November 23, 2026CodingSkillsOpen Source

app-store-compliance Gives Your Coding Agent an App Review Rehearsal

Coding agents made shipping an app fast. They did nothing for the week you lose when Apple sends back a one-line rejection. app-store-compliance is aimed squarely at that gap, and GitHub noticed: 617 stars on Sunday's daily trending board, close to 3,920 total, with v2.5.0 and v2.5.1 both landing on Sunday. https://github.com/mjmirza/app-store-compliance

It is two things in one repo. A playbook with 124 documented App Store and Google Play rejection patterns, each mapped to the guideline and the fix, plus 112 tracked deadlines across the EU AI Act, DSA, COPPA and age laws. And a bash guard you point at your Xcode or Android folder. It reads the project, flags privacy manifest and Data safety mismatches, and ends with a verdict: BLOCKED with file and line for each critical risk, or CLEAR. There is also a skill to install it into Claude Code, Codex, Gemini CLI, Cursor and others, so the check runs on every submit instead of when someone remembers.

The part worth respecting is how plainly it states its limits. CLEAR means no critical risk in your code, not an approval, and the README says outright that it does not guarantee approval. The top two modern rejection causes, a working demo account with a live backend and a privacy declaration that matches what every SDK really does at runtime, are things a scan cannot fully see, so they go into a checklist and a readiness file a human fills in. If the rule pack is more than 60 days old the report says so. Very few agent tools are this honest about what they cannot prove.

v2.5.0 folds in Apple's November 16 and 17 developer posts and roughly 300 replies from X, and the release notes say 14 Codex review rounds attacked the changes before release, until one found nothing. New deadlines include Apple's China mainland AI-generated-content declarations by January 31, 2027 and Google Play trader status by February 1, 2027. One caveat: the license is OpenRoots ORA 2.3, not a standard OSI license, so read it before you wire this into a commercial pipeline.

This is a good template for what an agent skill should be. Not a prompt that makes the model sound like a reviewer, but domain rules kept current, a hard verdict, and a clear line where the machine stops and a person takes over.
← Previous
Worker 755 Makes You the Subagent, and the AI Writes Your Review
Next β†’
Yuanshi Raises ~$42M Pre-A for an AI Virtual Cell That Picks Its Own Experiments
← Back to all articles

Comments

Loading...
>_