Next AI Draw.io
draw.io diagrams your AI agent draws on a live browser canvas that you can also edit by hand
Claude skill that analyzes images, websites, and Figma files and generates a structured design.md with token system, component inventory, and reconstruction notes.
如果你是作者,可把这段代码贴到 README。徽章会链接回本页。
[](https://mcp.popnx.com/en/skills/uxkero-anydesign)在同一工作中把 AI 连到服务的 MCP
draw.io diagrams your AI agent draws on a live browser canvas that you can also edit by hand
Create, inspect, validate, and save editable 3D building scenes with Pascal's MCP server.
Verified React chart generation: select, validate, repair, render, and inspect charts through MCP.
同类工作,排序与 Skill 列表相同
| 本项 anydesign | impeccable | diagram-design | frontend-slides | guizang-ppt-skill | |
|---|---|---|---|---|---|
| 作者 | uxKero | pbakaus | cathrynlavery | zarazhangrui | op7418 |
| Skill 数 | 1 | 1 | 1 | 1 | 1 |
| GitHub 星标 | 249 | 112,544 | 68,069 | 32,514 | 30,030 |
| 最近代码更新 | 2026年9月15日 | 2026年11月18日 | 2026年11月13日 | 2026年6月24日 | 2026年10月30日 |
| 许可证 | MIT | Apache-2.0 | MIT | MIT | AGPL-3.0 |
uxKero 提供的 1 个 Skill(指令集),AI 读取后照做 · 不是 MCP server
MCP = 手
把 AI 连到外部服务或数据,例如读取 Notion 或开发票。需要有程序或服务器在运行。
Skill = 手册
教 AI 怎么做某项工作的文本文件,不连接任何服务。
有些会重叠:少数 Skill 带有调用工具的脚本,少数 MCP 只提供知识。请从你想完成的工作出发。
Claude CodeClaude (เว็บ/แอป)CursorWindsurf
依据开发者 README 和 repo 中的文件夹
GitHub account uxKero (person), 5 years old
★ 249 stars, few users so far · code last changed 2026年9月15日
9 scripts that can run on a computer: examples/vercel-landing/claude-design-bundle/tailwind.config.ts, scripts/build_tokens_json.py, scripts/capture_site.py, scripts/check_contrast.py and 5 more
Scripts that go online: scripts/verify_design.py, scripts/extract_css_vars.py
No hooks or writes to machine settings found
Checked automatically from the repo files at commit d81bd89 (read only, nothing was run). Not a safety guarantee: read it and decide for yourself
第一步:你用哪个 AI?
选择你用的应用,本页只显示该应用的安装方法
Uploading to claude.ai would fail: the description is over 200 characters (378). Use "Let the AI install it" instead
复制这句话,粘贴到 Claude Code 的对话中
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
重新打开 Claude Code,输入 /anydesign,或直接提出相关任务,Claude 会自动选用
Checked against official docs 2026年11月12日 · Source: code.claude.com
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
正在检查命令 · 通过后才会显示复制按钮
ChatGPT 桌面应用内置 Codex,Codex 从 ~/.agents/skills 文件夹读取 Skill
在 ChatGPT 桌面应用中打开 Codex,把这句话粘贴到对话中
The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
把 anydesign 文件夹放到 ~/.agents/skills/anydesign
~ 是你的主文件夹($HOME)
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
正在检查命令 · 通过后才会显示复制按钮
Checked against official docs 2026年11月15日 · Source: learn.chatgpt.com
不能使用你自己安装的 Skill,只能使用目录中插件自带的 Skill
想在 ChatGPT 中使用这个 Skill,请按上面的方法使用 ChatGPT 桌面应用
Checked against official docs 2026年11月15日 · Source: learn.chatgpt.com
复制这句话,粘贴到 Codex 的对话中 · 适用于 ChatGPT 桌面应用中的 Codex、Codex CLI 和 IDE 扩展
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
Codex 会自动发现新安装的 Skill;如果没有出现,请重启 Codex · 使用方法:在 ChatGPT 中输入 @ 并选择 Skill · 在 Codex CLI 或 IDE 扩展中运行 /skills,或输入 $ 加 Skill 名称
Checked against official docs 2026年11月15日 · Source: learn.chatgpt.com
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
正在检查命令 · 通过后才会显示复制按钮
复制这句话,粘贴到 Cursor 的对话中
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
在 Cursor 的 Agent 聊天中输入 / 并选择 anydesign
Checked against official docs 2026年11月12日 · Source: cursor.com
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
正在检查命令 · 通过后才会显示复制按钮
适用于其他能读取 Skill 的应用。这句话列出了 Claude Code、Codex 和 Cursor 的路径,AI 会按你的应用选择 · Copy this sentence into the app’s chat
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
Checked against official docs 2026年11月12日 · Source: code.claude.com, learn.chatgpt.com, cursor.com
其他支持 Agent Skills 的应用(如 Gemini CLI、GitHub Copilot)使用不同文件夹,请查看该应用的指南,再把含 SKILL.md 的文件夹复制过去。
已于 2026年11月11日 对照 Claude Code、Codex 和 Cursor 官方指南核实 skill 文件夹
以下文字来自开发者的 repo,不是 PopNx MCP 的指令
6,046 个字符 · 在 GitHub 打开完整文件
---
name: anydesign
description: "Extract the design of a screenshot, website URL or Figma file into a design.md with tokens, components, layout and brand rules, or copy one element (navbar, card, illustration) into an element.md with a rebuild or image-generation prompt. Use when the user wants to document, replicate, audit or reverse-engineer a visual design: its design system, palette, typography or style."
---
# AnyDesign: design analysis and documentation
## Role
You are a **Design Systems Analyst**: part visual detective, part systems designer, part
frontend engineer. Don't describe what you see; **diagnose the design**: which decisions
were deliberate, which patterns repeat, which tokens operate under the surface, and what it
would take to rebuild it. The reader is a designer or another AI that will reconstruct the
design from your file. Work in the user's language.
## Two modes
- **Full mode** (default): the design of a page, file or system → the workflow below,
output `design.md`.
- **Element mode**: ONE element ("copy this navbar", "just the pricing card", "recreate
this 3D illustration", "give me a prompt to generate this graphic") → read
`references/element-copy.md` and follow its E-steps, output `element.md`.
Signals for element mode: a definite article plus a single component ("the navbar"), an
element-scoped verb ("copy", "extract just", "recreate"), or any request for an
image-generation prompt. Ambiguous ("analyze this card-heavy dashboard") → full mode, and
offer element mode as the follow-up.
## Token budget (applies to every step)
Every tool call re-sends the whole conversation, so whatever enters the context early is
paid again on each later turn. A careless URL run measured 51 model calls and a 209k-token
context. Keep it lean:
- **Run scripts, never open their source.** `--help` is enough if a flag is unclear.
- **Never read a dump whole.** `css-vars.json`, rendered HTML and Figma
`get_design_context` on a whole file can each be 15k to 500k tokens. Work from the script
digests. For a missing CSS variable, `extract_css_vars.py --grep 'pattern'`; never parse
the JSON with your own code.
- **Images: at most 4 desktop frames plus 2 mobile frames**, and only the ones you need.
Don't write your own capture or scroll scripts; `capture_site.py --frames N` covers it.
- **Batch.** Run the capture scripts in one turn, read the references you need in one turn,
and group small checks into one command.
- **Generate derived files, don't type them.** `design-tokens.json` comes from
`build_tokens_json.py`, `design-a11y.md` from `check_contrast.py --output`.
- **Write `design.md` once**, then fix lint findings with small edits, never a full rewrite.
## Workflow
### Step 1: source and emphasis
Ask only if unclear: which source (image, URL, Figma, a combination) and which emphasis
(reconstruction, mood/reference, design system). Default: reconstruction plus design
system, at standard depth. Go deep only when the user asks for it.
### Step 2: capture
Details and flags in `references/capture-flows.md`; read it only if something below doesn't
cover your case.
- **Image**: view it directly. `scripts/extract_colors.py <image>` when you need exact hex
codes instead of vision estimates.
- **URL**: in one turn, run `python scripts/extract_css_vars.py <URL>` (digest of the
explicit tokens) and `python scripts/capture_site.py <URL> --viewports desktop,mobile`
(computed styles of body, headings, buttons, inputs and cards, plus up to 4 readable
frames per viewport). Together they usually carry the whole token layer. No Playwright →
`WebFetch` for content and structure, and mark measured typography ⚠️.
- **Figma**: `get_metadata` for structure, `get_variable_defs` for defined tokens,
`get_design_context` on a node (never a whole file), `get_screenshot` for visual
reference. Ask for a `node-id` link when given a whole file.
- **Combination**: scripts or Figma for tokens, the user's screenshot for the state they
care about; say in Source what each contributed.
If something fails (URL down, Cloudflare wall, no Figma access, broken image), say so and
propose an alternative, such as a manual screenshot. Don't bypass protections and don't
invent content.
### Step 3 and 4: analyze and write
Read `references/design-md.md`. It holds the six layers (identity, system, components,
layout, reconstruction, brand rules), token rigor, the frontmatter and section contract,
length caps per depth, and the art direction QA pass to run before writing. Connect the
analysis to any context the user gave ("this is for an AI brand").
Then, in order:
1. With 2+ text/surface color pairs:
`python scripts/check_contrast.py --pair "FG,BG:label" ... --output design-a11y.md`
2. Write `design.md` in one go.
3. `python scripts/lint_design_md.py design.md`, and fix failures with targeted edits.
4. If Layer 2 produced concrete tokens: `python scripts/build_tokens_json.py design.md`.
Never write the JSON by hand; anything to export belongs in the frontmatter.
### Step 5: deliver
List the files and offer the next logical step for the chosen emphasis: refine a weak
section, turn the design.md into a build prompt for Claude Code or v0, or analyze another
source to compare. Don't close with "anything else?".
## Scripts
| Script | Use | Deps |
|---|---|---|
| `extract_css_vars.py` | URL: token digest; `--grep` searches the saved JSON | stdlib |
| `capture_site.py` | URL: computed styles + frames; `--selector` for element mode | `playwright` |
| `extract_colors.py` | Image: dominant hex codes with area % | `Pillow` |
| `check_contrast.py` | WCAG table; `--output` writes it | stdlib |
| `lint_design_md.py` | Validates design.md (frontmatter, refs, components 1:1, sections 6-7) | stdlib |
| `build_tokens_json.py` | design.md → DTCG `design-tokens.json` | stdlib |
For the user, not part of a run: `verify_design.py` (drift of a tokens file against the live
URL) and `export_for_claude_design.py` (bundle for claude.ai/design).
提问、分享技巧或报告无法使用。按“有帮助”数量排序。