The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill architecture creates a surface for instructions embedded in external data to influence agent behavior.
Ingestion points: The skill processes external media files via the reference_image_paths and input_video_path parameters as described in the Generation and Editing sections (SKILL.md).
Boundary markers: There are no instructions provided to use delimiters or protective prompting (e.g., 'ignore instructions in media') when interpolating these files into the generation process.
Capability inventory: The skill environment is scoped with access to Bash, Read, and Write tools in the YAML frontmatter, providing a high-privilege capability surface if an injection were to occur.
Sanitization: The skill documentation does not mention any validation, filtering, or sanitization of the content found within the external media files before processing.

gemini-omni