subagent-driven-development

# Subagent-Driven Development

Execute plan by dispatching fresh subagent per task, with **three-stage review** after each:
1. Spec compliance review
2. Code quality review (ios-reviewer or backend-reviewer)
3. UX review

**Core principle:** Fresh subagent per task + three-stage review = high quality with great UX

## When to Use

- Have an implementation plan with defined tasks
- Tasks are mostly independent (can be done in any order)
- Want to stay in the current session (no parallel sessions)

## The 3-Stage Review Process

```
┌─────────────────────────────────────────────────────────────┐
│                    PER-TASK WORKFLOW                        │
├─────────────────────────────────────────────────────────────┤
│  1. IMPLEMENT      Implementer subagent builds the feature  │
│         ↓                                                   │
│  2. SPEC REVIEW    Does code match requirements?            │
│         ↓          (./spec-reviewer-prompt.md)              │
│  3. CODE REVIEW    Is code quality good?                    │
│         ↓          (ios-reviewer or backend-reviewer)       │
│  4. UX REVIEW      Is user experience good?                 │
│         ↓          (ux-reviewer)                            │
│  5. COMPLETE       Task marked done                         │
└─────────────────────────────────────────────────────────────┘
```

### Stage Details

**Stage 1: Implementation**
- Dispatch implementer subagent with full task text
- Implementer asks questions if unclear
- Implementer implements, tests, self-reviews, commits

**Stage 2: Spec Compliance Review**
- Verify implementer built what was requested (nothing more, nothing less)
- If issues found → implementer fixes → re-review
- Only proceed when ✅ spec compliant

**Stage 3: Code Quality Review**
- Use `ios-reviewer` for Swift/SwiftUI code
- Use `backend-reviewer` for Supabase/TypeScript code
- Check patterns, performance, security
- If issues found → implementer fixes → re-review
- Only proc