garden-skills

Product Introduction

1. Definition: Garden-Skills is an open-source, curated repository of production-ready Agent Skills for AI coding assistants like Claude Code, Cursor, and Codex. It functions as a centralized knowledge base and toolkit for AI-augmented software development.
2. Core Value Proposition: It solves the problem of fragmented developer knowledge by aggregating essential, cutting-edge software development skills—from web design and AI image generation to knowledge retrieval—into a single, structured, and agent-executable format, dramatically improving developer efficiency and output quality.

Main Features

1. web-video-presentation Skill: This skill transforms scripts, articles, or product demos into cinematic, 16:9 web presentations built with Vite, React, and TypeScript. It works by mapping narration beats to full-screen visual scenes with click/keyboard navigation, includes 23 built-in design themes (e.g., editorial, terminal, Swiss International), and provides a pluggable architecture for Text-to-Speech (TTS) audio synthesis using providers like MiniMax, OpenAI, or ElevenLabs.
2. web-design-engineer Skill: This skill elevates AI-generated web artifacts from functional to polished by enforcing a "design engineer" workflow. It works by first creating a "Design Read" to assess context, then declaring a design system using one of 25 anchored style recipes (e.g., Stripe Press, Bloomberg Terminal, Mid-Century Modern). It includes an anti-cliché blocklist and provides concrete implementation rules for modern CSS (oklch(), container queries) and React patterns.
3. gpt-image-2 Skill: A focused image generation and prompt engineering skill for GPT-4 and compatible APIs. It operates in three distinct runtime modes: Garden local generation, host-native tool delegation, or advisor-only prompt writing. It includes 18 visual categories and 79 structured prompt templates, and manages a local library of prompts and generated images for versioning and reuse.
4. kb-retriever Skill: A local knowledge base retrieval skill designed for precise, context-aware Q&A from a directory of Markdown, text, PDF, and Excel files. It works using a layered search strategy, first navigating hierarchical data_structure.md index files before performing content extraction, thereby avoiding context window flooding. It enforces a "learn-before-process" rule for complex file types.
5. beautiful-article Skill: An editorial harness that converts any source (URL, PDF, DOCX, screenshot) into a polished, share-ready article. It works through a defined editorial loop: source → plan → confirmation → build → review → repair. It uses a "Reacticle" component protocol for semantic article structure and offers 11 authoring theme profiles (e.g., Tufte, Vignelli) with hard collaboration checkpoints for user control.

Problems Solved

1. Pain Point: The fragmentation and inaccessibility of advanced software development techniques and AI agent capabilities, leading to inefficient learning curves and inconsistent output quality from AI coding assistants.
2. Target Audience: Software developers, front-end engineers, technical content creators, AI prompt engineers, and tech enthusiasts who use AI coding agents (Claude Code, Cursor) and seek to produce production-grade code, designs, and content efficiently.
3. Use Cases: A developer needs to quickly create a professional product demo video from a spec document; a designer needs to generate a high-fidelity UI mockup following a specific brand's design system; a researcher needs to query a local archive of technical PDFs; a writer needs to transform a rough draft into a visually stunning, publishable article.

Unique Advantages

1. Differentiation: Unlike generic AI prompts or scattered online tutorials, Garden-Skills provides deterministic, production-tested workflows packaged as portable "Skills" that AI agents can directly execute. It moves beyond theory to provide concrete, themed templates and enforceable design rules.
2. Key Innovation: The "Skill" abstraction itself, which packages complex multi-step workflows (design, retrieval, generation) into a single, agent-discoverable unit (SKILL.md). Its focus on "anchored style recipes" and "hard collaboration checkpoints" ensures AI output is not just functional but deliberately styled and user-approved.

Frequently Asked Questions (FAQ)

1. What is the difference between Garden-Skills and Claude's built-in capabilities? Garden-Skills extends Claude Code's and other agents' native abilities with specialized, curated workflows for specific high-value tasks like video presentation building or systematic design, providing structured templates, style guides, and quality control steps that the base model lacks.
2. How do I install a specific Garden-Skills module into Cursor? You can install a Garden-Skills module into Cursor using the npx skills CLI command (e.g., npx skills add ConardLi/garden-skills -s web-design-engineer), which auto-detects the agent, or by manually copying the skill folder into your project's .agents/skills/ directory.
3. Can I use the gpt-image-2 skill without an OpenAI API key? Yes, the gpt-image-2 skill supports multiple runtime modes. In "Mode B: host-native delegation," it can leverage image generation tools already installed on your local machine. In "Mode C: advisor-only," it will generate detailed, structured prompts for you to use in any image generation interface.
4. Is Garden-Skills suitable for beginner programmers? While beginners can use it, Garden-Skills is optimally designed for developers and technical practitioners familiar with concepts like React, CSS, and CLI tools, as it provides advanced scaffolding and expects collaboration on technical implementation details.
5. How does the kb-retriever skill handle different file formats like PDFs? The kb-retriever skill uses a "learn-before-process" protocol. It first consults its internal reference documentation on tools like pdftotext and pdfplumber to understand extraction methods, then applies the correct technique to pull text from PDFs while maintaining source attribution for answers.