The open workshop of AppCubic

App Automaton

An open-source workshop for engineering with coding agents.

21 units · 4 shelves ↓
to filter
21 / 21 units
Act I · 06 units

SKILLs

Portable SKILL.md packs with three layers of disclosure. The same folder runs unchanged under Claude Code, Codex, Gemini, and OpenCode.

MCP · multi-agent
A-01

agent-designer

The skills workspace. Issue-driven workflows, an MCP tool catalog, and bridge skills for agent-to-agent delegation with session continuity.

uv · PEP 723
A-02

document-SKILLs

docx, xlsx, pptx, and pdf. Extraction, forms, formulas, and tracked changes. Runs on uv with PEP 723, no virtualenv.

PDF · PPTX
A-03

presentation

Consulting-quality decks, from strategy storyboarding to pixel-perfect PDF or editable PPTX.

Playwright · CDP
A-04

webmaton

Grounded web research. Citations, deterministic HTML-to-Markdown, and persistent sessions over Playwright, nodriver, and Chrome DevTools.

Android · Termux
A-05

playwright-skill

Browser automation that also runs on Android via a Termux launcher patch and headless Chromium.

arXiv · IEEEtran
A-06

latex-arxiv-SKILL

Turns a topic into an arXiv-ready ML review paper. Gated literature discovery, every citation verified, compiled to a two-column IEEEtran PDF.

Encyclopédie plate of theatrical cloud-machine rigging: a frame of painted clouds driven by cogwheel mechanisms
Intermission — Machines de théâtres, pl. XV: the cloud machineEncyclopédie, 1772
Act II · 06 units

Harnesses & runtimes

Stage gates, orchestration, and plumbing. The tooling that keeps an agent honest from plan to verify.

stage gates · markdown
B-01

automaton

A stage-gated harness. Frame, plan, review, execute, verify, resume. Installs as plain markdown.

tmux · worktrees
B-02

automux

Multi-agent orchestration in tmux or kitty, coordinating through files across parallel git worktrees.

JSON5 · Docker
B-03

openclaw-monorepo

A repo-local OpenClaw workspace with JSON5 config, plugins, and Docker sandboxes.

Go · PyPI
B-04

markmaton

HTML to Markdown for agent pipelines. A Go parser core wrapped in a Python CLI and API, on PyPI.

macOS · container CLI
B-05

docker-for-apple-container

A stateless docker shim over Apple’s native container CLI on macOS, with no Docker Desktop.

MCP · pinned upstream
B-06

pi-arcweld

A curated local layer welded to pinned upstream Pi. Extensions, system guidance, and MCP tooling join along one auditable seam, over a reproducible runtime.

Act III · 08 units

On-device MLX

Pure MLX on the Apple GPU. Speech, vision, video, 3D, and atomistic simulation that never leave the machine.

TTS · ASR
C-01

mlx-speech

Speech synthesis, voice cloning, dialogue, sound effects, and recognition, MLX-native on the Apple GPU.

Qwen3-ASR · TUI
C-02

tnt-asr

A terminal voice-to-text TUI. Qwen3-ASR transcribes in about a second, fully local.

LTX-2.3 22B · LoRA
C-03

ltx-video-mlx

Text- and image-to-video with synchronized audio on LTX-2.3 22B. On-device LoRA fine-tuning.

SAM 3 · boxes + masks + depth
C-04

mlx-cv

Open-vocabulary grounding, detection, depth and camera geometry, segmentation, and video tracking in pure MLX. Every model is checked against its upstream reference before it ships.

TRELLIS.2 · 3D
C-05

mlx-spatial

3D and spatial inference on device with SAM 3D Objects, TRELLIS.2, WorldMirror, and MapAnything.

DFT + MD · Metal
C-06

mlx-atomistic

Atomistic simulation on Apple silicon. A DFT and molecular-dynamics runtime on MLX and Metal, running on the GPU the Mac already has.

flow matching · 44.1 kHz
C-07

mlx-minimax-music3

MiniMax Music 3 in pure MLX. Lyrics and a structured caption to 44.1 kHz stereo, with phase-scoped model residency and no PyTorch at runtime.

MiniMax-H3 · video + audio
C-08

mlx-h3

MiniMax-H3 text to video and stereo audio, denoised together in one packed sequence. Phase-scoped model residency, no PyTorch at runtime.

Act IV · 01 units

Creative harnesses

The same stage-gated method, pointed at club music.

club music · stems
D-01

setloom

Producer-first co-production for club music. Musical intent into editable tracks, stems, renders, and listening notes, with the human keeping the taste gate.

Longitudinal architectural section of the Comédie-Française: stage, machinery loft, and auditorium, 1772The Comédie-Française in section — stage, loft & machines · 1772