The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from user prompts and external media files, creating an attack surface for indirect instructions.
Ingestion points: Data enters the agent's execution context through natural language prompts, reference_image_paths, and input_video_path as described in the SKILL.md documentation.
Boundary markers: The instructions do not define specific delimiters or "ignore previous instructions" warnings to help the model distinguish between administrative guidance and potentially malicious instructions embedded within the processed media or prompts.
Capability inventory: The skill leverages the gemini_omni_video tool, which performs external API operations for video generation and stateful editing based on the provided inputs.
Sanitization: There is no evidence of sanitization, validation, or content filtering mentioned for the user-supplied prompts or media files before they are passed to the model.

gemini-omni