Nano Banana Image Generation
Warn
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS] (MEDIUM): The skill relies on the npm package
@the-focus-ai/nano-banana, which is downloaded and executed at runtime. This package is not from a trusted organization as defined in the security guidelines. - [REMOTE_CODE_EXECUTION] (MEDIUM): The use of
npxto run an external package allows for the execution of arbitrary code from a remote registry. Without a locked version or a trusted source, this is a risk for remote code execution. - [COMMAND_EXECUTION] (LOW): The skill guides the agent to execute shell commands using the
nano-bananaCLI. This is a standard functional requirement but requires careful handling of user input. - [PROMPT_INJECTION] (LOW): The skill is vulnerable to indirect prompt injection (Category 8) because it processes untrusted user data within shell commands.
- Ingestion points: User-provided text prompts and file paths for image processing.
- Boundary markers: Prompt templates use double quotes (e.g., ""), which is insufficient to prevent command injection or payload execution if the input contains escaped characters.
- Capability inventory: The agent executes shell commands via
npxand has access to local files via the--fileand--outputflags. - Sanitization: No sanitization, validation, or escaping of user-provided content is performed before interpolation into the command line.
Audit Metadata