Anatomy is an agent skill that gets Claude to explain an idea by building it as a machine: a detailed isometric illustration, drawn in SVG, with parts you can drag. It is for people who have to make a system legible to someone else: engineers writing documentation, anyone preparing a talk or a lesson, a team that wants a picture of its caching layer that looks like a device instead of three boxes and an arrow. It is not an application you open; it is a set of instructions you drop into a coding agent.
What it does
You describe a concept and the agent returns a figure. The example the project leads with is a cache key: the thing that lets a build skip work when nothing has changed becomes a pin-tumbler lock that only turns when every pin lines up. From there the routine the project describes is consistent:
- it assembles the object from dozens of small shaded parts rather than a few flat shapes;
- it places the result in a framed card with a live readout;
- it adds restrained motion that follows your pointer;
- it audits every joint for mistakes before calling the job finished.
The repository describes the output as polished SVG, with WebGL shaders and 3D effects as options rather than defaults. Nothing in the material promises a library of ready-made diagrams: each figure is generated for the idea you asked about.
How it works
Anatomy ships as a skill in the Claude Code format and declares itself Agent Skills compatible; the README adds that it also runs in OpenCode and Codex. In practice that means the repository is mostly instructions and supporting JavaScript that the agent reads, so the thing doing the drawing is the model, and the repository is what tells it how. The badges claim zero dependencies, Node 20.10 or later, an MIT license and a CI workflow on GitHub Actions. Interactivity lives inside the generated artifact, which is why the result is draggable in a browser with no server behind it.
Worth keeping the dates in view: the repository was created on 8 October 2026 and last pushed on 13 October 2026, and the 9,080 stars and 653 forks on its page are GitHub's own counters from that same week. Those are popularity figures, not evidence of stability, and the project publishes no benchmark of its own.
Getting started
The repository is at github.com/wheresryan22/anatomy. The requirements it actually names are short: Node 20.10 or newer, and an agent to host the skill. There are no stated operating-system limits, and the project doesn't list hardware requirements, so no GPU, VRAM or memory figure is promised either way.
The cost question has a clear answer. The skill is MIT licensed and free to copy, but it does no work by itself: every figure is produced by a model, so you pay whatever your agent costs, a Claude subscription or an API key, and illustrations with this much detail are not cheap in tokens. The material says nothing about pointing it at a local model, so assume a hosted one.
The README offers three ways in: an installer invoked through npx (npx skills add, going by the install badge), a ZIP of the main branch straight from GitHub, and the project's own page at skills.wheresryan.sh/anatomy. The copy of the README available here is cut off before the full install section, so take the exact command from the repository rather than from this page.
When to use it / when not
The catch the project states plainly: true 3D is still beta. Treat the SVG and shader path as the reliable one and the 3D as something to try rather than something to ship. The second caveat is age. This is a repository days old at the time of writing, with no release history in evidence, so expect its instructions to keep changing under you.
Use it instead of just asking Claude for a diagram when the explanation is the point and a Mermaid flowchart would flatten it, or when you want one striking figure for a README, a slide or a lesson. Don't use it if you need neutral, conventional architecture diagrams, if you cannot run Node 20.10, or if you need 3D output you can depend on today.
Anatomy is worth taking seriously if you already work inside Claude Code, OpenCode or Codex and spend part of your week explaining systems to other people. It is a narrow skill with an opinion about how explanation should look, and that opinion is the product: dense, mechanical, hand-built figures instead of generic boxes. The commitment is small, the license is permissive, and the worst outcome is a figure you throw away and a few tokens spent. If your diagrams are fine as they are, or if the beta 3D is the part you came for, there is nothing here you need yet.