Skip to content

Latest commit

 

History

History
58 lines (40 loc) · 3.15 KB

CLAUDE.md

File metadata and controls

58 lines (40 loc) · 3.15 KB

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Project Overview

Microsoft Agent Framework (MAF) samples repository demonstrating AI agent patterns in .NET 10 with C#. Built on Microsoft.Agents.AI (preview) with Azure OpenAI as the LLM provider. Uses Azure credential-based authentication (Azure CLI, Visual Studio, or Managed Identity).

Build & Run Commands

# Build entire solution
dotnet build MafSamples.slnx

# Run tests
dotnet test src/Samples/Tests/TDD.csproj

# Run a single test
dotnet test src/Samples/Tests/TDD.csproj --filter "FullyQualifiedName~MethodName"

# Run the API
dotnet run --project src/Api/Api.csproj

# Run via Aspire orchestration
dotnet run --project src/AppHost/AppHost.csproj

Architecture

Solution structure: Shared (core) -> Samples (implementations) -> Api (host)

  • src/Shared/ - Core agent infrastructure: AgentFactory (creates agents via Azure OpenAI), AgentTools (tool definitions with AIFunctionFactory), CustomPromptAgentFactory (YAML-based agent creation), settings classes, and extension methods
  • src/Api/ - ASP.NET Core host. Program.cs registers services via DI; ApiExtensions.MapSamples() wires all sample endpoints
  • src/AppHost/ - .NET Aspire distributed app host for orchestrating multiple services
  • src/ServiceDefaults/ - Shared OpenTelemetry, resilience, and health check configuration
  • src/Samples/ - Three sample implementations:
    • AGUI/StateSnapShotEvents/ - Extends agents via DelegatingAIAgent to inject AG-UI STATE_SNAPSHOT events during streaming
    • Tools/ManualToolCall/ - Intercepts and manually executes tool calls from the LLM
    • A2A.Client/ & A2A.Server/ - Agent-to-Agent task delegation with card-based discovery

Key Patterns

  • Agent creation: AgentFactory.Create() builds agents from ChatClient with optional YAML templates and tools. Uses ChainedTokenCredential for Azure auth
  • Tool definition: Static AIFunctionFactory.Create() methods, stored in dictionaries. Use .AsDeclarationOnly() for tools that should be detected but not auto-executed
  • Agent extension: DelegatingAIAgent base class wraps agents to intercept/augment behavior
  • Configuration: Options pattern (IOptions<T>) with settings in appsettings.Development.json and User Secrets for credentials

Testing

  • Framework: xUnit with FluentAssertions and Moq
  • Test project: src/Samples/Tests/ - tests use real Azure OpenAI calls (integration tests)
  • Custom assertion extensions: AgentResponseHelper in Tests/Helpers/ provides fluent assertion chains for AgentResponse: .FunctionCalls(), .ShouldContainCall(), .ShouldHaveArgumentOfType<T>(), .ShouldHaveRequiredInputs()
  • Test helpers: SettingsHelper loads config from User Secrets, InfrastructureHelper creates template repositories, TravelPlanHelper builds test messages
  • YAML templates: Agent prompt templates in Tests/Templates/ (copied to output via PreserveNewest)

Package Management

Centralized in Directory.Packages.props - all NuGet versions are managed there, not in individual .csproj files.