Use RealityGate with Superpowers
Short answer: Superpowers improves how the agent works: brainstorming, plans, test-driven
development and code review all run as skills. RealityGate adds checks that don’t depend on the
agent’s judgment. check fails on stubs and invented packages even when a review missed them,
and selftest proves that the gate can fail at all. Install both plugins:
/plugin install superpowers@claude-plugins-official
/plugin marketplace add usmanmanii/realitygate
/plugin install realitygate@realitygate
Where they meet
| Superpowers skill | Add |
|---|---|
test-driven-development |
Nothing: keep red-green-refactor. RealityGate’s NO-TESTS and LOW-TEST-RATIO checks are a backstop |
verification-before-completion |
npx realitygate check must exit 0. For anything users touch, also run npx realitygate audit --exec, which actually runs the tests and a clean-install boot |
requesting-code-review |
npx realitygate review-pack <task-spec.md>, to get a second opinion from a different model family |
finishing-a-development-branch |
npx realitygate check --changed --base main before merge |
Setup
npx realitygate init
npx realitygate hooks install # block .env access, force-push and commits while the gate is red
npx realitygate selftest --gate "npx realitygate check"
Add to CLAUDE.md:
Verification before completion includes
npx realitygate check(exit 0) and, for features,npx realitygate audit --exec. A check marked NOT RUN is not a pass.
What each catches that the other doesn’t
A review skill reads the diff, and a convincing diff can talk it into approving. A pattern rule can’t be talked into anything, but it can’t judge design either. RealityGate catches the mechanical fakes:
throw new Error("not implemented")- a package that doesn’t exist
catch {}- an env var nobody documented
- an exported function nothing calls
Superpowers’ reviews catch the rest.