agentleFS
whinstudiowhinstudio/skills/game-release/SKILL.md

game-release

Skill0 starsChanged 3 days agoMITView on GitHub

Use when a Unity 6 mobile game generated from the cmpose template is about to go to Google Play or the App Store, when asked whether the game is ready to ship, for a store release checklist, or after launch for a post-mortem.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 1 is distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Every row below is one of two kinds:

    line 12
  2. Very specific:

    Do not mark a row PASS from memory, from the README, or because a similar row passed.

    line 20
  3. Very specific:

    No command here builds the signed archive.

    line 83
  4. Very specific:

    When app-store-rejection-checker or play-store-rejection-checker is installed (check each folder with test -d ~/.claude/skills/<name> and …

    line 107
  5. Very specific:

    Write qa/release-<date>.md (date +%F, with -2, -3 and so on appended when the file exists).

    line 115

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Game release

Checks that a game is ready for the stores by reading its real files and running exact commands, then writes a verdict. It writes only qa/release-<date>.md; row 6 also writes Builds/ and may touch Library/. The post-mortem is a separate mode: post-mortem.

Every row below is one of two kinds:

  • Checked: a command or file read in this project. The result is PASS, FAIL or NOT RUN. Rows 12 and 13 only state a value and have no failure: their result is stated. A row that does not apply to this game is n/a; WARN is a stated concern that is not a failure.
  • Needs the owner: a store console, a real device or legal text. It is NOT ASSESSED unless the user confirms it in this conversation; then it is CONFIRMED and says who confirmed.

Do not mark a row PASS from memory, from the README, or because a similar row passed.

Verdict

First match wins:

  1. NOT READY: any checked row is FAIL.
  2. NOT ASSESSED: any row is NOT RUN, or an owner row is unconfirmed. Name each one.
  3. READY: every row is PASS, PASS (debug-signed), WARN, stated, n/a or CONFIRMED.

WARN, stated and n/a never change the verdict. (PASS (debug-signed) counts as PASS; row 5 fails in that case, so rule 1 applies first.)

0. Project

The project is the current directory unless the request gives a path; cd into it first, because the commands below use relative paths. Assets/ and ProjectSettings/ must exist, and the editor version (ProjectSettings/ProjectVersion.txt) must start with 6000.; otherwise stop with NOT ASSESSED and write nothing. Read git status --short and git rev-parse --short HEAD; a dirty tree is stated in the report. The flags (ads, purchases, insights, ...) are the list under flags in .cmpose/generation.json. The shipped platforms come from the Stores line of design/brief.md, which game-brief writes as Google Play, App Store or both: Google Play or android → Android, App Store or ios → iOS, both → both (case does not matter). No brief, no Stores line or any other value → both, and the report says so and quotes the value it found.

1. Checked rows

#RowCheckFAIL when
1Smoke reportNewest qa/smoke-*.md by modification time (ls -t qa/smoke-*.md | head -1; a name sort misorders the -2, -3 runs). For each shipped platform its verdict line and Visual line. Its Commit field (on the Project: line) against git rev-parse --short HEAD, and find Assets ProjectSettings -type f -newer <report> -not -path 'Assets/Settings/*' -not -name ProjectSettings.asset -not -name GraphicsSettings.asset -not -name ShaderGraphSettings.asset -not -path ProjectSettings/AndroidResolverDependencies.xml -not -path ProjectSettings/GvhProjectSettings.xml -not -path 'Assets/Plugins/Android/*.aar*' -not -path 'Assets/Plugins/Android/*.jar*'. Unity and the External Dependency Manager write those excluded files by themselves on every test run and build (row 6 included): the resolver's dependency list and settings, and the .aar / .jar files (with their .meta) it copies into Assets/Plugins/Android/. So a change to them does not make the report stale. Row 1 is judged before row 6's build; a second release check run after an earlier check's build does not fail row 1 for the files that build wroteNo report, or stale (the Commit field differs from HEAD, ignoring a + uncommitted suffix, or the find prints any file; directories are not files), or a shipped platform's verdict is not PASS / PASS WITH WARNINGS, or its Visual is not DONE, or that platform is missing from the report (Overall is not tested)
2VersionbundleVersion and AndroidBundleVersionCode in ProjectSettings/ProjectSettings.asset; the iOS buildNumber map in the same file. Ask the user once in this run whether the game was released beforeA field other than the iPhone buildNumber is empty; or the game was released before and the version or code is the value of the last release (ask for it) or the template's 1.0.0 with code 1. First release: 1.0.0 with code 1 is PASS. No answer to the question (for example a non-interactive run): NOT RUN, and the report says so. An iPhone buildNumber of 0 or empty is a WARN (stated in the report, not a failure): set it to 1 or higher before archiving
3Package and bundle idapplicationIdentifier (Android and iPhone lines) in the same fileAn id starting com.example., the template's own id. The two ids differing is a WARN, stated in the report
4Scripting and ABIUnder scriptingBackend: the Android: value is 1 (IL2CPP), and AndroidTargetArchitectures includes ARM64 (bit 2)Android: is 0 or the key is missing (Mono), or no ARM64
5Android signingkeystore.properties exists, has non-empty storeFile, storePassword, keyAlias, keyPassword, and storeFile exists; when the file exists, git check-ignore keystore.properties prints it (with no file, skip this check: it prints the name anyway). Never print the passwordsMissing file or field: the build falls back to the debug key, which Google Play rejects. File not git-ignored
6Android bundleBuild with -cmposeAab (see below). The log has CMPOSE_BUILD_RESULT with result=Succeeded and the .aab exists. The evidence states the signing used (keystore or debug)Build failed or no .aab. Built without keystore.properties: the result is PASS (debug-signed) and the report says the .aab cannot be uploaded to Google Play
7Test IDs gone: adsIn Assets/_Project/Core/Ads/AdsConfig.cs check only the unit-id constants (Banner*, Interstitial*, Rewarded*, form ca-app-pub-<publisher>/<unit>) for the publisher 3940256099942544; ignore GoogleTestPublisher and the tests, which hold it on purpose. Separately, in Assets/GoogleMobileAds/Resources/GoogleMobileAdsSettings.asset check adMobAndroidAppId and adMobIOSAppId (form ca-app-pub-<publisher>~<app>, with a ~) for the same publisherAny such constant or app id carries it. State the form per file: unit (/) or app (~). Files absent because the game has no ads flag: n/a
8Test IDs gone: purchasesApiKeyApple starts appl_ and ApiKeyGoogle starts goog_ in Assets/_Project/Core/Purchases/PurchasesConfig.cs, and PremiumEntitlement is setA key that does not start so (the template ships RevenueCat's Test Store key, and release builds keep purchases off). No purchases flag: n/a
9Release log linesThe running game prints the markers of the flags whose code exists: CMPOSE_ADS available=true (no "test IDs" warning) with Assets/_Project/Core/Ads/, CMPOSE_PURCHASES available=true with Assets/_Project/Core/Purchases/, CMPOSE_INSIGHTS available=true with Assets/_Project/Core/Insights/; the batch build does not print them. A marker whose code is absent is n/a (a flag can be listed with no code behind it). Only a log from a release build on a device counts; editor play-mode and batch logs are ignored (they always print available=false). Look for a saved device log under qa/ or Builds/logs (grep the markers), no older than the last commit touching Assets/ or ProjectSettings/ (git log -1 --format=%cI -- Assets ProjectSettings). game-smoke keeps only Unity and build logs and PNGs, so such a log exists only when someone saved one, for example adb -s <serial> logcat -d -s UnityA fresh log shows available=false or the warning. No such log: NOT RUN, never PASS; say that the owner confirms them on a real device
10Privacy manifestAssets/Plugins/iOS/PrivacyInfo.xcprivacy exists and parses: plutil -lint <file>Missing or does not parse
11Firebase filesWith the insights flag and Assets/_Project/Core/Insights/ present: Assets/Firebase/google-services.json and Assets/Firebase/GoogleService-Info.plist exist, and their package name and bundle id match row 3Missing or mismatching. No insights flag, or no Insights code: n/a
12Build sizeSize of the .aab from row 6 (ls -l)Never. Result stated: the size, and the owner's budget when one is named
13Uncommitted workgit status --shortNever. Result stated; say it, because the smoke report (row 1) compares against commits

Android bundle command

<ns> is the full namespace after namespace in Assets/Editor/Build/BuildScript.cs, ending in .Editor.Build. Never when Temp/UnityLockfile exists (the editor has the project open): row 6 is then NOT RUN. Another Unity process on a different project is fine; the same project is not. The build can take ten minutes: run it in the background and poll the log for CMPOSE_BUILD_RESULT (the background job's own completion notice is not the build's). Do not use a foreground time limit; a clean IL2CPP build can pass ten minutes.

mkdir -p Builds/logs
"$UNITY" -batchmode -nographics -quit -projectPath "$PWD" -buildTarget Android -executeMethod <ns>.BuildScript.BuildAndroid -cmposeOut "$PWD/Builds/game.aab" -cmposeAab -logFile Builds/logs/release-android.log
grep CMPOSE_BUILD_RESULT Builds/logs/release-android.log

$UNITY is $UNITY_EDITOR, else /Applications/Unity/Hub/Editor/<version>/Unity.app/Contents/MacOS/Unity, with <version> read from m_EditorVersion in ProjectSettings/ProjectVersion.txt. Without keystore.properties this still builds, signed with the debug key; row 5 is the failure and row 6 is PASS (debug-signed).

iOS

No command here builds the signed archive. iOS rows are owner rows (below) unless the user gives you an archive to inspect. Builds/ios from game-smoke is a simulator build and says nothing about signing.

2. Owner rows

List each with what the owner must do. Unconfirmed means NOT ASSESSED.

RowWhat the owner confirms
Xcode archiveExported without -cmposeSimulator, opened Unity-iPhone.xcworkspace (.xcodeproj only when there is no Podfile), chose the team under Signing & Capabilities, archived and validated the archive
Google Play Data safetyForm filled to match what the game collects (with insights: crash logs, diagnostics, app interactions, device ids; with ads and purchases, their SDKs' lists)
App Store privacy answersApp Privacy answers filled; with ads, the App Tracking Transparency text and the IDFA explainer in AdMob
Consent messageWith ads: the AdMob consent message (Privacy & messaging) published
Account deletionOnly when the game has accounts or sign-in: the deletion path inside the app and the Play Console deletion URL
Privacy policyPublished, linked in both stores and from inside the game
Age ratingQuestionnaires answered in both consoles
Store images and listingApp icon (not Unity's default), title, description, screenshots and the Play feature graphic; screenshots for the sizes App Store Connect asks for
Store productsWith purchases: products, the premium entitlement and the shop offering set up in RevenueCat and in both stores
Real deviceThe release build installed and played on at least one real Android phone and one real iPhone, not only emulators
Native symbolsWith insights: Android symbols uploaded (firebase crashlytics:symbols:upload), as the template README describes

3. Store checkers

When app-store-rejection-checker or play-store-rejection-checker is installed (check each folder with test -d ~/.claude/skills/<name> and test -d .claude/skills/<name>; do not use a glob, which aborts the line in zsh when nothing matches), end the report with a line telling the user to run it as the last step. Do not run it here, and do not depend on it: when it is absent, say nothing about it.

4. Report

Write qa/release-<date>.md (date +%F, with -2, -3 and so on appended when the file exists). Do not commit it.

# Release check <date>

Project: <name> | Unity <version> | Commit <short sha, "+ uncommitted" if dirty>

| # | Row | Result | Evidence |
|---|---|---|---|
| 1 | Smoke report | FAIL | qa/smoke-2026-11-01.md: Commit abc1234 differs from HEAD def5678 |
| 5 | Android signing | PASS | keystore.properties complete, git-ignored |
| 12 | Build size | stated | game.aab 41 MB, no budget named |

## Owner rows
| Row | Status |
|---|---|
| Xcode archive | NOT ASSESSED |

iOS: <only when iOS is shipped: what was and was not checked, for example "nothing beyond rows 3, 10">
Verdict: NOT READY
Blockers: <each FAIL, one line, with the fix>
Not assessed: <each NOT RUN or unconfirmed owner row>

Final message to the caller: the verdict, the blockers and the not-assessed list, the report path, and the checker line from section 3 when it applies. A stale or missing smoke report is named as the blocker; do not re-run game-smoke from here, tell the user to run it.

Never

Edit tracked project files, change signing or keys, commit, upload to a store, or mark an owner row done on the user's behalf without them saying so.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.