AgentEval: Instructions file for GitHub Copilot

Sample Implementation Guidelines

Sample Naming Convention

Samples are numbered: Sample01_HelloWorld, Sample02_AgentWithOneTool, etc.

Sample Structure

Each sample should:

1. Have a clear single focus
2. Include time-to-understand estimate
3. Run against the configured model (AIConfig). With no provider, stop and say how to configure one. Never switch to a mock or scripted stand-in on your own: a mock runs only when the user asks (--mock), and every result it prints is labelled MOCK. Through 0.42 several samples switched silently and printed ✅ PASSED over canned replies.
4. Print clear step-by-step output

Sample Header Template

/// <summary>
/// Sample XX: Title - Short description
/// 
/// This demonstrates:
/// - Feature 1
/// - Feature 2
/// - Feature 3
/// 
/// ⏱️ Time to understand: X minutes
/// </summary>
public static class Sample_XX_Title
{
    public static async Task RunAsync()
    {
        PrintHeader();
        
        if (!AIConfig.IsConfigured)
        {
            AIConfig.PrintMissingCredentialsWarning();
            Console.WriteLine("   ⚠️  This sample needs a model provider.\n");
            return;
        }
        
        // Step 1, 2, 3...
    }
}

Console Output Pattern

Use consistent formatting:

Console.WriteLine("📝 Step 1: Creating agent...\n");
Console.WriteLine($"   ✓ Agent '{agent.Name}' created\n");

Console.WriteLine("📊 RESULTS:");
Console.WriteLine(new string('─', 60));

Using AIConfig

All samples should use AIConfig class for configuration. It resolves the provider AI_INFERENCE_PROVIDER selects (bitdeer, openai, foundry, azure, openai-compatible):

var chatClient = AIConfig.CreateChatClient(AIConfig.ModelDeployment);

Mock Agents (only on request)

A mock is an offline walkthrough the user asks for (--mock, read as AIConfig.UseMock). It is never a fallback for missing credentials. Gate the sample with AIConfig.StartModelSample() (it prints the MOCK banner, or stops when there is neither a provider nor --mock), name the mock agent "(Mock)", and append AIConfig.MockLabel to every pass/fail line:

if (!AIConfig.StartModelSample())
    return;
var agent = AIConfig.UseMock ? CreateMockAgent() : CreateRealAgent();

private static AIAgent CreateMockAgent()
{
    var mockClient = new MockChatClient("Expected response text");
    return new ChatClientAgent(mockClient, new ChatClientAgentOptions
    {
        Name = "MockAgent",
        ChatOptions = new() { Instructions = "You are a helpful assistant." }
    });
}