gemini-image-generator
Pass
Audited by Gen Agent Trust Hub on Apr 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The
SKILL.mdfile contains instructions that guide the user to modify shell configuration files (~/.zshrcand~/.bashrc) for persistent storage of environment variables. While presented as a setup step, modifying startup scripts is a persistence mechanism. - [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted data which is subsequently passed to an external AI model.
- Ingestion points: Text prompts provided via the
-p/--promptargument and image files provided via the-i/--input-imageor-r/--reference-imagesarguments inscripts/generate_image.py. - Boundary markers: Absent. User prompts are interpolated directly into the model's instructions without protective delimiters or filtering.
- Capability inventory: The skill performs network requests to the Google Gemini API (a well-known service) and reads/writes local image files.
- Sanitization: No validation, sanitization, or safety filtering is performed on the prompt text or image data before being sent to the external API.
Audit Metadata