Anatomy

wheresryan22/anatomyReviewed by AI

What it does

Anatomy is a skill you install inside an AI coding assistant (Claude Code, and also Antigravity, OpenCode, Codex and MiniMax Code). You ask it to explain an idea and it invents a machine whose working is that idea, then builds it as an isometric drawing in SVG, optionally with WebGL effects such as flame or water, as a standalone HTML page, a single SVG or React components you keep. It needs Node.js 20.10+ and Chrome for its own checks, and no API keys of its own. The author is clear that it is not a chart or diagram tool and does not draw people or maps.

How easy it is to install

C · install 1 or 2 tools · Windows

Windows C
macOS C
Linux C

The same on Windows, macOS and Linux: you must already have one of the AI coding assistants, then install Node.js and run one command, `npx skills add wheresryan22/anatomy -g`, or copy the folder in by hand with git or a ZIP; Chrome or Chromium also has to be present for the skill's own picture checks. After that it is a `/anatomy` command typed into the assistant, with no window of its own.

Numbers on GitHub

Stars: 23,364on Oct 31, 2026
Downloads: 2,420670 this week
Growth: +2,719stars in 6 days

Stars over time

Facts

Runs on: Windows, macOS, Linux
You need to install: Node.js, Other tools
Disk space: not determined
Accounts or keys: None
Cost: Free, API costs extra · The skill itself is free and MIT-licensed, with "No API keys and nothing to npm install". What costs money is the AI assistant that runs it: Claude Code or one of the other agents, on a paid plan or API credits, for sessions the README says are long on purpose.
License: MIT (Permissive)
Can replace: Illustrator
Last update: Oct 18, 2026

Risk signals (6)

AttentionVery new repository with many stars
23369 stars in 23 days
automatic check
Attentionasks you to loosen your agents sandbox
README, "Codex: if the browser checks are blocked": the skill "checks its own figures by starting headless Chrome, which writes a temporary profile and talks to it over a local port", and when Codex's sandbox stops that it suggests approving commands outside the sandbox, or a config with `sandbox_mode = "workspace-write"`, `network_access = true` and `writable_roots = ["/tmp"]`, or starting Codex with `--sandbox danger-full-access`, which it admits "lifts every limit for that session". It also notes the figure is still built without the checks.
AI review
Infothe install copies the whole repository and wants newer node
README, "With the skills CLI": "The CLI copies the whole repository, examples included (about 10 MB). It needs a recent Node.js: version 1.7.1 of the CLI asks for 22.20 or later" — above the Node 20.10+ the requirements table and the README badge state. The install command also runs a separate project, the `skills` CLI from vercel-labs, straight from npm.
AI review
Infoeach figure is a long paid session
README, Limits: "It is slow on purpose. A figure takes a long session: the parts list, the geometry, the audit, and close-ups of every joint in both themes. The Raptor's build takes a few minutes, and a full 3D orbit audit takes 15 to 20 minutes for 30 parts." The skill also "tells Claude to read every PNG it makes", so the assistant works through many images, which is what you pay for.
AI review
Infosharing a figure opens a pull request in your name
README, "Share a figure": asking the agent to publish "asks for a title and your GitHub name, shows what it will send, and after your yes opens a pull request on anatomy-community for you". It only happens when you ask, and it shows the content first, but the result is a public post carrying your GitHub name.
AI review
Info3d mode is beta and heavy
README note: "The 3D mode is beta. It is proved on its example and one test figure, not yet on a production figure. A small 3D figure runs at 60 fps; the 380-part stress hand measured about 20 fps at 1440 px wide and 13 to 15 fps on a throttled phone profile". The kit folder also lists turn-fixed.mjs, canvas-painter.mjs and gl-shared.mjs as "experimental variants, not yet documented or used by the examples".
AI review

Evidence is shown in its original language.

What Radar could not check

Whether your own request produces a figure as good as the examples in the gallery
How much a single figure costs in AI assistant usage, since the sessions are long by design
Whether the headless Chrome checks work on your machine, especially on Windows or inside a sandboxed agent
Disk use: the README says the install copies about 10 MB, which the automatic check could not measure
How 23369 stars in 23 days were gathered, and how the project behaves once the first attention passes