speech-to-text
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill provides instructions to download CLI installation metadata and add supplementary skills from the official inference-sh and belt-sh GitHub repositories. These are well-known sources for this skill's ecosystem.
- [INDIRECT_PROMPT_INJECTION]: The skill's primary function involves processing external audio and video data from user-provided URLs. This represents an indirect prompt injection surface where audio content could attempt to influence the agent's behavior.
- Ingestion points:
audio_urlandvideo_urlfields within JSON payloads passed to the transcription models inSKILL.md. - Boundary markers: None present in the provided command-line examples.
- Capability inventory: The skill is restricted to using the
beltCLI tool via Bash, limiting the impact of any potential injection. - Sanitization: Sanitization and safety filtering are handled by the underlying inference service and models (Whisper/ElevenLabs).
Audit Metadata