garden-skills

Garden Skills is a curated set of plug-in instructions, called "skills," designed to extend what AI coding assistants like Claude Code, Cursor, and Codex can do. A skill, in this context, is a structured set of directions that tells an AI agent how to handle a specific type of task more reliably than it would by default. Think of them as specialized playbooks that you drop into a project, and the agent follows the rules inside when you invoke a skill by name.

The collection currently has four skills. The first, web-video-presentation, turns an article or script into a screen-recording-ready web presentation. It builds a slide-like interface with a fixed 1920-by-1080 layout that can be stepped through with a keyboard, and it works through a set of checkpoints (script approval, theme selection, outline) before generating any visuals. The second skill, web-design-engineer, guides the agent through creating polished web pages and UI mockups, from gathering requirements through a defined six-step process to a finished, verified result. The third, gpt-image-2, handles image generation requests using OpenAI-compatible image APIs, with support for both generating new images and editing existing ones, across more than 80 prompt templates covering categories like posters, diagrams, and product mockups. The fourth, kb-retriever, lets the agent search a local folder of documents (Markdown, PDF, text, Excel) to answer questions, using a layered approach that avoids loading entire files into memory at once.

Installing a skill can be done several ways: running a command with npx, finding it in the Claude Code plugin marketplace, downloading a versioned zip file from the project's releases page, copying the files manually, or adding the repository as a git submodule. Each skill lives in its own subfolder with a README and a SKILL.md file that defines the behavior.

This project is aimed at developers who already use AI coding tools and want those tools to follow more consistent, deliberate processes for specific tasks rather than improvising each time. The skills are open source under the MIT license and the project accepts outside contributions.