将任何文档网站变成克劳德代理技能。

site2skill 是一个工具，可以抓取文档网站，将其转换为 Markdown，并打包为 Claude 代理技能（ZIP 格式），并带有一个合适的 SKILL.md 入口。

代理技能是动态加载的知识模块，Claude按需使用。它们可以跨Claude代码、Claude应用和API运行。

安装

从 PyPI（稳定版）安装

这安装了稳定的Python版本（0.1.x），需要Python 3.10+和wget。

试试Rust测试版

Rust 版本有 beta 版本。它快 ~ 4 倍，没有 wget 依赖，并且以单一二进制形式发布。

或者收藏版本：

确认你使用的是Rust版本：

稳定版没有 --version 标志，所以如果你看到版本字符串，说明你是在 Rust 测试版。

回到稳定版：

无安装运行

从源头构建

用途

CLI选项

它是如何运作的

1. Fetch：通过内置异步HTTP爬虫（robots.txt兼容，并发）爬取文档网站。
2. 转换：使用爬取器和 htmd 将 HTML 页面转换为 Markdown。
3. 规范化：解析相对链接到绝对链接。

site2skill

Turn any documentation website into a Claude Agent Skill.

site2skill is a tool that scrapes a documentation website, converts it to Markdown, and packages it as a Claude Agent Skill (ZIP format) with a proper SKILL.md entry point.

Agent Skills are dynamically loaded knowledge modules that Claude uses on demand. They work across Claude Code, Claude apps, and the API.

Installation

Install from PyPI (stable)

pip install site2skill

This installs the stable Python version (0.1.x), which requires Python 3.10+ and wget.

Try the Rust beta

A rewrite in Rust is available as a beta. It’s ~4x faster, has no wget dependency, and ships as a single binary.

pip install --pre site2skill

Or pin the version:

pip install 'site2skill==0.2.0b2'

Verify you’re running the Rust version:

site2skill --version
# site2skill 0.2.0-beta.2

The stable version has no --version flag, so if you see a version string, you’re on the Rust beta.

To go back to stable:

pip install 'site2skill<0.2'

Run without Installation

# Beta (Rust)
uvx --pre site2skill  

# Stable (Python, from legacy branch)
uvx --from 'git+https://github.com/laiso/site2skill@legacy' site2skill  

Build from Source

git clone https://github.com/laiso/site2skill.git
cd site2skill
cargo build --release
./target/release/site2skill --help

Usage

# Basic usage
site2skill  

# Example: Create a skill for PAY.JP
site2skill https://docs.pay.jp/v1/ payjp

# Example: Create a skill for uv documentation
site2skill https://docs.astral.sh/uv/ uv-docs

# Target specific agent (sets default output directory)
site2skill   --target claude-desktop

CLI Options

site2skill   [options]

Options:
  --target           Target agent (claude|claude-desktop|cursor|gemini|codex). Sets default output directory
  --output, -o       Base output directory for skill structure (overrides target default)
  --skill-output     Output directory for .skill file (default: .)
  --temp-dir         Temporary directory for processing (default: build)
  --max-concurrency  Maximum number of HTTP requests to run at once (default: 10)
  --skip-fetch       Skip the download step (use existing files in temp dir)
  --clean            Clean up temporary directory after completion
  --version          Print version (Rust beta only)

How it works

1. Fetch: Crawls the documentation site with a built-in async HTTP crawler (robots.txt compliant, concurrent).
2. Convert: Converts HTML pages to Markdown using scraper and htmd.
3. Normalize: Resolves relative links to absolute URLs.
4. Validate: Checks the skill structure and size limits.
5. Package: Generates SKILL.md and zips everything into a .skill file.

Output