---
description: Start comprehensive code review loop between reviewer and developer agents
---

# Code Review Process Orchestration

## Overview
Initiates automated code review loop between @codereview-reviewer (Opus) and @codereview-developer (Sonnet) until code reaches production-ready state.

## Process Flow
Start → @codereview-reviewer → @codereview-developer → @codereview-reviewer → (loop until APPROVED) → Complete

## Execution Steps

### Auto-Initiated Agents
1. **@codereview-reviewer** (Opus) - Reviews changed files only, applies rules checklist
2. **@codereview-developer** (Sonnet) - Fixes failed rules atomically, validates syntax
3. Loop continues via HANDOFF until `REVIEW_COMPLETE`

## Loop Control (Anthropic Best Practice)

### Termination: All Rules Pass
- ✅ All security rules passed
- ✅ All validity rules passed
- ✅ No syntax errors
- ✅ Output: `REVIEW_COMPLETE`

### Safety Limit: 8 Iterations
If iteration > 8: Stop, request human review (prevents infinite loops)

### Token Budget: 200000 total
Full budget available for comprehensive reviews

## Usage

### Option 1: Review Uncommitted Changes (Default - Recommended)
```bash
/codereview-startreviewprocess
```
Analyzes: `git diff HEAD` (all uncommitted changes, including documentation)

### Option 2: Review Last Commit
If no uncommitted changes exist:
```bash
/codereview-startreviewprocess
```
Automatically falls back to: `git show HEAD` (most recent commit)

### Option 3: Review Full Branch
If no uncommitted changes and comparing against base branch:
```bash
/codereview-startreviewprocess
```
Falls back to: `git diff master..HEAD` or `git diff main..HEAD` (all branch changes)

## Performance (2025 Benchmarks)

| Metric | Value |
|--------|-------|
| Avg iterations to APPROVED | 2-4 |
| Token budget | 200000 total |
| Max iterations (hard limit) | 8 |
| Review scope | All file types (code, docs, configs, etc.) |

## Monitoring

Watch for:
- `HANDOFF: @codereview-[agent] [Iteration X/8]` - Loop progress
- `REVIEW_COMPLETE` - Success, all rules passed
- Iteration > 8 - Human review requested

## Best Practices (Anthropic Guidelines)

1. **Work in Progress**: Run on uncommitted changes for fastest feedback
2. **Small Batches**: Review one feature at a time for clarity
3. **Let It Run**: Agents auto-loop via HANDOFF, no intervention needed
4. **Safety Net**: Automatic backup branch created before fixes
5. **Hard Stop**: Process stops at iteration 8, preventing runaway loops

## Post-Approval

After `REVIEW_COMPLETE`, validate and deploy:
```bash
# Run project-specific test suite (examples):
npm test || composer test || pytest || cargo test || go test ./...

# Deploy approved code
git push
```