video-frame-analysis

Pass

Audited by Gen Agent Trust Hub on Apr 18, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill facilitates an indirect prompt injection attack surface by extracting text from untrusted video files using OCR and returning the raw text to the agent. Malicious instructions visually embedded in a video could be interpreted and executed by the agent after the OCR process.
  • Ingestion points: Text extracted from video frames via tesseract in scripts/video-frame-analysis.sh and stored in ocr.txt and ocr.json.
  • Boundary markers: Frames are delimited by filenames in ocr.txt and organized as objects within a JSON array in ocr.json, providing structural boundaries.
  • Capability inventory: The skill executes local binaries (ffmpeg, ffprobe, tesseract, python3) and performs filesystem writes to the output directory.
  • Sanitization: No sanitization, filtering, or safety instructions are applied to the extracted OCR text before it is returned to the agent context.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 18, 2026, 02:44 AM
Security Audit — agent-trust-hub — video-frame-analysis