nano-banana-pro

Warn

Audited by Gen Agent Trust Hub on Feb 19, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION] (MEDIUM): Path traversal vulnerability in generate_image.py. The script determines the output path using Path.cwd() / args.filename. Because the filename argument is user-controlled and not sanitized for directory traversal characters (e.g., ../), it could be exploited to overwrite sensitive files outside of the intended working directory.
  • [DATA_EXFILTRATION] (LOW): The skill facilitates reading arbitrary local files through the --input-image parameter. While intended for image data, there are no file type or path restrictions, allowing the agent to potentially read and transmit non-image file contents to the external API if tricked by a prompt.
  • [PROMPT_INJECTION] (LOW): The skill is vulnerable to indirect prompt injection.
  • Ingestion points: Processes user-supplied text prompts and external image files via the --input-image parameter.
  • Boundary markers: None. Input content is interpolated directly into the API request without delimiters or 'ignore' instructions.
  • Capability inventory: The skill has the ability to read/write local files and communicate with external Google APIs.
  • Sanitization: No validation or sanitization is performed on the prompt text or the filename inputs.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 19, 2026, 03:56 PM