nanobanana
Warn
Audited by Gen Agent Trust Hub on Feb 26, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes the
nanobananaCLI via a shell command. \n - Evidence:
nanobanana <image-path> "<prompt>" [output-path]inSKILL.md.\n- [CREDENTIALS_UNSAFE]: The skill references a specific filesystem path for a sensitive API key. \n - Evidence: Mentions usage of
/run/agenix/gemini-api-keyfor theGEMINI_API_KEYinSKILL.md.\n- [PROMPT_INJECTION]: The skill incorporates untrusted user input directly into a command-line instruction, creating an attack surface for indirect prompt injection. \n - Ingestion points: User-provided edit prompt and image paths entering via the agent context. \n
- Boundary markers: Double quotes are used for the
<prompt>argument, but none are specified for<image-path>or[output-path]. \n - Capability inventory: Shell command execution via
bash(Category 4/10 behavior). \n - Sanitization: No explicit sanitization or validation of input paths or prompt content is mentioned in the skill instructions.
Audit Metadata