Claude Code + Anatomy: Turn Any Technical Idea Into an Interactive Machine You Can Poke

Vinay Bhaskarla

5 min read

·

Oct 13, 2026

--

2

Listen

Share

A free Claude skill that draws ideas as real, working objects. Here is how to set it up in 10 minutes, and the steering column crash test I built with it.

The best diagram I made this year has no boxes, no arrows and no labels.

It is a steering column crash test you can tilt, telescope and run. 106 solid parts, every number on screen true, and I did not draw a single line of it.

The tool is Anatomy, a free skill for Claude Code by Ryan (@wheresryan22). Version 1.0 shipped on October 10, 2026.

What Anatomy does differently

Most AI diagrams are boxes and arrows with nicer fonts. Anatomy asks Claude to invent a physical machine whose working is the idea, then build it.

Two examples from the README say it best:

• A cache becomes a door lock: every pin has to line up before it turns, the same way every check on the data has to pass before the cache skips the work.
• A graphics context becomes a plotter: one print head visiting 16 wells, the same way one engine feeds many screens.

A skill, if you have not used one, is a folder of instructions and code that Claude Code reads when a task matches it. Think of it as a craft manual Claude opens before starting work. Anatomy’s manual is strict: collect the real numbers first, write a parts list of at least 40 parts, build each one as a shaded solid, add a live readout and calm motion, then zoom in on every joint and fix what looks wrong.

It also ships its own building inspector. An audit script checks that nothing passes through anything, that every pipe ends on a real mount, and that wherever two parts overlap on screen, the nearer one is drawn on top. On my figure it sorted out 326 overlapping pairs at rest alone, then checked 5 more poses, and found zero problems.

Set it up in 10 minutes

You need Node.js, Chrome, Claude Code and the skill. No API keys, and nothing to npm install.

Step 1. Install Node.js 22.20 or newer.

Mac

Run the LTS installer from nodejs.org, or:

brew install node

Windows (PowerShell)

Run the LTS installer from nodejs.org, or:

winget install OpenJS.NodeJS.LTS

Close and reopen the terminal, then check:

node -v

Step 2. Install Google Chrome from google.com/chrome if you do not have it. The skill uses it in the background to check its own figures.

Step 3. Install Claude Code (you sign in with your Claude account).

Mac (Terminal)

curl -fsSL https://claude.ai/install.sh | bash

Windows (PowerShell)

irm https://claude.ai/install.ps1 | iex

If either command has changed, the current one is on code.claude.com.

Step 4. Install the skill. Same command on Mac and Windows:

npx skills add wheresryan22/anatomy -g -a claude-code

Step 5. Make a folder and start Claude Code in it.

Mac

mkdir -p ~/figures/first
cd ~/figures/first
claude

Windows (PowerShell)

mkdir $HOME\figures\first
cd $HOME\figures\first
claude

Step 6. Ask for a figure:

/anatomy explain how a token bucket rate limiter works, interactive

When it finishes, double-click the .html page it wrote. That page is the figure, no server needed.

Fair warning: Anatomy’s own README calls it slow on purpose. It audits and inspects its work before handing it over. Expect a long session, not a sketch.

What you can build: my example

I have a side project, a 3D explorer of a pickup truck’s steering system. I wanted one figure that explains why a steering column is built to collapse.

In the US the answer is a federal rule, FMVSS 203. A 75 lb body block shaped like a human torso flies into the steering wheel at 24 km/h. The force on its chest must stay under 11,120 N, about the weight of a small car, except for spikes adding up to 3 milliseconds.

Anatomy’s first version looked great and was wrong. It drew a hydraulic ram pushing slowly along the column, which is not the test the rule describes. So I asked Claude to rebuild it to the documents the rule points to: SAE J944 for the body block and NHTSA’s lab procedure TP-203–02 for the rig.

The result has a plywood body block drawn from J944’s figures, a pneumatic launcher, a laser speed trap within 3 inches of the wheel, and a load cell behind the wheel. Drag to tilt the column through 5 positions and telescope it 55 mm. Tap, and the test runs 25 times slower than life until the column collapses its full 80 mm.

Press enter or click to view image in full size

Play with it here.

That is the part nobody mentions about AI figures. When drawing gets cheap, what gets expensive?

Being right. My wrong draft cost nothing and nobody would have spotted it. Anatomy did not know the standards, but it made being specific so cheap that I could afford to be.

Prompts that get good figures

The README names three habits:

1. Name the idea, not the drawing. Give real sizes, counts and thresholds, and let Anatomy invent the object.
2. Or name the object, if you know the machine you want, and say what should move.
3. Say where it will live: a standalone HTML page, a lone SVG, or a React component, light or dark.

One of the skill’s own test prompts shows all three:

/anatomy a figure for our Next.js docs on database connection pooling: a pool of 10 connections, requests borrow one and give it back, and when all 10 are busy new requests wait in a queue. A React component plus a static HTML preview, dark theme.

It suits technical writers, engineers explaining a test or a failure, teachers, and anyone who needs a landing-page visual that explains instead of decorates. It draws machines, instruments and furniture, not people or maps.

See what others built, and share yours

Anatomy has a community gallery with a twist. Like a baking contest where you hand in the recipe and the judges bake it themselves, it accepts only source code. Its checks rebuild each figure in a locked sandbox, twice, and the two builds must match byte for byte before it goes live.

Mine went in on October 12. To share yours, ask Claude Code in your figure’s folder: “publish this figure to the Anatomy community gallery.” It shows you what it will send, and after your yes, it opens the pull request.

Press enter or click to view image in full size

• Browse the community gallery: skills.wheresryan.sh/anatomy/community
• Get the skill: github.com/wheresryan22/anatomy
• Gallery source and how to submit: github.com/wheresryan22/anatomy-community
• My figure: Steering Column Impact Test

Accuracy note: the figure shows how the FMVSS 203 test is set up. Its 80 mm stroke is the column’s design value, not a measured crash result, and nothing here is a compliance claim.

The machine was never the hard part to draw. Knowing which machine to draw is the one job Anatomy leaves to you.

Disclosure: This article was developed with the assistance of an AI writing program. The ideas, direction, research review, revisions, and final editorial decisions are my own.