speech-to-text
Pass
Audited by Gen Agent Trust Hub on Aug 8, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill refers to external resources for installation and functionality, specifically directing users to download the
beltCLI tool and associated skills from the officialinference.shGitHub organization and service domains. - [PROMPT_INJECTION]: The skill includes functionality that processes content from untrusted external sources, specifically audio and video URLs, creating a surface for indirect prompt injection.
- Ingestion points: Media URLs provided to the
audio_urlandvideo_urlparameters withinbelt app runcommand examples inSKILL.md. - Boundary markers: None present; the skill does not instruct the agent to distinguish between transcribed content and system instructions.
- Capability inventory: Transcribes audio/video to text, translates speech, and generates video captions via the
beltCLI. - Sanitization: The skill does not implement validation or filtering of the resulting transcription before it is ingested into the agent context.
Audit Metadata