AI Just Made Physics Interactive. This Is Insane. (Anatomy)

Highlights

• Anatomy is a new Claude skill that creates physical, interactive machines to visually explain abstract concepts instead of traditional boxes and arrows diagrams.

• A normal figure takes 30 to 90 minutes to generate, while hero pieces can take two to four hours and extensive token burning.

• The generation process starts with truth first by collecting real numbers and building a small pure model before drawing the readout on screen.

• The output lands in the project as a standalone HTML page, an SVG, or a React component that can be shipped under an MIT license.

• Self-checks automatically verify the geometry and alignment of the drawing, ensuring that objects do not pass through each other.

Timeline

1. 00:00 - 02:04

   Introduction to Anatomy and Concept Mechanics

   Standard diagrams fail to explain how system components actually function. Anatomy solves this by simulating physical counterparts, such as a pin tumbler lock for a cache key or a music box for debounce and throttle. After installation via the skill CLI or cloning into the skills folder, users can prompt the tool with a specific concept and the word interactive to generate a detailed model.

   • Traditional diagram tools rely on boxes and arrows that only indicate the existence of parts.
   • Anatomy directs the agent to invent a physical machine whose mechanism models a given concept.
   • The skill runs through platforms like Claude Code, Codex, Open Code, and Anti-Gravity with zero dependencies on Node 20.10 or newer.
2. 02:04 - 04:09

   Generation Process and Output Formats

   The tool operates in a structured sequence, prioritizing factual accuracy and behavioral simulation before rendering the visual elements. It uses an isometric camera, solids, tubes, and an optional WebGL layer for effects like fire and water. The resulting interactive figures can be customized for speed, light and dark modes, and 3D features, distinguishing it from static tools like Mermaid or Excalidraw.

   • The first step gathers real numbers and builds a pure behavioral model before drawing anything on screen.
   • Automated chrome processes check its own geometry to ensure that elements remain properly positioned and do not overlap.
   • Outputs are delivered as standalone HTML pages, SVGs, or React components with MIT licenses suitable for production sites.
3. 04:09 - 05:53

   Limitations, Costs, and Use Cases

   Substantial resource requirements and token consumption pose significant challenges for users on metered plans. Furthermore, because the visual explanation relies entirely on an invented physical analogy, users must review the underlying metaphor before trusting the drawing. Despite these caveats, the resulting interactive visualizations offer a novel way to explain complex technical products and codebases.

   • Normal figures require 30 to 90 minutes to render, while complex hero pieces can take up to four hours or even over 48 hours.
   • Self-checks verify physical placement and line endpoints rather than the factual truth of the underlying metaphor.
   • Technical writers, educators, and developers maintaining documentation or landing pages represent the primary audience for these explainer machines.