gemini-live-api-dev

Pass

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill recommends the installation of official SDKs from a well-known service provider, including 'google-genai' for Python and '@google/genai' for JavaScript.
  • [EXTERNAL_DOWNLOADS]: The skill references official documentation hosted on Google's 'ai.google.dev' domain, providing links for API references, tool usage, and session management.
  • [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of live user data streams (audio, video, text) via the 'send_realtime_input' method.
  • Ingestion points: Data received from end-users via the Live API connection.
  • Boundary markers: The skill instructions do not specify custom delimiters, as safety filtering is primarily handled by the Gemini model architecture.
  • Capability inventory: Real-time bidirectional streaming, function calling, and Google Search grounding.
  • Sanitization: No explicit sanitization logic is provided in the connection code snippets.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 19, 2026, 08:29 PM
Security Audit — agent-trust-hub — gemini-live-api-dev