The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the processing of external content from YouTube URLs and other media links (images/videos) provided by the user. This creates a potential surface for indirect prompt injection if the untrusted external data (e.g., transcripts or media metadata) contains malicious instructions designed to manipulate the agent. This risk is inherent to the generative nature and primary purpose of the skill.
Ingestion points: The url field in the /video/transcriptions endpoint and various media URL fields (image_url, video_url, reference_image_urls, etc.) in the /video/queue endpoint (SKILL.md).
Boundary markers: The skill does not explicitly provide delimiters or specific "ignore embedded instructions" warnings for the agent when processing these external data sources.
Capability inventory: The skill provides instructions for the agent to use shell commands (curl) to interact with the Venice AI API.
Sanitization: The documentation does not describe any specific sanitization or filtering of the content retrieved from external URLs.
[SAFE]: All network operations documented in the skill target the vendor's official API (api.venice.ai) or well-known services like YouTube. The skill correctly instructs the use of environment variables ($VENICE_API_KEY) for authentication rather than hardcoding credentials.

venice-video