Skill flagged — suspicious patterns detected

ClawHub Security flagged this skill as suspicious. Review the scan results before using.

m5stack-chrome-browser-control

v1.0.0

M5Stack | 通过 MCP chrome-devtools 协议控制用户本地 Chrome 浏览器,实现自动打开网页、操作任意 Outlook 邮箱、搜索内容等任务。使用前提:用户已在 Chrome 中开启远程调试(127.0.0.1:9222),且 OpenClaw MCP 已配置 chrome-devt...

0· 79·0 current·0 all-time
byXuanwu Yun@yuyun2000

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for yuyun2000/m5stack-chrome-browser-control.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "m5stack-chrome-browser-control" (yuyun2000/m5stack-chrome-browser-control) from ClawHub.
Skill page: https://clawhub.ai/yuyun2000/m5stack-chrome-browser-control
Keep the work scoped to this skill only.
After install, inspect the skill metadata and help me finish setup.
Use only the metadata you can verify from ClawHub; do not invent missing requirements.
Ask before making any broader environment changes.

Command Line

CLI Commands

Use the direct CLI path if you want to install manually and keep every step visible.

OpenClaw CLI

Bare skill slug

openclaw skills install m5stack-chrome-browser-control

ClawHub CLI

Package manager switcher

npx clawhub@latest install m5stack-chrome-browser-control
Security Scan
VirusTotalVirusTotal
Suspicious
View report →
OpenClawOpenClaw
Suspicious
medium confidence
Purpose & Capability
Name/description align with the runtime instructions: the SKILL.md explicitly describes using the Chrome DevTools Protocol (via an MCP adapter) to open pages, inspect and click elements, and read page content (Outlook, X, etc.). The required actions (enable remote debugging and configure MCP) are coherent with that purpose.
Instruction Scope
Instructions stay within the declared purpose (browse, snapshot, click, fill). They do require editing ~/.openclaw/openclaw.json and restarting the gateway, and they rely on reading sensitive content from logged-in browser pages (Outlook/X). Reading private user data is expected for the stated tasks but is intrinsically sensitive — the skill does not ask for unrelated system files or other credentials.
!
Install Mechanism
Although the skill itself has no install spec, it instructs the user to add an MCP server entry that runs `npx chrome-devtools-mcp@latest --autoConnect`. That causes an npm package to be downloaded & executed on the user's machine (using the `latest` tag with no integrity/pinning). Fetching and auto-running code from the public registry without pinning or provenance checks is a high-risk action and should be treated cautiously.
!
Credentials
The skill requests no environment variables, which is appropriate, but it requires enabling Chrome remote debugging on 127.0.0.1:9222 and auto-connecting an MCP adapter. Exposing the DevTools debugging port grants full programmatic control of the browser (including access to cookies, sessions, and page contents). That capability is necessary for the stated tasks but is highly sensitive and can be abused if the MCP adapter or other local processes are untrusted.
!
Persistence & Privilege
The instructions ask the user to persistently modify OpenClaw's config so the MCP adapter autoConnects. This makes a long-lived pathway to control the local browser (and can be invoked autonomously by the agent). Although the skill does not set always:true, adding an auto-connecting MCP entry is a persistent privilege that increases the blast radius if the adapter or skill is compromised.
What to consider before installing
This skill does what it says (control your local Chrome via the DevTools protocol) but it asks you to: 1) enable Chrome's remote debugging (127.0.0.1:9222) — doing so lets any local process control your browser and read logged-in pages (emails, social sites, etc.), and 2) add a config entry that runs `npx chrome-devtools-mcp@latest --autoConnect`, which will download and execute code from npm without a pinned version. Before installing: back up ~/.openclaw/openclaw.json; verify the identity and source of the `chrome-devtools-mcp` package (prefer a specific, audited version instead of @latest); consider using a separate browser profile or a throwaway profile for automation; only enable remote debugging when needed and disable it afterward; and be cautious about letting a persistent MCP auto-connect (it can be invoked autonomously). The skill's source/homepage is unknown — if you proceed, request or inspect the package source and prefer an explicit, pinned install rather than relying on `npx ...@latest`. If you are uncomfortable with these risks, do not enable the MCP entry or avoid using this skill.

Like a lobster shell, security has layers — review code before you run it.

latestvk973j841v2hqhvpmz90thj4pxh840d3f
79downloads
0stars
1versions
Updated 3w ago
v1.0.0
MIT-0

M5Stack | Chrome 浏览器控制 Skill

通过 MCP 协议远程控制用户本地 Chrome 浏览器,完成网页访问、Outlook 邮件查看、内容搜索等自动化任务。


一、用户需要做的准备工作

第一步:在 Chrome 中开启远程调试

  1. 打开 Chrome 浏览器(普通版 Chrome,不是 Chrome Beta
  2. 在地址栏访问:chrome://inspect/#devices
  3. 在页面顶部找到 "Remote debugging" 开关,点击开启
  4. 确认页面显示:"Server running at: 127.0.0.1:9222"

⚠️ 每次重启 Chrome 后需要重新开启此开关。

第二步:配置 OpenClaw MCP

编辑 ~/.openclaw/openclaw.json(Windows 下为 C:\Users\<用户名>\.openclaw\openclaw.json)。

只在已有 JSON 对象中新增 mcp 字段,绝对不要修改或删除文件中其他已有字段(如 plugins、gateway 等),否则 OpenClaw 将无法启动。

{
  "(其他已有字段保持不变)": "...",
  "mcp": {
    "servers": {
      "chrome-devtools": {
        "command": "npx",
        "args": [
          "chrome-devtools-mcp@latest",
          "--autoConnect"
        ]
      }
    }
  }
}

⚠️ 严禁

  • 不能加 --channel=beta(会寻找 Chrome Beta 安装路径,必然失败)
  • 不能覆盖整个文件,只能新增 mcp 字段
  • 修改前建议备份原文件

配置完成后在终端运行 openclaw gateway restart 重启网关使配置生效。


二、MCP 工具说明

所有工具均来自 chrome-devtools-mcp,通过 chrome-devtools__* 前缀调用。

chrome-devtools__list_pages

列出当前浏览器所有打开的标签页,返回每个页面的 pageId 和 URL。

  • 用途:确认 MCP 是否连接成功;查找已登录的目标标签页
  • 成功标志:返回页面列表
  • 失败标志Could not find DevToolsActivePort → 用户未开启调试模式

chrome-devtools__select_page(pageId)

切换到指定标签页,后续操作将作用于该页面。

  • 用途:当浏览器有多个标签时,切换到目标标签(如已登录的 Outlook 或 X)

chrome-devtools__new_page(url)

新建标签页并导航到指定 URL。

  • 用途:打开一个全新页面

chrome-devtools__navigate_page(type, url)

在当前选中标签页中导航到指定 URL。

  • type="url" → 跳转到新地址
  • type="reload" → 刷新当前页
  • 用途:在已登录的标签页内跳转到目标页面(避免登录状态丢失)

chrome-devtools__take_snapshot()

获取当前页面的无障碍树(a11y tree),返回所有可见元素及其 uid

  • 用途:读取页面内容(邮件列表、推文、正文等);获取可交互元素的 uid 供后续点击/输入使用
  • 注意:页面加载中时需等待后再调用

chrome-devtools__click(uid)

点击页面中指定 uid 的元素。

  • 用途:打开邮件、点击按钮、展开折叠内容

chrome-devtools__fill(uid, value)

向指定 uid 的输入框填入文本。

  • 用途:填写搜索框、表单输入

chrome-devtools__wait_for(text)

等待页面出现指定文本后再继续。

  • 用途:等待页面加载完成,避免在未加载时读取空内容

chrome-devtools__take_screenshot()

截取当前页面截图(返回图片)。

  • 用途:当 snapshot 无法准确反映页面状态时,用截图辅助判断

三、常见任务流程

查看 Outlook 邮件(支持任意 Outlook 账号)

1. chrome-devtools__list_pages
   → 查找 URL 含 "outlook.cloud.microsoft" 的标签页

2. 若存在已登录标签:
   chrome-devtools__select_page(pageId=N)
   chrome-devtools__take_snapshot()
   → 读取收件箱邮件列表

3. 若不存在:
   chrome-devtools__new_page(url="https://outlook.cloud.microsoft/mail/")
   → 提示用户手动登录,完成后告知 AI 继续

4. 点击目标邮件:
   chrome-devtools__click(uid="邮件元素uid")
   chrome-devtools__take_snapshot()
   → 读取邮件正文

搜索 X(Twitter)内容

⚠️ X 无法在未登录状态下访问:必须依赖浏览器中已保存的登录 Cookie,不能从零打开登录页。

1. chrome-devtools__list_pages
   → 查找已登录 X 的标签页(URL 含 "x.com")

2. 若存在:
   chrome-devtools__select_page(pageId=N)
   chrome-devtools__navigate_page(type="url", url="https://x.com/search?q=关键词&f=live")
   chrome-devtools__wait_for(text="推文关键词或用户名")
   chrome-devtools__take_snapshot()
   → 读取推文列表并总结

3. 若不存在:
   → 提示用户先在浏览器手动登录 X,完成后告知 AI 继续

四、故障排查

问题原因解决方法
Could not find DevToolsActivePortChrome 未开启调试重新在 chrome://inspect 开启远程调试
Could not find Chrome Beta executableMCP 含 --channel=beta去掉该参数,重启网关
gateway closed (1008): pairing requiredOpenClaw 网关配对失效重启 OpenClaw 或重新配对
Outlook 跳转到产品宣传页浏览器未登录或标签错误list_pagesoutlook.cloud.microsoft 标签并 select_page
X 跳转到登录页未登录或 Cookie 失效用户手动在浏览器登录 X 后告知 AI
页面内容为空页面尚未加载完成先调用 wait_for 等待关键文字出现

Comments

Loading comments...