openai-realtime
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFE
Full Analysis
- [DATA_EXFILTRATION]: The code samples correctly demonstrate secure handling of API credentials by using environment variables (OPENAI_API_KEY) rather than hardcoding sensitive values.- [EXTERNAL_DOWNLOADS]: The skill utilizes standard, official libraries for WebSocket communication and the OpenAI SDK, and connects only to well-known, trusted API endpoints.- [INDIRECT_PROMPT_INJECTION]: The skill processes external audio data through a structured WebSocket connection. It uses defined event patterns (e.g., response.audio_transcript.delta) to effectively separate user data from processing instructions, minimizing the surface for prompt injection.
Audit Metadata