Price terms as published
Each line is the vendor’s own wording, split from the price capture checked on 2026-10-20. Nothing here is rounded, converted or re-worded — the full sentences stay in the review below. Lines that carry only a price continue the plan named on the line above, in the vendor’s own order.
| # | Plan / term | Checked |
|---|---|---|
| 1 | Free (MIT) | 2026-10-20 |
| 2 | no paid tier, no hosted service | 2026-10-20 |
| 3 | the real cost is your agent's time (the official site's watch figure took 1h54m of Claude working in one pass) | 2026-10-20 |
One-liner: Anatomy is an agent skill, written for Claude Code and compatible with Antigravity, OpenCode, Codex and MiniMax Code, that turns a one-line idea into an interactive isometric figure — a real machine with moving parts, drawn as code you can open, ship and edit.
This entry is compiled from the project’s GitHub repository, its official site and two third-party write-ups, all read on 2026-10-20. It is not a hands-on test: no figure was generated for this page, no score is assigned, and every number below is sourced inline.
Verified facts (GitHub API, official site and README, captured 2026-10-20)
| Attribute | Value |
|---|---|
| Repository | wheresryan22/anatomy |
| Stars / forks | 17,003 / 1,295 (GitHub API, 2026-10-20) |
| Open issues | 59 (GitHub API, 2026-10-20) |
| Created / last push | 2026-10-08 / 2026-10-18 |
| Primary language | JavaScript |
| Licence | MIT (GitHub API licence field and LICENSE file agree) |
| Price | $0 — MIT, no paid tier, no hosted service, no price page on the official site |
| Figures published so far | 33, all interactive, per the official site |
| Install | npx skills add wheresryan22/anatomy -g, or unzip into ~/.claude/skills |
| Requirements | Node.js 20.10+, Chrome or Chromium for the visual checks; React 18+ only for React output |
| Runtime cost | Agent time, not money: the official site says its mechanical-watch figure took 1h54m of Claude working in one pass |
Two independent write-ups confirm the growth curve rather than only its endpoint: explainx.ai recorded 2,176 stars and 165 forks by 2026-10-11, and cellcog.ai recorded 4,103 stars and 323 forks four days after the launch, alongside an X post it read on 2026-10-12 at 1,396,709 views and 19,286 bookmarks. Both are third-party readings taken on their own dates — the star count above is the only one pulled directly from GitHub on 2026-10-20.
What it draws
The skill’s core rule, quoted on the official site from SKILL.md: “The object is the explanation.” You do not get a labelled diagram. You get an invented physical thing whose movement carries the idea:
- A prompt in the README’s own examples is
/anatomy explain how a token bucket rate limiter works, interactive— the output is a mechanism that visibly fills and drains, not a flowchart. - The flagship examples include a Raptor 2 rocket engine in a test stand with a WebGL shader plume, a mechanical watch movement you wind with the crown, and a gantry plotter for a GPU-canvas concept.
- Figures ship as standalone SVG/HTML pages, or as React components for a Next.js site, in a dark or light theme.
How one figure gets built
Four layers do the work, and the repository is unusually explicit about all four:
SKILL.mdis the workflow the agent follows: collect the true numbers first, invent the object, write a parts list, plan the world, build the geometry once with the kit, paint back to front, frame it, make it live, then verify it like a critic.kit/is the drawing library inside the repo — an isometric camera, solids and a parts library, round parts on any axis (lathe.mjs), pipes and cables (tube.mjs), a WebGL layer (gl.mjs), a 3D runtime in beta (turn*.mjs), plus React components.references/holds long-form guides the agent opens only when it needs them: craft, kit API, motion, React, verification, WebGL, a design walkthrough and the 3D mode.- Verification is code, not vibes:
kit/audit.mjschecks that nothing passes through anything, that every end sits on a mount, and that overlaps are drawn in depth order;scripts/lines.mjsfails any line that ends in mid-air; headless-Chrome scripts take screenshots and contact sheets that the agent is told to read before calling the figure finished.
How to install and run it
npx skills add wheresryan22/anatomy -gThe CLI asks which agents to configure (Claude Code, Antigravity, OpenCode, Codex, MiniMax Code); adding -a claude-code skips the question. The manual path is the same skill folder unzipped into ~/.claude/skills. Then, inside Claude Code, a prompt such as /anatomy explain how an LRU cache works, interactive starts a figure.
Requirements are light but real: Node.js 20.10+ for the kit and checks (built-in modules only — there is nothing to npm install, apart from React if you want React output), Chrome or Chromium for the headless visual checks, and a coding agent that reads the Agent Skills format. Windows works; the checks look for Chrome in the usual places, with CHROME_PATH as the override.
Price and licence: the figure is free, the time is not
The repository is MIT and there is no paid tier, hosted service or price page attached to it as of 2026-10-20 — the official site has an Install section, not a Pricing section. What you spend is agent time and tokens: the site’s own watch figure is quoted at 1h54m of Claude working on its own, and the README describes a Raptor build as taking minutes while a full 3D orbit audit runs 15 to 20 minutes for 30 parts. That is the honest cost model: if you already pay for a Claude or Codex plan, the marginal dollar cost is zero and the marginal cost is your session budget; if you run the agent against a metered API key, long builds are billed like any other long agent session.
How it differs from the alternatives it gets confused with
- Image generators (Midjourney, DALL·E, Flux and similar). Those return raster pixels: one picture, no parts, no live interactivity, and numbers in the image are whatever the model painted. Anatomy returns code — SVG/HTML/React — where each part is a separate object and every number shown on the card comes from the same model that moves the parts.
- Diagram tools (Mermaid, Excalidraw, draw.io). Fast and predictable, but they draw exactly the thing Anatomy’s rule rejects: boxes, arrows and labels. If your reader needs the shape of a process rather than the working of a mechanism, those tools are the cheaper answer.
- Hand-built Three.js or After Effects work. The closest output quality, and the gap is labour: you model and animate it yourself. Anatomy automates the build-and-verify loop, at the price of a long agent session per figure.
- Other image skills in this catalogue, such as Monocolor Print, aim at a still editorial image in one ink; Anatomy aims at a moving, inspectable object. They are complementary, not substitutes.
Honest limits
The README is candid, and these limits decide whether the skill fits your job:
- It is not a chart library. No axes, no series, no data plots. If you need a bar chart, use a chart library.
- It draws machines, instruments, tools and furniture — not people and not exact maps. An idea with no mechanism underneath will come out weaker.
- It is slow on purpose. A figure is a long session: parts list, geometry, audit, close-ups of every joint in both themes.
- Shaders need WebGL, and the 3D mode is beta: the README reports 60 fps for a small 3D figure but about 20 fps at 1440 px and 13–15 fps on a throttled phone profile for a 380-part stress scene.
- Install assumes a terminal and Node 20.10+; agents in sandboxed environments may need Chrome explicitly allowed for the screenshot checks.
Who it suits, and who it should skip
Use it if you explain mechanisms for a living — engineering posts, docs, course material, technical launches — and you already run Claude Code or Codex with enough session budget to let one figure build for an hour or two. Skip it if you need a dozen quick illustrations today, if the concept has no physical analogy, or if your publishing pipeline only accepts PNGs from an image model.
Sources
- GitHub repository
wheresryan22/anatomyand GitHub REST API, read 2026-10-20 (stars, forks, issues, dates, licence). - Raw
README.mdfrom the repository’smainbranch, read 2026-10-20 (requirements, install commands, limits, verification tooling, example prompts). - Official site
skills.wheresryan.sh/anatomy, read 2026-10-20 (33 live figures, install command, 1h54m watch build, craft rules). - explainx.ai write-up, read 2026-10-20 (third-party star/fork reading dated 2026-10-11).
- cellcog.ai write-up, read 2026-10-20 (third-party star reading four days after launch; X post views and bookmarks read 2026-10-12).