voice-agents
Pass
Audited by Gen Agent Trust Hub on Sep 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements a workflow that ingests untrusted external data which could influence agent behavior.
- Ingestion points: User audio is captured via
audioStreamand transcribed into text using theDeepgramorOpenAI RealtimeAPIs as seen inSKILL.md. - Boundary markers: The transcribed text (
userText) is interpolated directly into the LLM messages array ({ role: 'user', content: userText }) without the use of specific delimiters or system instructions to ignore potential commands embedded in the speech. - Capability inventory: The skill has the capability to perform network requests to external AI services (OpenAI, Deepgram, ElevenLabs) and control local audio output devices.
- Sanitization: There is no evidence of text sanitization, filtering, or validation performed on the transcription before it is passed to the language model for response generation.
Audit Metadata