subagent-driven-development

To execute multi-step implementation plans without context pollution, dispatch a fresh subagent per independent task with isolated context. This prevents prior session history from degrading outcome quality during iterative execution.

Subagent-driven development is a workflow pattern that executes multi-task implementation plans by dispatching fresh subagents per independent task. It applies to same-session development requiring spec verification, structured status handling, and maintainable code outcomes.

Verify spec compliance and maintainable code by running a two-stage post-task review. Execute a spec compliance review first to catch missing requirements, then run a code quality review to ensure maintainability after each task.

Handle subagent status during task decomposition using a clear status handling loop with escalation rules. Support DONE, DONE_WITH_CONCERNS, NEEDS_CONTEXT, and BLOCKED statuses to avoid retrying tasks without making actual changes.

Use isolated subagents for complex development workflows when you have an implementation plan with mostly independent tasks. It suits same-session development where tasks need iterative execution, spec verification, and maintainable code outcomes.

Select models for role-based prompting by using cheaper models for mechanical implementation work and more capable models for architecture, review, and judgment tasks. This optimizes cost while maintaining code quality review standards.