subagent-driven-development

To execute an implementation plan with isolated subagents, dispatch a fresh implementer subagent for each task so context remains minimal and focused, then run spec compliance and code quality reviews until both approve.

Two-stage code review in task orchestration enforces a spec compliance review first, then a code quality review, with re-dispatch loops for BLOCKED or NEEDS_CONTEXT statuses until both review stages approve the implementation.

When subagents return BLOCKED or NEEDS_CONTEXT statuses during development tasks, the orchestration workflow uses status-driven re-dispatch to automatically relaunch the subagent, allowing it to request missing context before continuing implementation.

Yes, subagent-driven development is ideal for executing execution-ready plans with mostly independent tasks, making it highly suitable for TDD-friendly workflows and acting as a parallel-safe controller for feature rollouts.

Model selection for subagent task orchestration uses provided model-selection guidance to match each subagent's role—implementer, spec reviewer, or code-quality reviewer—with the appropriate model based on cost efficiency and required capability.

You should avoid using per-task subagents when your implementation plan lacks execution-ready, mostly independent tasks, as the workflow relies on isolating context per task and enforcing sequential two-stage reviews for each item.