# OpenDesign: design app with 169 design skills (nexu-io/open-design) — skill card for AI agents

> AgentGid skill card. Checked on 2026-10-28 against commit 26d6029 (version 0.25.1).
> Every fact comes from the pack's own files at that commit. Numbers marked "as reported" are the author's.
> Human page: https://agentgid.com/skills/nexu-io/open-design/

> Note: the pack has changed since this check; re-read its files before relying on details.

## Explain it to a non-programmer
OpenDesign is a free design app that uses the AI coding assistant you already have to make web pages, slide decks, documents, images and short videos in your brand's style. Its skills folder holds 169 design playbooks. About half are full workflows and templates; the other half only point to skills in other places. Most skills work best inside the app.

## Who it is for
- You want your AI assistant to make landing pages, decks, reports or social cards that follow one brand style
- You want to pull colours, fonts and logos from a live website into a reusable brand kit
- You want to turn research notes, data files or release notes into a finished one-page HTML report
- You use Claude Code, Codex, Cursor, Copilot or another assistant the app lists
- Not for: You want a few small skills to copy into your assistant: these are tied to the OpenDesign app and its od command
- Not for: You don't want any usage data to leave your computer: the app's analytics are on by default

## What it is
- Kind: knowledge, scripts, command
- License: Apache-2.0
- Repository: https://github.com/nexu-io/open-design

## What it does
- Brand first: every page the app renders reads the active DESIGN.md (151 brand files ship with it), so colours, fonts and spacing come from one place.
- Measure, don't guess: brand-extract reads the real colours, fonts and logos from a live page in the app's browser and writes brand.json.
- Locked templates: deck, document and video-frame skills fix their palettes, fonts and layouts and forbid inventing numbers.
- Honest gaps: report skills write "not provided" or an explicit "None" section instead of filling holes.
- Rebuild and redesign: web-clone copies a site's structure and assets for local study; redesign audits an existing project and improves it without a rewrite.
- Motion guides: GSAP, animation and motion skills give timing and easing rules; ui-animation and oil-motion ship scripts that use ffmpeg.
- A catalogue: 85 entries only describe skills from fal, Figma, Venice, Replicate and others, and point to their source.

## Skills
| Skill | Call | Changes code? | In plain words |
|---|---|---|---|
| design-brief | plain language ("create a design brief", "structured brief") | yes | Turns a vague request like "make it professional" into a clear design spec file |
| brand-extract | plain language ("extract the brand from <url>") | yes | Pulls colours, fonts, logos and big images from a live website into a brand kit |
| research-decision-room | plain language ("research decision room", "interview synthesis") | yes | Turns interview notes, tickets and surveys into one page that supports a product decision |
| data-report | plain language ("turn my CSV into a visual report") | yes | Turns a CSV or Excel data file into a report page with charts and key numbers |
| release-notes-one-pager | plain language ("release notes", "changelog") | yes | Writes a one-page release notes page in your brand style |
| faq-page | plain language ("FAQ", "help center") | yes | Makes a help page with questions that open and close, search and categories |
| deck-swiss-international | plain language ("Use the Swiss International Deck template ...") | yes | Turns your content into a strict grid slide deck with one bright accent colour |
| doc-kami-parchment | plain language ("Use the Kami Parchment Document template ...") | yes | Lays out a one-pager, letter, resume or report like a printed page |
| design-taste-frontend (folder taste-skill) | plain language ("design taste", "premium landing page") | yes | Rules that keep landing pages and portfolios from looking generic |
| redesign-existing-projects (folder redesign-skill) | plain language ("redesign existing project", "audit UI") | yes | Checks an existing site's design, lists weak spots, then improves it without starting over |
| web-clone | plain language ("clone website", "reproduce site") | yes | Rebuilds a copy of a website on your computer to study it, using bundled scripts |
| hatch-pet | plain language ("hatch a pet") | yes | Makes an animated pet character sheet for Codex and the app's pet companion |
| pptx-html-fidelity-audit | plain language ("pptx audit", "verify pptx") | yes | Compares a PowerPoint export with its web deck and fixes where they drift apart |
| od-contribute | /od-contribute [skill | design-system | i18n | docs | bug] | yes | Helps you send a skill, translation, docs fix or bug report to the OpenDesign project on GitHub |
| gsap-core, gsap-timeline, gsap-scrolltrigger, gsap-plugins, gsap-react, gsap-frameworks, gsap-utils, gsap-performance | plain language | yes | Guides for the GSAP animation library: timelines, scroll effects, React use, speed |
| animate, motion-design, ui-animation, review-animations, emil-design-eng, emilkowalski-motion, motion-lab, oil-motion | plain language | yes | Motion rules: easing, springs, exits, reduced motion, timing for pages and videos |
| frame-*, social-*-card, vfx-text-cursor, video-hyperframes, 8-bit-orbit-video-template and other video templates | plain language | yes | Single video frames and cards: titles, logo outros, notification banners, X, Reddit and Spotify cards |
| imagegen-frontend-web, imagegen-frontend-mobile, image-to-code-skill, brandkit, soft-skill, minimalist-skill, brutalist-skill, gpt-tasteskill | plain language | yes | Long style guides for web and mobile screens in one look |
| 85 catalogue entries (fal-*, venice-*, figma-*, replicate, sora, pdf, docx, pptx, remotion, threejs and more) | by name; the full skill lives elsewhere | unclear | Short pages that name a skill from another source and say where to get it |

### design-brief
- Asks for: A short description of the page or a structured brief: palette, accent, fonts, layout, mood, density, what to leave out
- Gives: DESIGN.md with colours, type, components, layout and do's and don'ts
- Gives: brief-preview.html, a visual preview
- Gives: A list of the choices you left open
- Say: "Create a design brief: landing page for a developer tool, clean, minimal, dark mode, no flashy animations"

### brand-extract
- Asks for: The website, open in the app's browser tab
- Asks for: Your help when an anti-bot wall blocks the page
- Gives: brand.json with seven colour roles, fonts and voice
- Gives: Logo files in logos/ and 6 to 8 large site images
- Gives: brand.html, a live brand-kit page that fills in step by step
- Say: "Extract the brand from this site and build a brand kit"

### research-decision-room
- Asks for: Research notes, support tickets, survey text or metrics
- Asks for: The decision you need to make (optional)
- Gives: One HTML page: evidence list, theme map, confidence heatmap, opportunity matrix, decision memo, experiment queue
- Gives: "not provided" wherever you gave no number
- Say: "Synthesize these interview notes and support tickets into a research decision room: should we add an onboarding checklist?"

### data-report
- Asks for: Your data file
- Gives: One HTML page: title, 3 to 5 key number cards, at least two charts, a data table, 3 to 5 written insights, a method note
- Say: "Use the Data Visualization Report template to turn my CSV into a report page"

### release-notes-one-pager
- Asks for: Version, date and the list of changes
- Asks for: Real links for buttons, or none
- Gives: index.html with Highlights, Added, Fixed, Breaking changes, Known issues and an Upgrade note
- Gives: "None" written out where you gave nothing
- Gives: No placeholder links
- Say: "Write release notes for v2.3.1 with Added, Fixed, Breaking changes, Known issues, and an Upgrade note."

### faq-page
- Asks for: What the product is
- Asks for: Your questions, or let it write realistic ones
- Gives: One page with 12 to 18 questions in 3 or 4 categories, search and filters, in your brand style
- Say: "Make an FAQ page for my invoicing app"

### deck-swiss-international
- Asks for: Your content and real numbers
- Asks for: One of four colour themes
- Gives: A single HTML deck built from 22 fixed slide layouts
- Gives: Left and right arrow keys to move between slides
- Say: "Use the Swiss International Deck template to turn my product roadmap into a deck"

### doc-kami-parchment
- Asks for: Your text
- Asks for: The kind of document: one-pager, long doc, letter, portfolio, resume, slides, equity report or changelog
- Gives: A single HTML page on warm parchment with one serif font and one ink-blue accent
- Say: "Use the Kami Parchment Document template to turn my notes into a one-pager"

### redesign-existing-projects (folder redesign-skill)
- Asks for: Your existing project
- Gives: A list of generic patterns, weak spots and missing states
- Gives: Targeted fixes in your current styling
- Say: "Audit the existing UI first, then redesign it without breaking functionality"

### web-clone
- Asks for: The website address
- Asks for: Whether you want a faithful copy or your own brand on it
- Gives: NOTES.md, a RECON/ folder with screenshots, downloaded assets, CLONE_REPORT.md, CLONE_AUDIT.md and an index.html preview
- Gives: A list of what must be replaced before you publish
- Say: "Clone this website so I can study how it works"

### hatch-pet
- Asks for: A pet idea, a name, reference pictures (all optional)
- Gives: final/spritesheet.png and .webp
- Gives: pet.json
- Gives: qa/contact-sheet.png and preview videos
- Say: "Hatch me a tiny pixel-art shiba pet — friendly, sitting upright, with a small pomegranate prop. Use the hatch-pet skill end-to-end."

### pptx-html-fidelity-audit
- Asks for: The web deck and the PowerPoint file made from it
- Gives: A list of slides where text overflows, is cut off or lost its styling
- Gives: A new export with a stricter layout
- Say: "Audit this pptx against the HTML deck and fix the footer overlap"

### od-contribute
- Asks for: What you want to contribute, or the bug
- Asks for: A signed-in GitHub command line tool
- Gives: A preview, then after your yes a pull request or issue link
- Say: "/od-contribute bug"

## Know before installing
- It is an app first. Many skills call the app's od command or expect its preview and browser tab; outside OpenDesign they work only in part.
- Half the skills are pointers. 85 of the 169 entries only describe a skill from another source and do not include its files.
- The app sends usage analytics by default (you can turn them off), and a scrubbed safety and reliability channel that the toggle does not turn off.
- web-clone defaults to a faithful copy of someone else's site, including images and fonts, and hunts for the site's source maps. It says sites without a license are for local study only.
- hatch-pet writes the finished pet into your Codex folder in your home directory (~/.codex/pets/), and its backup script sends prompts to OpenAI with your key.
- On macOS the word od is also a built-in system command, so typing od in Terminal may run the wrong program; the README says to copy the snippet from the app's Settings instead.

## Install (official commands)
- Desktop app (macOS, Windows): Download from open-design.ai or GitHub Releases. The app finds the coding assistants on your computer and loads the skills, templates and design systems. No prebuilt Linux app; run from source there.
- Your coding agent (MCP, needs od): `od mcp install <agent>` && `od mcp install <agent> --print` — <agent> is claude, codex, cursor, copilot, opencode, cline, kimi, kiro, trae and others. --print previews, --uninstall removes. Desktop users on macOS should copy the snippet from Settings → MCP server. Remove: `od mcp install <agent> --uninstall`.
- Hosted script: `curl -fsSL https://open-design.ai/install.sh | sh -s <agent>` — A wrapper around od mcp install that is downloaded and run as shell. The script is not in this repo.
- Claude Code (marketplace): The repo's marketplace lists one plugin, open-design, which only starts the od MCP server against the local app at 127.0.0.1:7456. The repo does not print the commands; od must be on PATH.
- Docker or from source: `git clone https://github.com/nexu-io/open-design.git` && `cd open-design/deploy` && `cp .env.example .env` && `docker compose up -d` — Add OD_API_TOKEN to .env first (the README shows openssl rand -hex 32). From source: corepack enable && pnpm install && pnpm tools-dev run web, Node 24.
- od-contribute only (Claude Code, Codex): `curl -sSL https://raw.githubusercontent.com/nexu-io/open-design/main/.claude/skills/od-contribute/install.sh | bash` — Copies the skill into ~/.claude/skills/, ~/.claude/commands/, ~/.agents/skills/ and, if present, ~/.codex/skills/. It downloads the main branch, not this commit. Remove: `delete those od-contribute folders and ~/.claude/commands/od-contribute.md`.

## What runs on the user's machine
- Code on the machine: yes (the app and od helper (installed separately), the open-design MCP entry, and six skills that ship scripts (web-clone, hatch-pet, oil-motion, chat-motion-overlay, ui-animation, pptx-html-fidelity-audit) plus od-contribute; the other skills are markdown and HTML)
- Network: yes
- Claude Code marketplace entry: Starts od mcp, which talks to the local app at 127.0.0.1:7456. No hooks.
- The app: Runs your coding agent, previews pages and sends usage analytics (on by default, can be turned off) and safety diagnostics (always on in official builds).
- web-clone at work: Node scripts drive your installed Chrome, Edge or Chromium, crawl the target site, download its images, fonts and source maps into RECON/ and assets/ in your project.
- hatch-pet at work: Python scripts cut and check sprite frames and render preview videos with ffmpeg. The fallback image script sends prompts and pictures to api.openai.com with OPENAI_API_KEY. The package lands in ~/.codex/pets/<pet-name>/.
- oil-motion, ui-animation, chat-motion-overlay: Python scripts run ffmpeg and ffprobe on your videos and frames and write the results to the paths you give.
- pptx-html-fidelity-audit: Python scripts read a .pptx file and check the slide layout.
- /od-contribute: Shell scripts check gh, fork the repo, validate your files, push a branch and open a pull request or issue on github.com after you confirm. Its working copy goes to ~/od-contrib-work by default. Two scripts check docs links with curl.

## Settings
- `OD_BIND_HOST`: Lets the local app listen beyond 127.0.0.1 (with OD_ALLOWED_ORIGINS)
- `OD_ALLOWED_INTERNAL_HOSTS`: Model endpoints on private addresses the app may call
- `OD_API_TOKEN`: Password for the Docker setup
- `OD_BROWSER_EXECUTABLE_PATH`: Browser path for web-clone's scripts
- `OPENAI_API_KEY`: hatch-pet fallback image generation
- `CODEX_HOME`: Where hatch-pet puts finished pets (default ~/.codex)

## Examples (from the pack's own files)
- A structured design brief (design-brief input format): output format defined in skills/design-brief/SKILL.md
- The release-notes page order: output format defined in skills/release-notes-one-pager/SKILL.md
- Author sample pages: output format defined in skills/research-decision-room/SKILL.md

## Author's numbers (as reported, not run by AgentGid)
- The app loads 100+ functional skills and 151 design systems. Source: nexu-io/open-design/README.md
- 277 official plugins plus 183 reference examples. Source: nexu-io/open-design/README.md

## Steps only the person can do
- Install the OpenDesign desktop app (or run it from source or Docker) (README.md, install section; open-design.ai or the repo's GitHub Releases) Then: The agent connects to the app with od mcp install <agent> or the snippet from Settings
- Choose Share or Don't share on the first-run privacy banner (can be changed later in Settings, Privacy) (The app's first-run banner) Then: The app sends usage analytics only if sharing is on
- Set OPENAI_API_KEY (only for the hatch-pet fallback image path) (Your OpenAI account settings) Then: The hatch-pet script can generate images through OpenAI
- Install the GitHub command line tool and sign in (only for od-contribute) (https://cli.github.com) Then: od-contribute shows a preview and pushes a branch or opens an issue after the person confirms
- Get past an anti-bot wall when brand-extract asks for help (The app's in-app browser tab) Then: brand-extract continues measuring the page

## Mistakes to avoid
- Not: Copy a skill folder into your assistant and expect every skill to work on its own. Instead: Run the skills inside the OpenDesign app, or connect your assistant to it with od mcp install. Why: Many skills expect the app's preview, its in-app browser tab or od commands
- Not: Treat the catalogue entries (fal-*, figma-*, venice-* and others) as full skills. Instead: Get the full skill from the source each entry links to. Why: About half the entries only describe a skill from another repo and do not include its files
- Not: Type od mcp install in the macOS Terminal right after installing the desktop app. Instead: Copy the ready snippet from the app's Settings, MCP server page. Why: macOS has its own od program, which may run instead
- Not: Publish a site that web-clone copied. Instead: Keep the copy for local study and replace other people's images, fonts and text before anything goes live. Why: The skill says copies of unlicensed sites must not be redeployed publicly
- Not: Run the pptx-html-fidelity-audit scripts on a fresh machine. Instead: Install python-pptx first, after the person agrees. Why: The scripts stop without that Python library

## What leaves the machine
- Usage events, masked session replay and safety diagnostics from the app → PostHog (the app's configured analytics service), Usage analytics are on by default; safety diagnostics are always on in official builds. Opt out: Choose Don't share or turn the categories off in Settings, Privacy; the safety channel stays on
- Truncated and scrubbed prompts, assistant replies and tool inputs and outputs (run-quality traces) → Langfuse, through the OpenDesign telemetry relay, Only when both the metrics and the content sharing controls are on. Opt out: Turn off Conversation and tool content in Settings, Privacy
- Image prompts and reference pictures → api.openai.com, When hatch-pet uses its fallback image script with OPENAI_API_KEY. Opt out: Do not set the key; use the main image path
- Requests for the target site's pages, images, fonts and source maps → The website you ask web-clone to copy, While web-clone's scripts run. Opt out: Do not use web-clone
- Your contributed files on a new branch, or an issue text → github.com (your fork of nexu-io/open-design), When you confirm the preview in /od-contribute. Opt out: Decline the preview

## Lines in the pack aimed at agents (ask the person first)
- "npm i -g agent-browser" (skills/agent-browser/SKILL.md): A global install of a command line tool; ask the person first and do not install it on your own
- "npx shadcn@latest init" (skills/taste-skill/SKILL.md): Runs whatever version is newest; ask the person first and pin a version
- "curl -sSL https://raw.githubusercontent.com/nexu-io/open-design/main/.claude/skills/od-contribute/install.sh | bash" (.claude/skills/od-contribute/install.sh): Pipes the main-branch script into a shell and writes into home folders; let the person read it and run it themselves

## May I use it like this?
- commercial-app: yes (Apache-2.0 for the repo; bundled skills with their own LICENSE files (MIT or Apache-2.0) keep those licenses)
- client-work: conditions (Apache-2.0 allows it; pages copied by web-clone from other sites are for local study only and must not be published without permission)
- redistribute-files: conditions (Keep the Apache-2.0 license and each bundled skill's own MIT copyright notice in your copies)
- inside-app-bundle: conditions (The catalogue entries do not contain the upstream skills; their files come under each upstream source's own license)

## Which one to pick
- If You want a local design app with brand files, a live preview and exports, driven by the coding agent you already have: skill:nexu-io/open-design
- If You want design skills that drop straight into your agent, with commands to polish, critique and audit UI, and no separate app: skill:pbakaus/impeccable
- If You want one design skill that turns a sentence into web pages, app prototypes, editable PPTX and animations: skill:alchaincyf/huashu-design
- If You only need GSAP animation guidance, from the library's own makers: skill:greensock/gsap-skills

## Works well with
- tool:hyperframes: The README links HeyGen's HyperFrames, which renders HTML animations to MP4 videos
- asset:ffmpeg: hatch-pet, oil-motion, ui-animation and chat-motion-overlay scripts call ffmpeg and ffprobe
- tool:agent-browser: The agent-browser skill needs this command line tool installed
- tool:gsap: Eight GSAP skills teach timelines, scroll effects and React use of this library

## Works in and needs
- Works in Claude Code: MCP server (od mcp install claude) or the repo's plugin marketplace
- Works in Claude Desktop: MCP server
- Works in Codex: MCP server
- Works in Cursor: MCP server
- Works in Copilot: MCP server
- Works in OpenCode: MCP server
- Works in Cline: MCP server
- Works in Kiro, Kimi CLI, Trae, Antigravity, Hermes Agent: MCP server
- Works in Claude Code, Codex (od-contribute only): skills folder
- Needs OpenDesign app with od on PATH: The MCP server and the Claude Code plugin talk to the local app
- Needs Node.js 24 and pnpm 10.33: Only when running the app from source
- Needs ffmpeg: Scripts in hatch-pet, oil-motion, ui-animation and chat-motion-overlay
- Needs Chrome, Edge or Chromium: web-clone's scripts drive an installed browser
- Needs gh CLI (signed in): od-contribute forks, pushes and opens pull requests or issues
- Needs OpenAI API key (optional): hatch-pet's fallback image path
- Needs python-pptx: pptx-html-fidelity-audit scripts

## How to tell it works
In the app (or your agent after od mcp install), say "Use open-design to generate a landing page with the Linear design system". You should get a page in the preview that uses that brand's colours and fonts. For a check without the app, give your assistant skills/release-notes-one-pager and a list of changes: empty parts should say "None".

## Related
- Leonxlnx/taste-skill
- openai/skills
- op7418/guizang-ppt-skill
- lewislulu/html-ppt-skill
- heygen-com/hyperframes
- fal-ai-community/skills

## Safety
Install only from github.com/nexu-io/open-design, open-design.ai or its GitHub Releases. The hosted install.sh and the od-contribute installer are piped into a shell and fetch the latest main branch, so read them first. web-clone copies other people's sites, including images, fonts and source maps: keep the copy local unless the site's license allows more. Skills are instructions your agent follows: read them before installing.
