yt-viral

Pass

Audited by Gen Agent Trust Hub on Nov 14, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (YouTube video titles and channel metadata) to identify patterns and suggest content strategies, creating a surface for potential instruction injection.
  • Ingestion points: The swipe.py script reads video data from a JSON file (e.g., collected.json) containing information sourced from external YouTube pages.
  • Boundary markers: The instructions do not define clear boundaries or markers to prevent the agent from following instructions potentially embedded in video titles during analysis.
  • Capability inventory: The skill uses a script to rank videos and then asks the agent to categorize them and suggest new content titles based on the findings.
  • Sanitization: Video titles are processed through regex matching and displayed to the agent without sanitization or escaping.
  • [EXTERNAL_DOWNLOADS]: The instructions recommend using yt-dlp, a third-party tool, to collect data from YouTube. While this is an optional manual step and the tool is not a direct dependency, the skill's primary function relies on ingestion of external data.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 14, 2026, 05:51 AM