gemini-image
Fail
Audited by Gen Agent Trust Hub on Mar 5, 2026
Risk Level: HIGHCREDENTIALS_UNSAFECOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill is built around executing a series of bash commands via the
camofoxCLI to automate complex browser workflows on Google Gemini. - [CREDENTIALS_UNSAFE]: The skill's authentication flow (Step 3.2) explicitly instructs the agent or user to pass sensitive credentials, including Google account passwords, directly as plaintext arguments in CLI commands (e.g.,
camofox type e2 "password"). This practice exposes credentials to shell history, process monitors, and system logs. - [COMMAND_EXECUTION]: The skill provides a fallback mechanism using
camofox evalto execute arbitrary JavaScript code within the browser context. This allows for dynamic code execution that could bypass standard safety boundaries if the input prompt is manipulated. - [DATA_EXFILTRATION]: The skill uses
curlto retrieve generated images from a local API endpoint (http://localhost:9377). While targeting localhost, the use of network tools to move data between the automation environment and the local filesystem is a monitored pattern.
Recommendations
- AI detected serious security threats
Audit Metadata