Overview
REQUIRED SUB-SKILL: Invoke superpowers:executing-plans — it owns the actual execution/review-checkpoint process. This skill is a required layer on top: how mechanically to trust the plan, and what not to do around commits.
Rules
- Reread the plan before executing, and check it for contradictions with the repository's current state — the repo may have moved since the plan was written.
- A detailed plan is ready for mechanical execution. Do not ask a subagent to redesign the solution, or reread the whole plan itself, when the plan already contains exact steps and interfaces — that's re-litigating a decision already made at plan-writing time.
- Execute tasks in the order the plan defines, not reordered for convenience.
- Run the plan's single final verification pass after all tasks — not per-task, not skipped. This is the full build/type-check/lint/whole-suite pass; it's separate from TDD's own per-step scoped test runs (verify-red/verify-green) inside each task, which are required throughout, not deferred to this pass — don't skip those while waiting for the final one.
- No
git add/git commit/PR creation without a direct request from the user — see therepo-workflow-conventionsskill for the general rule this instance follows.
Common Mistakes
| Mistake | Fix |
|---|---|
| Starting execution without rereading the plan first | Reread it — the repo may have drifted since it was written |
| Asking an implementer subagent to redesign a step the plan already specifies exactly | Trust the plan; only escalate if it's genuinely wrong for the current repo state |
| Running the full build/lint/test suite after each task | Wait for the plan's single final verification pass — but still run TDD's own scoped single-test verify-red/verify-green inside each in-scope task, that's not the same thing |
| Committing after finishing all tasks without being asked | Leave changes uncommitted; the user commits or asks explicitly |