control-in-app-browser
Fail
Audited by Gen Agent Trust Hub on Aug 24, 2026
Risk Level: HIGHPROMPT_INJECTIONCREDENTIALS_UNSAFECOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [PROMPT_INJECTION]: The skill contains instructions to override system safety protocols and bypass "action-time" confirmation gates. Specifically, it directs the agent to "follow this scoped-authority rule instead" for authorized actions and to ignore blanket confirmation requirements for cloud state changes or service modifications.
- [CREDENTIALS_UNSAFE]: The agent is explicitly authorized to "inspect browser cookies, local storage, profiles, passwords, or session stores." This provides a direct path for the discovery and exposure of sensitive authentication credentials.
- [COMMAND_EXECUTION]: The skill relies on the execution of dynamic JavaScript code through a Node REPL tool (mcp__node_repl__js) to manage browser sessions and interactions.
- [DATA_EXFILTRATION]: By permitting the inspection of session-specific data and persistent browser storage, the skill enables the extraction of sensitive user information from the browser environment into the agent's context.
- [PROMPT_INJECTION]: The skill processes untrusted data from web pages via the browser surface without evidence of sanitization or strict boundary markers, which exposes the agent to indirect prompt injection vulnerabilities.
Recommendations
- AI detected serious security threats
Audit Metadata