subagent-driven-development

You can execute a multi-step development plan by dispatching fresh subagents for each task. This preserves isolated context per task, preventing cross-contamination while running tasks in parallel or sequentially within a single session.

A two-stage review assesses completed subagent tasks sequentially. The first stage checks spec compliance to verify requirements, and the second stage evaluates code quality to ensure high-quality outcomes before final handoff.

To prevent context cross-contamination, dispatch a fresh subagent for each independent task. This approach isolates the working context per task, ensuring parallel problem-solving does not interfere with other ongoing implementations.

Yes, you can run iterative implementations with automatic task decomposition in a single session. The workflow decomposes your plan into independent tasks and dispatches subagents sequentially or in parallel until completion.

Yes, you need to provide the implementation plan to start dispatching subagents. The system uses your plan to automatically decompose tasks and execute them with structured two-stage reviews and a clear handoff workflow.

You should not use isolated subagents for development workflows when tasks require shared memory or continuous dependent context. The system is designed for independent tasks, so highly coupled steps lose the benefits of context isolation and two-stage reviews.