Browser Automation

PassAudited by VirusTotal on May 7, 2026.

Overview

Type: OpenClaw Skill Name: tianxi-browser-automation Version: 1.0.0 The skill bundle provides standard browser automation capabilities such as navigation, form filling, and element interaction using the 'tai-app-browser-mcp-server'. While it includes powerful tools like 'browser_evaluate' (JavaScript execution) and 'browser_install' (software installation), the SKILL.md instructions explicitly mandate user consent for installations and include security constraints against accessing sensitive data or performing unauthorized actions. The behavior is well-documented and aligns with the stated purpose of automating web tasks.

Findings (0)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

What this means

The browser could approve a warning or confirmation that the user did not intend to accept.

Why it was flagged

The skill directs the agent to accept browser dialogs by default. Dialogs can be confirmations for posting, submitting, deleting, payments, or other account-changing actions.

Skill content
操作过程中遇到弹窗,优先使用 `browser_handle_dialog` 处理,默认选择"确认"
Recommendation

Require the agent to read dialog text aloud and ask the user before confirming any dialog that changes data, sends content, purchases, deletes, or affects an account.

What this means

The skill may disrupt browser sessions, log the user out of sites, or remove session state more broadly than expected.

Why it was flagged

Cookies are browser session material tied to logged-in identities. The instruction is broad and automatic, without limiting which site cookies are cleared or requiring separate user approval.

Skill content
操作过程中若涉及用户隐私数据,操作完成后自动清理浏览器缓存和Cookie
Recommendation

Use a dedicated browser profile where possible, and change this behavior to ask before clearing cookies or to clear only the specific site involved.

What this means

The real browser-control behavior depends on external software not reviewed here.

Why it was flagged

The skill relies on an external MCP server and may invoke a browser installation flow, but those components are not included in the reviewed artifacts.

Skill content
- **依赖MCP**: tai-app-browser-mcp-server ... 调用 browser_install 进行安装
Recommendation

Verify the MCP server and browser installer source before use, and avoid installing them from untrusted channels.

What this means

On sensitive or untrusted pages, custom JavaScript could expose or change more data than a normal click-based workflow.

Why it was flagged

Executing custom JavaScript in a page is a powerful automation capability. It is disclosed and purpose-aligned, but it can alter page state or extract page data.

Skill content
执行JS代码:使用 `browser_evaluate` 可执行自定义JavaScript
Recommendation

Use JavaScript execution only when necessary, on trusted pages, and after confirming what the script will do.

What this means

Private webpage contents, email pages, account pages, or request details may be exposed to the agent/tooling during automation.

Why it was flagged

The skill sends page snapshots, screenshots, and network-request information through the browser MCP/tool boundary; these can include private page content.

Skill content
`browser_snapshot` 获取页面元素快照 ... `browser_take_screenshot` ... `browser_network_requests` 可查看页面加载的所有接口请求
Recommendation

Avoid using it on highly sensitive pages unless you trust the MCP server and understand its logging/storage behavior.