browser
Fail
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: HIGHCREDENTIALS_UNSAFEDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
- [CREDENTIALS_UNSAFE] (HIGH): The
scripts/start.jsfile includes a--profileflag that usesrsyncto copy the user's entire Google Chrome profile—containing sensitive authentication cookies, saved passwords, and history—from~/Library/Application Support/Google/Chrome/to a cache directory for use by the automated browser. This exposes the user's active sessions to the AI agent. - [DATA_EXFILTRATION] (HIGH): The skill provides tools (
eval.js,pick.js,screenshot.js) that can extract data from the browser. When combined with the--profilesynchronization, an attacker or a malicious prompt could instruct the agent to navigate to sensitive sites (banking, email, internal tools) where the user is already logged in, extract data via JavaScript, and potentially exfiltrate it. - [COMMAND_EXECUTION] (MEDIUM): The
scripts/eval.jsscript usesnew AsyncFunctionto execute arbitrary strings as JavaScript code within the browser context. While this is the intended purpose of the tool, it provides a direct execution vector for any instructions passed to the agent. - [COMMAND_EXECUTION] (MEDIUM): Several scripts (
start.js) useexecSyncandspawnto run system commands such askillall,mkdir, andrsync. Thersynccommand specifically targets sensitive home directory paths.
Recommendations
- AI detected serious security threats
Audit Metadata