speech-to-text
Pass
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references installation commands for external CLI tools (
belt-sh/cli) and supplementary skills from theinference-shrepository. It also provides a link to an external markdown file on GitHub for installation instructions. These references are essential for the skill's intended functionality of providing transcription services. - [INDIRECT_PROMPT_INJECTION]: The skill's primary function involves processing external audio/video files and generating transcripts that are then used in subsequent workflows, such as video captioning. This introduces a surface where malicious content hidden in the audio stream could be interpreted as instructions by the agent.
- Ingestion points: The skill accepts
audio_urlandvideo_urlas inputs for transcription tasks inSKILL.mdexamples. - Boundary markers: The instructions do not specify any boundary markers or delimiters for the generated transcription data when passed between tools.
- Capability inventory: The skill utilizes the
beltCLI tool for command execution, file writing via redirection (> transcript.json), and remote inference. - Sanitization: There is no evidence of sanitization or validation of the transcribed text before it is consumed by downstream apps like
infsh/caption-videos.
Audit Metadata