Subagent-Driven Development
Execute plan by dispatching fresh context per task, with two-stage review after each: spec compliance review first, then code quality review.
Core principle: Fresh context per task + two-stage review = high quality, fast iteration
When to Use
- Have implementation plan? ✓
- Tasks mostly independent? ✓
- Want fast iteration? ✓
The Process
┌─────────────────────────────────────────────────────────────┐
│ Per Task Cycle │
├─────────────────────────────────────────────────────────────┤
│ │
│ 1. Read task from plan │
│ 2. Execute task (implement, test, commit) │
│ 3. Self-review for spec compliance │
│ 4. Review for code quality │
│ 5. Fix any issues found │
│ 6. Re-review until approved │
│ 7. Move to next task │
│ │
└─────────────────────────────────────────────────────────────┘
Two-Stage Review
Stage 1: Spec Compliance Review
Questions to answer:
- Does the implementation match the spec?
- Are all requirements addressed?
- Did the task do what it was supposed to do?
Must pass before Stage 2.
Stage 2: Code Quality Review
Questions to answer:
- Is the code clean and readable?
- Are there any bugs or edge cases?
- Does it follow project conventions?
- Are tests comprehensive?
Task Execution Flow
For each task in plan:
- Announce: "Starting Task N: [description]"
- Execute: Follow TDD (write test, watch fail, implement, watch pass)
- Commit:
git commit -m "feat: [task description]" - Self-Review Stage 1: Check spec compliance
- Self-Review Stage 2: Check code quality
- Report: "Task N complete. Summary: [what was done]"
Common Mistakes
- Skipping the spec compliance review
- Starting code quality review before spec compliance passes
- Moving to next task with open issues
- Skipping the TDD cycle
- Not committing after each task
Integration
Required skills:
test-driven-development— Follow TDD for each taskwriting-plans— Creates the plan this skill executesrequesting-code-review— Template for review process