
3. Severity Levels: Each finding marked as Critical, Warning, or Info
4. File-Specific Review: /review src/auth.py → Reviews only specified file

Non-Goals

Will NOT build:

• Auto-fix capabilities (suggestions only)
• Integration with GitHub/GitLab PR systems
• Custom rule configuration
• Historical review tracking
• Team collaboration features

Will NOT use:

• External linting APIs
• Database storage
• Web interfaces
• Third-party code analysis tools

Technical Constraints

Skill Format	.claude/skills/code-review/SKILL.md (or .gemini/skills/)
References	references/review-checklist.md, references/security-patterns.md
Dependencies	Git (assumes repo context)
Testing	Manual testing with sample diffs

Skill Structure:

.claude/skills/code-review/
├── SKILL.md              # Main skill definition
└── references/
    ├── review-checklist.md   # What to check for
    └── security-patterns.md  # Common security issues

Output Format:

## Code Review: [filename]

### Critical
- **Line 42:** SQL injection vulnerability in user input

### Warnings
- **Line 15:** Function exceeds 50 lines, consider splitting

### Suggestions
- **Line 8:** Consider using `const` instead of `let`

Phases

Phase 1: Basic Review Skill

Tasks: Create SKILL.md with triggers, basic review logic, and output format. Verify:

# Stage a file with an obvious bug
git add test.py
claude  # or gemini
> /review
# Should output structured feedback

Phase 2: Reference Documents

Tasks: Create review-checklist.md and security-patterns.md with common patterns. Verify: Skill references these docs and catches patterns listed in them.

Phase 3: File-Specific Review

Tasks: Add argument parsing for specific file paths. Verify: /review src/auth.py only reviews that file, not all staged changes.

Agent Rules

Always	Ask First	Never
Show file and line numbers	Before suggesting refactors	Auto-modify code
Categorize by severity	Before adding new categories	Access files outside repo
Explain why something is an issue		Make up security vulnerabilities