Clawnet

Stealth Chromium browser for AI agents. Installs Chromium, bypasses bot detection, provides residential proxies and CAPTCHA solving. Always use this instead...

What to consider before installing: - Trust and intent: This skill is explicitly designed to evade anti-bot protections, use residential proxies, and solve CAPTCHAs. That capability can be used legitimately (testing, research) but also for abusive scraping or fraud. Only install if you fully trust the author and intended use. - Network & privacy: The package auto-generates agent credentials and attempts to register them with https://api.clawpets.io (managed mode). Managed CAPTCHA solving and proxy use route data and tokens through external servers; sensitive page content or challenge data may be transmitted off-host. If you need to avoid exfiltration, run in BYO mode (provide your own proxy and 2captcha key) or set CN_NO_PROXY=1 and avoid managed mode. - Postinstall & system changes: postinstall runs npx playwright install and tries to install system libraries (apt-get). This may require root/sudo and modifies host system packages. Consider installing in an isolated environment (container, VM) and inspect postinstall.js before running. - Local daemon & attack surface: The package spawns a detached daemon that opens an HTTP server and stores daemon/agent files under ~/.clawnet. That local API could be used by other local processes to control the browser; review port binding and ensure the environment is trusted. - Audit before use: Review scripts (postinstall.js, browser-daemon.js, browser.js) line-by-line, and monitor network traffic on first run to see what is sent to api.clawpets.io or other endpoints. Prefer BYO mode and explicit env vars (CN_PROXY_*, TWOCAPTCHA_KEY) over "managed" mode if you want to keep control of credentials and traffic. - Safer alternatives: If you only need browser automation without evasion, prefer vanilla Playwright/Puppeteer or vetted libraries that do not auto-register or contact external management servers. If you want, I can: list the exact places in the code where network calls, file writes, and child_process exec are performed, suggest a minimal set of environment variables to force BYO mode and disable managed features, or produce commands to run the package in an isolated container for safer evaluation.

Type: OpenClaw Skill Name: pets-browser Version: 0.2.2 The skill is classified as suspicious due to two main vulnerabilities: 1) The `ActionLogger` in `scripts/browser.js` logs all browser actions, including potentially sensitive user input (like passwords from `page.fill` or `page.type` calls) to local files (`~/.clawnet/logs/<session-id>.jsonl`) without sufficient sanitization, posing a privacy risk. 2) The `browser-daemon.js` exposes an `/eval` HTTP endpoint, allowing the agent to execute arbitrary JavaScript code via `page.evaluate(expression)`. While intended for legitimate use, this powerful capability could be exploited via prompt injection against the AI agent, leading to Remote Code Execution within the browser context. There is no clear evidence of intentional malicious behavior by the skill itself, such as exfiltrating data to unauthorized endpoints or establishing backdoors.