让 Claude 通过你自己的 Chrome 会话,只读浏览 X(Twitter)并整理出带证据链接的情报素材。
browse-x 是一个可放进 .claude/skills/ 的 Agent Skill。它使用本地 Chrome DevTools Protocol(CDP)和 Playwright,读取 X 的 For You、Following 或 Explore 页面,按时间窗口和关键词筛选帖子,并可对高频作者做主页深扫。
公开仓库只包含这一个 Skill:没有其他 Agent、账号凭据、抓取结果或个人 Profile 数据。
- 读取主页推荐流、关注页或 Explore 趋势内容;
- 按最近小时数和关键词过滤;
- 对主 feed 中出现频率较高的作者做可选深扫;
- 输出原帖文本、作者、时间、X 链接、点赞数、转发数和来源;
- 同时支持 JSON 文件输出和 stdout 标记输出,方便交给 Agent 继续总结。
它是只读工具,不会点赞、转发、发帖、关注、私信,也不调用 X API。
- Node.js 18 或更高版本;
- 已安装 Chrome,并以 CDP 方式监听本机回环地址;
- 一个已登录 X 的专用 Chrome Profile;
- 能访问
x.com的网络环境。
脚本不需要 X API Key,也不会读取 Cookie 文件。首次使用建议按照平台指南启动独立 Profile,并在窗口内手动登录:
把这个仓库放到目标项目的 .claude/skills/browse-x/,确保 SKILL.md 位于该目录根部:
REPO_URL="https://github.com/<owner>/browse-x.git"
git clone "$REPO_URL" .claude/skills/browse-x
cd .claude/skills/browse-x/scripts
npm ci如果你是从压缩包或本地目录安装,直接将整个仓库目录复制到 .claude/skills/browse-x/ 即可。然后在 Claude 对话中使用“浏览 X”“看 X 动态”“X 情报”等表达触发 Skill。
先准备 CDP 和登录状态,再执行:
cp examples/config.home.json /tmp/x-scrape-config.json
cd scripts
npm ci
node scraper.mjs --config=/tmp/x-scrape-config.jsonWindows 用户可将配置中的 outputPath 改为类似 C:/Users/<用户名>/AppData/Local/Temp/x-scrape-output.json 的路径。更完整的配置说明见 配置与输出。
mode |
页面 | 说明 |
|---|---|---|
home |
https://x.com/home |
For You 推荐流 |
following |
https://x.com/home |
尝试切换到 Following 标签,失败时回退到主页 |
explore |
https://x.com/explore |
Explore 趋势内容 |
当前版本不支持线程、单用户主页和站内搜索模式。不要把这些能力当作已实现功能。
使用前请确认你对相关账号、浏览器会话和数据处理拥有授权,并遵守 X 的服务条款与适用法律。
- 只连接
127.0.0.1/localhost上的 CDP;不要把调试端口暴露到局域网或公网; - 使用独立 Chrome Profile,不要复制真实 Profile 的
Cookies、Login Data或Local State; - 不要把抓取 JSON、页面截图、临时 Profile 或个人账号信息提交到仓库;
- X 的 DOM 选择器可能变化,抓取结果应视为“当前页面可见内容”,不能替代官方 API 或独立核验。
cd scripts
npm ci
npm testnpm test 当前执行脚本语法检查。真实抓取需要本机 Chrome CDP、X 登录会话和网络环境,因此不放在 CI 中自动运行。
贡献前请阅读 CONTRIBUTING.md。
browse-x is a read-only Agent Skill that lets Claude inspect X/Twitter through a user-owned Chrome session over local CDP. It supports the For You, Following, and Explore pages, filters recent posts by keywords, optionally deep-scans frequent authors, and returns structured JSON with source links and engagement metrics.
It does not use the X API, store credentials, or perform write actions such as posting, liking, retweeting, following, or messaging. See the platform setup guides and configuration docs before running it.
MIT. See LICENSE.