Prompt collections that teach coding agents taste
Structured workflow packs that push AI coding agents past generic output toward polished presentations, deliberate web design, and local knowledge retrieval.

What it does
Garden Skills is a set of four production-ready “skill” packs for AI coding agents such as Claude Code, Cursor, and Codex. The visible README thoroughly documents two of them: one turns scripts and articles into cinematic, screen-record-ready web presentations with fixed 1920×1080 stages and narration beats, while another guides agents through a six-step design workflow to produce polished, non-generic web pages using concrete style recipes. The collection also lists skills for image generation and local knowledge retrieval, though their documentation was truncated in the provided sources.
The interesting bit
Instead of hoping an LLM invents good taste, the project hard-codes creative briefs and anti-cliché blocklists directly into the agent context. The video skill forces hard checkpoints for script, theme, and audio before implementation, while the design skill ships 25 anchored recipes — from Bloomberg Terminal dashboards to Aesop apothecary pages — complete with palettes, typography, and anti-patterns to paste into the design-system declaration.
Key highlights
web-video-presentation: 23 built-in themes, pluggable TTS with built-in MiniMax and OpenAI runners, and a click-driven(chapter, step)cursor for narration pacing.web-design-engineer: Six-step workflow covering requirements through verification, plus 25 style recipes such aslinear,swiss-ikb, andmid-century-modern.- Practical implementation rules for
oklch()color work, container queries, inline React + Babel, and reduced-motion handling. - Also includes
gpt-image-2(image generation / prompting) andkb-retriever(local knowledge retrieval) skills per the gallery table. - MIT licensed with documentation in English, Chinese, and Japanese.
Caveats
- Only four skills exist so far, and the README details two thoroughly while summarizing the others.
- This is primarily a documentation and prompt-template repository, not an installable framework or library.
Verdict
Grab it if you live in Claude Code or Cursor and want to stop beige AI output. Skip it if you are hunting for npm packages or runtime tools; these are structured context packs, not dependencies.
Frequently asked
- What is ConardLi/garden-skills?
- Structured workflow packs that push AI coding agents past generic output toward polished presentations, deliberate web design, and local knowledge retrieval.
- Is garden-skills open source?
- Yes — ConardLi/garden-skills is open source, released under the MIT license.
- What language is garden-skills written in?
- ConardLi/garden-skills is primarily written in CSS.
- How popular is garden-skills?
- ConardLi/garden-skills has 9.7k stars on GitHub and is currently cooling off.
- Where can I find garden-skills?
- ConardLi/garden-skills is on GitHub at https://github.com/ConardLi/garden-skills.