GitHub Project BAMA Promises to Build Any macOS App Using AI‑Driven Agent Skills

Overview

The open‑source repository BAMA (Build Any Mac App) announced on GitHub that it can *research, clone, design, build, optimise, and test* native macOS applications using AI‑powered agent skills written in Swift and SwiftUI. The project’s README describes the tool as a “no‑Xcode‑GUI” workflow that lets a developer point the agent at an existing app and receive a fully functional copy that can be dropped into the Dock.

How It Works

According to the repository documentation, the agent follows a multi‑step pipeline:

• App discovery – The agent opens the target app, captures screenshots of every window, and creates a *screen index*.
• User‑story generation – Screens are turned into user stories with explicit acceptance criteria, grouped into milestones.
• Incremental commits – Each story is built in a separate Git commit. Before committing, the agent snapshots the UI in both light and dark modes and runs automated tests.
• Skill orchestration – Users invoke the process with the command `/bama`. The agent then selects the next appropriate skill from a library of pre‑written scripts.

The README notes that the workflow works with several large‑language‑model back‑ends, including Claude Code, Codex, Cursor, and Gemini CLI. For reverse‑engineering native macOS binaries, the documentation recommends using Claude Pro or Claude Max for “computer use” capabilities.

Ready‑Made Build Plans

The `specs/` folder ships with 12 starter plans for small‑scale macOS utilities, such as:

• Window snapper
• Window switcher
• Application launcher
• Clipboard history manager
• Screenshot tool
• Screen recorder

These plans were extracted from the author’s own menu‑bar app, Onyx, and can be copied into an empty directory. The agent then builds the chosen plan automatically.

Technical Requirements

• Xcode 26 (or later) with the Swift 6.2 toolchain – the toolchain is used behind the scenes; developers never open the Xcode UI.
• skills.sh – a helper script that installs all agent skills or individual ones via the `--skill` flag.
• A macOS machine capable of running the required LLM back‑ends.

Related Projects & Inspirations

BAMA’s “any” promise is traced to the REA (Reverse Engineer Anything) concept by morluto, which demonstrated that an AI agent could reverse‑engineer arbitrary software. BAMA adapts that idea specifically for macOS. The skill format follows conventions set by Matt Pocock, and the project is reported to work well with SwiftUI Pro (Paul Hudson) and SwiftUI Expert (Antoine van der Lee).

Practical Implications

• Rapid prototyping – Developers can generate a working macOS prototype from an existing app in minutes, potentially accelerating UI redesigns.
• Learning aid – New SwiftUI learners can inspect the generated code to understand how common UI patterns are implemented.
• Automation – Teams could embed BAMA in CI pipelines to keep a reference implementation of a UI in sync with design changes.

Verifying the Claim

The repository is publicly available at https://github.com/daveebbelaar/bama** Interested users can:

1. Clone the repo and run `./skills.sh --install`.
2. Follow the “Quick start” instructions in the README to point the agent at a simple macOS app.
3. Inspect the generated Swift/SwiftUI source and the automated test suite.

If the output builds and runs without manual Xcode interaction, the core claim is substantiated.

Limitations & Unknowns

• The README does not provide performance benchmarks or success rates for complex, multi‑window applications.
• No independent third‑party review has been published yet, so the reliability of the LLM‑driven reverse‑engineering step remains unverified.
• Licensing of the generated code is not explicitly discussed; developers should review the repo’s MIT license and any third‑party dependencies.

*The article is based solely on the information available in the BAMA GitHub repository and its README.*