Design & Motion · anthropics/skills

canvas-design

Name: canvas-design
Author: anthropics/skills

Create beautiful visual art in .png and .pdf documents using design philosophy. You should use this skill when the user asks to create a poster, piece of art, design, or other static piece. Create original visual designs, never copying existing artists' work to avoid copyright violations.

L2 Local · Restricted workspaceV1 Commit-PinnedWrite FileDual-pass Audit

Author anthropicsGitHub anthropics/skills ↗Stars 174801License Apache-2.0（skill 自带 LICENSE.txt 全文为 Apache License 2.0；仓库根无 LICENSE 文件，GitHub API license 返回 null，授权按 skill 目录各自携带）Commit 41bbe19d1a

Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.

1How It Works & Mechanism

Pure instruction (prompt orchestration) skill: Contains only SKILL.md + LICENSE + a fonts directory, without any scripts or executable code—all "canvas rendering" capabilities are delegated to the host model, which writes code on the fly for host execution.

canvas-design/SKILL.md

These are instructions for creating design philosophies - aesthetic movements that are then EXPRESSED VISUALLY. Output only .md files, .pdf files, and .png files. (L7) …… Take a second pass. Go back to the code and refine/polish further to make this a philosophically designed masterpiece. (L126)

Full-text grep (word boundaries) yields zero hits for invocation surfaces like python/curl/wget/exec/terminal/shell/token/api; L126 "go back to the code" confirms that rendering/refinement operates via model-generated code executed by the host—the skill itself serves merely as an "artistic field manual" for the model.

Same-session "Manifesto → Canvas" two-stage relay: First outputs a design philosophy .md (kept in the dialogue context without isolating to disk), and then the "same version" executes visual expression based on that manifesto, which provides aesthetic context while deliberately leaving interpretative room.

canvas-design/SKILL.md

Complete this in two steps: 1. Design Philosophy Creation (.md file) 2. Express by creating it on a canvas (.pdf file or .png file) (L9-11) …… Then, the same version receives the philosophy and EXPRESSES IT VISUALLY - creating artifacts that are 90% visual design, 10% essential text. (L25) …… Remain specific about the aesthetic direction, but concise enough that the next Claude has room to make interpretive choices also at a extremely high level of craftmanship. (L49)

The two steps relay within the same context (rather than re-injecting the .md as an external file), ensuring subsequent steps "read" the manifesto; simultaneously emphasizes leaving room for downstream interpretation to prevent cookie-cutter templating.

Countering the "AI-generated look" via structured prompt engineering: Provides no visual templates, instead embedding high-density craftsmanship terminology + a "final refinement" step simulating user dissatisfaction + a STOP rule against clutter to drive self-iteration.

canvas-design/SKILL.md

The user ALREADY said "It isn't perfect enough. It must be pristine, a masterpiece if craftsmanship, as if it were about to be displayed in a museum." …… If the instinct is to call a new function or draw a new shape, STOP and instead ask: "How can I make what's already here more of a piece of art?" (L124) …… Take a second pass. Go back to the code and refine/polish further to make this a philosophically designed masterpiece. (L126)

The FINAL STEP assumes the user is dissatisfied with the first draft, forcing the model into a second refinement pass rather than adding new elements—encoding "human designer rework habits" into the prompt, which is the most valuable implementation technique to emulate.

Driving composition via an "aesthetic worldview" rather than layout templates: Provides 5 condensed philosophical paradigms and recommends a "scientific diagram / systematic observation" visual grammar (repeating motifs, perfect shapes, dense markings, clinical typography, systematic reference markers) to depict abstract themes as "fictional discipline schematics."

canvas-design/SKILL.md

Generally use repeating patterns and perfect shapes. Treat the abstract philosophical design as if it were a scientific bible, borrowing the visual language of systematic observation—dense accumulation of marks, repeated elements, or layered patterns that build meaning through patient repetition and reward sustained viewing. Add sparse, clinical typography and systematic reference markers that suggest this could be a diagram from an imaginary discipline (L106)

A three-tier progression of "Philosophy → Paradigms → Visual Grammar": Establishes abstraction first before offering actionable composition syntax, preventing the model from producing generic poster templates.

Type as visual elements: Bundles 54 TTFs across 27 font families (all under SIL OFL licenses), instructing that text use varied typefaces and be "drawn" onto the canvas rather than treated as standard digital typesetting, while permitting on-demand downloads of additional fonts.

canvas-design/SKILL.md

IMPORTANT: Use different fonts if writing text. Search the `./canvas-fonts` directory. Regardless of approach, sophistication is non-negotiable. (L108) …… Download and use whatever fonts are needed to make this a reality. Get creative by making the typography actually part of the art itself -- if the art is abstract, bring the font onto the canvas, not typeset digitally. (L110)

canvas-fonts/ contains 81 verified files (54 TTF + 27 OFL texts). Families: ArsenalSC, BigShoulders, Boldonse, BricolageGrotesque, CrimsonPro, DMMono, EricaOne, GeistMono, Gloock, IBMPlexMono, InstrumentSans, Italiana, JetBrainsMono, Jura, LibreBaskerville, Lora, NationalPark, NothingYouCouldDo, Outfit, PixelifySans, PoiretOne, RedHatMono, Silkscreen, SmoochSans, Tektur, WorkSans, YoungSerif.

Dual built-in originality/copyright guardrails: The description explicitly forbids copying existing artists' works; the manifesto deliberately de-thematizes content (omitting explicit creative intent) so outputs serve as abstract art pieces rather than literal thematic diagrams.

canvas-design/SKILL.md

Create original visual designs, never copying existing artists' work to avoid copyright violations. (L3 description) …… Keep the design philosophy generic without mentioning the intention of the art, as if it can be used wherever. (L85)

Reinforced by L89-96's "subtle reference" step: Themes are woven into the canvas strictly via obscure conceptual DNA (analogous to a jazz musician quoting another track), systematically mitigating direct imitation risks.

2Core Capabilities

01Two-phase art workflow: Produces a 4–6 paragraph "art movement manifesto" design philosophy .md first, followed by the canvas creation based on it.

02Decodes "subtle references" from user requests into the canvas—connoisseurs notice, while others simply see an abstract masterpiece (jazz-quote principle).

03Museum/magazine-grade canvas composition guidelines: Scientific diagram visual grammar, limited color palettes, and self-checking discipline to prevent out-of-bounds or overlapping elements.

04Typesetting text as visual elements: Built-in library of 27 OFL font families + allows on-demand font downloads, rendering typography as part of the visual art itself.

05Built-in "human-grade craftsmanship" second-pass refinement: Simulates user dissatisfaction, enforces STOP rules against clutter, and exclusively polishes existing compositions.

06Multi-page extensions: Generates narrative series under the same philosophy (coffee-table book concept), packaging them into a single PDF or multiple PNGs.

07Strict output-type whitelist: Emits only .md / .pdf / .png files, delivered as single-file downloadable artifacts.

3External Dependencies

Kind	Dependency / Endpoint
network	按需字体下载（模型可选项，来源未指定）

4Recommended Governance & Evidence

Suggested Controls for Safe Operation

For Agent Runtime

Restrict working directory

For Administrators

Retain operation logs

Evidence & Triggers (File Paths & Excerpts)

本地工作目录: File I/O and artifact output: Outputs delivery files: design philosophy .md, final canvas .pdf/.png (L7/L116, the core function of the skill).

Audit Notes: Pure prompt instruction with zero scripts, zero hardcoded network calls, and zero credential access; deterministic behavior = reading bundled fonts + writing .md/.pdf/.png delivery files (strictly local file I/O). Sole gray area: L110 permits the model to download fonts autonomously, but no endpoints exist in the source (defaults to the bundled 54 fonts); rendering depends on host execution of model-generated code, falling under host platform responsibilities.

5Second-Pass Verification

[ok] 无脚本、无硬编码网络调用、无凭证读取 — SKILL.md 全文单词边界 grep 零命中调用 token；glob+git ls-tree 显示目录仅 SKILL.md/LICENSE.txt/canvas-fonts，无隐藏脚本
[ok] canvas-fonts 目录存在且与 SKILL.md 引用一致 — L108 引『Search the `./canvas-fonts` directory』；实测目录 81 文件（54 TTF + 27 OFL 文本）
[ok] 字体文件真实有效 — file 命令：WorkSans-Regular.ttf / Tektur-Medium.ttf 均报 TrueType Font data, 16 tables
[ok] LICENSE.txt 为 Apache-2.0 — LICENSE.txt 首段『Apache License / Version 2.0, January 2004 / http://www.apache.org/licenses/』
[ok] 关键引语与行号可定位 — L7/L9-11/L25/L49/L85/L89-96/L106/L108/L110/L116/L124/L126/L128-130 逐条 grep 命中且与原文一致
[ok] 元数据（commit/stars/last_push） — HEAD==41bbe19d1a1a7eaab5e7bb9050a417e5c6cffc8f（2026-09-03）；gh api：stars=174801，pushed_at=2026-09-03T16:37:14Z
[ok] official_desc 原文引用 — 与 SKILL.md frontmatter L3 description 逐字一致
[unlocatable] 「Download fonts」的网络端点 — L110 允许下载字体但全文无任何 URL/域名；定为模型可选项与潜在面，不计为确定性调用

6Verdict & Scenarios

Zero dependencies and zero scripts; minimal attack surface and fully ready out of the box.

Encodes human designer rework habits into the prompt workflow: Simulates user dissatisfaction + STOP rules against clutter + second-pass refinement to systematically eliminate generic AI draft aesthetics.

Ships with 54 OFL fonts across 27 families, ensuring consistent typesetting across environments without relying on host system fonts or network downloads.

Dual built-in copyright/originality guardrails: Description explicitly bans reproducing artists' works; manifestos deliberately de-thematize and rely solely on subtle conceptual references.

Portable and composable workflow: The manifesto .md can be reused for multi-page series and coffee-table-book narrative extensions.

Best for: Users of Claude-family hosts equipped with canvas/code rendering capabilities; ideal for creating static art objects such as posters, event visuals, covers, abstract typographic art, and brand visual explorations; perfect for creators who embrace a "worldview-first, execution-second" workflow and seek non-templated outputs.

Not for: Plain-text/non-rendering hosts; print workflows requiring precise corporate VI/brand template reproduction or editable source files (.ai/.figma/.psd); scenarios demanding real photography/3D assets instead of programmatic graphics; requests to directly replicate the style of living artists (explicitly prohibited by the skill).

Install Agent-Ready

① Verified Mirror 2026-09-06

Option A · Direct DownloadDownload canvas-design.tar.gz

sha256: 9c25c6d4b005a6ec…

Option B · Copy Guide for Agent

install.json

Agents automatically download and verify mirrors from this JSON guide.

② Upstream GitHub

GitHub Repo ↗

Pinned to commit 41bbe19d1a.