Garden Skills — A Curated Collection of Agent Skills for AI Coding Agents
Production-ready Agent Skills for Claude Code, Cursor, Codex, etc., covering web design, video presentations, image generation, and local knowledge retrieval.
Evidence shows: skills include user confirmation checkpoints (e.g., collaboration checkpoints in web-video-presentation and beautiful-article), but least privilege principle is not explicitly stated; data flow transparency is limited, no explanation of how data is processed or stored; sensitive data handling not mentioned; dependency security lacks dependency list or audit; external effects (e.g., network calls, file writes) not clearly specified; rollback mechanism not mentioned; source attribution only via README links and download links, not verified. Deductions: lack of explicit permission declarations, data flow explanations, and dependency security information.
Evidence shows: README and skill descriptions are consistent across multiple language versions; skill structure is clear; dependency availability not verified, but external tools like pdftotext, pandas are mentioned; failure messages not evident in static files. Deductions: dependency availability and failure handling not explicit in source.
Evidence shows: explicitly targets AI coding agents like Claude Code, Cursor, Codex; scenarios cover broad range; capability boundaries defined via skill descriptions and README categories; trigger precision not explicit, but skills have clear SKILL.md; environment fit via multiple installation methods and compatibility notes. Deductions: trigger precision and specific environment fit details insufficient.
Evidence shows: clear information architecture with table of contents and categories; detailed installation instructions with multiple methods; naming stable, skill names consistent; examples and FAQ via case studies in README; known limitations not explicitly listed; license is MIT; versioning and changelog via GitHub Releases and tags; maintenance responsibility via contribution guide and PR welcome. Deductions: known limitations not explicit, version changelog not directly in source.
Evidence shows: output usability demonstrated via cases and demos; marginal value via rich features and themes; cost-benefit not explicit, but free open-source license provided. Deductions: cost-benefit not quantified, output usability not tested.
Evidence shows: README claims have some link support, but no independent verification; cross-source corroboration limited, mainly self-documentation; facts and inferences not clearly separated. Deductions: lack of independent verification and fact-inference separation.
- Static review cannot verify actual runtime behavior; all scores are inferred from source and documentation.
- Dependency security lacks dependency list or audit; check dependencies before use.
- Data flow and sensitive data handling not explicit; be cautious with sensitive data.
- External effects (e.g., network calls, file writes) not explicit; review skill behavior before use.
What does this agent do, and when should you use it?
Garden Skills is ConardLi's open-source collection of Agent Skills that follow Anthropic's SKILL.md specification and work with Claude Code, Cursor, Codex, and other AI coding agents. The repository contains five skills: web-design-engineer polishes AI-generated web artifacts into deliberate, visually memorable front-end work with a design system and 25 style recipes; web-video-presentation turns scripts and articles into record-ready 16:9 presentations with 23 themes and pluggable TTS; gpt-image-2 offers 79 structured prompt templates for image generation and editing across three runtime modes; kb-retriever provides progressive, source-cited retrieval from local knowledge directories; beautiful-article transforms URLs, PDFs, or notes into share-ready, beautifully typeset articles. Each skill is self-contained, with install options including npx CLI, Claude Code plugin market, pinned ZIPs, or git submodules, all released with versioned artifacts. Licensed under MIT, it serves as a reference for production-grade skill development.
Garden Skills provides loadable capability modules for AI coding agents via standard SKILL.md files. web-design-engineer conducts a design read, declares a design system, selects from 25 anchored style recipes (e.g., Linear, Aesop, Bloomberg), then builds and verifies React prototypes. web-video-presentation converts scripts into narration outlines, constructs a 1920×1080 Vite + React stage, and optionally synthesizes narration audio via MiniMax or OpenAI TTS. gpt-image-2 detects its execution mode (local Garden generation, host-native delegation, or advisor-only prompting) and applies structured templates to generate or edit images, saving artifacts under garden-gpt-image-2/. kb-retriever navigates local knowledge/ directories using hierarchical data_structure.md indexes, combines keyword search with windowed reads, and processes PDF/Excel via pdftotext/pdfplumber and pandas, returning answers with sources. beautiful-article runs an editorial loop from source extraction through plan, double-confirmation, build, review, and repair, outputting self-contained HTML (with optional PDF).
- A frontend developer wants to transform a ChatGPT-generated landing page mockup into a polished, production-ready design with a coherent design system.
- A tech writer needs to turn a detailed tutorial into a screen-recordable video presentation, choosing among 23 built-in themes.
- A product designer requires professional UI mockups, product visuals, or infographics, and wants structured prompt templates with reusable outputs.
- A team with an internal documentation knowledge base wants an AI assistant that can answer questions with cited sources from Markdown, PDF, and Excel files.
- An editor wants to convert a URL, PDF, or set of notes into a long-form, beautifully formatted article with options for ten article types and eleven authoring themes.
What are this agent's strengths and limitations?
- Modular design adhering to the SKILL.md spec, enabling portability across multiple agents with minimal adaptation.
- web-design-engineer provides 25 detailed style recipes with anti-cliché guidance, yielding distinctive, high-quality output beyond generic AI UI patterns.
- web-video-presentation includes 23 themes and a pluggable TTS runner with two built-in providers, streamlining video production.
- gpt-image-2 offers 79 prompt templates and three runtime modes, adapting to different execution environments.
- Every skill ships with versioned ZIP artifacts and checksums, ensuring reproducible installations in CI pipelines.
- Dependency on third-party agent platforms (e.g., Claude Code) to support the Skills feature; compatibility with all agents is not fully guaranteed beyond the six listed.
- Some features (e.g., TTS in web-video-presentation) require external services (MiniMax or OpenAI), incurring potential costs.
- gpt-image-2's advisor-only mode (Mode C) generates prompts only, requiring manual execution for image creation.
- beautiful-article is at version 0.1.0, indicating early-stage maturity and possible instability.
- kb-retriever caps retrieval at 5 search rounds; for very complex knowledge bases this may limit depth of exploration.
How do you install or deploy this agent?
Install via npx skills CLI: npx skills add ConardLi/garden-skills to install all skills, or use -s to install a specific one (e.g., npx skills add ConardLi/garden-skills -s web-design-engineer). For Claude Code, subscribe to the plugin marketplace: /plugin marketplace add ConardLi/garden-skills, then /plugin install <pack>@garden-skills. For CI or air-gapped environments, download the pinned ZIP with SHA-256 checksum from GitHub Releases and unzip into your agent's skills directory (e.g., .claude/skills/). You can also manually copy the skill folder after git clone, or add the repo as a git submodule.
How do you use this agent?
After installation, the skill becomes available when the agent scans the workspace. For example, with web-design-engineer in Claude Code, simply ask: "Apply a Linear-style design to this landing page." For web-video-presentation, provide a script or article; the skill generates a presentation framework and optionally calls TTS (requires MiniMax or OpenAI API keys). gpt-image-2 in Garden mode saves outputs under garden-gpt-image-2/. For kb-retriever, maintain a knowledge/ directory locally and ask questions; the skill returns citations. For beautiful-article, supply a source URL or file and follow the interactive checkpoints.
How does this agent compare with similar options?
Related projects include Anthropic's official skills repository and community collections like travisvn/awesome-claude-skills and obra/superpowers. Garden Skills differentiates itself with specialized, production-oriented skills for web design, video presentation, and image generation, offering unique style recipe galleries and theme systems not found in generic skill lists.