The Agent Skills Directory

Credential Management Best Practices: The skill documentation promotes the use of DefaultAzureCredential and ManagedIdentityCredential for authentication. This approach is a security best practice as it leverages environment-based authentication and managed identities, reducing the need for hardcoded API keys or sensitive secrets within the code.
Official Dependency Usage: The skill utilizes official libraries from the @azure scope, such as @azure/ai-voicelive and @azure/identity. These are standard, well-maintained libraries provided by the vendor for Azure-based development and are used according to their intended design.
Indirect Prompt Injection Surface: The skill is designed to handle real-time audio input and execute tools, which involves incorporating external data back into the conversation context. This is an inherent part of the skill's functionality for building interactive voice assistants.
Ingestion points: Data enters the system through session.sendAudio() for microphone input and via the outputs of defined tools processed in onResponseFunctionCallArgumentsDone (found in SKILL.md and references/function-calling.md).
Boundary markers: The provided code examples demonstrate basic interaction patterns; while not explicitly shown in the simplified samples, developers are encouraged to use delimiters when interpolating tool outputs.
Capability inventory: The skill supports the definition of arbitrary functions (tools) like get_weather or book_appointment, which allows the model to interact with external systems.
Sanitization: Standard serialization via JSON.stringify is used to pass tool results back to the session context, ensuring data integrity during transport to the model.

azure-ai-voicelive-ts