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 official installation instructions and configuration from the inference-sh organization's GitHub repository.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process audio from external URLs for transcription. Processing untrusted data is an inherent surface for indirect prompt injection, although the skill's current scope is limited to returning text data. 1. Ingestion points: Audio data enters the context through the
audio_urlfield in the command input (SKILL.md). 2. Boundary markers: The instructions do not currently include specific delimiters or warnings to isolate transcribed content from agent instructions. 3. Capability inventory: The skill's capabilities are restricted to executing thebeltCLI for remote inference as defined in the frontmatter of SKILL.md. 4. Sanitization: Transcribed text is returned directly to the agent without additional filtering, escaping, or sanitization.
Audit Metadata