youtube-research
Pass
Audited by Gen Agent Trust Hub on Sep 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, specifically YouTube video titles, descriptions, and transcripts fetched via the TubeLab API. This content is then used as input for AI analysis and report generation, which creates a standard surface for indirect prompt injection attacks where malicious instructions hidden in video metadata could attempt to influence the agent.
- Ingestion points: Video metadata and transcripts are retrieved in
scripts/find_outliers.pyand stored inoutliers.jsonfor subsequent processing. - Boundary markers: The instructions in
SKILL.mddo not define specific delimiters or instructions to the AI to ignore potentially malicious content embedded within the video transcripts. - Capability inventory: The skill has the capability to write files (JSON data and Markdown reports) and make network requests to the TubeLab and Gemini APIs.
- Sanitization: The provided scripts do not show evidence of sanitizing or filtering the fetched video content before it is processed by the AI model or written to report files.
Audit Metadata