video-perception

Pass

Audited by Gen Agent Trust Hub on Aug 13, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill consists of markdown instructions for an AI agent to use specific video analysis tools. It contains no executable scripts, hardcoded credentials, or unauthorized data exfiltration patterns.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external media and transcriptions, creating an inherent attack surface for indirect prompt injection.\n
  • Ingestion points: Video files (.mp4, .mov, etc.), audio transcriptions, and metadata files processed by the video_watch and video_analyze tools.\n
  • Boundary markers: The instructions do not specify any delimiters or safety warnings to ignore embedded instructions within the media content.\n
  • Capability inventory: Tools for metadata retrieval (video_info), frame extraction (video_watch), and dependency installation (video_setup).\n
  • Sanitization: No sanitization or filtering of external content is described in the provided instructions.\n- [EXTERNAL_DOWNLOADS]: The video_setup tool is mentioned as a means to check or install dependencies. This is part of the primary functional purpose of the skill and does not involve specific unknown URLs in the instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 13, 2026, 08:11 PM
Security Audit — agent-trust-hub — video-perception