A Claude skill that makes Claude act as a human Apple App Review reviewer: it drives the iOS Simulator with computer-use tools, installs and launches a real build, and exercises the app hands-on the way App Review actually does — then writes the verdict in Apple's rejection-letter style.
The point is to catch what a static code review can't: crashes on launch, broken or placeholder screens, dead-end logins, permission prompts shown with no context, missing in-app account deletion, missing report/block on user-generated content, paywall/IAP problems, and layouts clipped behind the notch / Dynamic Island / home indicator.
SKILL.md— the skill: prerequisites (getting a current build onto a booted simulator), the review loop, severity mapping, and the App-Review-style report format.runtime-checklist.md— the guideline-by-guideline hunting list: what to actually do on the simulator and what would-reject behavior looks like for each (Guidelines 1.2, 2.1, 3.1, 4.8, 5.1.1, etc.).
Drop the folder into a skills directory your agent reads:
- Claude Code / Cowork: copy into
.claude/skills/apple-reviewer-simulator/(keepSKILL.mdat the root of that folder). - It triggers on requests like "review my app the way Apple would", "test my app on the simulator like a reviewer", "will this pass App Review?", or "act as the App Store reviewer."
A real reviewer never reads your source — they hold the build and try to break it. This skill reproduces that: request access to the Simulator, cold-launch the app, walk every screen like a suspicious first-time user, deliberately drive the high-risk flows (login + Sign in with Apple, each permission prompt, account deletion, UGC report/block, purchases), screenshot the evidence, and report each finding with its guideline number, repro steps, and a binding/metadata/note severity.
Because building the app needs Xcode, the build/boot step is run by you (e.g. npx expo run:ios); the skill takes over from the running simulator.
MIT