subagent-driven-development

Executing implementation plans with isolated subagents requires dispatching a fresh subagent per task with complete context. This approach prevents context bleed across independent tasks, ensuring clean task isolation and reducing cross-task contamination during code execution.

A two-stage review process first checks spec compliance to verify the implementation meets requirements, then assesses code quality. This ensures subagent-driven development produces functional code that adheres to both the original plan and quality standards.

To delegate independent tasks, use delegate_task to dispatch each task with complete context to a fresh subagent. Follow this with spec-compliance and code-quality reviews, plus final integration checks to ensure the overall implementation maintains high standards.

Subagent-driven development works with test-driven development and writing-plans workflows to align planning and execution. It orchestrates subagents to execute the plan while applying two-stage reviews, ensuring TDD specs are met and code quality is maintained.

Use fresh subagents for task delegation when working with implementation plans that have multiple independent tasks. This approach is ideal when two-stage reviews are desired and when fresh context per task is needed to reduce cross-task contamination.

Dispatching fresh subagents per task is limited to plans with multiple independent tasks. It is not suitable for highly interdependent tasks where context sharing is required, as the isolation mechanism prevents cross-task contamination but also restricts shared context.