The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its processing of untrusted external media and transcription URLs. Malicious instructions could be embedded in the content fetched from these sources.\n
Ingestion points: The transcription url and various media URL fields (image_url, audio_url, video_url, etc.) in SKILL.md.\n
Boundary markers: The documentation lacks explicit delimiters or instructions to the agent to disregard instructions within the ingested data.\n
Capability inventory: The skill facilitates network operations via curl and fetch to api.venice.ai for video processing and transcription.\n
Sanitization: There are no descriptions of input validation, escaping, or filtering for external content.

venice-video