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:

bash

git status --porcelain

Decision:

•If changes exist: Review uncommitted changes via git diff HEAD (staged + unstaged)
•If no changes: Review latest commit via git log -1 and git show HEAD
•If user specified a commit hash: Review that commit via git show <hash>

Empty diff guard: If no diff output is produced, report "No changes to review" and stop.

2. Exclusion Patterns

Filter out the following from diff analysis and agent prompts:

•Lock files: package-lock.json, pnpm-lock.yaml, yarn.lock, go.sum, Cargo.lock
•Generated/build: dist/, build/, *.generated.*, *.min.*
•Source maps: *.map
•Binary/media: images, fonts, compiled artifacts

Use git diff HEAD -- . ':!package-lock.json' ':!pnpm-lock.yaml' ':!yarn.lock' ':!go.sum' ':!Cargo.lock' ':!dist' ':!build' to exclude these.

3. 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, *.proto, API route files)
•Configuration files (package.json, go.mod, pyproject.toml, Cargo.toml, build.gradle, docker-compose.yml)
•Infrastructure files (*.tf, Dockerfile, K8s manifests, CI/CD configs)
•Security-sensitive files (auth/, **/middleware/auth*, *.pem, *.key)
•New directories or modules

4. Invoke code-reviewer Agent

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

Provide structured context to avoid redundant discovery:

code

## Review Context

- **Target**: {uncommitted changes / commit <hash>}
- **Work Type**: {feature/bugfix/refactor/chore/prototype}
- **Commit Message**: {message from git log or commit_message.md}

## Scope Boundary

Architecture concerns (component boundaries, module coupling, system scalability, deployment architecture) are out of scope. Do NOT review these.

## Diff

{see diff size rules below}

Diff size rules:

•≤ 500 lines: Include full diff inline
•> 500 lines: Include git diff --stat summary only, and instruct: "Read specific files using Read tool as needed for detailed review."

5. Conditional architect-reviewer Invocation

Scope Classification Criteria (score each criterion met):

Criterion	Threshold	Score
Files Modified	≥ 8 files	+1
Total Lines Changed	≥ 300 lines	+1
Directories Affected	≥ 3 directories	+1
New Module/Package	New directory with ≥3 files	+1
API Contract Changes	Modified: schema/, .graphql, .proto, api/, routes	+2
Database Schema	Modified: migrations, schema files, ORM models	+2
Config/Infrastructure	Modified: docker-compose, Dockerfile, K8s, CI/CD, .tf	+1
Dependency Changes	Modified: package.json, go.mod, requirements.txt	+1
Security-Sensitive	Modified: auth/, middleware/auth, .pem, *.key	+2

Decision:

•Score < 3 → Invoke code-reviewer only
•Score ≥ 3 → Invoke both code-reviewer and architect-reviewer in parallel (two Task calls in one message)

architect-reviewer structured context:

code

## Review Context

- **Target**: {uncommitted changes / commit <hash>}
- **Work Type**: {feature/bugfix/refactor/chore/prototype}
- **Commit Message**: {message}
- **Scope Score**: {score} (triggered: {list of criteria met})
- **Key Change Areas**: {directories/modules with most impact}

## Scope Boundary

Line-level code quality (naming, formatting, individual function logic, test coverage) is out of scope. Do NOT review these.

## Diff

{same diff size rules as code-reviewer}

Error Handling

•Empty diff → Report "No changes to review" and stop
•Git command failure → Report the error with the failed command
•Agent failure → Include partial results from successful agent + note which agent failed
•Agent reports no issues → Include "No issues found" in the relevant section

Output Format

Generate consolidated review summary:

markdown

# 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}

### Warning (Should Fix)

- [ ] {issue 3}

### Suggestion (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

bash

# Review uncommitted changes
/code-review

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