voice-ai-development
Pass
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides patterns for transcribing user speech via Speech-to-Text (STT) and passing the resulting text directly to an LLM for processing.
- Ingestion points: Audio streams processed in
process_audio_stream,vapi_webhook, andon_transcriptcallback functions. - Boundary markers: None identified in the provided templates to distinguish between instructions and data.
- Capability inventory: The skill patterns include network operations via WebSockets/HTTP and execution of function calls (e.g.,
get_weather,check_order). - Sanitization: No input validation or sanitization of transcribed text is implemented before it is appended to the
conversation_historyand sent to the LLM. - [SAFE]: API keys and secrets are handled correctly via environment variables (
os.environ["OPENAI_API_KEY"]) or designated placeholders, following security best practices. - [SAFE]: All network operations (WebSockets for OpenAI/Deepgram, REST for Vapi/ElevenLabs) are directed to legitimate, well-known service providers associated with the skill's primary purpose.
Audit Metadata