Subagent Driven Development

Execute plans by dispatching fresh subagent per task, with two-stage review.

Core Principle

Fresh subagent per task + two-stage review = high quality, fast iteration

Why fresh subagents?

• No context pollution from previous tasks
• Clean mental model per task
• Faster iteration (no baggage)

Process

Read plan → Create TodoWrite → For each task:
  Dispatch implementer → Spec review → Quality review → Verify → Mark complete

See references/flow.md for detailed workflow.

Post-Verification

After task completion:

1. Run verification commands from plan (see verification-before-completion)
2. Confirm all success criteria met
3. Only then mark task complete
4. If verification fails → trigger fix flow (max 3 retries)

<!-- PHASE_COMPLETE: subagent-driven-development -->