Code Review Gate Skill | Agent Skills

Code Review Gate

You are a Code Review Gate responsible for ensuring code quality, security, and correctness BEFORE commits reach version control.

Purpose

Problem: Bugs, security issues, and quality problems slip into commits because reviews happen after code is already committed.

Solution: Mandatory pre-commit review using code-reviewer agent with structured feedback and approval workflow.

When to Use This Skill

Manual invocation ONLY: /code-review

When to invoke:

Before running git commit
After completing feature implementation
Before creating pull request
When user asks "review my changes"

NOT auto-triggered (disable-model-invocation: true)

The 5-Phase Review Workflow

Phase 1: IDENTIFY CHANGES

Goal: Determine what needs review

Git commands:

# See all changes
git status

# Staged changes (will be committed)
git diff --cached --stat
git diff --cached

# Unstaged changes (might want to stage first)
git diff --stat
git diff

# Recent commits (if reviewing after commit)
git log -1 --stat
git show HEAD

Categorize changes:

New files: Complete review needed
Modified files: Focus on changes only
Deleted files: Check dependencies (use impact-analysis)
Renamed files: Verify refactoring correctness

Deliverable: List of files to review with change types

Phase 2: LAUNCH REVIEWER

Goal: Get structured code review from code-reviewer agent

Prepare review prompt:

Task(
  subagent_type="code-reviewer",
  description="Pre-commit code review",
  prompt="""
Review these changes for bugs, security issues, performance problems, and code quality:

**Files Changed**:
[List files with brief description of changes]

**Context**:
- Purpose: [what this code does]
- Related: [related files or systems]

**Focus Areas**:
- [BUG]: Logic errors, edge cases
- [SECURITY]: Vulnerabilities, input validation
- [PERF]: Performance issues, N+1 queries
- [QUALITY]: Code smells, maintainability
- [STYLE]: Formatting, naming (low priority)

Provide structured feedback with:
1. Issues categorized by severity
2. Specific line references
3. Suggested fixes
4. Verdict: APPROVE or REQUEST_CHANGES
"""
)

Wait for review results

Deliverable: Structured code review with categorized issues

Phase 3: ANALYZE FEEDBACK

Goal: Prioritize issues and plan fixes

Issue Categories (from code-reviewer):

CRITICAL (Must fix before commit):

[BUG] - Logic errors, crashes, data corruption
[SECURITY] - Vulnerabilities, injection risks, auth bypasses

WARNINGS (Should fix before commit):

[PERF] - Performance issues, scalability problems
[QUALITY] - Code smells, maintainability issues

SUGGESTIONS (Optional, can defer):

[STYLE] - Formatting, naming conventions
[DOCS] - Missing or unclear comments

Verdict Analysis:

APPROVE:

No critical issues found
No warnings that block commit
Code meets quality standards
→ Proceed to commit (Phase 5)

REQUEST_CHANGES:

Critical issues present (MUST fix)
Warnings that should be addressed
→ Fix issues (Phase 4)

NEEDS_DISCUSSION:

Unclear requirements
Architectural concerns
→ Consult user or brainstorm-coordinator

Deliverable: Prioritized issue list with fix plan

Phase 4: FIX ISSUES

Goal: Address critical and warning issues

Fixing Strategy:

For each CRITICAL issue:

Read file with issue
Understand the problem
Apply minimal fix
Verify fix doesn't break tests
Mark as resolved

For each WARNING:

Assess impact (breaking change? refactor needed?)
Apply fix if straightforward
Create TODO if complex (defer to separate commit)
Mark as resolved or deferred

Example fix workflow:

# Read file
Read(file_path="src/auth/login.ts")

# Apply fix
Edit(
  file_path="src/auth/login.ts",
  old_string="if (user.password == password)",  # BUG: == instead of proper comparison
  new_string="if (await bcrypt.compare(password, user.password))"
)

# Run tests
npm test src/auth/login.test.ts

# Verify fix
Read(file_path="src/auth/login.ts", offset=40, limit=10)

Deliverable: All critical issues fixed, warnings addressed

Phase 5: RE-REVIEW & APPROVE

Goal: Verify fixes and get final approval

Re-review process:

If fixes were applied:

Task(
  subagent_type="code-reviewer",
  description="Re-review after fixes",
  prompt="""
Re-review changes after addressing previous feedback:

**Previous Issues**:
[List issues that were fixed]

**Files Changed** (after fixes):
[Updated file list]

Focus on:
- Verify previous issues resolved
- No new issues introduced by fixes
- Overall code quality

Provide verdict: APPROVE or REQUEST_CHANGES (if still issues)
"""
)

Wait for verdict

Verdict Handling:

APPROVE:

✅ All issues resolved
✅ No new issues from fixes
→ Proceed to commit

REQUEST_CHANGES (still issues):

❌ Return to Phase 4 (fix remaining issues)
❌ Maximum 2 iterations (prevent infinite loop)
❌ After 2 iterations, ask user for guidance

Deliverable: Final APPROVE verdict

Phase 6: COMMIT (After Approval)

Goal: Commit reviewed, approved code

Pre-commit checklist:

# Ensure all changes staged
git add <fixed-files>

# Final status check
git status

# Run tests one more time
npm test

# Commit with descriptive message
git commit -m "$(cat <<'EOF'
feat: Add user authentication

- Implement JWT-based auth
- Add login/logout endpoints
- Secure password hashing with bcrypt

Fixes: #123

Co-Authored-By: Claude Sonnet 4.5 (1M context) <noreply@anthropic.com>
EOF
)"

Post-commit:

# Verify commit
git show HEAD

Deliverable: Clean commit with reviewed, approved code

Integration Points

With Other Skills

After /workflow: Review work before committing
After /debug: Review bug fixes for correctness
After /refactor-safely: Verify refactoring didn't break anything
Before push: Final quality gate

With Pre-Tool-Use Hook

Suggested integration:

# In hooks/pre-tool-use.sh
if [[ "$TOOL" == "Bash" && "$COMMAND" =~ "git commit" ]]; then
  echo "💡 Suggestion: Run /code-review before committing"
fi

With Capsule Context

File operations and sub-agent results are captured automatically by Capsule hooks. No manual logging needed.

Examples

Example 1: New Feature Review (Issues Found)

Phase 1: Identify Changes

git status
# Modified: src/auth/login.ts, src/auth/register.ts
# New: src/middleware/auth.middleware.ts

Phase 2: Launch Reviewer

Reviewing 3 files for authentication feature...

Phase 3: Analyze Feedback

CRITICAL ISSUES:
[BUG] src/auth/login.ts:42 - Password comparison uses == instead of bcrypt.compare()
[SECURITY] src/auth/register.ts:28 - No password strength validation

WARNINGS:
[QUALITY] src/middleware/auth.middleware.ts:15 - Error handling missing for invalid tokens

Verdict: REQUEST_CHANGES

Phase 4: Fix Issues

// Fix 1: Password comparison
- if (user.password == password)
+ if (await bcrypt.compare(password, user.password))

// Fix 2: Password validation
+ if (password.length < 8) {
+   throw new Error('Password must be at least 8 characters')
+ }

// Fix 3: Error handling
+ try {
    const decoded = jwt.verify(token, SECRET)
+ } catch (error) {
+   throw new UnauthorizedException('Invalid token')
+ }

Phase 5: Re-review

All issues resolved ✓
Fixes look good ✓
No new issues introduced ✓

Verdict: APPROVE

Phase 6: Commit

git add src/auth/*
git commit -m "feat: Add user authentication with security fixes"

Example 2: Bug Fix Review (Clean)

Phase 1: Identify Changes

git diff --cached
# Modified: src/services/order.service.ts (1 file, 3 lines changed)

Phase 2: Launch Reviewer

Reviewing bug fix in order service...

Phase 3: Analyze Feedback

No issues found ✓

The fix correctly:
- Adds null check before property access
- Matches the root cause (TypeError prevention)
- Minimal change (good)
- Test coverage exists

Verdict: APPROVE

Phase 4: Fix Issues Skipped (no issues)

Phase 5: Re-review Skipped (already approved)

Phase 6: Commit

git commit -m "fix: Add null check in order service (fixes TypeError)"

Example 3: Large Refactor Review

Phase 1: Identify Changes

git diff --cached --stat
# 12 files changed, 450 insertions(+), 320 deletions(-)

Phase 2: Launch Reviewer

Reviewing large refactoring across 12 files...

Phase 3: Analyze Feedback

WARNINGS:
[QUALITY] Inconsistent error handling in 3 files
[DOCS] Missing JSDoc for new public methods
[STYLE] Some files use var instead of const/let

No CRITICAL issues ✓
Refactoring is structurally sound ✓

Verdict: APPROVE (with suggestions for future cleanup)

Phase 4: Fix Issues

Warnings are minor, can be addressed in follow-up PR.
Creating TODO issue for cleanup.

Phase 5: Re-review Skipped (approved with minor suggestions)

Phase 6: Commit

git commit -m "refactor: Modernize authentication module

- Extract auth logic to separate services
- Improve testability with dependency injection
- Update imports across 12 files

TODO: Follow-up cleanup for error handling consistency"

Success Criteria

Review Execution

✅ Code-reviewer agent consulted (not manual review) ✅ All CRITICAL issues fixed before commit ✅ WARNINGS addressed or acknowledged ✅ APPROVE verdict received before committing ✅ Tests pass after fixes

Quality Signals

Zero Bugs in Review: No logic errors found
Security Checked: No vulnerabilities introduced
Performance OK: No obvious scalability issues
Clean Commit: Reviewed code only (no debug logs, TODOs)

Anti-Patterns

❌ Committing without review: Always review first ❌ Ignoring REQUEST_CHANGES: Fix issues before committing ❌ Fixing only some issues: Address all CRITICAL, acknowledge WARNINGS ❌ Infinite fix loop: Max 2 re-reviews, then ask user ❌ Skipping tests: Always test after fixes

Advanced Patterns

Selective Review (Large Changes)

For commits with 20+ files changed:

# Review critical files only
Task(subagent_type="code-reviewer", prompt="Review these 5 critical files: ...")

# Defer non-critical files to PR review
# Focus on: security-sensitive, complex logic, core functionality

Security-Focused Review

For auth, payments, data handling:

Task(subagent_type="security-engineer", prompt="Security review of authentication changes")
Task(subagent_type="code-reviewer", prompt="General code review")

# Both must APPROVE before commit

Performance Review

For database, API, algorithm changes:

Task(subagent_type="code-reviewer", prompt="Review with focus on performance and scalability")

# Look for: N+1 queries, inefficient algorithms, missing indexes

Remember: Code review BEFORE commit is 10x cheaper than fixing bugs in production. Use this skill religiously.

Agent Skills: Code Review Gate

Install this agent skill to your local

Skill Files

Code Review Gate

Purpose

When to Use This Skill

The 5-Phase Review Workflow

Phase 1: IDENTIFY CHANGES

Phase 2: LAUNCH REVIEWER

Phase 3: ANALYZE FEEDBACK

Phase 4: FIX ISSUES

Phase 5: RE-REVIEW & APPROVE

Phase 6: COMMIT (After Approval)

Integration Points

With Other Skills

With Pre-Tool-Use Hook

With Capsule Context

Examples

Example 1: New Feature Review (Issues Found)

Example 2: Bug Fix Review (Clean)

Example 3: Large Refactor Review

Success Criteria

Review Execution

Quality Signals

Anti-Patterns

Advanced Patterns

Selective Review (Large Changes)

Security-Focused Review

Performance Review