Design Distill logo

Design Distill

Community
Muluk-m
design-distill

Extract a design system from a website URL or local project into a structured token set (tokens.json, canonical) + a rendered DESIGN.md (Google Stitch-compatible) with semantic roles, essence, and a WCAG audit. Saves to the global library only (~/.config/design-distill/<name>/), never to the project root. When invoked without arguments, lists all saved design systems.

Overview

PublisherMuluk-m
Repositorydesign-distill
Skill namedesign-distill
Stars
11
Forks
2
Bundled files
1
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 1 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by Muluk-m on GitHub. Read the source before you install it.

Installation

Install the Design Distill AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/Muluk-m/design-distill/tree/main/skills/design-distill
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/Muluk-m/design-distill.git /tmp/design-distill
mkdir -p .claude/skills
cp -r /tmp/design-distill/skills/design-distill .claude/skills/design-distill
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Design Distill in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Design Distill on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Design Distill is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Design Distill

No-Argument Behavior: List Styles

When invoked without arguments, list all saved styles with agent-native file ops:

bash
ls -1 ~/.config/design-distill/ 2>/dev/null   # honors $DESIGN_DISTILL_HOME

If the library directory is empty or missing, suggest running node scripts/setup.mjs to seed the bundled styles (github, linear, notion, stripe, vercel).

Display the result and ask what the user wants to do next.


Distill from URL

Step 1: Confirm Source

If the user provided a URL, use it directly. Otherwise ask:

  • Website URL?
  • Style name? (default: derive from domain, e.g., https://linear.app → linear)

Step 2: Bootstrap & Probe

The skill is self-contained — it does not depend on any external /browse skill. It ships bundled primitives under scripts/ (run via node) backed by a single browser dependency (Playwright/Chromium).

First ensure the environment and learn the capability tier:

bash
node scripts/setup.mjs --probe       # → { browser, extractor, source, tier }
# If browser is false and you want visual capture:
node scripts/setup.mjs               # reuses system Chrome/Edge or a CDP endpoint if present;
                                     # only downloads Chromium when no browser is found. Also seeds bundled styles.

The skill reuses a browser you already have — set DESIGN_DISTILL_CDP=<ws-endpoint> to attach to a running browser, or just have Chrome/Edge installed. No forced 150MB download when a usable browser exists.

  • tier: "full" → extract + screenshots (best fidelity)
  • tier: "token-only" → extraction only; tell the user visual validation was skipped

Step 3: Extract Tokens + Capture Pixels

  1. Extract tokens (source precedence MCP → pinned dembrandt → native fallback, all handled inside the primitive):

    bash
    node scripts/extract.mjs <url>      # → stable token-set JSON on stdout

    The JSON contains colors (with confidence), typography, spacing, radius, shadows, components.

    Comprehensive capture (opt-in — heavier, only when the request needs it):

    bash
    node scripts/extract.mjs <url> --crawl 3   # multiple pages, merged
    node scripts/extract.mjs <url> --dark-mode # also capture the dark palette (kept as a variant)
    node scripts/extract.mjs <url> --mobile --slow --cookie "<s>" --header "<h>"

    Cost rule of thumb (each multiplies the dembrandt run): default single-page ≈ 1×; --crawl N ≈ N×; --dark-mode adds a second full run; --slow ≈ 3× timeouts. Default to single-page light unless the task needs breadth.

  2. Capture screenshots for visual ground truth (full-tier only) with the bundled primitive — not /browse:

    bash
    node scripts/screenshot.mjs <url> --out <dir> --viewports desktop,mobile --schemes light
    • Observe the real background color (light or dark? don't guess from the brand name)
    • Capture 1-2 representative subpages (e.g., /pricing, /docs) by re-running with their URLs
  3. Combine both sources:

    • Use the extracted exact values (hex colors, px/rem sizes, font names) for the data
    • Use screenshots to validate: overall lightness/darkness, visual density, design personality
    • When they conflict, trust the screenshot

If the site has /design, /brand, /style-guide, or /storybook, capture those too.

Screenshots are primary. On token-only tier (no browser), proceed from token data alone and clearly note that visual validation was skipped.

Step 4: Normalize + Generate (structured-first)

The canonical artifact is a structured token set (tokens.json); DESIGN.md is rendered from it. Pipe the extracted tokens through build-design, which applies semantic normalization (raw → roles + decision rules), derives the essence, and renders the document:

bash
node scripts/extract.mjs <url> \
  | node scripts/build-design.mjs --name <name> --source-url <url> --out ~/.config/design-distill/<name>
# writes ~/.config/design-distill/<name>/tokens.json (canonical) + DESIGN.md (view)

build-design records every role assignment / override in tokens.json.decisions and the rendered "Normalization decisions" section, so the result is inspectable.

Key principles:

  • Every value must be concrete — #1a1a2e, not "a dark blue"
  • Visual judgments (light/dark, personality) come from screenshots — read them and correct the rendered tone/essence if the data disagrees
  • tokens.json is canonical; never hand-edit DESIGN.md as the source — re-render it
  • source_url is recorded for design-apply to re-screenshot

Step 5: Save to Global Library Only

Derive default style name:

  • URL source: extract domain without TLD (e.g., https://linear.app → linear)
  • Confirm with user

Check for collision with agent-native file ops (the library lives at ~/.config/design-distill/, overridable via DESIGN_DISTILL_HOME):

bash
ls ~/.config/design-distill/<name>/DESIGN.md 2>/dev/null

If the style exists, ask: "Style <name> already exists. Overwrite or choose a different name?"

Save — write the DESIGN.md directly:

bash
mkdir -p ~/.config/design-distill/<name>
# Write DESIGN.md content to ~/.config/design-distill/<name>/DESIGN.md

Save screenshots to ~/.config/design-distill/<name>/screenshots/ for archival. Previous screenshots are overwritten on re-distill.

IMPORTANT: Do NOT write DESIGN.md to the project root directory. The distilled design system belongs to the global library, not the current project.

Completion message:

"Saved to global library as <name>. Use design-apply 用 <name> 做个 <页面> to generate with this style."


Distill from Local Project

Step 1: Read Design Sources

Read in priority order:

bash
# 1. Tailwind config (richest source)
find . -name "tailwind.config.*" -not -path "*/node_modules/*" | head -3

# 2. CSS custom properties
grep -r "^:root\|^html" --include="*.css" --include="*.scss" -l | head -10

# 3. Design token files
find . -name "tokens.*" -o -name "design-tokens.*" -o -name "theme.*" \
  -not -path "*/node_modules/*" | head -5

# 4. Component sampling
find . -path "*/components/*" \( -name "Button*" -o -name "Card*" \) | head -5

Step 2: Generate and Save

Same as URL flow Step 4-5, but:

  • Style name defaults to directory name (e.g., ./my-app → my-app)
  • No source_url in header (local projects don't have one)
  • No screenshots to archive

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Design Distill AI skill do?

Extract a design system from a website URL or local project into a structured token set (tokens.json, canonical) + a rendered DESIGN.md (Google Stitch-compatible) with semantic roles, essence, and a WCAG audit. Saves to the global library only (~/.config/design-distill/<name>/), never to the project root. When invoked without arguments, lists all saved design systems.

Why use Design Distill on TypingMind?

Because you install it once and use it with any model. Design Distill is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Design Distill in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/Muluk-m/design-distill/tree/main/skills/design-distill. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Design Distill?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Design Distill?

As many as you like. As long as a model supports skills, you can use Design Distill with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Design Distill AI skill free?

Yes. It is published on GitHub by Muluk-m under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇