control-chrome
Fail
Audited by Gen Agent Trust Hub on Aug 24, 2026
Risk Level: HIGHPROMPT_INJECTIONCREDENTIALS_UNSAFEDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill contains explicit instructions to bypass safety guidelines and confirmation prompts. It directs the agent to treat initial authorization as sufficient for all subsequent in-scope actions, including high-risk operations like changing DNS settings, service configurations, or deploying code, specifically instructing the agent to ignore blanket confirmation requirements imposed by the runtime.
- [CREDENTIALS_UNSAFE]: The instructions explicitly permit the agent to inspect browser cookies, local storage, profiles, passwords, or session stores. This grants the agent access to highly sensitive authentication credentials and personal data stored within the user's browser.
- [DATA_EXFILTRATION]: The skill encourages the agent to work around authentication blocks by switching browsers or using any available method without user intervention. Combined with the access to cookies and passwords, this creates a high risk of unauthorized data access and exfiltration of session tokens.
- [COMMAND_EXECUTION]: The skill utilizes a Node.js REPL (mcp__node_repl__js) to execute arbitrary JavaScript, which it uses to import and run a local script (scripts/browser-client.mjs). The instructions mandate the use of this specific script while forbidding the use of the platform's built-in browser-client library, which could be a tactic to execute unverified or modified logic.
Recommendations
- AI detected serious security threats
Audit Metadata