When asking an agent to explain a technical concept, instead of drawing boxes and arrows, it invents a physical machine whose "operating principle is exactly this concept" (for example, depicting a cache key as a pin tumbler lock that only turns when all pins align). It assembles dozens to hundreds of small parts into an interactive isometric diagram and outputs a standalone HTML page, where the readings on the diagram and the movement of the parts are driven by the same model.
Security Audit: Not audited
- Not yet scanned
Rationale
- The workflow first defines the concept and parts list before drawing geometry; it establishes scale before starting: Figure scale involves 40–80 parts and is done directly; Hero scale involves 80–150 parts and takes about 2–4 hours, requiring notification before starting; Epic scale takes over 6 hours, does not start directly, and first provides an estimate and a Hero-scale proposal for the user to choose from.
- Includes 8 complete examples (dial indicator, desktop computer, arcade machine, rocket engine test stand, ripple tank, etc.), requiring the user to open the closest one first; examples generate standalone pages using
node examples/\<name\>/build.mjs, with--lightfor a light theme version. - Has self-check requirements: before delivery, zoom in to 4x to add parts where needed; some examples include
--audit, which the author claims checks for part intersection, ensures every endpoint lands in its mounting position, and verifies correct occlusion order. - Optional 3D mode and print mode (freezing poses into vector graphics arranged into A4 / Letter PDFs); the author notes that 3D mode is still in beta.
- The README specifies MIT license, Node 20.10 or higher, and zero dependencies; the skills.sh page shows the repository has 22.9K stars (indicating only popularity).
Usage
- Installation commands will be displayed after the audit passes; the usage provided in the README is to enter
/anatomy explain how a token bucket rate limiter works, interactivein Claude Code, or use$anatomyin Codex. - No account or external services are required; local Node 20.10 or higher is needed to run the skill's built-in build scripts.
Risks
- This edition has not been audited, is fixed at C-level, and does not display installation commands; SKILL.md instructs the agent to run the skill's built-in
kit/*.mjsand examplebuild.mjsscripts, which have not been scanned; the Gen Agent Trust Hub audit on skills.sh is marked as Warn. - Time-consuming and high token consumption: The author estimates Hero scale takes 2–4 hours, and SKILL.md also requires not reporting time to users for Figure scale; if a quick diagram is needed, specify Figure scale first.
- Requires "every diagram to be highly detailed," making it unsuitable for cases where only a simple flowchart is desired; the final output quality was only assessed from README screenshot descriptions, without opening the example pages.