youtube-research-digest
Pass
Audited by Gen Agent Trust Hub on Aug 25, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from multiple external platforms, creating a surface for indirect prompt injection.
- Ingestion points: Video transcripts, titles, and descriptions from YouTube, as well as data from TikTok, Reddit, and other social media/e-commerce platforms (SKILL.md, Steps 1, 4, and 6).
- Boundary markers: The agent is not instructed to use delimiters or protective prompting when handling the external content.
- Capability inventory: The skill uses tools to fetch and analyze external data via the Scavio API.
- Sanitization: No explicit sanitization or filtering of external text is performed.
Audit Metadata