Research

The Anatomy Skill for Claude Code: A Practical Guide, Tested on Everyday Objects and Seven Models

Anatomy is a Claude Code skill that explains an idea by building it as an interactive isometric machine. How to install it, how to prompt it, what worked and what didn't, with 24 everyday objects you can play with, from a toilet flush to a lift's safety brake, and the same prompt run through seven AI models.

Five isometric figures built by Opus 5.5 with the Anatomy skill: a toilet cutaway, a fridge with its side panel off, a lift shaft, a hand coffee grinder in exploded view and a zip in a jig.
Five everyday objects, built by Opus 5.5 with the Anatomy skill on 12 October 2026. All playable below.

Anatomy is a Claude Code skill that takes a one-line idea and builds it as a small interactive machine you can turn, poke and take apart. A developer called R (@wheresryan22) released it on 10 October 2026. By the time we tested it, the launch post had passed 1.3 million views and the repo had 4,000 stars. This guide covers what the skill does, how to install and use it, and which prompts worked for us, then shows 24 everyday objects we built with it and the same prompt run through seven AI models.

How we ran it. On 12 October 2026 we used Anatomy at commit 689dcf9, one run per figure, with no help once a run started. The original five objects and comparison runs below are the model's own files, unedited. The later everyday-object series was continued and repaired after interrupted builds. Click "Try it" to play with it. The numbers in each readout come from a small model inside the page: they're the model's illustration, not measurements.

Choosing a model for your own product? We compare models on your prompts and data, check their outputs against your requirements, and help you decide what is ready to ship. Explore our evaluation services or tell us what you are building.

What the Anatomy skill does

Ask for a figure that explains something, and Anatomy has the agent invent a physical object whose working is the idea, then build it from dozens of small shaded isometric parts in SVG. The figure sits in a card with a title, a hint and a live readout, and it moves under the pointer and the keyboard. WebGL is used for things lines can't draw, such as fire or water, and there is a 3D mode in beta. The agent then screenshots its own figure and zooms in on the joints to check it. The result is source you own: a Node build script and a standalone HTML page, a lone SVG file, or React components. Its README is clear that it is not a chart library: it draws machines, instruments and tools, not data plots, people or maps.

How to install the Anatomy skill

Anatomy is a standard Agent Skill: a folder with a SKILL.md at its root. These steps come from the install section of its README, checked on 2 November 2026.

  • For you, in Claude Code: npx skills add wheresryan22/anatomy -g -a claude-code. It installs to ~/.claude/skills/anatomy.
  • For one project: npx skills add wheresryan22/anatomy without -g. For Claude Code it installs to .claude/skills/anatomy; commit that folder and your team gets it too.
  • With git instead: git clone https://github.com/wheresryan22/anatomy.git ~/.claude/skills/anatomy. Run git pull in that folder to update.
  • Other agents: swap -a claude-code for -a codex, -a opencode, -a antigravity or -a minimax-code. Leave -a out and the CLI asks which agents to set up.

You need Node.js 20.10 or later (the README says version 1.7.1 of the skills CLI wants 22.20 or later), and Chrome or Chromium for the screenshot checks. There are no API keys and nothing to npm install. The CLI copies the whole repository, examples included, which is about 10 MB. Claude Code picks up a new skill in a running session; if the skills folder didn't exist when the session started, run /reload-skills.

How to use the Anatomy skill

Type /anatomy and say what the figure should explain. This is the README's own example:

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

In Codex the command is $anatomy; OpenCode takes /anatomy as in Claude Code. You don't have to name the skill: Claude Code also loads it when you ask for an isometric illustration, an explanatory figure for docs or a blog post, or an interactive explainer. That's worth knowing the other way round too: to get a quick plain drawing, turn the skill off (see the comparison below).

The README suggests three habits: name the idea and give the real numbers, or name the object you want and say what should move, and say where the figure will live (HTML page, SVG file or React component, light or dark). Every brief we ran was one line about the subject, followed by the same ending:

  • interactive
  • light theme
  • the skill's smallest size, "Figure"
  • work without asking questions

Prompt patterns that worked, and ones that didn't

  • Worked: a question about an everyday object. One-line briefs such as "how does a toilet flush?" or "what keeps a lift from falling?" gave the strongest figures. The object already has a mechanism, so the skill has something real to build. See the toilet and fridge.
  • Worked: a product-page brief. For the coffee grinder we asked for a product page instead of an explainer, and got an exploded view with a ring that clicks through grind settings.
  • Worked: a process you can stage as a machine. A product feed going into Google Shopping became a conveyor past three checks, with bins for rejected items.
  • Weaker: abstract ideas. Next-word sampling and prompt caching came out as neat boxes that need their caption to make sense. For next-word sampling, the plain bar chart Opus 5.5 drew without the skill was easier to read.
  • Weaker: tiny parts at the smallest size. The zip is small inside its jig, and the lift's safety wedges are hard to see, so the readout carries the story.

Tips and limitations

  • It's slow. Each of our first five figures took 22 to 37 minutes. Much of that is the skill screenshotting and zooming into its own work to check it. Start it and do something else.
  • Long runs get interrupted. Our seven-model runs were stopped by mistake about 25 minutes in, and each one resumed from its own session. Several of the second everyday-object series needed continuing and repairing before they were right.
  • Click through every control. All our figures loaded without page errors, but the lift's Space-to-play is cut off by the figure's idle tour before the catch, so you have to drag instead. Try each interaction before you publish.
  • Check the numbers. Each readout is driven by the same small model that moves the parts, so it's consistent. But the model wrote that model, so check the facts before you put a figure in front of customers.
  • The skill sets the look. Every model we tried produced the same white bench-instrument style. That's useful if you want a consistent set. The second series shows it can also give each object its own palette.

Worked examples: five everyday objects

We asked Opus 5.5 to explain five objects most people use without knowing how they work. Each got a one-line brief and the skill's smallest size ("Figure").

How does a toilet flush? (34 min)

Isometric cutaway of a toilet cistern and bowl, with the float, fill valve and flush valve visible.

This is a cutaway of the cistern and bowl. Click to flush and watch the sequence:

  1. The flush valve lifts and water drops into the bowl.
  2. The S-bend fills until the siphon pulls everything down the trap. The readout follows the water ("siphon · 2.8 L/s down the trap").
  3. The float drops, and the fill valve tops the cistern back up until the float closes it.

How does a fridge work? (37 min)

Isometric fridge with its side panel off: food on the shelves, the cold plate inside, hot coils on the back and the compressor underneath.

The fridge has its side panel off, and you can see what each part does:

  • food on the shelves, with a cold plate at the back
  • the coils on the back, giving off heat you can see rising
  • the compressor underneath

Click to open the door, and use the arrow keys to turn the thermostat. The readout shows the inside temperature, how long the compressor has been running and the plate temperature. It's the clearest answer we've seen to "why is the back of my fridge warm?"

What keeps a lift (elevator) from falling? (28 min)

Isometric lift shaft with the car on its guide rails, a speed governor pulley at the top, a tension pulley and buffers in the pit.

The full lift shaft is drawn: the car on its guide rails, a speed governor at the top, a tension pulley in the pit and buffers below. Drag across to scrub through the moment the ropes part. Within a few hundredths of a second the governor trips, a lever on the car lifts the wedges, and the car brakes against its rails until the readout says "held on the rails". The wedges are tiny at this size, so it's the car slowing and the readout that tell the story. Space is meant to play it too, but the figure's idle tour cuts in before the catch, so drag instead.

How does a zip (zipper) work? (22 min)

A metal zip held in a jig, with the slider half way along and its Y-shaped channel cut away to show the teeth hooking together.

A No. 5 metal zip is clamped in a jig. Pull the slider and its Y-shaped channel tips each tooth's hook into the hollow of the one beside it. Pull back and a wedge prises them apart. The caption explains why you can't zip it by pushing the two rows together by hand. This is the weakest of the five at a glance, because the zip is small inside the jig.

How does a hand coffee grinder work? (24 min)

Exploded view of a hand coffee grinder: crank, body, conical burrs, adjustment ring and catch cup, with a magnifier on the burr gap.

We gave this one a product-page brief instead of an explainer. The grinder hangs in exploded view, with a magnifier on a stand pointed at the burr gap:

  • crank
  • body
  • conical burrs
  • adjustment ring
  • catch cup

Turn the ring and it clicks through the settings, from the zero point through espresso (0.27 mm) to cold brew at 1.2 mm. It's easy to picture this on a shop's product page in place of a static photo.

More everyday objects, with their own colour palettes

The second series started as parallel Opus 5.5 builds with the same Anatomy kit. We continued unfinished builds and repaired the figures before publishing them on 13 October. Each object gets its own palette. We check the mechanism, geometry and interactions before adding it here. Dimensions, forces, temperatures and timings in the readouts are illustrative.

Why does a click pen stay out after one click and retract after the next?

Cream and teal cutaway of a retractable pen: a coral rotating cam, alternating barrel channels and notches, and the spring around the refill.

The refill's spring pushes against a rotating cam. The button drives the cam clear of the barrel's ribs; sloping teeth turn it into the next position. On release, its lugs land either in a shallow notch, holding the tip out, or a deep channel, letting the spring retract it. This illustrated design turns the cam 45 degrees per click. Click the drawing or press Space to watch it slowly. This is one common design; the original mechanism drawings show how the plunger, ratchet teeth and barrel slots work together.

What actually lines up when you enter a padlock code?

A matte black combination lock cut open to show silver number rings, hidden cams, yellow gates and the spring-loaded shackle bar.

Each number ring turns a hidden cam. The shackle's notched bar can rise only when all four gates face it; any solid cam rim holds it down. This is one gate-and-bar design, enlarged so you can see the contact. Click a wheel to turn it, or use the arrow and number keys. Its code is 3 8 1 5.

Why does the plate in a microwave turn?

Mint and cream microwave cutaway showing its magnetron, waveguide, turntable and a dish whose heating pattern changes as it rotates.

Reflections inside the metal cavity create a pattern of stronger and weaker electric fields. Turning the dish moves the food through that pattern and helps spread the heating. Stop the turntable to see the uneven heating accumulate. The heat map is a simplified illustration, rather than a simulation of a particular oven; the FDA's explanation of microwave heating describes how microwave energy is absorbed by food.

How do the driven wheels turn at different speeds?

A blueprint-blue open differential on a test stand, showing the carrier, two side gears and the small spider gears between them.

The carrier turns both side gears. In a straight line, the spider gears travel around with it; in a bend, they also spin on their own pins, letting one wheel turn faster and the other slower. Their speeds still average to the carrier speed. Move across the drawing or use the arrow keys to steer.

What drives the spray out of an aerosol can?

Lilac and plum aerosol can cutaway with yellow liquid, a dip tube, spring-loaded valve and spray nozzle.

In this liquefied-propellant design, vapour pressure pushes the product up the dip tube. Pressing the button opens the stem valve; the product escapes through the nozzle and breaks into droplets as dissolved propellant flashes to gas. Release it and the spring closes the valve. Compressed-gas and bag-on-valve cans work differently.

What holds a Rubik's cube together?

Classic colourful cube opened around its six-armed core, with interlocking edge and corner feet and a spring-loaded centre.

The six centres attach to a core. Feet on the edges and corners fit behind neighbouring pieces, retaining them as a layer turns. Take it apart to see those hidden shapes, or pull a layer to feel the model's spring tension. This is an illustrated classic screw-and-spring construction; modern speed cubes vary.

Why does a soft-close toilet seat fall so slowly?

Sage toilet seat and porcelain bowl with brushed-steel hinges, showing an oil-filled rotary damper and its vane.

The seat turns a vane inside an oil-filled rotary damper. Closing forces oil through a narrow passage, resisting the motion; lifting opens a bypass. Drain the model's oil to compare the fall, or inspect the hinge for a closer view. This illustrates a vane design; some dampers resist motion through oil shear.

How does a pencil advance its lead without losing it?

Yellow, graphite and pink mechanical-pencil cutaway showing its eraser, spring, three clutch jaws, ring and tip retainer.

Closed clutch jaws carry the lead forward. At the front stop, the ring releases them and a rubber tip retainer keeps the lead from slipping back. The jaws return, then grip again. Click the eraser or press Space to follow one cycle. The illustrated advance is 0.6 millimetres per click.

Why does a plane toilet make that sudden whoosh?

Icy-blue aircraft toilet cutaway with a fresh-water rinse line, flush valve, holding tank, pressure gauge and vacuum blower.

A pressure difference pulls the waste through a valve into a holding tank. At altitude, low outside pressure helps create it; near the ground, a blower supplies the vacuum. Fresh water rinses the bowl. Change the altitude and flush to compare the two modes. Waste stays in the tank; the exterior vent carries air.

Why does a seatbelt lock when you yank it?

Navy and steel seatbelt retractor cutaway showing a webbing-sensitive flyweight lock and a separate vehicle-sensitive pendulum.

Slow withdrawal lets the reel turn. A sudden webbing acceleration swings its flyweight into a lock ring; vehicle braking can instead move a weighted pendulum and engage a lever with the spool ratchet. Try the three controls to compare them. Dimensions and trip thresholds belong to this illustrative retractor.

How does a vending machine know your snack fell?

Arcade-purple vending machine cutaway with a blue delivery motor, wire spiral, pink infrared curtain and a falling snack packet.

This equipped machine watches for a falling packet to break its infrared curtain. If the motor turns but no delivery is detected, the illustrated controller waits, then makes one further turn. Try a normal vend and a stuck snack. Infrared is drawn pink; this detector and retry policy are not features of every machine.

What makes a bike tick while you coast?

Forest-green freehub cutaway with three chrome pawls, bent leaf springs and a ratchet ring; an inspection handle drives the sprocket carrier.

While pedalling, spring-loaded pawls catch the steep faces of the ratchet teeth and drive the wheel. When you coast, the ring overruns the driver and pushes the pawls inward; each spring lets its pawl click out again. Inspect the hub for a closer view. This is a freehub; a traditional screw-on freewheel carries its mechanism in the sprocket assembly.

How do two small valves make a soap pump work?

Aqua glass lotion bottle with amber soap, a hollow pump chamber, two ball valves, a return spring and a measuring cup.

Pressing the piston closes the inlet valve and opens the outlet, pushing soap up the stem and through the spout. On release, the spring raises the piston; lower pressure closes the outlet and draws a fresh charge past the inlet. Hold Space or the pump to try it. The test fixture's dimensions, forces and fluid properties set its illustrated pressure and volume.

How do two threads become a lockstitch?

Mint and blush sewing machine cutaway showing the needle, rotating hook, bobbin thread, take-up lever and feed dogs.

The needle brings the top thread through the fabric. As it rises, a loop forms; the rotating hook carries that loop around the bobbin thread. The take-up lever then draws the interlock into the fabric. Step the cycle or use the arrow keys to follow it. The thin thread paths illustrate loop formation and reserve length; they are not a simulation of every contact in flexible thread.

Why must an airbag inflate, then let gas out?

Red airbag on a steel test fixture, showing an inertia sensor, inflator, opening cover flaps and a padded carriage loading the vented bag.

This slowed test stand shows sensing, ignition, cover opening, inflation and loading. Gas fills the bag before the padded carriage reaches it; vents let gas escape as the carriage compresses it. Scrub the timeline to follow the sequence. The older electromechanical sensor, approximate timings and simplified gas model illustrate one design.

How does smoke make light reach a hidden sensor?

Photoelectric smoke alarm cutaway with a light emitter, angled sensor, baffles, piezo horn and red light paths.

In clean air, a baffle blocks the direct light path to the photodiode. Smoke scatters some of each pulse towards it, raising the sensor reading. Change the smoke level to compare clean air and an alarm. The light is drawn red for visibility; voltages, thresholds and timing are illustrative.

How do escalator steps stay level and disappear below the landing?

Steel-blue escalator cutaway showing the step chain, separate roller tracks, curved landings and the return run beneath the stairs.

Each step's front and rear rollers follow different tracks. Their separation keeps the tread level on the incline; the tracks bring the steps into a flat landing, then guide them around the hidden return run. Scrub the loop or pause it to inspect the transitions. This compact cutaway illustrates the track geometry rather than a particular installed escalator.

What releases the spring that makes toast pop up?

Diner-red and cream toaster with a chrome cutaway, glowing heater, thermal latch, return springs and rising toast carriage.

This thermal-timer design holds the carriage down while the bread heats. A warming bimetal strip releases the latch; the springs raise the toast and open the heater contacts. Adjust the browning control to change the model's release point. Other toaster timers use electronics. Its temperature and browning curves are illustrative; the fine spring and electrical lines show their route and motion.

How does a kettle know when the water has boiled?

Copper kettle cutaway showing the water, steam passage, snap-disc switch and separate dry-boil cut-out in its charcoal handle and base.

Steam travels to a bimetal disc in the handle. When the disc snaps, it releases the switch and opens the heater contacts. A separate thermal cut-out protects the heater if there is too little water. Watch the water reach boiling, then inspect the switch as it trips. The heat and steam model illustrates the sequence; its dimensions, temperatures and timings are illustrative.

Bonus: how a product feed gets into Google Shopping

An isometric conveyor: product cards pass a barcode reader, a price comparator with a dial gauge and an image camera, with reject bins below and a tray of live ads at the end.

For something closer to our own work, we asked for a product feed going into Google Shopping. Products ride a conveyor past three checks:

  • a barcode reader that checks the GTIN's check digit
  • a price comparator that sets the feed price against the price on the product page
  • a camera that rejects small images and images with promotional text

Disapproved items drop into bins and the rest end up as live ads. Press F to run the corrected feed. The checks match how Merchant Center disapproves items, though the real list is much longer.

Anatomy skill vs a plain prompt

As a control, we turned the skill off and asked Opus 5.5 for "an interactive isometric illustration" of how an LLM picks its next word. It took 3 minutes and made a colourful 3D bar chart: a tower per candidate word, with temperature and top-p controls. With the skill, the same model spent about half an hour building a machine instead: a gantry that draws from a row of blocks sized by probability.

Opus 5.5 without the skill: a colourful 3D bar chart of candidate next words.
Opus 5.5, no skill (3 min)
Opus 5.5 with the skill: a gantry over a row of blocks sized by probability.
Opus 5.5 + Anatomy

The skill's own rules call charts and labelled boxes "too abstract" and ask for a machine you'd recognise. On a physical object like the toilet, that's exactly right. On an abstract idea like this one, we found the bar chart easier to read, because the machine needs its caption to make sense.

Does Anatomy work with GPT, Gemini, Kimi and Grok?

The README says Anatomy also runs in Codex and OpenCode. So we gave the next-word request to seven models, each in its own agent:

All seven finished a working figure with no page errors.

GPT-6.1 Sol's figure: pillars and gates, with a prompt box, an odds table and sliders.
GPT-6.1 Sol · Codex
Grok 4.7's figure: labelled keys for each candidate word and a stamped output.
Grok 4.7 · OpenCode
Kimi K3's figure: a roulette-style dial with candidate words printed on its slices.
Kimi K3 · OpenCode
Haiku 5.5's figure: a roulette wheel with unlabelled pockets and a gauge bank.
Haiku 5.5 · Claude Code
Gemini 3.8 Flash's figure: a dense rig of stations with a dial gauge and a gantry.
Gemini 3.8 Flash · OpenCode
Fable 5.1's figure: a long trough under a gantry.
Fable 5.1 · Claude Code
  • GPT-6.1 Sol broke furthest from the house style. It added a prompt box, a table of odds before and after the cut-off, and real sliders. It's the most explanatory figure, and the least like an Anatomy one.
  • Grok 4.7 printed the words on the keys of its machine, which makes it the easiest to read at a glance.
  • Haiku 5.5 and Kimi K3 both built a roulette wheel, where each pocket's width is a word's chance. That's the best metaphor of the lot. Kimi labelled its pockets; Haiku didn't.
  • Gemini 3.8 Flash built the densest rig, around the sentence "The robot gripped the…".
  • Opus 5.5 and Fable 5.1 made clean, fairly sparse gantries.

The bigger finding is how alike they look. Every one is a white instrument on a base plate with a ruler and a dial gauge. The skill sets the style far more than the model does. That's good news if you want one consistent look across a set of figures.

Method

  • The brief. Every run got the same ending: interactive, light theme, the skill's "Figure" size, and work without asking questions.
  • The agents. Claude Code 2.1.285, Codex 0.159.2 (GPT-6.1 Sol at its configured ultra reasoning) and OpenCode 1.18.33.
  • The interruption. We stopped the seven-model runs about 25 minutes in by mistake and resumed each from its own session. So we don't report their build times.
  • The checks. We opened every figure, clicked through it and screenshotted it before writing about it.
  • More. More small model tests live in our playground, and the overall rankings are on the leaderboard.

Quick answers

What is the Anatomy skill in Claude Code?

An open-source, MIT-licensed agent skill by @wheresryan22, released on 10 October 2026. It has Claude Code explain an idea by building it as an interactive isometric machine in SVG, with optional WebGL and 3D. It ships with a drawing kit, eight worked examples and scripts that screenshot and check the figure.

How do I install the Anatomy skill?

Run npx skills add wheresryan22/anatomy -g -a claude-code to install it for your user in Claude Code, or leave out -g to install it into the current project. You can also git clone https://github.com/wheresryan22/anatomy.git into ~/.claude/skills/anatomy. It needs Node.js 20.10 or later, and Chrome or Chromium for its checks.

How do I use the Anatomy skill?

Type /anatomy followed by what the figure should explain, for example "/anatomy explain how a token bucket rate limiter works, interactive". In Codex the command is $anatomy. Claude Code also loads the skill without the command when you ask for an isometric illustration or an interactive explainer. In our tests, physical objects with a clear mechanism gave the best figures.

Does the Anatomy skill work with models other than Claude?

Yes. On 12 October 2026, GPT-6.1 Sol in Codex, and Gemini 3.8 Flash, Kimi K3 and Grok 4.7 in OpenCode, all finished a working figure from the same prompt. They all came out in the same white bench-instrument style, because the skill decides the look more than the model does.

How long does an Anatomy figure take?

Our five everyday-object figures took Opus 5.5 between 22 and 37 minutes each, including the skill's own screenshot checks. Without the skill, Opus 5.5 answered a similar request in 3 minutes.

Model pages: Claude Opus 5.5 · Claude Fable 5.1 · Claude Haiku 5.5 · GPT-6.1 Sol · Gemini 3.8 Flash · Kimi K3 · Grok 4.7

More research
All posts →