Implement plan.md tasks in order by delegating both coding and AC review to subagents. Use when plan.md already exists and each task should loop implement -> review -> rewrite task-n-review.md until COMPLETE, where n must be replaced with the actual task number, without running final test-plan verification.
plan.md.test-plan.md should not run yet.plan.md is identified.task-n-review.md report written or overwritten by a review subagent, and n must be replaced with the actual task number.COMPLETE.plan.md to use. If more than one candidate fits, ask the user instead of guessing.runSubagent for each implementation attempt.runSubagent for each review attempt.COMPLETE, run another implementation subagent for that same task, then another review subagent, and overwrite the same report file.test-plan.md verification in this skill.task-n-review.md files at its root, keep that convention, but replace n with the actual task number when creating or overwriting the file.reviews/task-n-review.md inside the same spec folder, with n replaced by the actual task number.plan.md. If ambiguous, ask the user.plan.md and the sibling goal.md.task-n placeholder with the actual task number in both file path and visible document title.COMPLETE, move to the next task.COMPLETE, feed the blocking findings back into a new implementation subagent for the same task, then rerun the review subagent and overwrite the same report.COMPLETE, then continue.test-plan.md here.plan.md first instead of silently doing a mega-task.task-n-review.md report with n replaced by the actual task number in the file path and visible title.COMPLETE.