garden-skills

garden-skills is a collection of Agent Skills put together by GitHub user ConardLi (code秘密花园 on X). It holds six Skills; you can install all of them or just the one you need. web-video-presentation turns an article or voiceover script into a 16:9 web presentation where each click advances one line of narration, with 23 built-in themes; screen-record it and you have a video, with optional voiceover. motion-showreel makes 5 to 60 second motion videos in code, such as product teasers, logo stings and kinetic type, with picture and music on one shared timeline, and delivers an MP4 plus editable source; the README says it ships 28 styles as starting points. web-design-engineer has the AI work like a design engineer on pages, dashboards, prototypes and HTML slide decks, settling a design direction before building, with 25 style recipes and a list of AI clichés to avoid. gpt-image-2 writes prompts for GPT Image 2 and can generate the images; the README lists 18 categories and 79 structured templates covering posters, UI, infographics, academic figures, comics, storyboards and more. beautiful-article turns a URL, PDF, Word document or notes into a single-file web article that opens offline, with optional PDF export. kb-retriever searches a local knowledge-base folder of Markdown, text, PDF and Excel files by working through index files layer by layer, caps itself at five search rounds, and cites its sources. What sets it apart is that the content-making Skills build human sign-off into the workflow: the narrated video, for example, stops for you to approve the script, outline and theme, and won't go past chapter one until you've checked it; gpt-image-2 first detects whether your setup can generate images at all, and only hands over prompts when there is no image tool.

Who it's for:creators who make tutorial videos, product demos, web designs or knowledge articles, or who need answers pulled from a pile of local documents. Most of these Skills set up projects and run commands on your computer, so you'll need an AI coding tool that runs commands, such as Claude Code or Codex.

Before you start: install a single Skill with npx skills add ConardLi/garden-skills -s plus its name, add it from the Claude Code plugin marketplace, or download a pinned release zip with a SHA-256 checksum; the README lists Claude Code, Claude.ai on the web, Cursor, Codex CLI, Gemini CLI and OpenCode as tested. Most Skill instructions are in Chinese; web-design-engineer is in English. The narrated video, motion video and article Skills create projects in your folder, run commands, and go online to install dependencies such as Vite, React, Playwright and Chromium from npm; beautiful-article pulls the latest reacticle release, so versions aren't pinned. Voice and images cost money. web-video-presentation's voiceover (TTS) uses MiniMax's mmx-cli by default and can switch to OpenAI TTS (needs OPENAI_API_KEY), ElevenLabs, Azure, Google and others; your script is sent to the service you choose and billed at its rates, and edge-tts or the built-in macOS say are the free options. gpt-image-2 only generates images directly after you turn on its switch and provide OPENAI_API_KEY; prompts and reference images then go to OpenAI or a third-party compatible gateway you set, billed by that provider. Without that setup it hands the prompt to your AI tool's own image feature, or just gives you the prompt. Some image templates cover avatars, fan merchandise and pictures made from real people's photos or celebrities; the MIT license gives you no rights to third-party people, brands or characters. The repo is MIT licensed.

Based on the project's public documentation. Not tested by AI Inspiration World.