One painting, rigged with @rig-c/parts, compiled and rendered by rig-c. How it was made →
2D character rigs, authored by an AI agent that cannot see them, and verified before anything is written. One painting goes in; a rigged, animated character that plays in the browser comes out. Every stage refuses by name what it cannot verify, so a wrong file never reaches the disk.
rig-c makes no images: the painting and its layer split come from image models you run yourself.
create sets up the project with the user's direction, an interview and a signed plan. parts makes the pieces, core compiles and checks the rig, and play or adapters run it. pose adds motion, and review is where a person looks at the result.
| Repository | Package | What it does |
|---|---|---|
| create | @rig-c/create |
Project direction, an interview and a signed plan before rigging |
| parts | @rig-c/parts |
Turns one painting's layers into rig-ready parts, meshes, bones, physics and an idle |
| core | rig-c |
Compiles a rig and motion spec into skeleton data, checked before a byte is written |
| play | @rig-c/play |
Plays the compiled model in the browser, as DOM elements or one canvas |
| adapters | @rig-c/pixi · @rig-c/phaser · @rig-c/three |
Plays a model inside PixiJS, Phaser or Three.js, with no Spine runtime |
| pose | @rig-c/pose |
Puts motion from a video, a dataset or hand-placed points onto a rig |
| review | @rig-c/review |
A local workspace to inspect a rig and export feedback bound to a frame, bone or slot |
| Also here | |
|---|---|
| getting-started | The guide for people setting the toolchain up |
| examples | The public example corpus |
| exam | The open-book exam for AI agents |
| trailer | The trailer, with one command to rebuild it |
rig-c/exam gives an agent one painting and the installed packages, and nothing else.
Basic: 14 of 17 models passed every graded line, from each vendor's lighter tier to its heaviest. Advanced: 5 of 6 passed every graded row in all three rounds.
Claude Sonnet and Gemini Flash are already enough for this work; Claude Haiku and GPT Luna miss some lines. See every result, or sit it with yours.
The tools are MIT licensed; the exam's content is CC BY-NC 4.0. rig-c is also published as @rig-c/core, and names such as rig-create, rig-parts and spine-rigc are aliases of the same packages.