The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content from video files and YouTube URLs, which is a known attack surface.
Ingestion points: Untrusted video data enters the agent context through the results of the analyze_video.py script as described in SKILL.md.
Boundary markers: The documentation does not provide specific delimiters or "ignore previous instructions" warnings to help the agent distinguish between video analysis results and its own system instructions.
Capability inventory: The skill is granted access to high-privilege tools including Bash, Read, and Write, which could be abused if malicious instructions embedded in a video are executed by the agent.
Sanitization: The instructions do not mention sanitization or validation of the video analysis output before it is processed by the agent.
[EXTERNAL_DOWNLOADS]: The skill facilitates fetching content from YouTube and interacts with Google's AI Studio and Vertex AI platforms. These references to well-known and trusted services are documented neutrally and do not represent a direct threat.

gemini-video-understanding