App Store Review Skill

Install

npx skills add ElxMaj/app-store-review-skill

Then ask your agent:

Audit this iOS app before submission. Report first and do not edit files.

The skill works with Codex and Claude Code, detects Xcode, Expo, React Native, and Flutter projects, and changes zero files on its first pass. See every install method, including Tessl and the Claude Code marketplace.

Why a review gate?

A build can work perfectly on your phone and still give App Review a broken path. A widget may have no privacy manifest. A camera call may not match its purpose string. Account deletion may stop inside the app. Review notes may omit the only usable test account. A polished interface may still look like an interchangeable template.

The skill collects those signals before offering advice. Every finding answers four questions:

Unknowns stay unknown. The report arrives before any proposed fix group.

Three review modes

01 · Pre-submission audit

Checks native targets, purpose strings, privacy manifests, Required Reason APIs, purchases, restore paths, account deletion, Sign in with Apple, ATT, third-party AI consent, typed metadata, placeholders, shipped internal files, reviewer access, iPad behavior, and release paths.

02 · Rejection recovery

Preserves Apple’s exact message and classifies the next move as FIX, CLARIFY, APPEAL, or REQUEST INTERPRETATION. It traces the likely cause, keeps published rules separate from community evidence, and drafts a Resolution Center reply with exact navigation and build details.

03 · Human-craft audit

Reviews product distinction, provenance, visual identity, microcopy, missing states, accessibility, motion, and App Store product-page specificity. Grades are DISTINCT, CREDIBLE, GENERIC, HIGH RISK, or UNVERIFIED.

The artifact

The reviewed JSON becomes a self-contained editorial report, not a fake dashboard. It includes the release verdict, evidence confidence, reviewer checklist, App Review Notes draft, and approval-required fix groups. The same evidence is available as stable JSON for CI, issue reports, and later audits.

Use it

/app-store-review Audit this Expo app before submission. Report first and do not edit files.
/app-store-review Check this release archive for iOS 27 engineering, design, and product-page readiness.
/app-store-review Apple rejected build 42 under 4.3(a). Find the cause and draft my reply.
/app-store-review Run the human-craft audit. Show what feels generic or unfinished.
/app-store-review Audit this iPad app's Apple design quality, including motion and gestures.

For deterministic scanning outside an agent:

python3 scripts/app_store_review_scan.py <project-path> --format all --output-dir <report-directory>

Add typed metadata with --metadata FIELD[:LOCALE]=PATH, Fastlane trees with --metadata-root PATH, and an IPA, ZIP or .xcarchive with --archive PATH.

Coverage

Apple’s published guidelines do not name AI-written code as a rejection category. They do assess the product that ships. Read the source-based guide to Guidelines 4.2.6 and 4.3.

iOS 27 coverage

Version 2.0.0 adds engineering, design and product-page review across 23 technology families. The readiness guide and Apple source register cover Foundation Models/PCC, Core AI, Siri/search, native interfaces, widgets, background assets, media, games, health, identity, tracking and commerce.

The scanner reads each app/extension plist from an IPA, ZIP or .xcarchive, checks dated SDK/Xcode/deployment gates and SDK 27 launch-screen declarations, and routes runtime, scenes, accessibility, adaptive/Duo layouts and store assets to manual verification. Optional API adoption is not a submission requirement.

JSON schema 1.2 adds platform_review and dated release_watchlist context; earlier reports remain readable. All formats preserve beta/future scope and print coverage. No project code, build, app or network request is executed; static evidence cannot prove runtime compatibility, signing or asset approval.

Keep the evidence current

references/research-prompt.md is a quarterly research prompt for checking official policy changes and refreshing rejection evidence. New cases must include the guideline, Apple’s wording when available, the response, outcome, date, and source URL.

Project map

SKILL.md                          Workflow and review rules
references/                       Policy, recovery, craft, and framework guidance
scripts/app_store_review_scan.py  Read-only deterministic scanner
scripts/ios_platform_review.py    Archive gates and iOS 27 evidence routing
scripts/render_app_store_report.py Self-contained report renderer
scripts/tests/                    Regression tests
evals/                            Behavior and trigger cases
.claude-plugin/ · .codex-plugin/  Claude Code and Codex manifests

Contributing

After changing canonical resources, run python3 .github/scripts/package_skill.py to refresh the Copilot package and portable archive, then run the test suite. The packaging step is deterministic and does not publish a release.

Real rejection cases and false-positive reports are welcome. Read CONTRIBUTING.md first, and never publish credentials, signing material, personal data, or private app details.

Built by Elie Majorel · Published by Maj Labs · MIT licensed

This project checks public Apple guidance and clearly labeled community evidence. It cannot guarantee approval, does not contact Apple, and is not affiliated with or endorsed by Apple Inc.