AugmentClaude

App Review Board

Predict App Store approval odds before submission using mock reviewers.

Installation

  1. Make sure Claude is on your device and in your terminal.

    Skills load from ~/.claude/skills/ when Claude Code starts up — so you need it on your machine first. If you don't have it yet, install it once with the command below, then run claude in any terminal to verify.

    One-time setup
    npm i -g @anthropic-ai/claude-code

    Already have it? Skip ahead.

  2. Paste into Claude Code or into your terminal.

    This copies the whole skill folder into ~/.claude/skills/app-review-board-savannahfeder/ — the SKILL.md plus any scripts, reference docs, or templates the skill ships with. Works for every skill.

    What this command does
    1. Makes a new, empty temporary folder.
    2. Downloads the skill's folder, .claude/skills/app-review-board, from github.com/savannahfeder/agentbox (branch main), the GitHub repository it's listed from. You get the version that is there now, not a copy we keep.
    3. Copies that folder to ~/.claude/skills/app-review-board-savannahfeder/.
    4. Deletes the temporary folder.

    Installing runs none of the skill's files; it only downloads and copies them. When you later use the skill, Claude follows its SKILL.md and may run scripts that come with it. Read them first.

    How we check skills, and what we don't · Report this skill

    Faster alternative (instruction-only skills)

    Skips the clone and grabs only the SKILL.md file. Don't use this if the skill ships Python scripts, reference markdowns, or asset templates — they won't be downloaded and the skill will fail when it tries to load them.

    Quick install (SKILL.md only)
    Sign up to copy
  3. Restart Claude Code.

    Quit and reopen Claude Code (or any other agent that loads from ~/.claude/skills/). New skills are picked up on startup.

  4. Just ask Claude.

    Skills auto-activate when your request matches the skill's description — no slash command needed. Trigger phrases live in the skill's own frontmatter; you can read them in the “What this skill does” section above.

Prefer to read the source first? Open on GitHub.

When Claude uses it

Judge whether the phone app's App Store submission will be accepted first time, and refuse to submit under 90%. Runs independent mock Apple reviewers (Codex and Claude) on the store build's source and checks live what a reviewer meets (privacy, terms and Help pages, the demo account). Use before any App Store submission or resubmission of the phone app, after a rejection, or when asked for the odds of approval.

What this skill does

App review board

scripts/app-review-board.mjs gives the odds that the phone app's next App Store submission is accepted first time, and says Submit only when all of these hold:

  • the odds are 90% or more,
  • at least two reviewers answered,
  • every live check passed. A check that could not run counts as failed.

The odds are the reviewers' judgement, anchored to Apple's published rejection rate. They are not a measurement, so say so when you report them.

Run it

REVIEW_EMAIL=… REVIEW_PASSWORD=… node scripts/app-review-board.mjs --codex 2 --claude 2 --out <dir>
  • On the founder's Mac, Codex runs on the second login: set CODEX_HOME=~/.codex-2.
  • Never print the demo password or write it to a file in the repository.
  • Where claude -p is not signed in (a cloud run), pass --claude 0. Then give <dir>/prompt.txt (made by --prompt-only) to two Claude helpers. Have each write its answer to a file, and add each file with --with <file>.
  • <dir>/report.md is the verdict: the risks, highest first, each with its fix. The exit code is 0 only for Submit.

The review computer

While Apple reviews the app, the demo account needs a computer signed in to it. On the cloud alone, a thread goes quiet 20 minutes after an answer.

  • scripts/app-review-computer.mjs on makes one small AWS server that does this. It reads the demo password from this Mac's Keychain.
  • login signs Codex in on the server. It prints a code to type into a browser signed in to the Codex account the founder keeps for this. Never use the founder's own Claude or Codex account: "we're not giving other people access to my account".
  • status shows the server and its end date. off deletes it at once.
  • What the first one (2026-11-08) taught:
    • Only the desktop app sends threads to the phone. The headless npx agentbox-app copy never does. So the server runs the desktop app on a virtual screen (xvfb), built from the public repository.
    • It signs in to the demo account only when ~/.agentbox/team.config.json holds the cloud's public address. The password comes from AGENTBOX_TEAM_TEST_LOGIN.
    • The desktop app has no screen to click, so make the project and its first thread in the headless app over an SSH tunnel to port 7777. Then set accountId in ~/app/zero.config.json to the account that holds them.
    • Give git a demo identity on the server, or the agent's first answer asks the reviewer for a name and email.
    • The proof that it works: reply on the demo account from app.agentbox.ac, and an answer comes back.
  • It always ends. The end date is 14 days unless --days says otherwise, and 30 at most. The machine shuts itself down then, AWS deletes it on shutdown, and an AWS schedule deletes it too.
  • The founder asked for it to be turned off for certain. So when you turn it on:
    1. File a thread with runAt at the end date plus an hour: "Check the App Review computer is gone". It runs status, and off if anything is left.
    2. Run off the day the app is approved or rejected for good, and say so on the App Store thread.
    3. Never run on again without saying so on that thread.

Use it as a loop

  1. Run the board.
  2. Fix every risk at 5% or more that code, the listing or the review notes can fix.
  3. Run the board again with fresh reviewers. Do not re-ask the same ones.
  4. Repeat until it says Submit or what is left is only the person's to do.
  5. The synthetic reviewer on the iOS simulator is one more reviewer. What it hits counts as a risk of at least 30% until fixed.

Related skills