subagent-driven-development

To prevent context pollution during multi-step development plans, delegate independent implementation tasks to fresh subagents. This isolates each task's execution context, keeping work focused and preventing previous instructions from corrupting subsequent task logic.

A two-stage code review workflow for subagent task delegation first checks specification compliance to ensure requirements are met, then evaluates code quality. This enforces review checkpoints before marking any delegated implementation task complete.

Coordinate sequential implementation tasks by breaking the plan into isolated subagent tasks and executing them one by one. The controller manages progress, questions, and fixes across tasks, ensuring no parallel collisions occur during single-session development.

Subagent task delegation is designed for single-session development plans where tasks execute sequentially. It is not suitable for parallel workflows, as the architecture requires isolated context and explicitly prevents parallel collisions across tasks.

Avoid using fresh subagents for implementation plans when tasks require shared context, parallel execution, or continuous collaboration. This approach requires tasks to be decomposable into isolated steps that can be executed sequentially without collisions.