BD TECH.

Open source · App Factory

Skills that take an app from code to the store

42 Claude Code skills that build, describe, ship and announce apps — to the App Store and Google Play, and outside them, on the Mac and on Windows. You turn on the tracks a repo needs; a session opened there carries them, and a session opened anywhere else does not.

Install Source on GitHub Download .zip
  • MIT licence
  • Claude Code plugins
  • 42 skills · 8 tracks
  • Python 3.9+ · standard library

Install

A marketplace, then the tracks one repo needs

Terminal
claude plugin marketplace add yybd/app-factory
cd ~/code/my-app && python3 ~/.claude/plugins/marketplaces/app-factory/factory/enable.py
# then the install lines it prints

enable.py reads the repo, proposes the tracks that fit it, and prints the lines that turn them on. The quickstart walks the whole path in fifteen minutes, with one app and a check after every step.

Built for Claude Code, and tested only there. The scripts the skills run are ordinary command-line tools that any agent can call, but the loading (tracks, the marketplace, enable.py) is Claude Code's plugin system.

The problem

Nothing is global

The usual way to steer a coding agent is one large instruction file. It does not hold: the file grows past what stays in view, rules contradict each other, and the agent proposes an Android deploy while you are editing a website. The failure is not that the agent forgets. It is that everything was told to it at once, whether or not it applied.

So each skill is a document the agent loads when its description matches what you asked for, and only the descriptions are always in context. Having all 42 available costs about 8,400 tokens of descriptions, and only in a repo where you turned them on.

The tracks

Eight tracks, and one that sets them up

apple-track

The App Store, the Mac App Store, and direct distribution outside both.

android-track

Google Play, and getting a build onto a device.

shared-track

The work that is the same either way: naming, copy, release order, prices.

web-track

A site's structure, an audit of it, and whether search finds it.

capacitor-track

Web-in-a-shell apps, where the bugs live in the seam.

design-track

Visual craft for any HTML interface.

windows-track

A Windows installer that updates itself, outside any store.

announce-track

The posts that announce a release or an article, checked per channel — and never posted for you.

factory-setup

On everywhere. It says when a repo looks like an app and no track is on.

The full list, with where each skill loads, is in SKILLS.md.

How a release runs

Executors do one thing. Orchestrators keep the order.

Executors

play-store-ship builds, signs and uploads an AAB. app-icon-generator writes every icon size from one image. One job each, done the same way every time.

Orchestrators

prepare-app-release drives identity, compliance, media, copy, the version bump, the build and the upload, and hands the website step off explicitly rather than doing it silently. A release rarely fails inside one step; it fails in the order, and in the step nobody remembered.

Before you enable it

What it will do to your repos

  • It writes files in the repo you point it at, and reads credentials in place without copying or printing them.
  • It uploads to the stores only at the end of the skills whose whole job that is.
  • It never submits an app for review, changes a price, or pushes.
  • Nothing is needed to install it. The Apple track needs macOS; the Android, web, design and cross-store tracks also run on Linux and Windows. Individual skills need what they need (Xcode for Apple, a JDK for Android, fastlane for either store), and each says so and stops rather than failing halfway.

The complete list is TRUST.md, written to be read before you enable anything. The licence is MIT, except two skills that came from elsewhere and keep their own.

Writing

Articles about App Factory

  1. An app factory: 41 skills that take an app from code to the App Store and Google Play

    Why one big instructions file stops holding past the third app, and what replaced it. October 2026.

    he
  2. From vibe coding to spec-driven: an AI-native architecture built on specs, skills and self-verification loops

    Specification, skills that run deterministic checks, hooks that enforce them, and a loop that repairs before a human reviews. April 2026.

    he

More open source

From the same desk

  • RepoGuards

    Claude Code hooks

    The companion for work that spans several repos: it refuses a blind commit or a push into a repo the session is not standing in. Optional; nothing here requires it.

    Project page

  • App Store Reviews to Telegram

    Self-hosted · Node.js

    A dashboard and Telegram bot that tells you the moment a new review of your Mac or iOS app is posted.

    Project page

Contact

Questions, bugs, or a skill you want

Issues and pull requests are welcome on GitHub. Email reaches the developer directly.

support@bdtech.app