speech-to-text
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill documentation correctly instructs users to manage sensitive credentials like
ELEVENLABS_API_KEYusing environment variables. It also highlights secure practices for client-side streaming by recommending the generation of single-use tokens on the backend rather than exposing API keys directly. - [INDIRECT_PROMPT_INJECTION]: The skill provides capabilities to ingest and transcribe audio/video content from local files and remote URLs. This is the primary intended function of the skill. While it creates a surface for indirect prompt injection (malicious instructions potentially spoken in the audio), this risk is inherent to transcription services and is mitigated by the agent's internal safety guardrails.
- Ingestion points: Audio stream URLs and file read operations described in
SKILL.mdandreferences/realtime-server-side.md. - Boundary markers: None explicitly defined for handling the transcribed text output.
- Capability inventory: The skill uses official SDKs to communicate with ElevenLabs API; no dangerous system-level code generation was observed.
- Sanitization: The skill relies on the service provider's transcription engine and standard application-level processing.
Audit Metadata