pull/this/repo
RepoAI agent skills#claude-code#agent-skills#ui-design#design-systems

Dimensional Drawing: an AI skill that turns your UI into Apple-style blueprints

By Pull This RepoPosted 4 min read

Short answer

Dimensional Drawing is a free, MIT-licensed pair of AI agent skills. Point your agent at a UI component, such as a card or a dialog, and it draws a one-page technical sheet with every padding, gap, corner radius and layer measured and labelled. A second skill animates the sheet into a short film. It draws structure only: no text and no colour.

Dimensional Drawing gives an AI coding agent a new job: draw your interface components the way Apple draws its devices for case makers. You point the agent at a card, a dialog or a button, and it returns one technical sheet where every padding, gap, corner radius and layer is measured and labelled. A second skill animates that sheet, so the lines draw themselves while a camera pulls back and the layers lift apart.

What is Dimensional Drawing?

It is a repo by Enes Ozturk holding two agent skills:

  • dimensional-drawing reads one UI component and draws a one-page sheet: a title block with the key variables, notes, a plan view, an exploded view of the layers, detail circles (nested corners, radii, the body well) and variable tables.
  • dimensional-drawing-motion turns a finished sheet into a film of about 10 seconds: a close-up on the header as its lines draw in, a pull back to the plan view, the layers lifting into an exploded view, then the full sheet.

The repo was created on 23 October 2026 and had 2,194 stars when we checked on 18 November 2026. It still has a single commit.

What makes the sheet useful in a handoff is that it uses variables, not raw numbers. Padding is S1, a corner radius is R1, and relationships are written out, such as H1 = 2·S1 + H3. The tables give each code its value. In the repo's own card example, the card padding S1 is 14, the minimum gap S4 is 12 and the card radius R1 is 16.

How do you install Dimensional Drawing?

For Claude Code, the README's install is two lines:

terminal
git clone https://github.com/enesozturk/dimensional-drawing
cp -r dimensional-drawing/skills/* ~/.claude/skills/

On Windows PowerShell the copy step is Copy-Item -Recurse dimensional-drawing\skills\* $HOME\.claude\skills\. Restart Claude Code so it finds the two new skills.

For Codex, Cursor, Gemini CLI and other agents, use the skills CLI, or tell the agent to read the two SKILL.md files:

terminal
npx skills add enesozturk/dimensional-drawing

On claude.ai or the Claude desktop app, zip one skill folder (for example skills/dimensional-drawing) and add it under Settings, Capabilities, Skills.

How do you use it?

Ask for a drawing and point at the real source: a file in your repo, a Storybook story, a URL or a screenshot. The README's two examples:

text
Make a dimensional drawing of the Card component in src/components/Card.tsx
Draw the anatomy of our settings dialog, same style as the examples

Prompts that follow the skill's own rules work well too:

  • "Measure every instance of Card in this repo first, then draw one sheet whose variables reflect the whole system, and list which values differ."
  • "Make a dimensional drawing of the modal at http://localhost:3000/settings, reading padding, gap, radius and border from computed styles."
  • "List every [TBD] on this sheet as questions for the design team, one line each."

Then ask it to "animate this drawing". On claude.ai the motion skill builds the film in the Motion artifact and you export an MP4. Without Claude, the headless renderer needs Node 18 or newer and ffmpeg:

terminal
cd skills/dimensional-drawing-motion/adapters/headless
npm install
node render.mjs --scene ../../templates/scene-dialog.js --css ../../templates/base.css --out dialog.mp4

The README adds that it works for physical objects too, using third-angle views, millimetres and published specs only.

What is the catch?

  • Structure only, on purpose. There is no copy, no text widths and no colour. Slots carry names like TITLE SLOT · FLEX and ACTION · HUG. It is a spec of the architecture, not a pixel mock-up.
  • Monochrome and light only. Black line on white paper, greys only, by design.
  • It never invents values. Anything it cannot read stays [TBD], so a screenshot gives a thinner sheet than source code.
  • The film is tuned for Claude's Motion artifact. The README notes that Remotion, Motion Canvas, GSAP and Lottie are not tested.
  • It is very new. One commit and no releases yet, so pin the commit you install.

Who is Dimensional Drawing for?

It fits design system teams who hand components between design and engineering, front-end developers documenting a component library, and anyone who wants a striking, accurate diagram of a component for a post or a doc. It is not a tool for colour, typography or content decisions.

How does it compare with similar tools?

ToolWhat it doesWhen to pick it
Dimensional Drawingagent skill that draws and animates measured component sheetshandoff specs and explainer films of one component
Storybookworkshop for building and documenting components in isolationliving component docs; also a source Dimensional Drawing can read
Motion CanvasTypeScript explainer animations with a live editorcustom animation beyond one fixed choreography
HyperFrameswrite HTML, render video, built for agentsdropping a finished sheet into a longer video with captions

What did we run for the video?

We did not install or run the skills for our video. The footage was the repo's own published film and sheets plus its GitHub page, and two explainer animations we drew ourselves. The install and prompt steps above come from the README and skill files.

If you design screens, FigCraft is a free design app whose MCP server lets an agent build the components in the first place. For another skill that turns an agent into a careful measurer, see Geo Sleuth.

Pull it. Stash it.

Questions people ask

What does Dimensional Drawing make?

One dense, self-contained HTML sheet per component: a title block with the key variables, notes, a plan view, an exploded view of the layers, detail circles for corners and radii, and tables that give each code, such as S1 or R1, its value. The motion skill can then turn the sheet into a film of about 10 seconds.

Which AI agents can use it?

The README gives a two-line install for Claude Code, a zip upload for claude.ai and the Claude desktop app, and the skills CLI or a pointer to the SKILL.md files for Codex, Cursor, Gemini CLI and other agents.

Is it made by Apple?

No. It is an independent project by Enes Ozturk. Its README credits the dimensional drawings Apple publishes for accessory makers as the style it follows.

What if the skill cannot measure something?

It does not guess. The README's house rules say unknown values stay marked [TBD], so a richer source such as the component's code, a Storybook story or a live URL gives a fuller sheet than a screenshot.

Can I render the film without Claude?

Yes. The motion skill has a headless renderer that needs Node 18 or newer and ffmpeg. The README says it is tested with Playwright Chromium and ffmpeg, while other animation libraries are not tested.

Sources

  1. 1dimensional-drawing repository and README
  2. 2dimensional-drawing skill file
  3. 3dimensional-drawing-motion skill file
  4. 4GitHub API: enesozturk/dimensional-drawing (stars, licence, language, created date, checked 2026-11-18)
  5. 5Apple accessory dimensional drawings (the style the README credits)
  6. 6Storybook
  7. 7Motion Canvas
  8. 8HyperFrames

Get tomorrow's repo in your feed

One free, open source repo a day as a short video, shown working, with the catch named. Pull it. Stash it.