analyze-youtube

Pass

Audited by Gen Agent Trust Hub on Oct 26, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes content from public YouTube videos, which serves as an ingestion point for untrusted data that could potentially contain instructions designed to manipulate the agent's analysis.
  • Ingestion points: Untrusted data enters the agent context via YouTube video content resolved in scripts/gemini_youtube.py through the Gemini API.
  • Boundary markers: The script employs structured prompts (PROMPTS dictionary in scripts/gemini_youtube.py) to constrain and format the model's output.
  • Capability inventory: The skill possesses network communication capabilities (via urllib.request) and can write analysis results to local files if the --output flag is used.
  • Sanitization: The validate_url function in scripts/gemini_youtube.py implements a domain check to ensure only HTTPS links from youtube.com or youtu.be are processed.
  • [DATA_EXFILTRATION]: The skill transmits video URLs and prompts to Google's official Gemini API endpoint (generativelanguage.googleapis.com) to perform its primary function. This communication is documented as a standard requirement for the skill.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 26, 2026, 05:17 PM