subagent-driven-development

To prevent context pollution when implementing multi-step plans, dispatch a fresh isolated subagent per task so each worker sees only its specific task scope and cannot carry assumptions from other tasks.

The best way to enforce quality in task orchestration is using a two-stage review pipeline, requiring a separate spec compliance review followed by an independent code quality review before any task is accepted.

Test-driven development verification with isolated subagents works by requiring actual test output, performing local verification, and tracking commits for each implementation task before routing any failures back for iterative fixes.

Yes, you can execute independent tasks in parallel from a written plan by dispatching a fresh implementer subagent for each task, ensuring isolated, verifiable, and reproducible code changes across the session.

When a code review fails during subagent-driven development, the workflow routes fixes back to the implementer subagents and re-runs the reviews until all spec and quality criteria are met.

You should not use isolated subagents for task implementation when your written plan lacks independent tasks or when the tasks cannot be completed within the same session, as the workflow requires isolated implementations.