subagent-driven-development

You orchestrate subagents by defining a plan with independent tasks, dispatching a fresh implementer subagent per task, and running a two-stage review loop for spec compliance and code quality. This isolates context and accelerates iteration.

A two-stage code review loop evaluates subagent output by first checking spec compliance to ensure correct results, then assessing code quality for maintainability. This mechanism coordinates planning, dispatch, review, and completion within a single session.

To prevent context leakage and cross-task contamination, assign each independent task to a fresh, task-specific subagent. This isolates the implementation context and ensures transparent, traceable task progression without prior session interference.

Yes, you can use in-session task orchestration to execute complex implementation plans. It coordinates planning, dispatch, two-stage reviews, and completion tracking within a single session to distribute work and accelerate iteration.

Use a dedicated subagent per task when your implementation plan contains independent tasks. This approach isolates context, prevents cross-task contamination, and allows an orchestration mechanism to track tasks and results through a two-stage review process.

This approach requires per-task subagent dispatch and a two-stage review process, meaning it is limited to sessions with independent tasks. Tasks requiring shared context or sequential dependencies are not suitable for this orchestration mechanism.