util-youtube-analyzer

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • EXTERNAL_DOWNLOADS (LOW): The skill documentation and scripts facilitate the download of a Whisper model file (~141MB) from HuggingFace (huggingface.co). While this is a standard practice for AI tools, HuggingFace is not on the predefined trusted organization list.
  • PROMPT_INJECTION (LOW): The skill is susceptible to Indirect Prompt Injection (Category 8) because it processes untrusted data from external sources.
  • Ingestion points: Transcripts are fetched from YouTube captions or generated from video audio via yt-dlp and whisper-cpp in scripts/yt-transcript.sh.
  • Boundary markers: The script wraps transcripts in basic markdown headers (e.g., # Video Title), which do not provide a secure boundary against embedded instructions.
  • Capability inventory: The agent reads the resulting transcript file and performs analysis, which involves high-level reasoning over potentially adversarial text.
  • Sanitization: No sanitization is performed on the transcript content. An attacker could include malicious prompt instructions in video captions to subvert the agent's analysis.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 06:21 PM