subagent-driven-development

Coordinate autonomous subagents by dispatching focused implementer, spec reviewer, and code-quality reviewer subagents per task, using standardized statuses like DONE or BLOCKED to guide workflow continuation and prevent context leakage.

The two-stage review process sequentially validates spec compliance and code quality for each task. Subagents first check implementation against the plan, then review code quality, ensuring structured oversight before reporting a final status.

Prevent context leakage in parallel subagents by using per-task prompts that isolate each subagent's context. This ensures subagents operate independently without cross-task contamination during the implementation workflow.

Use subagent workflow orchestration when tasks are largely independent and require structured two-stage reviews. It suits complex features across multiple modules where maintaining code quality and plan fidelity within a single session is critical.

Subagents report task completion or blockers using standardized statuses: DONE, DONE_WITH_CONCERNS, BLOCKED, or NEEDS_CONTEXT. These statuses guide the controller to continue, rework, or escalate tasks within the workflow.

No external dependencies are required to run subagent-driven development workflows. The orchestration operates autonomously within a single session, managing implementer and reviewer subagents without additional component installations.