What is the canvas-design skill?
The canvas-design skill is a SKILL.md in anthropics/skills. It makes an AI agent write a design philosophy of 4 to 6 paragraphs, name a movement in one or two words, then render a PNG or PDF that is about 90 percent visual. It is for posters and art, not web UI.
The canvas-design skill is one of the creative skills in the anthropics/skills repository. It produces visual art as PNG or PDF files, such as posters and single-page design pieces. It is not a tool for building web interfaces.
It works in two steps. First the agent writes a design philosophy as a markdown file. Then it expresses that philosophy on a canvas, with about 90 percent visual and 10 percent text.
What changes in your agent's output
- The agent names a design movement in one or two words, then writes a 4 to 6 paragraph philosophy such as Concrete Poetry or Geometric Silence.
- It deduces a subtle reference from your request and weaves it into the piece instead of stating it.
- Output defaults to a single page, with multi-page as an option, and a refine pass asks for polish with no new graphics.
- Fonts come from the bundled canvas-fonts folder, which holds 54 .ttf files with their OFL licenses.
Who the canvas-design skill is for
Best for
- Posters, cover art and single-page visual pieces from a short brief.
- Teams that want a written design rationale alongside the artwork.
- Agents that otherwise produce text-heavy graphics.
Not for
- Web pages, components or CSS. Use frontend-design for interfaces.
- Brand tokens and color ramps. brandkit covers brand foundations.
- Text-heavy documents, since the skill pushes toward visual expression.
Use it with UiChemy
Canvas-design makes artwork, not WordPress pages. If you want the site, UiChemy builds it: Composer writes real HTML, CSS and JS inside Elementor or Gutenberg, and Figma to WordPress converts a design. Click Get Started on uichemy.com.
How to install Canvas Design Skill: Posters and Visual Art From an Agent
git clone https://github.com/anthropics/skills.gitClone the repository
Run
git clone https://github.com/anthropics/skills.git. The skill is the skills/canvas-design folder.Copy the skill folder
Copy skills/canvas-design into your agent's skills folder. Keep every file in the folder, including LICENSE.txt and any bundled resources the SKILL.md refers to.
Invoke it
Ask your agent to use canvas-design and describe the piece, for example a poster for a design conference. Review the philosophy file first, then the rendered PNG or PDF.
Check that it loaded
Ask for one poster. You should get a markdown philosophy that names a movement, then a single-page PNG or PDF with very little text. If you get a web page instead, the skill did not load.
Key features
- Two-step processA written philosophy comes first and the visual piece second, so the artwork follows a stated idea.
- Named movementThe philosophy begins with a movement name of one or two words, for example Chromatic Language or Analog Meditation.
- Philosophy lengthThe design philosophy runs 4 to 6 paragraphs and is saved as a markdown file.
- Subtle referenceThe agent infers a reference from the request and embeds it quietly in the work.
- Visual-first ratioAbout 90 percent of the piece is visual and 10 percent is text.
- PNG or PDF outputSingle-page output is the default, and a multi-page piece is optional.
- Bundled fontsThe canvas-fonts folder ships 54 .ttf files and their OFL license texts.
- Refine passA second pass improves the piece without adding new graphics.
- Apache 2.0 license fileThe skill folder carries its own LICENSE.txt, which is Apache 2.0.
Frequently asked questions
What is the canvas-design skill?
It is the skills/canvas-design folder in anthropics/skills. It makes an AI agent write a design philosophy, then render a visual piece as PNG or PDF, using bundled fonts. The result is poster or art output rather than interface code.
How do I install the canvas-design skill?
Clone anthropics/skills with git, then copy skills/canvas-design into your agent's skills folder with every file in it. Invoke the skill and describe the piece. Claude Code users can also add the repository as a plugin marketplace, per the README.
Can it design a website?
No. It outputs single-page PNG or PDF artwork. For web interfaces use frontend-design, and for a WordPress build use UiChemy. Treat canvas-design strictly as a poster and art tool, not a site builder.
What formats does it produce?
PNG and PDF. A single page is the default and a multi-page piece is optional. A refine pass improves the result without adding graphics, so ask for it when the first render is close.
What is the design philosophy step?
The agent names a movement in one or two words and writes 4 to 6 paragraphs describing it. Examples in the skill include Concrete Poetry, Chromatic Language, Analog Meditation, Organic Systems and Geometric Silence.
Which AI agents can use canvas-design?
The skill is a SKILL.md folder, so any agent that loads skills from a folder can use it, including Claude Code, Cursor and Codex. It needs a way to run code and write files, since it renders the output to PNG or PDF.
What is a better fit for brand work?
Use brandkit to build brand color and type foundations. For a polished interface with distinctive style, high-end-visual-design gives layout and visual rules. Canvas-design suits one-off artwork and posters only, never brand systems.