subagent-driven-development

Read the plan once, extract every task into a todo list, then dispatch a fresh implementer subagent per task using delegate_task with the full task text and TDD instructions. After each implementation, run a spec compliance review and then a code quality review before marking the task complete.

The first stage checks spec compliance against the original task requirements, and the second stage reviews code quality such as style, error handling, and test coverage. Spec review must pass before quality review starts, and any issues trigger a fix-and-re-review loop.

Use it when you have a written implementation plan with mostly independent tasks and quality matters. Fresh subagents per task avoid context pollution, and automated reviews catch issues early, though it costs more subagent invocations than manual execution.

Inspect the live working tree before redispatching, since timeouts often leave valid partial edits. Preserve good work, resume the session once if useful, and otherwise start a fresh narrower worker with only the remaining invariant rather than restarting blindly.

Each task should represent roughly two to five minutes of focused work, such as creating one model or one endpoint. Large items like implementing an entire authentication system should be split into smaller tasks that can each be tested and reviewed independently.