subagent-driven-development

Iterative development with AI agents is automated by breaking down implementation plans into independent tasks, dispatching a dedicated subagent for each task, and running two-stage reviews for spec compliance and code quality. Feedback loops manage iterative refinement until all criteria are met.

Subagent dispatch works by assigning a fresh AI subagent to each independently decomposed task from a development plan. This ensures clean context per task before proceeding to mandatory spec compliance and code quality reviews.

The best way to ensure code quality and spec compliance is implementing a mandatory two-stage review process where a reviewer agent checks spec compliance first, then code quality, managing feedback loops with the implementer agent until tasks pass.

Yes, you can use AI agents to execute an implementation plan found in a markdown file. The skill reads the plan file, decomposes it into independent tasks, and dispatches subagents to handle each task within a single session.

Manual code review is minimized because the workflow mandates automated two-stage reviews for spec compliance and code quality. Reviewer agents run iterative feedback loops with implementer agents to automatically refine tasks until they meet all criteria.

Independent subagents are used for each task to ensure clean context isolation. This approach prevents context bleeding across tasks, allowing dedicated agents to focus on specific API endpoints or UI components before multi-stage reviews validate integration.