Who garden-skills is for#
Developers building web presentations and video walkthroughs
Load web-video-presentation to turn articles, scripts, or product demos into 16:9 Vite + React presentations. The skill handles theme selection, scene mapping, narration beats, and optional TTS, so the agent does the production work rather than the developer.
Skip if:
If your presentation tool of choice is PowerPoint, Keynote, or Slides, and you want slide-format output rather than a browser-rendered stage, this skill targets a different output format. The 1920x1080 stage is built for screen recording, not export to slide formats.
Frontend developers who want polished AI-generated UI
Load web-design-engineer to move the agent's output beyond generic AI UI patterns. The 25 style recipes (Linear, Stripe Press, Bloomberg Terminal, and more) give the agent a concrete design vocabulary, and the anti-cliché blocklist prevents default-looking output.
Skip if:
If you are building production backend services, APIs, or infrastructure code, this skill targets the visual frontend layer only. It does not improve the agent's code quality for server-side work.
Content creators needing AI-generated visual assets
Load gpt-image-2 to generate posters, UI mockups, infographics, technical diagrams, and product visuals through your AI coding agent. The 79 structured prompt templates cover 18 visual categories, reducing the prompt engineering needed per image.
Skip if:
If you are working with a non-OpenAI image API, check whether your provider is compatible with GPT Image 2 mode first. The skill includes an advisor-only mode for environments without direct image tool access, but results may vary.
Teams standardizing creative workflows across AI agents
Use Garden Skills as a shared skill library across a team that runs multiple agents. Because skills are version-pinned and installable via git submodule, every team member and CI environment can load the same skill definitions for consistent output across sessions.
Skip if:
If your team's creative production is already handled by a dedicated design tool like Figma or a presentation tool with its own templating, adding an AI agent skill layer may introduce complexity rather than reduce it.
The problem it solves#
AI coding agents are capable of generating code quickly, but their default output for creative tasks tends to be visually generic. Ask Claude Code or Cursor to build a web presentation or landing page and you get something functional, but you spend significant time prompting, revising, and adjusting before it looks intentional. The same problem repeats for image generation, article formatting, and design work: each session starts from scratch with no shared design vocabulary.
The deeper challenge is that creative production tasks require structured workflows: deciding on a design direction, checking visual consistency, verifying a theme before building the full experience. Without an explicit skill loaded, most AI coding agents skip these steps and produce output that looks auto-generated. Garden Skills encodes those workflows into installable SKILL.md files that any compatible agent can load and follow.
How it solves it#
web-video-presentation: click-driven 16:9 presentations
Turns scripts, articles, or product demos into screen-recordable 16:9 presentations built with Vite, React, and TypeScript. Includes 23 built-in themes (editorial, terminal, Swiss International, and more), a (chapter, step) cursor with one narration beat per visual step, and pluggable TTS support with two built-in providers: MiniMax mmx-cli and OpenAI TTS.
web-design-engineer: design-system-aware frontend generation
Turns AI-generated web artifacts from functional into polished frontend work. Ships a Design Direction Advisor with six differentiated schools and 25 anchored style recipes (Linear, Aesop, Stripe Press, Bloomberg Terminal, and more), each with concrete palette, typography, signature moves, and anti-patterns. Runs an anti-cliché blocklist to push output beyond generic AI UI patterns.
gpt-image-2: structured image generation across 79 templates
A focused image-generation skill for GPT Image 2 and OpenAI-compatible APIs. Supports three runtime modes: local Garden generation, host-native delegation, and advisor-only prompt writing. Provides 18 visual categories and 79 structured prompt templates for posters, UI mockups, infographics, technical diagrams, and more.
beautiful-article: any source to a formatted article
Transforms any source material into a formatted article. Listed alongside the web design, presentation, and image generation skills in the collection; it handles the layout and formatting decisions that otherwise require manual post-processing after bare agent output.
Five install paths, including npx CLI and git submodule
Install via the npx skills CLI, the Claude Code plugin marketplace, pinned versioned zip from GitHub Releases, manual file copy, or git submodule. Each skill is versioned independently, so you can pin web-design-engineer to v1.3.0 while updating other skills separately.
Strengths and trade-offs#
Strengths
- MIT license, no subscriptions requiredEvery skill in the collection is MIT licensed, so you can use, modify, and ship them commercially with no restrictions. There are no per-seat fees, API usage charges from the skill layer itself, or vendor lock-in. The skills load into your existing AI agent, whichever one you already run.
- Strong community traction: 10,900+ stars in under four monthsThe repository reached over 10,900 GitHub stars with 1,388 forks between April 2026 (creation) and July 2026 (latest push). PRs are open for contribution, and the skill spec is documented at agentskills.io, giving contributors a shared standard to build against.
- Agent-agnostic: works with Claude Code, Cursor, and CodexSkills are not tied to one agent. The same SKILL.md files load into Claude Code, Cursor, Codex, and any other agent that follows the SKILL.md spec. Teams that switch agents or run multiple tools in parallel can reuse the same skill collection without reinstalling.
- Per-skill versioned releases, independently installableEach skill ships as an independently versioned release. You can download web-video-presentation v1.2.2 or web-design-engineer v1.3.0 as standalone zip files, pin them in git, or install them separately via the npx skills CLI. No need to take the entire collection to get one skill.
Trade-offs
- -Requires a compatible AI coding agent already runningGarden Skills is not a standalone tool. It has no UI, no server, and no API of its own. To use it you need an existing AI coding agent (Claude Code, Cursor, or Codex) configured and functional. Teams without an agent setup get no value from the skill files alone.
- -Creative skills only, not code completion or debuggingThe current skill set is focused on web design, presentation building, image generation, and article formatting. It does not extend the agent's code completion, debugging, or refactoring capabilities. Developers expecting broader agent enhancements beyond creative production will need to look elsewhere.
- -Small catalog at this stage of developmentThe README badge shows 5 skills at the time of cataloging. This is a young repository created in April 2026 with the latest push in July 2026. The collection is actively growing, but teams that need skills outside the current categories have no drop-in option yet.
garden-skills vs alternatives#
Garden Skills vs GitHub Copilot
GitHub Copilot is a paid AI coding assistant from GitHub that provides inline code completion, chat, and agent mode inside editors like VS Code and JetBrains. Garden Skills is not a code completion tool; it is a collection of SKILL.md files that extend what any AI coding agent can do with structured creative workflows.
| Feature | Garden Skills | GitHub Copilot |
|---|---|---|
| License | MIT | Proprietary |
| Cost | Free (self-hosted) | $10-$19/user/month |
| Code completion | No | Yes |
| Web design skills | Yes (25 style recipes) | No |
| Presentation generation | Yes (23 themes) | No |
| Image generation | Yes (79 prompt templates) | No |
Garden Skills is not a replacement for GitHub Copilot. Where Copilot excels at inline code suggestions and editor integration, Garden Skills targets a different problem: giving any AI coding agent a design vocabulary, a presentation workflow, and an image generation capability that no general-purpose agent ships with by default. Teams using Copilot for code completion can still load Garden Skills into Claude Code or Cursor for creative production tasks.
The case for Garden Skills is strongest when you run an AI coding agent outside the GitHub ecosystem, or when your workflow involves web design, video content, and visual asset production. If your primary need is autocomplete and code chat inside a GitHub-integrated editor, Copilot is the stronger fit for that specific workflow.
Garden Skills vs Cursor and Replit AI
Cursor is a paid AI code editor with built-in AI assistance. Replit AI is the equivalent in the Replit environment. Both are commercial platforms with subscription pricing. Garden Skills is designed to work inside Cursor (the README lists it as a supported agent), making these tools complementary rather than competing.
| Feature | Garden Skills | Cursor | Replit AI |
|---|---|---|---|
| License | MIT | Proprietary | Proprietary |
| Self-hosting | Yes (no server needed) | No | No |
| Built-in skill library | Yes | No | No |
| Creative design workflows | Yes | Via ad-hoc prompting | Via ad-hoc prompting |
| Code completion | No | Yes | Yes |
Where Cursor and Replit AI require ad-hoc prompts for creative tasks, Garden Skills encodes those workflows into structured SKILL.md files. The result is repeatable output: the same web-design-engineer skill produces design-system-aware work every session, using a consistent workflow, regardless of which agent loads it. Replit AI is still the stronger choice if your work is primarily code-only inside the Replit environment.
Install and self-host#
Install individual skills by cloning the repository or adding it as a git submodule.
```bash
git clone https://github.com/ConardLi/garden-skills
```What it's built on#
- Languages
- JavaScriptPythonTypeScript
- Frameworks
- React
FAQ#
What is Garden Skills and how does it work with Claude Code?
Garden Skills is a collection of SKILL.md files that AI coding agents like Claude Code load and execute. When a skill is installed in your project, the agent reads its instructions and follows a structured workflow instead of generating output from scratch. Each skill encodes a specific production task: web design, presentation building, image generation, or article formatting.
Is Garden Skills free to use?
Yes. The entire collection is MIT licensed and free to use, modify, and distribute. There are no usage fees or API charges from Garden Skills itself. You do need your own API keys for any external services a skill calls (such as an OpenAI key for gpt-image-2), but the skill files themselves cost nothing.
Which AI coding agents are compatible with Garden Skills?
The README lists Claude Code, Cursor, and Codex as compatible agents. Any agent that follows the SKILL.md spec at agentskills.io should work. The skills are agent-agnostic files, not agent-specific plugins, so compatibility depends on whether your agent implements the SKILL.md standard.
How do I install a skill from Garden Skills?
There are five install options: the npx skills CLI, the Claude Code plugin marketplace, a pinned versioned zip from GitHub Releases, manual file copy into your project, or a git submodule. Each skill is versioned independently, so you can install specific skills without taking the whole collection.
Does Garden Skills replace GitHub Copilot or Cursor?
No. Garden Skills is a skill collection that runs inside your existing AI coding agent, not a replacement for one. It extends what Claude Code, Cursor, or Codex can do with specialized creative workflows (design, presentations, image generation). It does not provide code completion, inline suggestions, or debugging capabilities of its own.
Similar open-source tools#
codex
OpenAI's terminal coding agent, Apache-2.0 licensed
fx
Native Zig CLI coding agent, 7.8 MiB, any AI model.
kilocode
Open source AI coding agent. 500+ models at zero markup.
free-claude-code
Route 9 AI coding agents through 1.3B+ free monthly tokens
prime-agent
Automate coding workflows with a self-improving agent
t3code
Control your coding agents from one interface