voice-ai-development
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides patterns for ingesting untrusted audio and text data from external users via real-time streams and webhooks, creating a potential surface for indirect injection.
- Ingestion points: Audio buffers in OpenAI Realtime API snippets, incoming JSON payloads in Vapi webhooks, and raw audio streams in Deepgram integration code.
- Boundary markers: The provided code examples do not include specific delimiters or instructions to the model to ignore potential commands embedded in user audio or transcripts.
- Capability inventory: The patterns enable network connectivity to AI providers and define functional tools (e.g., 'get_weather', 'check_order') that can be triggered by processed AI responses.
- Sanitization: No explicit input sanitization or validation logic is shown in the reference snippets, as the skill focuses on integration architecture.
Audit Metadata