subagent-driven-development

Executing multipart implementation plans without context pollution requires dispatching isolated subagents per task. This approach provides fresh per-task context, preventing cross-task interference and maintaining deterministic execution quality.

The two-stage review process for test-driven development requires spec compliance review first, then code quality review. Each task undergoes rework loops until both review stages pass, ensuring implementation meets specifications and maintains code quality standards.

Enforcing test-driven development across independent implementation tasks involves dispatching a new implementer subagent per task. Each subagent writes tests first, implements the task, and undergoes spec and code quality reviews before the task is marked complete.

Subagents for task orchestration support model selection guidance by selecting the least-powerful model that can handle each task. When a subagent is blocked, clear escalation paths allow switching to more capable models to complete the work.

Handling status and rework loops during spec compliance reviews requires a structured pass-fail gate before code quality review. If spec compliance fails, the task enters a rework loop until the implementation matches the plan specifications before proceeding.

Subagent-driven development for feature plan execution should not be used when implementation tasks lack independence or cannot be individually tested and reviewed within the same session. Tasks requiring shared mutable state or tight coupling are unsuitable for isolated subagent dispatch.