Key Summary
- Visualization code that works from a one-line idea
- From explanations you read to explanations you manipulate
- Cost should be measured in review and revision time
- Starting from a single concept explained over and over
The open-source skill 'Anatomy' for Claude Code takes a one-line idea and helps AI design a device that explains the underlying principle, producing visualization code that runs in the browser. Developer Ryan (@wheresryan22) revealed the project via X and his blog on October 11, 2026 (local time); the core idea is explaining concepts through objects that move.
How are abstract concepts shown?
A message queue—a buffer that stacks data or task requests between programs—is represented as a device where beads wait in order, while an LRU cache that removes items unused for a long time is depicted as a bookshelf where book positions shift. Users can manipulate the device to observe how the concept actually works.
- Default output: isometric SVG-based shapes combining component motion with user interaction
- Optional feature: WebGL effects to render fire, water, light, and similar phenomena
- Beta feature: true 3D with parts rotating in actual depth
Published examples include a mechanical watch movement, the Rolls-Royce Trent 970 turbofan engine, and SpaceX's Raptor 2 engine. In the turbofan visualization, moving the thrust lever changes the rotation speed of the fan and shaft, with the on-screen thrust figures updating in sync. The developer explained that component movement and displayed values were built to operate from the same underlying model.
Is it finished with just a one-line command?
No. The process requires AI to design the structure, write the code, and then review and revise while checking the output on screen. According to the developer's published production log, the mechanical watch took about 2 hours, while the complex engine examples took over 10 hours including review and revisions.
Anatomy is not an official Anthropic feature but an agent skill distributed by the developer under the MIT license. The repository includes installation and usage instructions not only for Claude Code but also for Codex and OpenCode. The developer notes that Codex's sandbox may restrict browser execution needed for screen verification, and provides a separate execution guide for that case. A companion library called 'anatomy-figures' offers 52 interactive diagrams as React components and custom elements, for use on websites once the visualizations are created.
Vercel CEO Guillermo Rauch commented on X, "This is genuinely amazing—this is the future of education," while Claude Code developer Boris Cherny said, "There's a mountain of work left to do to apply LLMs to education, and we've only just taken the first step."
Implications for Korean Enterprise AX
For organizations that must repeatedly explain equipment principles or system architecture to new hires or customers, this approach offers an alternative to static documentation—an explanatory tool users can directly manipulate to understand. However, the cost benchmark should be review and revision time, not the one-line input itself. Since the claim that motion and figures are linked through the same model comes from the developer, anyone using this for internal training materials needs a designated person to verify that the figures match reality.
If considering internal adoption, it's worth checking two things first, in order: whether this MIT-licensed release (not an official feature) complies with your organization's open-source usage policies, and whether the screen-verification step can run without issue in your organization's approved coding tools and execution environment.
This week, try picking one concept your organization explains most frequently, and start by determining who would review it for accuracy once it's turned into a manipulable device.
Source: "Drew a 3D machine that runs on a single sentence"... Claude Code's 'Anatomy' unveiled