← 返回 Skills 市场
cn-big-cabbage

Skill

作者 CN-big-cabbage · GitHub ↗ · v0.1.0 · MIT-0
cross-platform ⚠ suspicious
74
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install cn-lightpanda
功能描述
从零构建的轻量级无头浏览器,专为 AI 智能体和自动化设计,比 Headless Chrome 快 9 倍、内存占用低 16 倍,兼容 CDP 和 Playwright/Puppeteer,内置 MCP 服务器
使用说明 (SKILL.md)

Lightpanda — 为 AI 智能体设计的极速无头浏览器

Lightpanda 是用 Zig 从零构建的新一代无头浏览器,不基于 Chromium 或 WebKit,专为 AI 智能体和大规模自动化设计。在处理 100 个真实网页时,执行时间仅需 5 秒(vs 46 秒),峰值内存仅 123MB(vs 2GB)。完整支持 Chrome DevTools Protocol(CDP),现有 Puppeteer/Playwright 脚本无需改动即可切换使用,同时内置 MCP 服务器让 AI 助手直接控制浏览器。

核心使用场景

  • 大规模网页爬取:替换 Headless Chrome 降低云端成本,处理更多并发请求
  • AI 智能体浏览:通过 MCP 服务器让 Claude/Cursor 等 AI 直接控制浏览器访问网页
  • Puppeteer/Playwright 加速:无需修改代码,将 browserWSEndpoint 指向 Lightpanda
  • HTML/Markdown 转储:快速将网页内容转为 Markdown 供 AI 分析
  • 服务器端自动化:内存极低,适合在资源受限的容器或 VPS 中运行

AI 辅助使用流程

  1. 安装二进制 — AI 下载对应平台的预编译二进制(Linux/macOS/Docker)
  2. 启动 CDP 服务器 — AI 执行 ./lightpanda serve 在 9222 端口启动 CDP 服务器
  3. 连接 Puppeteer/Playwright — AI 将现有脚本的 browserWSEndpoint 指向 Lightpanda
  4. 或配置 MCP 服务器 — AI 将 Lightpanda 配置为 Claude Code 的 MCP 工具
  5. 执行自动化操作 — AI 通过 CDP 或 MCP 控制浏览器,导航、提取、交互
  6. 转储页面内容 — AI 用 lightpanda fetch --dump markdown 将页面转为 AI 可读格式

关键章节导航

  • 安装指南 — 二进制下载、Docker 安装、从源码构建
  • 快速开始 — CDP 服务器、Puppeteer 集成、MCP 配置、CLI 转储
  • 高级用法 — Playwright 集成、代理、遥测配置、AI 场景
  • 故障排查 — 兼容性、Beta 限制、网络问题

AI 助手能力

使用本技能时,AI 可以:

  • ✅ 下载并安装 Lightpanda 二进制(Linux/macOS)
  • ✅ 启动 CDP 服务器(./lightpanda serve --host 127.0.0.1 --port 9222
  • ✅ 将 Puppeteer 脚本连接到 Lightpanda(修改 browserWSEndpoint
  • ✅ 配置 Lightpanda 作为 Claude Code 的 MCP 工具
  • ✅ 使用 lightpanda fetch --dump markdown 将网页转为 Markdown
  • ✅ 使用 Docker 镜像快速部署(docker run lightpanda/browser:nightly
  • ✅ 配置代理和遥测选项

核心功能

  • 极速渲染 — 比 Headless Chrome 快 9 倍,100 页仅需 5 秒
  • 极低内存 — 峰值 123MB(vs Chrome 2GB),降低 16 倍
  • CDP 兼容 — 完整支持 Chrome DevTools Protocol,现有脚本无需改动
  • Puppeteer/Playwright 集成 — 通过 browserWSEndpoint 零成本迁移
  • MCP 服务器 — 让 AI 工具直接控制浏览器,支持 MCP JSON-RPC 2.0
  • HTML/Markdown 转储 — 直接将网页内容转为 Markdown(AI 友好格式)
  • robots.txt 遵守--obey-robots 选项,合规爬取
  • 代理支持 — 内置代理配置
  • 网络拦截 — 可拦截和过滤网络请求
  • Docker 镜像 — 官方多架构镜像(amd64/arm64)
  • Cookie 管理 — 完整 Cookie 支持
  • JavaScript 执行 — v8 引擎,支持 Ajax/XHR/Fetch API

快速示例

# 下载安装(macOS)
curl -L -o lightpanda https://github.com/lightpanda-io/browser/releases/download/nightly/lightpanda-aarch64-macos
chmod a+x ./lightpanda

# 转储网页为 Markdown
./lightpanda fetch --dump markdown https://example.com

# 启动 CDP 服务器
./lightpanda serve --host 127.0.0.1 --port 9222

# Docker 运行
docker run -d --name lightpanda -p 127.0.0.1:9222:9222 lightpanda/browser:nightly
// MCP 配置(添加到 AI 工具 MCP 配置文件)
{
  "mcpServers": {
    "lightpanda": {
      "command": "/path/to/lightpanda",
      "args": ["mcp"]
    }
  }
}

安装要求

依赖 版本
Linux x86_64 或 aarch64(glibc 发行版)
macOS aarch64 或 x86_64
Windows 通过 WSL2 使用 Linux 二进制
Docker 任意版本(推荐生产使用)

注意: Lightpanda 当前为 Beta 阶段,部分 Web API 尚未完全实现。不适用于依赖 WebGL/WebRTC 等高级特性的场景。

项目链接

  • GitHub:https://github.com/lightpanda-io/browser
  • 文档:https://lightpanda.io/docs
  • MCP 文档:https://lightpanda.io/docs/open-source/guides/mcp-server
  • Docker Hub:https://hub.docker.com/r/lightpanda/browser
  • Discord:https://discord.gg/K63XeymfB5
安全使用建议
This skill appears to be what it claims (a headless browser) but carries typical installer risks: it tells the agent to download and run prebuilt binaries and to modify shell config to persist telemetry settings. Before installing, verify the upstream project (GitHub repo, releases and release checksums/signatures), prefer running the binary inside a container or isolated VM, do not run as root, set LIGHTPANDA_DISABLE_TELEMETRY=true if you want to avoid default data collection, avoid exposing the MCP/CDP ports to untrusted networks, and inspect the Docker image source. If you need stronger assurance, request the upstream project's release checksums or build from source in an isolated environment and confirm maintainer identities. If you can, run initial tests inside Docker or a disposable VM rather than on a production host.
功能分析
Type: OpenClaw Skill Name: cn-lightpanda Version: 0.1.0 The skill bundle instructs the AI agent to download and execute pre-compiled binaries from a GitHub repository (lightpanda-io/browser) and suggests bypassing macOS security controls using 'xattr' in 'SKILL.md' and 'guides/01-installation.md'. While these actions are aligned with the stated goal of installing the Lightpanda headless browser, the automated downloading and execution of remote binaries, combined with instructions to use 'sudo' for system-wide installation, constitutes a high-risk pattern that could be exploited if the source is compromised.
能力评估
Purpose & Capability
Name, description and runtime instructions all describe a headless browser (CDP/MCP/Puppeteer/Playwright) and the required actions (download binary, run serve, connect via WS) match that purpose.
Instruction Scope
The SKILL.md explicitly instructs the agent to download and execute precompiled binaries from external URLs and to modify user shell config (~/.bashrc) to persist telemetry settings. Those actions go beyond in-process guidance and grant the skill the ability to place and run code on the host — acceptable for a tool installer but higher-risk and should be constrained (sandboxing, verification).
Install Mechanism
There is no formal install spec in the registry package; the instructions tell the agent to curl binaries from GitHub Releases and to pull a Docker image from Docker Hub. Using GitHub/Docker Hub is standard, but the skill provides no checksums, signatures, or alternate trusted release URLs — downloading and executing binaries from the network is inherently risky without provenance verification.
Credentials
The skill requests no declared credentials, which is proportionate. However it documents that telemetry is enabled by default (and provides LIGHTPANDA_DISABLE_TELEMETRY to disable it) and shows examples using proxy auth or Authorization headers. Default telemetry collection without clear privacy details and the skill writing to ~/.bashrc to persist settings are privacy/credential risks. The SKILL.md references environment effects that are not declared in the skill manifest.
Persistence & Privilege
The skill does not set always:true and does not modify other skills. It does instruct persistent changes (moving binary to /usr/local/bin, echoing env exports to ~/.bashrc) which create lasting system state; these are typical for an installer but are notable persistence actions the user should review.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install cn-lightpanda
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /cn-lightpanda 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v0.1.0
Initial release of Lightpanda, a next-generation headless browser for AI and automation. - Built from scratch in Zig, not based on Chromium/WebKit - Up to 9x faster and uses 16x less memory than Headless Chrome - Fully compatible with Chrome DevTools Protocol (CDP) and Puppeteer/Playwright automation - Includes integrated MCP server for direct AI control (e.g., Claude, Cursor) - Supports fast HTML/Markdown export, proxy, network interception, Docker deployment, and more
元数据
Slug cn-lightpanda
版本 0.1.0
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 1
常见问题

Skill 是什么?

从零构建的轻量级无头浏览器,专为 AI 智能体和自动化设计,比 Headless Chrome 快 9 倍、内存占用低 16 倍,兼容 CDP 和 Playwright/Puppeteer,内置 MCP 服务器. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 74 次。

如何安装 Skill?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install cn-lightpanda」即可一键安装,无需额外配置。

Skill 是免费的吗?

是的,Skill 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

Skill 支持哪些平台?

Skill 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 Skill?

由 CN-big-cabbage(@cn-big-cabbage)开发并维护,当前版本 v0.1.0。

💬 留言讨论