Subagent-Driven Development

Executes a written plan by dispatching a clean subagent for every task, then running two reviewers: one for spec compliance, one for code quality. The controller stays lean to coordinate and curate context. Implementers never read the plan file directly. They get exactly the context the task needs.

Use it when a plan has mostly independent tasks and the work stays in one session. Don't use it when tasks are tightly coupled, when exploration is still active, or when multiple subagents would edit the same files. Prefer `executing-plans` when work should run in a parallel session instead.

Outputs commits against the plan, a final-pass code review, and a branch handed off to `finishing-a-development-branch`. Model selection per role is part of the skill (cheap for mechanical, standard for integration, most capable for review). Pairs with `writing-plans` (upstream), `test-driven-development` (what subagents follow), and `using-git-worktrees` (the isolated workspace).