2.0 KiB
Superpowers Rules (Always-On)
These rules apply to ALL work unless the user explicitly opts out.
1) Plan gate for non-trivial work
If the task is anything beyond a tiny change, do NOT edit code immediately. You MUST:
- Brainstorm briefly (goal, constraints, risks, acceptance criteria)
- Write a step-by-step plan with verification steps
- Ask the user to approve the plan Only after approval may you implement.
Execute-plan gate (Superpowers parity)
After the user approves a plan, do NOT begin implementation automatically.
You MUST pause and instruct the user to run: /superpowers-execute-plan
Only begin implementation after /superpowers-execute-plan is invoked,
unless the user explicitly says to proceed without it.
What counts as "tiny"?
- single-file change
- obvious edit
- low risk Even then: do a mini-plan (3–5 steps) and include verification.
2) Verification is mandatory
After implementation, you MUST provide:
- exact commands to verify (tests/lint/run)
- and results if you were able to run them
3) Prefer TDD / regression tests
- If fixing a bug: add a regression test if practical
- If adding behavior: add/adjust tests when practical If tests aren’t feasible, provide a concrete alternative verification path.
4) Review pass required
Before final response, do a review pass and list issues by severity:
- Blocker / Major / Minor / Nit
5) Safety
- Never log secrets
- Add timeouts, retries, and idempotency for API automations
- Fail safe (no silent data loss)
Artifact persistence (mandatory)
Any brainstorm, plan, review, or finish output must be written to disk under:
artifacts/superpowers/
Do not leave these as IDE-only documents. After writing, confirm the file exists.
Persistence enforcement
When a workflow requires saving an artifact to artifacts/superpowers/, you MUST ensure the file exists on disk.
Preferred method: use python .agent/skills/superpowers-workflow/scripts/write_artifact.py --path <...>.
If you cannot execute commands, instruct the user to save the output manually.