subagent-driven-development

You coordinate subagents by dispatching a fresh subagent per independent task to avoid context pollution and enable rapid iteration. Each subagent receives explicit prompts, executes the task, and performs a self-review before returning results.

The two-stage review process evaluates each completed task sequentially. The first stage checks spec compliance against the original requirements, and the second stage reviews code quality to ensure standards are met before merging.

To avoid context pollution during multiple iterations, dispatch a fresh subagent for each independent task rather than reusing a single subagent. This approach keeps the context window clean and allows rapid iteration without cross-task interference.

Yes, you can use a dedicated task board for plan-driven subagent workflows. The board supports per-task dispatch, explicit prompt generation, self-review tracking, and a final merge workflow to maintain quality and traceability across the session.

You should use fresh subagents when your development tasks are largely independent and require rapid iteration. This approach prevents context buildup from previous tasks and enforces structured two-stage reviews for spec compliance and code quality.