feat: add multiple messages pattern to dotnet agent-framework sample

docs/multiple-messages-tracking.md

@@ -0,0 +1,42 @@
+# Multiple Messages Per Prompt — Issue #268 Tracking
+
+Tracking implementation of the multiple-messages pattern across all Agent365-Samples.
+
+**Pattern**: Send an immediate ack message → typing indicator loop → LLM response, all as separate discrete Teams messages.
+
+> **Reference**: [GitHub Issue #268](https://github.com/microsoft/Agent365-devTools/issues/268)
+
+---
+
+## Progress
+
+| Sample | Implementation File | Code | README | Committed | Tested (Agents Playground) |
+|---|---|---|---|---|---|
+| `dotnet/agent-framework` | `Agent/MyAgent.cs` | ✅ | ✅ | ✅ | ✅ |
+| `dotnet/semantic-kernel` | `Agents/MyAgent.cs` | 🔧 | 🔧 | ⏳ | ✅ |
+| `python/agent-framework` | `host_agent_server.py` | 🔧 | 🔧 | ⏳ | ❌ Blocked — `agent_framework` SDK API break (`ChatAgent` removed); pre-existing env issue |
+| `python/openai` | `host_agent_server.py` | 🔧 | 🔧 | ⏳ | ⏳ |
+| `python/claude` | `host_agent_server.py` | 🔧 | 🔧 | ⏳ | ⏳ |
+| `python/crewai` | `host_agent_server.py` | 🔧 | 🔧 | ⏳ | ⏳ |
+| `python/google-adk` | `hosting.py` | 🔧 | 🔧 | ⏳ | ⏳ |
+| `nodejs/openai` | `src/agent.ts` | 🔧 | 🔧 | ⏳ | ✅ |
+| `nodejs/claude` | `src/agent.ts` | 🔧 | 🔧 | ⏳ | ⏳ Needs ANTHROPIC_API_KEY |
+| `nodejs/langchain` | `src/agent.ts` | 🔧 | 🔧 | ⏳ | ⏳ |
+| `nodejs/langchain/quickstart-before` | `src/agent.ts` | 🔧 | 🔧 | ⏳ | ✅ |
+| `nodejs/vercel-sdk` | `src/agent.ts` | 🔧 | 🔧 | ⏳ | ⏳ Needs ANTHROPIC_API_KEY |
+| `nodejs/devin` | `src/agent.ts` | 🔧 | 🔧 | ⏳ | ⏳ Needs Devin credentials |
+| `nodejs/perplexity` | `src/agent.ts` | 🔧 | 🔧 | ⏳ | ⏳ Needs PERPLEXITY_API_KEY |
+| `nodejs/copilot-studio` | `src/agent.ts` | 🔧 | 🔧 | ⏳ | ⏳ |
+
+**Legend**: ✅ Done · 🔧 Modified locally, not committed · ⏳ Pending
+
+---
+
+## Notes
+
+- `python/google-adk` uses a different hosting pattern — `MyAgent(AgentApplication)` class in `hosting.py` rather than `host_agent_server.py`. Pattern applied to `message_handler` directly.
+- `nodejs/copilot-studio` was missing the `InstallationUpdate` handler in the constructor — added as part of this work.
+- `nodejs/langchain/quickstart-before` is a pre-refactor snapshot — fixed pre-existing TypeScript errors (`instructions` → `systemPrompt` for langchain 1.2.32+, added `@types/express`/`@types/node`).
+- Node.js samples use a manual `setInterval` typing loop (~4s) even though `startTypingTimer: true` is set in the constructor. The manual loop is necessary for long-running LLM calls that exceed the ~5s typing indicator timeout.
+- Python samples use `asyncio.create_task` for the typing loop since all aiohttp handlers run on the same event loop.
+- C# (`dotnet/semantic-kernel`) uses a single typing indicator (no loop) sent before agent initialization; the streaming informative update takes over as the progress indicator once the LLM call starts. Unlike agent-framework which runs a full `Task.Run` typing loop alongside streaming.

dotnet/agent-framework/sample-agent/Agent/MyAgent.cs

@@ -185,6 +185,11 @@ await AgentMetrics.InvokeObservedAgentOperation(
         /// <returns></returns>
         protected async Task OnMessageAsync(ITurnContext turnContext, ITurnState turnState, CancellationToken cancellationToken)
         {
+            if (turnContext is null)
+            {
+                throw new ArgumentNullException(nameof(turnContext));
+            }
+
             // Log the user identity from Activity.From — set by the A365 platform on every message.
             var fromAccount = turnContext.Activity.From;
             _logger?.LogDebug(
@@ -208,7 +213,6 @@ protected async Task OnMessageAsync(ITurnContext turnContext, ITurnState turnSta
                 ObservabilityAuthHandlerName = ToolAuthHandlerName = OboAuthHandlerName;
             }
 
-
             await A365OtelWrapper.InvokeObservedAgentOperation(
                 "MessageProcessor",
                 turnContext,
@@ -219,7 +223,33 @@ await A365OtelWrapper.InvokeObservedAgentOperation(
                 _logger,
                 async () =>
             {
-                // Start a Streaming Process to let clients that support streaming know that we are processing the request. 
+                // Send an immediate acknowledgment — this arrives as a separate message before the LLM response.
+                // Each SendActivityAsync call produces a discrete Teams message, enabling the multiple-messages pattern.
+                // NOTE: For Teams agentic identities, streaming is buffered into a single message by the SDK;
+                //       use SendActivityAsync for any messages that must arrive immediately.
+                await turnContext.SendActivityAsync(MessageFactory.Text("Got it — working on it…"), cancellationToken).ConfigureAwait(false);
+
+                // Send typing indicator immediately on the main thread (awaited so it arrives before the LLM call starts).
+                await turnContext.SendActivityAsync(Activity.CreateTypingActivity(), cancellationToken).ConfigureAwait(false);
+
+                // Background loop refreshes the "..." animation every ~4s (it times out after ~5s).
+                // Only visible in 1:1 and small group chats.
+                using var typingCts = CancellationTokenSource.CreateLinkedTokenSource(cancellationToken);
+                var typingTask = Task.Run(async () =>
+                {
+                    try
+                    {
+                        while (!typingCts.IsCancellationRequested)
+                        {
+                            await Task.Delay(TimeSpan.FromSeconds(4), typingCts.Token).ConfigureAwait(false);
+                            await turnContext.SendActivityAsync(Activity.CreateTypingActivity(), typingCts.Token).ConfigureAwait(false);
+                        }
+                    }
+                    catch (OperationCanceledException) { /* expected on cancel */ }
+                }, typingCts.Token);
+
+                // StreamingResponse is best-effort: in Teams with agentic identity the SDK may buffer/downscale it.
+                // The ack + typing loop above handle the immediate UX; streaming remains for non-Teams / WebChat clients.
                 await turnContext.StreamingResponse.QueueInformativeUpdateAsync("Just a moment please..").ConfigureAwait(false);
                 try
                 {
@@ -252,7 +282,16 @@ await A365OtelWrapper.InvokeObservedAgentOperation(
                 }
                 finally
                 {
-                    await turnContext.StreamingResponse.EndStreamAsync(cancellationToken).ConfigureAwait(false); // End the streaming response
+                    typingCts.Cancel();
+                    try
+                    {
+                        await typingTask.ConfigureAwait(false);
+                    }
+                    catch (OperationCanceledException)
+                    {
+                        // Expected: typingTask is canceled when typingCts is canceled; no further action required.
+                    }
+                    await turnContext.StreamingResponse.EndStreamAsync(cancellationToken).ConfigureAwait(false);
                 }
             });
         }

dotnet/agent-framework/sample-agent/README.md

@@ -66,6 +66,56 @@ The handler is registered twice in the constructor — once for agentic (A365 pr
 
 To test with Agents Playground, use **Mock an Activity → Install application** to send a simulated `installationUpdate` activity.
 
+## Sending Multiple Messages in Teams
+
+Agent365 agents can send multiple discrete messages in response to a single user prompt in Teams. This is achieved by calling `SendActivityAsync` multiple times within a single turn.
+
+> **Important**: Streaming responses are not supported for agentic identities in Teams. The SDK detects agentic identity and buffers the stream into a single message. Use `SendActivityAsync` directly to send immediate, discrete messages to the user.
+
+The sample demonstrates this in `OnMessageAsync` ([MyAgent.cs](Agent/MyAgent.cs)) by sending an immediate acknowledgment before the LLM response:
+
+```csharp
+// Message 1: immediate ack — reaches the user right away
+await turnContext.SendActivityAsync(MessageFactory.Text("Got it — working on it…"), cancellationToken);
+
+// ... LLM processing ...
+
+// Message 2: the LLM response (via StreamingResponse, buffered into one message for Teams agentic)
+await turnContext.StreamingResponse.EndStreamAsync(cancellationToken);
+```
+
+Each `SendActivityAsync` call produces a separate Teams message. You can call it as many times as needed to send progress updates, partial results, or a final answer.
+
+### Typing Indicators
+
+For long-running operations, send a typing indicator to show a "..." progress animation in Teams:
+
+```csharp
+// Typing indicator loop — refreshes every ~4s for long-running operations.
+using var typingCts = CancellationTokenSource.CreateLinkedTokenSource(cancellationToken);
+var typingTask = Task.Run(async () =>
+{
+    try
+    {
+        while (!typingCts.IsCancellationRequested)
+        {
+            await turnContext.SendActivityAsync(Activity.CreateTypingActivity(), typingCts.Token);
+            await Task.Delay(TimeSpan.FromSeconds(4), typingCts.Token);
+        }
+    }
+    catch (OperationCanceledException) { /* expected on cancel */ }
+}, typingCts.Token);
+
+try { /* ... do work ... */ }
+finally
+{
+    typingCts.Cancel();
+    try { await typingTask; } catch (OperationCanceledException) { }
+}
+```
+
+> **Note**: Typing indicators are only visible in 1:1 chats and small group chats — not in channels.
+
 ## Running the Agent
 
 To set up and test this agent, refer to the [Configure Agent Testing](https://learn.microsoft.com/en-us/microsoft-agent-365/developer/testing?tabs=dotnet) guide for complete instructions.