subagent-driven-development

Read the plan once, extract all tasks into a todo list, then dispatch a fresh delegate_task subagent per task with the full task text and project context inline. After each implementation, run the TDD gate, spec review, and quality review in that order before marking the task complete.

The TDD gate inspects the git log for test commits preceding feat commits as RED-GREEN evidence, runs the full test suite to confirm no regressions, and checks that tests exercise real behavior. A TDD_FAIL verdict invalidates the implementation and triggers a re-dispatch with stricter TDD instructions.

Avoid it for single-file fixes, tasks that share files due to conflict risk between parallel subagents, and exploratory prototyping without a written plan. It is designed for decomposed multi-task plans where independent tasks benefit from isolated context and automated review.

The implementer subagent or a new fix subagent addresses the specific feedback, then the same reviewer re-runs. The loop repeats until the gate passes, and the workflow never advances to the next task or next gate with open critical or important issues.

Fresh subagents prevent context pollution from accumulated state, giving each task clean focused context without confusion from prior tasks' code or reasoning. The trade-off is more subagent invocations, but early issue detection is cheaper than debugging compounded problems later.