The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted text prompts and media assets provided by users, which are then passed to generative models, creating a surface for indirect prompt injection attacks.
Ingestion points: The skill accepts arbitrary text prompts and file paths for images and videos via command-line arguments in all generation scripts, including text_to_video.py, image_to_video.py, reference_to_video.py, edit_video.py, and extend_video.py.
Boundary markers: There are no explicit delimiters or specific instructions provided to the agent or the underlying model to distinguish between instructional system context and potentially malicious embedded data in the user-supplied prompts.
Capability inventory: The skill possesses network capabilities to interact with Google's GenAI API endpoints and file-system write access to save generated video files (e.g., to .mp4 format).
Sanitization: The provided scripts do not implement sanitization, validation, or filtering of the input prompt text before it is interpolated into the model request.

veo-use