azure-ai-voicelive
Warn
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- EXTERNAL_DOWNLOADS (MEDIUM): The skill requires the installation of
azure-ai-voicelive. This is not a standard or widely documented official Microsoft Azure SDK package for Python (standard packages useazure-cognitiveservices-speechorazure-ai-openai), suggesting it could be an unverifiable dependency or a potential typosquatting risk.\n- PROMPT_INJECTION (LOW): (Category 8: Indirect Prompt Injection) The skill's architecture for processing live audio and executing tools creates a surface for indirect prompt injection.\n - Ingestion points: Untrusted audio data is ingested into the session context via
conn.input_audio_buffer.appendinSKILL.md.\n - Boundary markers: There are no boundary markers or delimiters defined in the instructions to separate audio input from system-level commands.\n
- Capability inventory: The skill enables high-impact capabilities including function calling (
toolsinSessionResource) and session updates.\n - Sanitization: The provided documentation and code do not include mechanisms for sanitizing the audio stream or its transcript before it influences model responses.
Audit Metadata