subagent-driven-development

Subagent-driven development enforces spec compliance by dispatching a fresh implementer subagent per task, then validating results through a two-stage review for spec compliance and code quality before proceeding. This prevents spec drift and ensures incremental progress is validated.

Automate TDD by instructing implementer subagents to write failing tests first, implement minimally, and verify with pytest. The execution loop runs fixes until both spec compliance and code quality reviewers approve the task.

The two-stage review process works by first checking implemented task results against the original plan for spec compliance, followed by a code quality review for robustness, tests, and security. Both stages must approve before moving to the next task.

Yes, you can use subagent delegation for refactoring with guardrails. Each refactoring task is handled by a separate implementer subagent, and the two-stage review pipeline ensures consistent quality and correctness throughout the process.

Implementation drifts from the original plan due to context pollution and missed requirements. Using a fresh implementer subagent per task prevents context pollution, while the spec compliance reviewer validates results against the original plan to catch drift.

Yes, you need to read and parse the full implementation plan once before starting. The process requires running delegate_task with complete task context and project constraints, including TDD with pytest, for each independent task in the plan.