azure-ai-voicelive-py

Pass

Audited by Gen Agent Trust Hub on Apr 21, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs the installation of azure-ai-voicelive, aiohttp, and azure-identity. While azure-identity is an official library, azure-ai-voicelive is a third-party community package not managed by the primary service provider.
  • [CREDENTIALS_UNSAFE]: Documentation identifies the use of AZURE_COGNITIVE_SERVICES_KEY for authentication but utilizes safe placeholder values (<api-key>) and explicitly recommends the more secure DefaultAzureCredential for production deployments.
  • [PROMPT_INJECTION]: The skill implements a real-time voice interface that processes audio transcripts, presenting a surface for indirect prompt injection.
  • Ingestion points: The skill ingests untrusted data from event.transcript, event.delta, and event.arguments via a WebSocket connection in SKILL.md.
  • Boundary markers: Examples use basic system instructions without explicit delimiters or warnings to ignore instructions embedded in the audio stream.
  • Capability inventory: The skill enables bidirectional WebSocket communication, model-triggered function execution, and session state modification.
  • Sanitization: There is no evidence of transcript sanitization or validation before the data is processed by the agent.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 21, 2026, 12:19 PM