subagent-driven-development

To execute an implementation plan without context pollution, dispatch a freshly instructed subagent for each independent task. This provides every subagent with only the specific data it needs, keeping coordination parallel-safe and preventing stale context from affecting subsequent tasks.

Subagent-driven development is a workflow orchestration technique that dispatches isolated subagents for independent implementation tasks. Use it when working through a detailed plan in one session to keep coordination fast and prevent context bleeding between separate code changes.

To ensure spec compliance and code quality during task orchestration, apply a two-stage quality gate after each implementer report. Run a spec compliance reviewer followed by a code-quality reviewer to catch missing requirements and structural issues before advancing.

Yes, you can match model capability to task complexity by pairing mechanical implementation work with fast models and complex judgments with the most capable models. This model selection guidance makes the subagent workflow cost-effective yet thorough.

To track progress when dispatching subagents for modular tasks, log the status in TodoWrite as each implementer completes. Finish the workflow with a final branch review so the whole implementation plan can ship smoothly.

You should not use a subagent workflow when implementation tasks are highly dependent on each other. This approach is designed for mostly independent implementation tasks where parallel-safe coordination and avoiding context pollution are the primary goals.