Open source — the studio's workshop
App Automaton,the studio's open workshop.
Where AppCubic builds in public: a GitHub organization of MIT-licensed tooling for AI coding agents, introduced at appautomaton.github.io and developed at github.com/appautomaton.
Scroll — the cube comes apart
Coding agents are capable. They dotheir best work when you hand them clearworkflows, honest constraints, and good tools.
App Automaton packages those things in forms you can clone, symlink, and use the same day — with Claude Code, Codex, Gemini, or OpenCode. Everything is plain text first, portable across agents, and shaped by daily use inside the studio.
Skills
Skills that teach agents workflows.
The flagship is agent-designer, a portable skills workspace with issue-driven planning and bridge scripts that let one agent consult another without losing the session. It has its own deep-dive page. Around it sit document-SKILLs for Word, Excel, PowerPoint, and PDF work, presentation for consulting-quality decks, and webmaton for web research with grounded citations. A newer addition, latex-arxiv-SKILL, turns a research topic into an arXiv-ready review paper — it gates the outline for review, writes issue by issue, and verifies every citation, with its own project page.
Harnesses
Harnesses that keep agents honest.
automaton is a stage-gated harness that walks an agent from framing through planning, review, execution, and verification, with durable state in the repository. automux orchestrates several agents at once in tmux, each in its own git worktree, coordinating through files. Smaller tools handle the sharp edges: markmaton turns HTML into clean Markdown, and docker-for-apple-container runs docker and docker compose on Apple's native container CLI with no Docker Desktop, with its own project page.
Models
Models that run on the laptop.
A third shelf is pure MLX work for Apple Silicon, with no cloud and no PyTorch. mlx-speech covers speech synthesis, voice cloning, and recognition. tnt-asr is a terminal voice-to-text tool that transcribes in about a second. ltx-video-mlx generates video with synchronized audio on device. mlx-spatial reconstructs 3D from images, turning photos into meshes, Gaussian splats, and point clouds. mlx-cv adds detection, segmentation, and open-vocabulary grounding. The newest work, mlx-atomistic, takes MLX into science, running molecular dynamics and density functional theory on the Apple GPU, with its own documentation site.
The inventory — eight places to start
- appautomaton/document-SKILLs
Claude Code and Codex SKILLs for PDF, Excel, Word, and PowerPoint work: extraction, forms, formulas, and tracked changes, adapted from Anthropic's skills.
- appautomaton/presentation
Claude Code and Codex SKILLs for consulting-quality decks: strategy storyboarding, brand-system identity, and pixel-perfect PDF or editable PPTX output.
- appautomaton/webmaton
Portable web-research and browser-automation SKILLs for Claude Code, Codex, and OpenCode: Playwright, Chrome DevTools, nodriver, and HTML to Markdown.
- appautomaton/mlx-speech
Pure-MLX speech synthesis, voice cloning, dialogue, sound effects, and ASR for Apple Silicon: Fish S2 Pro, VibeVoice, LongCat, MOSS, Step-Audio, and Cohere ASR.
Project page - appautomaton/tnt-asr
A terminal voice-to-text TUI built on mlx-speech. Qwen3-ASR on the Apple GPU, fully local, no PyTorch, transcribes in about a second.
Project page - appautomaton/latex-arxiv-SKILL
A portable Agent Skill for Claude Code and Codex that turns a topic into an arXiv-ready review paper. Gated literature discovery, an issue-by-issue writing loop, verified citations, and a two-column IEEEtran PDF.
Project page - appautomaton/mlx-atomistic
An Apple-Silicon molecular dynamics and density-functional-theory runtime on MLX and Metal. It runs on the Mac GPU with no CUDA, server, or cloud.
Project page - appautomaton/docker-for-apple-container
A stateless docker shim over Apple's native container CLI on macOS. It runs docker and docker compose with no Docker Desktop.
Project page