transcribe-refiner
Pass
Audited by Gen Agent Trust Hub on Aug 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and refine untrusted external data from various captioning services (Zoom, YouTube, Teams, etc.).
- Ingestion points: The skill processes user-provided raw caption files in .txt, .vtt, and .srt formats, as well as raw text pastes.
- Boundary markers: The instructions do not define explicit delimiters or boundary markers for the agent to use when ingesting the untrusted transcript content, which is a common practice to prevent the agent from confusing input data with instructions.
- Capability inventory: The skill has no identified dangerous capabilities; it performs purely text-based transformations and does not request tool access, network connectivity, or file system modifications.
- Sanitization: The skill instructions focus on removing technical noise (timestamps, filler words) but do not explicitly direct the agent to sanitize or ignore potential instruction-like strings that might be embedded within the source transcripts.
Audit Metadata