Garden Skills: A Curated Pack of Agent Skills for Claude Code, Cursor, and Codex
ConardLi's open-source Skills collection, featuring web design, knowledge retrieval, image generation, and more.
At a glance
- What is it?
- Garden Skills is a collection of production-ready Agent Skills for AI coding agents, covering web design, video presentations, image generation, and article writing. The repo is a practical kit, not a framework, and its value depends on how well the skills match your workflow.
- Who is it for?
- Adopt Garden Skills if you use Claude Code, Cursor, or Codex and need ready-made, structured skills for web design, video presentations, image generation, or article writing. Skip it if you prefer to build your own prompts or need deep customization beyond the provided recipes.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 65 days ago.
- What is it written in?
- Mainly CSS, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
What Garden Skills Actually Solves
Garden Skills addresses a specific pain: AI coding agents produce generic output unless you give them detailed, structured instructions. The repo packages those instructions as Agent Skills, a format that Claude Code, Cursor, and Codex can load. Each skill is a folder with a SKILL.md file and supporting assets, and the README presents four skills: web-design-engineer for front-end work, web-video-presentation for click-driven 16:9 presentations, gpt-image-2 for image generation, and beautiful-article for turning any source into an article. The target user is a developer or content producer who wants to skip the prompt-engineering grind and drop in a proven workflow. The repo does not claim to be a framework or a library. It is a set of opinionated instructions that shape how the agent behaves.
Inside web-design-engineer: From Brief to Design System
The web-design-engineer skill treats the agent as a design engineer rather than a code generator. The workflow starts by converting the brief into a five-dial Design Read that covers variance, motion, density, asset dependence, and brand fidelity. That structure forces the agent to think about design dimensions before writing code. The skill also distinguishes three redesign modes: extension, preserve, and overhaul, so the agent knows whether to build on existing work or start fresh. A notable feature is the Design Direction Advisor with six schools and 25 anchored style recipes, each carrying concrete palette, typography, signature moves, and anti-patterns. The recipes are named after real references like Linear, Aesop, Pentagram, Bloomberg, Stripe Press, and Mid-Century. The skill includes an anti-cliché blocklist to push beyond generic AI UI patterns, and it covers inline React with Babel, CSS tokens, oklch() color, container queries, and reduced-motion handling. One limitation is that the skill runs browser acceptance only when the user explicitly requests it, so the agent may not verify visual output by default.
web-video-presentation: Building Record-Ready Presentations
This skill turns scripts, articles, lessons, and demos into 16:9 web presentations that can be screen-recorded as videos. The output is a Vite + React + TypeScript project with a fixed 1920x1080 stage that scales to the viewport. The interaction model is a click or keyboard driven (chapter, step) cursor, with one narration beat per visual step. That design keeps the recording stable because the presenter controls pacing manually. The skill enforces hard collaboration checkpoints for script, theme, outline, implementation mode, and optional audio, which means the agent stops and asks for approval before proceeding. Hidden hover-only progress controls keep the stage clean during recording. The theme system includes 23 built-in themes, each with a design signature such as editorial, terminal, engineering, and Swiss International. TTS is pluggable and provider-agnostic, with two built-in providers: MiniMax mmx-cli and OpenAI TTS via curl. The repo also provides snippets for ElevenLabs, edge-tts, Azure, Google Cloud, and macOS say. The audio synthesis is optional and happens after the visual outline is approved, which is a sensible way to avoid wasted TTS calls.
Getting It Running: Five Installation Paths
The README lists five ways to install the skills. The first is a skills CLI via npx, which is the most direct for CLI users. The second is a Claude Code plugin marketplace, which fits users who already manage plugins. The third is a pinned .zip from Releases, which gives you a specific version, such as web-design-engineer-v1.3.0 or web-video-presentation-v1.2.2. The fourth is manual copy into your project, which is the most transparent because you can see exactly what files you add. The fifth is a Git submodule, which keeps the skills linked to the upstream repo for updates. The README also includes a compatibility section, but the truncated material does not detail which agent versions are supported. That is a gap worth checking before adoption. The releases show active maintenance, with web-design-engineer moving from v1.2.2 to v1.3.0 between June and July 2026, and web-video-presentation at v1.2.2.
The Trade-Off: Structured Workflows vs. Flexibility
The main limitation is that these skills are opinionated. The web-design-engineer skill forces a five-dial Design Read and a design-system declaration before building. That structure is useful for consistent output, but it adds overhead for simple tasks like a quick mockup. The anti-cliché blocklist and style recipes are prescriptive, so if you have your own design language, you may fight the skill's defaults. Similarly, web-video-presentation requires hard checkpoints, which can slow down a fast ideation loop. The skills are also tied to the Agent Skills format, so they only work with agents that support that standard. If you use a tool that does not implement Agent Skills, you cannot use these directly. The README does not mention a fallback for non-compatible agents. Another concern is that the repo is a collection, not a unified tool, so each skill has its own maintenance cadence. The release list shows separate version numbers for each skill, which means you must track updates per skill.
Alternatives: Custom Prompts vs. a Full Design System
The obvious alternative is to write your own prompts or use a general coding agent without skills. That gives you full control but requires you to invest time in prompt engineering. A more structured alternative is to use a design-system tool like Storybook or a style-guide generator, but those are not Agent Skills and do not integrate with the agent's workflow. Another alternative is to use a dedicated presentation tool like Marp or reveal.js, which are static slide frameworks, but they do not include the TTS integration or the checkpoint workflow that web-video-presentation offers. The key difference is that Garden Skills embeds the design and presentation logic into the agent's decision process, whereas these alternatives require you to manually orchestrate the output. If you already have a robust prompt library, Garden Skills may be redundant. If you are starting from scratch, the curated recipes save time.
Maintenance, License, and Upgrade Cost
The repo is not archived, and the last push is from July 2026, which suggests active maintenance. The MIT license is permissive, so you can use, modify, and redistribute the skills freely, but note that the license covers the skill definitions, not the output they generate. The upgrade cost is per skill because each has its own version. To update, you would re-run the skills CLI or re-download the pinned .zip for the new version. The README does not document a migration path between versions. If you manually copied skills into your project, you must track changes yourself. The Git submodule option helps with that because it pins a specific commit, but you still need to review changes before updating. The repo also includes a contributing section, which implies the maintainer expects community input, but the README does not specify contribution guidelines beyond that.
Editorial conclusion
Adopt Garden Skills if you use Claude Code, Cursor, or Codex and need ready-made, structured skills for web design, video presentations, image generation, or article writing. Skip it if you prefer to build your own prompts or need deep customization beyond the provided recipes. Before adopting, verify that the skill versions in the repo match your agent's Skill support, and check the pinned .zip releases for stability. The real test is whether the five-dial Design Read or the 23-theme presentation engine saves you more time than writing your own instructions.
Community notes