ugc-street-interview-director
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill operates on untrusted data from multiple sources, creating an attack surface for indirect prompt injection.\n
- Ingestion points: Data enters the process via the
ad-strategybrief, human owner instructions and feedback, and raw audio content used for transcription (SKILL.md).\n - Boundary markers: The instructions do not define technical delimiters for untrusted input, but they establish mandatory manual checkpoints, including a human Stage-3 gate and explicit owner approval steps (SKILL.md).\n
- Capability inventory: The agent is empowered to generate storylines, visuals (stills and video), audio transcripts, and burn subtitles into media files (SKILL.md).\n
- Sanitization: No specific sanitization or filtering of external input is described beyond standard QA and human review.
Audit Metadata