New app: one page in plain English about what it is, who it is for and what done looks like. Existing app: connect the repo, and the engine reads it the way a new engineer would.
It breaks the work into milestones and small, checkable pieces, sizes them, and keeps refining the plan as the app grows.
Several AI builders take pieces at once. Tests run first; a second AI reviewer must approve every meaningful change before it lands.
Spending, anything the public will see, and open questions park for your tap. Answer from your phone. Everything else keeps moving.
TestFlight and Google Play testing tracks, quick updates between store builds, and a QA agent tapping through the real app before each release.
Every piece of work is proposed to you first. Good for your first week, or your first app.
Work runs unasked; anything touching money or the public, and any open question, waits for you.
Your standing consent: every gate clears, spend stays inside the caps you set.
Skip this if you just want an app. Read it if you want to know what you are trusting.
Expo / React Native projects. Real git commits on real branches; the engine never rewrites history. Bring one repo or a whole portfolio, each with its own engine.
Lint and tests before review. An adversarial reviewer on every non-trivial diff. A baseline test gate that blames only new failures. Per-change and daily spend caps you set.
Builds go through your Apple, Google and Expo accounts; the AI runs on your Claude subscription. Nothing is resold, nothing is metered, nothing leaves your name.