Code Review

Code Review Command

Purpose

Perform comprehensive code review of current changes or latest commit. Automatically invokes:

• code-reviewer agent: Always (code quality, security, maintainability)
• architect-reviewer agent: Conditionally (architectural impact assessment)

Workflow

1. Determine Review Target

Check for uncommitted changes:

git status --porcelain

Decision:

• If changes exist: Review uncommitted changes via git diff
• If no changes: Review latest commit via git log -1 and git show HEAD

2. Analyze Change Scope

Extract metrics from git output:

Count:

• Files changed
• Lines added/removed
• Directories affected
• New files vs modifications

Detect:

• Database migration files (migrations/, schema.prisma, *.sql)
• API contract files (*.graphql, openapi.yaml, API route files)
• Configuration files (package.json, go.mod, docker-compose.yml)
• New directories or modules

3. Invoke code-reviewer Agent

Always execute using Task tool with subagent_type: "code-reviewer":

Perform comprehensive code review of the current changes.

Focus areas:
- Code quality: readability, naming, structure
- Security: exposed secrets, input validation, vulnerability patterns
- Error handling: edge cases, error messages, recovery
- Performance: obvious bottlenecks, inefficient patterns
- Testing: coverage for critical paths
- Coding standards compliance

Changes to review:
{paste git diff or git show output}

Provide feedback organized by:
- Critical issues (must fix before merge)
- Warnings (should fix)
- Suggestions (consider improving)

Include specific code examples and fix recommendations.

The code-reviewer agent has access to Read, Write, Edit, Bash, and Grep tools and will analyze the code thoroughly.

4. Conditional architect-reviewer Invocation

Scope Classification Criteria (score each criterion met):

Decision:

• Score < 3 → Invoke code-reviewer only
• Score ≥ 3 → Invoke both code-reviewer and architect-reviewer

Request architect-reviewer using Task tool with subagent_type: "architect-reviewer":

Perform architectural review of these changes.

Change scope:
- Files changed: {count}
- Lines: +{added} -{removed}
- Directories affected: {list}
- Scope score: {X}/8
- Triggered by: {list of criteria met}

Evaluate:
- System design impact
- Scalability implications
- Technology choice justification
- Integration pattern soundness
- Security architecture
- Technical debt introduced/resolved
- Evolution path clarity

Provide strategic recommendations prioritized by:
- Critical architectural risks
- Design improvement opportunities
- Long-term maintainability concerns

Changes to review:
{paste git diff or git show output}

The architect-reviewer agent has access to Read, Write, Edit, Bash, Glob, and Grep tools for comprehensive analysis.

Output Format

Generate consolidated review summary:

# Code Review Report

**Generated**: {timestamp}
**Scope**: {uncommitted changes / latest commit: {hash}}
**Files Changed**: {count}
**Reviewers Invoked**: code-reviewer{, architect-reviewer if applicable}

---

## 📊 Change Summary

- **Lines**: +{added} -{removed}
- **Files**: {count} ({new_count} new, {modified_count} modified)
- **Directories**: {affected directories}

---

## 🔍 Code Reviewer Feedback

{consolidated output from code-reviewer agent}

---

## 🏗️ Architecture Reviewer Feedback

{if invoked, consolidated output from architect-reviewer agent}
{if not invoked, state: "Architectural review not required for this change scope"}

---

## ✅ Review Checklist

Based on agent feedback, generate action items:

### Critical (Must Fix)

- [ ] {issue 1}
- [ ] {issue 2}

### High Priority (Should Fix)

- [ ] {issue 3}

### Suggestions (Consider)

- [ ] {improvement 1}

---

## 📝 Next Steps

{recommended actions based on review results}

Important Notes

• Never modify code directly - this command only performs review
• Agent autonomy: code-reviewer and architect-reviewer may read files, run tests, or analyze dependencies as needed
• Coding guidelines: code-reviewer checks for coding standard violations
• Incremental reviews: For large changes (>20 files), agents may focus on high-impact areas first
• Git safety: All git commands are read-only (status, diff, log, show)

Example Usage

# Review uncommitted changes
/code-review

# Review specific commit (pass commit hash in conversation)
# User: "Review commit abc123"
# Then: /code-review

Execution Instructions for Claude

1. Run git status --porcelain to detect changes
2. If changes exist:
   • Run git diff to get uncommitted changes
   • Run git diff --stat for metrics
3. If no changes:
   • Run git log -1 --oneline to get latest commit
   • Run git show HEAD to get commit changes
   • Run git show HEAD --stat for metrics
4. Calculate scope score against 8 criteria
5. Always invoke code-reviewer using Task tool with subagent_type: "code-reviewer"
6. If score ≥ 3 invoke architect-reviewer using Task tool with subagent_type: "architect-reviewer"
7. Wait for agent(s) to complete analysis
8. Consolidate feedback into structured report
9. Generate actionable checklist

Token optimization:

• Use git diff --stat and git show --stat for metrics (avoid full diff parsing)
• Only pass full diff content to agents, not to your own analysis
• Let agents handle file reading - don't pre-read all changed files