This page answers the questions developers asked most after the playbook spread on X in November 2026. About 300 replies were read and sorted by theme on 19 November 2026. Every answer points at a file in this repo, so you can check it rather than trust it.
A clean scan and a ready submission are different things. The guard reads your project. A reviewer reads your running build, your store listing, and your review notes. Most painful rejections live in the second half.
| Half | What it covers | How you prove it |
|---|---|---|
| Checked by the scan | Selected patterns in source code, missing purpose-string keys in Info.plist, the privacy manifest, AndroidManifest permissions, Gradle and Xcode build settings, SDK names in your dependencies, placeholder text, signs of non-store payment code. Whether a purpose string fits its feature is a manual check | bash agent-os/hooks/app-store-compliance-guard.sh "/path/to/app" |
| Only a person can prove | A demo account that signs in, a backend that is live and holds test data, review notes that say where each feature is, screenshots of this build for each device family you support, account deletion working on a real device, privacy labels that match real traffic, an in-app purchase the reviewer can reach | .app-review-readiness.json, read by the same guard |
Since version 2.5.0 the guard prints both halves on every run. With no readiness file it says plainly which checks it could not see. Copy the template into your app folder, test each item on the release build, write pass, fail, n/a, or todo, and set checked_on to that day.
fail always blocks, so a broken demo login stops the submission the same way a missing purpose string does."enforce": true in the file, or APP_STORE_GUARD_REQUIRE_READINESS=1, and any check still todo blocks too. Use this in CI.n/a counts only with a reason under notes for that check. The privacy labels check never accepts n/a, because every app has one.checked_on date older than 30 days, missing, or in the future is flagged, because a pass on last month’s build says nothing about this one.The full manual walk is the pre-submission checklist. For the review notes themselves, use the review notes template.
A stale checklist gives false confidence, and several people said that is worse than none. So every report now opens with a line like this.
Rule pack. v2.5.0. Its 109 dated sources were verified between 2026-07-13 and 2026-11-21 (newest 0 days ago). The checks run against this dated snapshot, not the live guideline pages.
Rule pack. 2 of 109 dated sources were last verified more than 60 days ago, so their rules may have changed since. Each one carries its date as verified_on in data/regulatory-deadlines.json.
source_url it was read from and the verified_on day.Rule pack STALE and the command to update. Change the window with APP_STORE_GUARD_STALE_DAYS. If the date cannot be read or compared, it says freshness was not checked rather than staying quiet.scripts/monitor*.py watch Apple’s news feed and the Play policy pages.The playbook covers the Apple App Store and Google Play. People asked whether Play is an afterthought. It is not, as the counts below show (taken from this repo on 21 November 2026).
| Apple | Google Play | Both | |
|---|---|---|---|
Rejection patterns in data/rejection-patterns.json |
62 | 43 | 13 |
| Distinct code checks in the guard | 34 | 38 (20 Play policy, 18 Android build) | 5 |
| Rule map | APPLE.md | GOOGLE-PLAY.md | MISTAKE-PATTERNS.md |
On Play, the guard reads every AndroidManifest.xml for sensitive permissions (background location, SMS and call log, all files, accessibility, photos and videos, foreground service special use), checks target SDK and 64-bit ABIs in Gradle, looks for Play Billing on digital goods, and flags analytics or ad SDKs so you verify the Data Safety form. A screenshot that does not match the build is a manual check on both stores.
The Mac App Store shares Apple’s guidelines and is covered where the code is shared. Notarization and Developer ID for apps outside the store are in PLATFORM-MECHANICS-2026.md.
Not covered by the guard. the rules of Huawei AppGallery and the other Chinese Android stores, Samsung Galaxy Store, and the rest. Their review rules are summarised in OTHER-STORES.md for reading, with no code checks behind them. A new personal Google Play account also needs a closed test with at least 12 testers for 14 days in a row before production. The guard cannot see Play Console, so that one is a checklist item.
Many replies asked about 4.3, often after eight or ten rejections. A static scan cannot decide 4.3. Apple judges it on whether your binary, metadata, and concept look like apps already in the store, including apps from other developers, and that store is not in your repo.
What the playbook does for 4.3 and 4.2.
What it cannot do. Tell you that your concept is too close to someone else’s. Treat any tool that claims it can with suspicion.
Taken from rejection reasons developers described in public replies in November 2026. They are the developers’ own accounts, not verified rejection letters. Where a reason was vague, the row says so.
| Reported reason | Caught by | Strength |
|---|---|---|
| Missing privacy manifest, purpose strings that did not match | Guard. APPLE-PRIVACY-MANIFEST-MISSING and APPLE-5.1.1-MISSING-USAGE-DESCRIPTION for a missing key, plus scripts/validate-privacy-manifest.py. Whether the wording fits the feature is not checked |
Scan for missing keys, a person for the wording |
| Onboarding button said “Allow permissions” before the location prompt, Apple wanted “Continue” | Guard. APPLE-5.1.1-PRE-PROMPT-ALLOW-BUTTON, new in 2.5.0, when the button and the permission request sit in the same file | Scan, medium, reported as possible |
| Reviewer could not log in | Guard. APPLE-2.1-STAGING-BACKEND for a staging or localhost backend, plus demo_account in the readiness file |
Scan flags part, a person proves the rest |
| Reviewer could not find the in-app purchase | iap_reachable_by_reviewer and review_notes in the readiness file |
Person |
| Five of six rejections were account information and the privacy policy | Guard. APPLE-5.1.1-MISSING-PRIVACY-POLICY and GOOGLE-MISSING-PRIVACY-POLICY, scripts/metadata-audit.py for the listing URL |
Partial |
| Naming Google Play or an Android API inside the iOS app | scripts/metadata-audit.py APPLE-2.3-CROSS-PLATFORM-REFERENCE on listing text. Strings inside the binary are not checked |
Partial |
| AI-built web wrapper bounced under 4.2 | Guard. IONIC-4.2-THIN-WRAPPER for WebView shells | Scan, heuristic |
| 4.3(a) and 4.3(b) spam, duplicate app, saturated niche | Not decidable by a scan. See section 4 | Guidance only |
| UI not “1:1 and consistent across screens” | Not covered. The reason as reported is too vague to map to one guideline | None |
| iPad layout and iPad screenshots | screenshots_match_build in the readiness file. Layout on iPad is not scanned |
Person |
| Trial or intro offer text shown to an Apple ID that already used it | Not scanned. Check Product.SubscriptionInfo.isEligibleForIntroOffer (StoreKit 2) before you show trial wording, and state what is charged after the trial (3.1.2(a)) |
Guidance only |
| Account deletion that let the account sign back in | Guard. APPLE-5.1.1-NO-ACCOUNT-DELETION and APPLE-ACCOUNT-DELETION-WEAK, plus account_deletion_device_test in the readiness file |
Scan finds the flow, a person proves it works |
| Permissions pulled in by a third-party SDK | Guard. APPLE-5.1.2-MISSING-ATT for tracking SDKs, APPLE-PRIVACY-MANIFEST-MISSING for required reason SDKs, GOOGLE-DATASAFETY-MISMATCH for analytics and ad SDKs | Partial |
Rows marked None or Guidance only are the honest gaps. If you were rejected for one of them, open an issue with the guideline number from the letter and it becomes a check or a documented pattern.
CLEAR. means no critical risk in what the scan can see. It does not mean Apple or Google will approve. Reviewers are people, and a submission can still be rejected on a judgment call. What the playbook saves is the extra round after a rejection you could have prevented, not the queue time itself.
On queue time, Apple’s App Review page says at least 50 percent of submissions are reviewed in under 24 hours and 90 percent in under 48 hours (Apple). Developers in the replies saw anything from same day to five weeks, longest for new apps and for the round after a rejection. If a critical bug fix is stuck, Apple accepts expedited review requests with the steps to reproduce the bug.
Run by hand, the report ends with a verdict, and the exit code matches it. A Next. line with the step to take can follow it.
| Exit | Verdict line | Meaning |
|---|---|---|
| 0 | CLEAR. |
No critical risk in what the scan can see. Not an approval |
| 2 | BLOCKED. |
A critical risk in the code, or a manual check recorded as failed |
| 1 | NOT CHECKED. |
No iOS or Android project was found, so nothing was proven. Treat it as a failure in CI, never as a pass |
APP_STORE_GUARD_OK=1 turns a block into exit 0 with the line APP_STORE_GUARD_OK set., and the run is logged. Use it only for a finding you have checked by hand.
As a hook inside an AI coding tool, the guard runs only when that tool is about to run a submit command (fastlane, eas submit, xcrun altool, a Gradle release bundle, and the others in the README). Two things follow.
NOT CHECKED result lets the command through by default, so a hook never blocks work in a folder that is not an app. Set APP_STORE_GUARD_BLOCK_UNCHECKED=1 to block it instead.bash /path/to/app-store-compliance/agent-os/hooks/app-store-compliance-guard.sh "$PWD"
In Fastlane, the same command in a before_all block with sh(...) stops the lane, because sh fails on a non-zero exit.
Read the scripts before you trust them, whatever the star count. Here is what to look for.
python3 only for local helpers in this repo (the deadline list, the readiness file, a manifest parser).scripts/monitor*.py do fetch Apple and Google pages, and only when you run them.scripts/install.sh writes to your AI tool’s config folder and backs up any settings file it changes. Run bash scripts/install.sh --dry-run first to see every write.Found a security issue in the scripts? Report it privately as described in SECURITY.md.