Build

This orchestrates TDD execution by spinning up parallel subagents (test writers, implementers, doc writers) to knock out plan tasks in batches, with hard validation gates between each batch and at the end. It's the muscle behind Step 5 of a larger workflow, taking an approved plan from /review and dispatching the actual work while you stay hands off. The interesting bit is how it handles failure modes: subagents report DONE, DONE_WITH_CONCERNS, BLOCKED, or NEEDS_CONTEXT, and it won't blindly retry without changing something. Also works for quick hotfixes without the full orchestration overhead. Falls back gracefully if you're in Codex or Cursor where subagents aren't available.