azure-ai-agents-persistent-dotnet

Pass

Audited by Gen Agent Trust Hub on Mar 1, 2026

Risk Level: SAFE
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill references the installation of the Azure.AI.Agents.Persistent and Azure.Identity packages via the NuGet package manager. These are official Microsoft libraries, which are well-known and trusted sources.
  • [DATA_EXPOSURE]: The skill follows security best practices by using Environment.GetEnvironmentVariable to handle sensitive configuration data, such as resource endpoints and connection IDs, preventing the exposure of hardcoded secrets.
  • [PROMPT_INJECTION]: The skill provides an architecture for processing external data, which inherently introduces a surface for indirect prompt injection.
  • Ingestion points: Data enters the agent's context through user messages created via client.Messages.CreateMessageAsync and files uploaded using client.Files.UploadFileAsync.
  • Boundary markers: The provided code snippets demonstrate the structural separation of roles (User/Assistant), though they do not explicitly show content delimiters or instruction-guarding prompts.
  • Capability inventory: The agent is granted capabilities to execute code (via CodeInterpreterToolDefinition), perform web searches (via BingGroundingToolDefinition), and execute local application logic (via FunctionToolDefinition).
  • Sanitization: While the SDK provides the framework, the example code does not demonstrate specific sanitization of the arguments passed to local functions like ExecuteFunction; this responsibility is deferred to the developer implementing the tool logic.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 1, 2026, 04:46 PM