subagent-driven-development

To automate plan execution, dispatch a fresh implementer subagent for the first task. Once it finishes, trigger spec-reviewer and code-quality-reviewer subagents, manage rework cycles, and repeat this automated tracking flow until all tasks are complete.

You prevent context pollution by dispatching a fresh subagent per task. This isolates the working context for each individual task, maintaining focus and enabling faster iteration without degrading performance from accumulated session history.

Each task undergoes a two-stage review workflow: a spec compliance review followed by a code quality review. Automated per-task self-reviews and rework cycles are enforced before proceeding to the next implementation step.

You should avoid using fresh subagents when tasks are highly interdependent rather than largely independent. The isolated context approach requires strict review gates and rapid iteration, which breaks down if tasks require continuous shared state.