ConardLi/garden-skills

Project Overview

Garden Skills is a curated collection of six Agent Skills packaged in the portable SKILL.md format for AI coding agents such as Claude Code, Cursor, Codex CLI, Gemini CLI and OpenCode. Each skill is a self-contained folder containing a SKILL.md description, optional reference documents, scripts and assets that the agent loads on demand when the task matches the skill's trigger description What is a Skill?.

The current collection covers distinct work areas:

• motion-showreel — renders code-driven short and chapter-based videos with a deterministic render(t) pipeline, shared audio/picture timeline and a large starting library of motion styles.
• web-video-presentation — scaffolds Vite + React + TypeScript 16:9 presentations with 23 themes and pluggable TTS, aimed at screen-recorded video.
• web-design-engineer — guides an agent through design-engineering workflow with a five-dial brief, style recipes (Linear, Aesop, Bloomberg Terminal, etc.) and browser acceptance checks.
• gpt-image-2 — runs image generation and editing through GPT Image 2 or OpenAI-compatible APIs, with 18 categories and 79 prompt templates.
• beautiful-article — turns URLs, PDFs, DOCX, Markdown and notes into polished share-ready articles using a source → plan → confirm → build → review loop.
• kb-retriever — searches a local knowledge/ directory progressively through index files, limited to five search rounds, with source-aware answers.

If you work in an AI coding agent and want prebuilt capabilities for design, media, editorial and retrieval tasks, this collection is intended for that use.

Quick Start

The quickest path is the skills CLI, which auto-detects your agent and installs the skill into the expected directory:

# Install every skill in the collection
npx skills add ConardLi/garden-skills

# Or install a single skill
npx skills add ConardLi/garden-skills -s web-design-engineer

This installs to ./.skills per project; add --global to install to ~/.skills. To target a specific agent, pass -a claude-code. For a pinned version, point at a tag-scoped tree/ URL:

npx skills add ConardLi/garden-skills/tree/web-design-engineer-v1.0.0/skills/web-design-engineer

After installation, the agent discovers the skill the next time it scans the workspace. Run npx skills list to confirm what is installed.

Alternative install paths documented in the Install section include the Claude Code plugin marketplace (/plugin marketplace add ConardLi/garden-skills), pinned .zip files from GitHub Releases with SHA-256 verification, manual git clone copying and Git submodules.

Usage Constraints

The compatibility matrix lists Claude Code, Claude.ai web, Cursor, Codex CLI, Gemini CLI and OpenCode as tested Compatibility, with different skill directories for each — .claude/skills/, .agents/skills/, .codex/skills/ and .opencode/skills/. The SKILL.md format is described as portable, so other agents that support skills should generally work, but the listed matrix is the extent of testing shown.

Installation defaults to the latest commit on main. For CI, air-gapped environments or reproducible setups, use a pinned .zip from Releases and verify its SHA-256 checksum; the README recommends this explicitly for unattended installs. Some skills also depend on external tools: kb-retriever uses grep, pdftotext, pdfplumber and pandas; gpt-image-2 requires access to GPT Image 2 or an OpenAI-compatible image API; TTS in web-video-presentation and motion-showreel requires configured providers.

The project is MIT licensed License. The README's Connect section links to the author's social and video accounts, which are informational rather than a requirement for use.

HotRepo's Take

On the 2026-11-04 JavaScript daily ranking, Garden Skills reached 4th place with 184 stars gained that day, bringing the repository to 13,774 stars and 1,615 forks. There is no all-language daily rank recorded for this project and no star-growth time series beyond that single snapshot, so the trajectory cannot be judged from HotRepo data alone.

What the repository does show is a fairly substantial surface for a skill collection: six skills, each with its own README, SKILL.md, downloadable pinned release zips, a plugin marketplace manifest and a compatibility matrix covering six agent runtimes. The design and image skills ship their own demo sites and galleries, and beautiful-article depends on a separate React library. The strong one-day attention is consistent with a launch push, but it should not be read as evidence of long-term adoption or quality.