The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data derived from video transcripts and visual scene analysis, which constitutes a vulnerability surface for indirect prompt injection attacks.
Ingestion points: Video content is ingested via tool calls like generate_transcript, get_transcript, and understand_video which fetch data from external sources (SKILL.md).
Boundary markers: No specific delimiters or instructions to ignore embedded commands within the video data are provided to the agent.
Capability inventory: The agent has access to the videodb SDK to perform searches, ask questions, and generate video clips based on its interpretation of the video data (SKILL.md).
Sanitization: There is no evidence of sanitization or filtering of the transcript text or visual descriptors before they are processed by the LLM.
[DATA_EXPOSURE_EXFILTRATION]: Safe secret management is observed. The skill instructs the agent to save user-provided API keys into a .env file, which is the recommended practice for protecting credentials in a local development environment.
[EXTERNAL_DOWNLOADS]: The skill references and uses the videodb Python package, which is the official SDK provided by the vendor for its services.

video-search