.Net: Repeated ChatHistory.AddStreamingMessageAsync(GetStreamingChatMessageContentsAsync) throws ClientResultException

[Hassmann]

Describe the bug

When using the streaming API with automatic tool calls like this:

// Create Kernel
var kernel = Kernel.CreateBuilder()
	.AddAzureOpenAIChatCompletion(deployment, client)
	.Build();

kernel.ImportPluginFromObject(new Functions());

// Prepare Chat
var chatService = kernel.GetRequiredService<IChatCompletionService>();

OpenAIPromptExecutionSettings settings = new()
{
	FunctionChoiceBehavior = FunctionChoiceBehavior.Auto()
};

ChatHistory chatHistory = new("You are to test the system");

for (int i = 0; i < 2; i++)
{
	try
	{
		chatHistory.AddUserMessage("Please test the system");

		var results = chatHistory.AddStreamingMessageAsync(chatService
			.GetStreamingChatMessageContentsAsync(chatHistory, settings, kernel)
			.Cast<OpenAIStreamingChatMessageContent>()
		);

		await foreach (var result in results)
		{
			Console.Write(result.ToString());
		}

		Console.WriteLine($"Call #{i} OK");
	}
	catch (Exception e)
	{
		// On 2nd call...
		Console.WriteLine("ERROR: " + e.Message);
	}
}

a second call after the assistant has actually invoked a function will throw a ClientResultException with:

HTTP 400 (invalid_request_error: )
Parameter: messages.[5].role

An assistant message with 'tool_calls' must be followed by tool messages responding to each 'tool_call_id'. The following tool_call_ids did not have response messages: call_MnrWYfXsn1cz6RgiUi44acoy

To Reproduce
I have attached a small Console app that reproduces the issue.
ToolCallBug.zip

Platform

• OS: Windows, Azure OpenAI (GPT4o)
• IDE: Visual Studio
• Language: C#
• Source: Microsoft.SemanticKernel 1.25.0

[evchaki]

@Hassmann - thanks for reporting this, we will take a look.