subagent-driven-development

Subagent-driven development executes implementation plans by dispatching an isolated subagent per task. This prevents context leakage and enables parallel-safe orchestration across multiple implementation tasks within a single session.

Spec compliance and code quality are enforced through a two-stage review loop. A spec compliance review runs first, followed by a code quality review, with the implementer applying fixes and re-reviews until both stages approve.

Test-driven development is enforced via TDD-aligned prompts sent to the implementer subagent. The subagent must follow TDD practices and complete a self-review before submitting its work for the staged review process.

Yes, subagent-driven development provides built-in integration guidance for git worktrees and task tracking. This allows each isolated subagent to execute its implementation task within a separate worktree branch.

Use this approach when tasks are largely independent and work should proceed within the same session across multiple implementation tasks. It is ideal for executing multi-task feature plans extracted from a markdown plan document.

A fresh implementer subagent is dispatched per task to avoid context leakage. This isolation ensures each task starts with a clean state, maintaining high-quality implementation without prior task interference.