Nanosaur2-670M
well9472 · Kit
Makes illustrations from text prompts with a small 670M-parameter diffusion model, run through ComfyUI. Built as a low-budget research model for people who wan…
New
Overview
ConardLi’s six free skills for AI coding agents: code-rendered motion video, click-through web presentations, web design, GPT Image 2 prompts and images, beautiful HTML articles and local knowledge-base answers.
Garden Skills is an open-source project on GitHub in the Developer category, listed by Mindket. It is free. This is version 2, last updated on . Install it from GitHub as its README says. Get Mindket’s setup notes on your Mindket account, as a zip.
At a glance
- Runs in
- Claude Code: through the plugin marketplace, the skills CLI or a copied folder
- Cursor: the skill’s folder goes in .agents/skills/
- Codex CLI: .codex/skills/ in the README, and .agents/skills/ in the skills CLI’s docs
- OpenCode: .opencode/skills/
- Start with
npx skills add ConardLi/garden-skills -s web-design-engineer
Platform
Claude Code
Cursor
Codex
OpenCode
Gemini CLI
Claude
About
What it is
Garden Skills is ConardLi’s collection of six skills for AI coding agents. Each skill is a folder the agent loads when your request fits it. motion-showreel renders motion videos from code, up to 20 minutes. web-video-presentation turns an article or a narration script into a click-through web presentation that you screen-record. web-design-engineer builds and critiques web pages and prototypes. gpt-image-2 writes prompts for GPT Image 2 and can run its image scripts. beautiful-article turns a source into a single-file HTML article. kb-retriever answers questions from a local folder of documents. Your agent does the work, and you answer its questions at the stops each skill sets.
- You ask your agent for a video, page, image or article
- The agent asks what it still needs from you
- You approve the plan or style frames at its stops
- The agent builds and checks the result
Who it suits
Developers, designers and creators who already work with an AI coding agent and want finished things from it. That means a video, a page, an image prompt or a long article, not only code. Most skills run scripts and need Node.js, so they suit people who are comfortable with a terminal. web-design-engineer needs none of that.
| Who | Skill | Example |
|---|---|---|
| Product teams | motion-showreel | A team gives three screenshots and asks for a 20-second teaser with music. The team confirms the requirements and style frames, then approves the export. |
| Video makers | web-video-presentation | A video maker gives an article and gets a script, an outline and a click-through presentation to screen-record. |
| Designers and front-end builders | web-design-engineer | A builder asks for a dashboard prototype and reacts to an early draft before the full build. |
| Marketers | gpt-image-2 | A marketer asks for a poster and gets a saved prompt, and an image file when an image route is set up. |
| Analysts and writers | beautiful-article | An analyst gives a long PDF and gets a share-ready article as one HTML file. |
| Support and operations teams | kb-retriever | A support lead asks a question over a folder of product docs and gets an answer that names the files used. |
Where to use it
Any agent that reads skills. The README lists six as tested.
- Claude Code: through the plugin marketplace, the skills CLI or a copied folder
- Cursor: the skill’s folder goes in
.agents/skills/ - Codex CLI:
.codex/skills/in the README, and.agents/skills/in the skills CLI’s docs - OpenCode:
.opencode/skills/ - Gemini CLI: an extension manifest, which the README does not detail
- Claude.ai: Settings, then Capabilities, then Skills
How to set it up
Get it saves a zip of Mindket’s setup notes in your downloads folder, with every route the README gives.
| Need | Detail |
|---|---|
| An AI coding agent | One that reads skills. The README tests Claude Code, Claude.ai, Cursor, Codex CLI, Gemini CLI and OpenCode |
| Node.js | For the skills CLI route, 22.20 or later per its own README. Several skills also run Node scripts, and motion-showreel asks for Node 18+ and ffmpeg |
| Git | Only for the copy and submodule routes |
| Tools for one skill | Only for the skill you use: ffmpeg for motion-showreel, an image API key for gpt-image-2’s own image route, PDF and Excel tools for kb-retriever |
| Accounts and keys | None to install. Narration and image routes use your own keys. The license is MIT |
- Check that Node.js is recent enough. The skills CLI’s own README asks for 22.20 or later. If your machine has none, install Node.js first.
- Run the install line for one skill. The skills CLI detects your agent and puts the skill in its skills folder. Add
--globalfor a user-level install. - Ask for what the skill covers. If the skills folder is new, run
/reload-skillsin Claude Code first. Some skills need one more setup step, listed in the kit’s notes.
npx skills add ConardLi/garden-skills -s web-design-engineer
To install all six skills, leave off the skill name.
npx skills add ConardLi/garden-skills
In Claude Code, the plugin marketplace is another route. You run these inside a session, and the plugin’s skills then run under the plugin’s name.
/plugin marketplace add ConardLi/garden-skills /plugin install web-design-skills@garden-skills
motion-showreel needs one more step. You install Node 18+ and ffmpeg yourself, then run this from the skill’s folder.
sh scripts/setup.sh
How to use it
Ask in plain words for what a skill covers. Each skill stops at set points to ask you, and the agent does the building between them.
| Skill | Ask it for | What comes back, and where you step in |
|---|---|---|
| motion-showreel | A short motion video, such as a logo sting, a product teaser, kinetic type or a narrated explainer up to 20 minutes | An MP4 plus editable source. You confirm the requirements and the style frames, then watch a preview and approve the export. |
| web-video-presentation | An article or a narration script to turn into a click-through 16:9 web presentation | A script, an outline and a Vite project. You confirm five choices at one checkpoint, and may add narration audio. |
| web-design-engineer | A page, dashboard, prototype, slide deck or UI mockup, or a critique of one | A design system you confirm, an early v0 draft you react to, then the full build. |
| gpt-image-2 | A poster, UI mockup, infographic, diagram or an edit of an image you give it | A saved prompt, and an image file when an image route is set up. Otherwise a prompt for you to paste into an image tool. |
| beautiful-article | A URL, PDF, DOCX, Markdown file or text to turn into a share-ready article | One self-contained HTML file, with an optional PDF, after you confirm the plan, the first spread and the delivery. |
| kb-retriever | A question about a folder of Markdown, PDF and Excel files | An answer that names the files it used, found in at most five search rounds. |
gpt-image-2 starts with a mode check. With the two Garden environment variables set, it makes images itself. With an image tool already in your agent, it uses that tool. With neither, it hands you a prompt.
Example
You have no style in mind and want a pricing page prototype for a small budgeting app.
Design a pricing page for my budgeting app. I have no brand yet, so show me directions.
web-design-engineer offers three directions from three different design schools, each with a reason it fits, and you pick one.
After you pick, it states a design system for you to confirm, then shows an early v0 draft in that direction.
For newcomers and experts
If you are new. A skill is a folder of instructions your agent follows when your request fits it. Start with web-design-engineer, which needs no setup. Then try gpt-image-2 with no keys, so the agent writes a prompt and you paste it into the image tool you already use. Add the heavier skills when you need a video or an article.
For experts. Pin a release when a build must not change under you. Use the README’s release zip with its SHA-256 file. Its short ConardLi/garden-skills/tree/<tag>/... line for the skills CLI does not pin, since the CLI reads tree/<tag> there as a folder.
gpt-image-2 reads ENABLE_GARDEN_IMAGEGEN and OPENAI_API_KEY, and takes OPENAI_BASE_URL and OPENAI_IMAGE_MODEL as options, so an OpenAI-compatible image gateway can stand in. motion-showreel captures a page frame by frame and joins the frames with ffmpeg, and its narration takes only a script you supply, spoken by Gemini, MiniMax or an adapter you write.
Good to know
Five of the six SKILL.md files are written in Chinese. web-design-engineer’s is English, and the README links Chinese and Japanese versions.
- Setup is yours: motion-showreel’s setup script never installs system software, so you install Node.js and ffmpeg, and the PDF tools kb-retriever uses
- Narration: motion-showreel never writes the narration for you, and the agent reads only a script you supply
- Image keys: gpt-image-2 calls OpenAI’s image API or the gateway you name with your own key. Without one, it hands the prompt to your agent’s own image tool, or to you
- Sources first: beautiful-article asks for your material and does not invent an article from nothing
- Browser tests: web-design-engineer runs them only when you ask for acceptance or QA
- Photoreal people, filming and general video editing are outside motion-showreel
- Two README details differ from the skills CLI’s docs: where a global install goes, and which folder Codex reads. Look in your agent’s folder after installing
- Maker: ConardLi, MIT licensed. A community project, not made by Mindket
README and files
# Garden Skills: setup notes This kit holds setup notes written by Mindket. It does not contain the skills, their scripts, themes or templates. Those install from ConardLi’s repository: https://github.com/ConardLi/garden-skills License: MIT. The repository’s LICENSE file (https://github.com/ConardLi/garden-skills/blob/main/LICENSE) reads “MIT License” and “Copyright (c) 2026” with no name after the year, and the README ends “MIT © ConardLi”. Garden Skills is a community project by ConardLi. Mindket lists it and did not write the skills. The pages these notes came from: the README (https://github.com/ConardLi/garden-skills/blob/main/README.md), each skill’s README and SKILL.md under https://github.com/ConardLi/garden-skills/tree/main/skills, and the plugin file https://github.com/ConardLi/garden-skills/blob/main/.claude-plugin/marketplace.json. Facts about where Claude Code loads skills and plugins come from Claude Code’s own docs, named where they are used. Facts about the skills CLI’s folders come from its docs (https://github.com/vercel-labs/skills). Mindket copied every command. It ran only the release zip’s download and checksum steps, to test them, and installed nothing (setup/pinned-zip.md). ## What it is Six skills for AI coding agents. Each one is a folder with a SKILL.md and, for most, scripts, references and templates. | Skill | What it does | |---|---| | motion-showreel | Code-rendered motion videos, from logo stings to films up to 20 minutes, as MP4 plus editable source | | web-video-presentation | Turns an article or a narration script into a click-through 16:9 web presentation you screen-record | | web-design-engineer | Builds or redesigns polished web pages, dashboards, prototypes and slide decks | | gpt-image-2 | Prompt templates and scripts for GPT Image 2 image generation and editing | | beautiful-article | Turns a URL, PDF, DOCX, Markdown file or text into a single-file HTML article | | kb-retriever | Answers questions from a local folder of Markdown, PDF and Excel files | Source: the README’s skills gallery and each SKILL.md. ## Files in this kit - setup/skills-cli.md: the README’s Option A, the skills CLI (`npx skills add`) - setup/claude-code.md: Option B, the Claude Code plugin marketplace, and Option D for Claude Code, copying a folder by hand - setup/other-agents.md: Option D for Cursor and other agents, Claude.ai, and Option E, a git submodule - setup/pinned-zip.md: Option C, a pinned release zip with a checksum - setup/what-each-skill-needs.md: what each skill needs on your machine, and what you do versus what the agent does ## What you need first - An AI coding agent that reads skills: the README tests Claude Code, Claude.ai, Cursor, Codex CLI, Gemini CLI and OpenCode. - Node.js for the skills CLI route. The skills CLI’s own README says “Requires Node.js 22.20 or later.” - Git for the copy-by-hand and submodule routes. - Per skill, more: motion-showreel needs Node 18+ and ffmpeg, gpt-image-2 needs an image API key for its own image route, and kb-retriever needs PDF and Excel tools. See setup/what-each-skill-needs.md. ## What installs where - Skills CLI: the skills go into your agent’s skills folder, project-level by default and user-level with `--global`. - Claude Code plugin: Claude Code keeps the plugin’s files itself, and its skills run under the plugin’s name. - Copy by hand: a folder per skill in `.claude/skills/` or `.agents/skills/` of your project. ## Update and remove The README gives `npx skills update` and `npx skills remove <skill>` for the skills CLI route. It gives no update or removal step for the plugin or copy routes. Claude Code’s own docs say what to do for those (setup/claude-code.md). ## Your data Skills are instructions. Several also run scripts on your machine, such as setup.sh for motion-showreel and the image scripts for gpt-image-2. Those scripts can call outside services you configure, such as a text-to-speech service or an image API, with your own keys. Mindket read the README and the SKILL.md files for this. It did not audit the scripts.
README.md 4 KB setup/claude-code.md 4 KB setup/other-agents.md 3 KB setup/pinned-zip.md 3 KB setup/skills-cli.md 3 KB setup/what-each-skill-needs.md 7 KB
What’s new
Updated by Mindket to the current version of this kit.
Reviews
What people say
Posts and comments that name this project on Hacker News, Bluesky, X, Reddit, Facebook and YouTube, found by a weekly search and on Trendshift. Mindket doesn’t write or check them. Open one to read it where it was posted.
ぷーじClaude CodeやCursorなどのAIエージェントで使える実践的なスキル集ツールっぽい 開発作業の効率化や自動化に役立ちそうかも yug1224 starred ConardLi/garden-skills https://github.com/ConardLi/garden-skills2 likes · 1 reply@latreon.bsky.socialUp 332 stars over seven days. Instead of writing custom instructions for every AI assistant task, Garden Skills provides ready-to-use agent skills for Claude Code and Cursor. https://github.com/ConardLi/garden-skillsGitHub Trending🔥 Hot Repo！ 🔥 (100+ new stars) 📦 ConardLi / garden-skills ⭐ 13,947 (+181) 🗒 JavaScript ConardLi's open-source Skills collection, featuring web design, knowledge retrieval, image generation, and more.GitHub Trending JS/TS🔥 Hot Repo！ 🔥 (100+ new stars) 📦 ConardLi / garden-skills ⭐ 13,779 (+184) 🗒 JavaScript ConardLi's open-source Skills collection, featuring web design, knowledge retrieval, image generation, and more.@latreon.bsky.socialInstead of hand-writing prompt tools for AI agents, use Garden Skills to equip Claude Code and Cursor with pre-built capabilities. https://github.com/ConardLi/garden-skillstechnews4869github.com/ConardLi/garden... GitHub - ConardLi/garden-skills: ConardLi's open-source Skills collection, featuring web design,...@latreon.bsky.socialGarden Skills saves you from building custom agent capabilities from scratch by providing ready-to-use skills for AI coding agents. 451 new stars this week. https://github.com/ConardLi/garden-skills #CSS #AI #AIAgents #OpenSourceGitHub Trending8. https://github.com/DietrichGebert/ponytail - JavaScript, 1610 stars today 9. https://github.com/calesthio/OpenMontage - Python, 1284 stars today 10. https://github.com/rohitg00/ai-engineering-from-scratch - Python, 547 stars today 11. https://github.com/ConardLi/garden-skills - CSS, 413 stars […]GitHub Trending🔥 Hot Repo！ 🔥 (100+ new stars) 📦 ConardLi / garden-skills ⭐ 10,754 (+136) 🗒 CSS ConardLi's open-source Skills collection, featuring web design, knowledge retrieval, image generation, and more.
More from Mindket
See all from Mindketwaqf-ocr-hand-written-v1
Waqf-AI · Kit
Reads handwritten Arabic text from images and returns it as text. A fine-tuned PaddleOCR-VL model for people digitizing Arabic handwriting.
New
Wan2.2-TI2V-5B-Diffusers
Wan-AI · Kit
Makes video clips from a text prompt or an image plus text, at up to 720P and 24 frames per second, in the Diffusers format for a single consumer GPU.
New
Similar projects
See all in DeveloperDify
langgenius · Kit
LangGenius’s open-source platform for building LLM apps: visual workflows, RAG, model management and agents with a sandbox. Its difyctl tool and skill let Clau…
New161k
LobeHub
lobehub · Kit
LobeHub’s workspace for a team of AI agents, under its own community license. You build agents, group them, give them tasks and schedules, and keep editable me…
New84.4k
Wan2.2-S2V-14B
Wan-AI · Kit
Generates cinematic video of a character driven by an audio track, and can also lip-sync existing video. For video creators, animators and researchers with a l…
New