The Agent Skills Directory

[PROMPT_INJECTION]: The skill possesses an indirect prompt injection attack surface because it processes untrusted user data (text prompts and reference images) and passes them to the Gemini model to perform image generation, editing, and composition.
Ingestion points: The skill accepts untrusted data through the instruction and prompt arguments in scripts/generate_image.py, scripts/edit_image.py, and scripts/compose_images.py.
Boundary markers: There are no explicit boundary markers or specialized system instructions used to delimit user input or warn the model about embedded instructions in the processing scripts.
Capability inventory: The skill has the capability to read local files (via Image.open), write local files (via image.save), and perform network requests to the Gemini API endpoint.
Sanitization: No input validation, escaping, or filtering is applied to the user-provided prompt strings or image content before processing.

gemini-imagegen