gemini-live-audio
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill provides numerous references to official Google AI and Google Cloud documentation (e.g., ai.google.dev, cloud.google.com). These resources are from a well-known service and are used to provide legitimate technical specifications for the Gemini Live API.- [INDIRECT_PROMPT_INJECTION]: The skill describes architecture for agents that process untrusted data through live audio streams and multimodal inputs.
- Ingestion points: Real-time microphone audio, streamed media, and camera/screen frames are ingested as described in SKILL.md.
- Boundary markers: The skill recommends using specific system instructions to enforce short turns and concise responses as logical delimiters for the agent's behavior.
- Capability inventory: Documented capabilities include WebSocket session management, Google Search grounding, and custom tool/function calling (e.g., get_order_status in SKILL.md).
- Sanitization: The skill explicitly advises requiring verbal user confirmation before the agent executes high-impact tool actions.
Audit Metadata