Update .NET Copilot SDK cookbook for current SDK releases (#3548)

* Update .NET Copilot SDK cookbook: fix On<T> breaking change, add in-process runtime, tool search, and message provenance recipes

- Fix breaking change: Session.On is now generic (On<T> where T : SessionEvent).
  Updated all recipes that used the old non-generic session.On(evt => ...) pattern:
  error-handling, pr-visualization, managing-local-files, accessibility-report,
  ralph-loop (both .md and .cs where applicable).
- Add new 'In-Process Runtime' recipe (in-process-runtime.md + recipe/in-process-runtime.cs)
  demonstrating RuntimeConnection.ForInProcess(), live-tested.
- Document tool search / deferred loading (SessionConfig.ToolSearch / ToolSearchConfig)
  in pr-visualization.md, live-tested.
- Document typed message provenance (MessageOptions.Source / MessageSource) in
  error-handling.md, live-tested.
- Fix dotnet/README.md recipe index: it was missing Ralph Loop and Accessibility
  Report entries; also added the new In-Process Runtime entry.
- Add In-Process Runtime entry to the top-level cookbook/copilot-sdk/README.md
  .NET section.
- Update dotnet/recipe/README.md runnable-examples table with the new recipe.

All existing .NET cookbook recipes were rebuilt and rechecked against the current
GitHub Copilot SDK (Copilot CLI 1.0.85); no other breaking changes were found.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Update cookbook model names to current supported models

- Replace stale/nonexistent model IDs (gpt-5, claude-sonnet-4.5,
  claude-opus-4.6, gpt-5.1-codex-mini) with current values from the
  supported AI models list
- Prefer Model = "auto" wherever a specific model isn't load-bearing
  to the recipe's teaching point
- ralph-loop: pin gpt-5.3-codex (current codex-family GA model), since
  a dedicated coding model is a deliberate, documented choice for the
  autonomous loop
- multiple-sessions: reshape the model narrative so it demonstrates a
  real explicit-override use case (A/B testing via claude-sonnet-5)
  instead of two sessions sharing the same hardcoded model

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Use SendAndWaitAsync in in-process runtime recipe

Replace the event-based TaskCompletionSource pattern with SendAndWaitAsync so the example waits for the final assistant response, propagates session errors, and uses the SDK timeout behavior.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

cookbook/copilot-sdk/README.md

@@ -13,6 +13,7 @@ This cookbook collects small, focused recipes showing how to accomplish common t
 - [PR Visualization](dotnet/pr-visualization.md): Generate interactive PR age charts using GitHub MCP Server.
 - [Persisting Sessions](dotnet/persisting-sessions.md): Save and resume sessions across restarts.
 - [Accessibility Report](dotnet/accessibility-report.md): Generate WCAG accessibility reports using the Playwright MCP server.
+- [In-Process Runtime](dotnet/in-process-runtime.md): Run the Copilot runtime in-process instead of launching a separate CLI process.
 
 ### Node.js / TypeScript
 

cookbook/copilot-sdk/dotnet/README.md

@@ -9,6 +9,9 @@ This folder hosts short, practical recipes for using the GitHub Copilot SDK with
 - [Managing Local Files](managing-local-files.md): Organize files by metadata using AI-powered grouping strategies.
 - [PR Visualization](pr-visualization.md): Generate interactive PR age charts using GitHub MCP Server.
 - [Persisting Sessions](persisting-sessions.md): Save and resume sessions across restarts.
+- [Ralph Loop](ralph-loop.md): Run an unattended agentic loop that iterates on a task until it's done.
+- [Accessibility Report](accessibility-report.md): Audit a web page for accessibility issues and stream the findings live.
+- [In-Process Runtime](in-process-runtime.md): Run the Copilot runtime in-process instead of launching a separate CLI process.
 
 ## Contributing
 

cookbook/copilot-sdk/dotnet/accessibility-report.md

@@ -62,7 +62,7 @@ Console.WriteLine("Please wait...\n");
 // Create a session with Playwright MCP server
 await using var session = await client.CreateSessionAsync(new SessionConfig
 {
-    Model = "claude-opus-4.6",
+    Model = "auto",
     Streaming = true,
     OnPermissionRequest = PermissionHandler.ApproveAll,
     McpServers = new Dictionary<string, McpServerConfig>()
@@ -80,7 +80,7 @@ await using var session = await client.CreateSessionAsync(new SessionConfig
 // Wait for response using session.idle event
 var done = new TaskCompletionSource();
 
-session.On(evt =>
+session.On<SessionEvent>(evt =>
 {
     switch (evt)
     {
@@ -226,7 +226,7 @@ This gives the model access to Playwright browser tools like `browser_navigate`,
 Unlike `SendAndWaitAsync`, this recipe uses streaming for real-time output:
 
 ```csharp
-session.On(evt =>
+session.On<SessionEvent>(evt =>
 {
     switch (evt)
     {

cookbook/copilot-sdk/dotnet/error-handling.md

@@ -24,12 +24,12 @@ try
     await client.StartAsync();
     var session = await client.CreateSessionAsync(new SessionConfig
     {
-        Model = "gpt-5",
+        Model = "auto",
         OnPermissionRequest = PermissionHandler.ApproveAll
     });
 
     var done = new TaskCompletionSource<string>();
-    session.On(evt =>
+    session.On<SessionEvent>(evt =>
     {
         if (evt is AssistantMessageEvent msg)
         {
@@ -53,6 +53,13 @@ finally
 }
 ```
 
+> `Session.On` is now generic: `On<T>(Action<T> handler) where T : SessionEvent`. The type
+> argument can no longer be inferred from a lambda that only pattern-matches inside the body, so
+> calls like `session.On(evt => { if (evt is AssistantMessageEvent msg) ... })` fail to compile
+> with `CS0411`. Either specify `On<SessionEvent>` and pattern-match inside (as above), or
+> subscribe directly to the concrete event type you care about, e.g.
+> `session.On<AssistantMessageEvent>(evt => Console.WriteLine(evt.Data?.Content))`.
+
 ## Handling specific error types
 
 ```csharp
@@ -79,14 +86,14 @@ catch (Exception ex)
 ```csharp
 var session = await client.CreateSessionAsync(new SessionConfig
 {
-    Model = "gpt-5",
+    Model = "auto",
     OnPermissionRequest = PermissionHandler.ApproveAll
 });
 
 try
 {
     var done = new TaskCompletionSource<string>();
-    session.On(evt =>
+    session.On<SessionEvent>(evt =>
     {
         if (evt is AssistantMessageEvent msg)
         {
@@ -113,7 +120,7 @@ catch (OperationCanceledException)
 ```csharp
 var session = await client.CreateSessionAsync(new SessionConfig
 {
-    Model = "gpt-5",
+    Model = "auto",
     OnPermissionRequest = PermissionHandler.ApproveAll
 });
 
@@ -159,7 +166,7 @@ await client.StartAsync();
 
 var session = await client.CreateSessionAsync(new SessionConfig
 {
-    Model = "gpt-5",
+    Model = "auto",
     OnPermissionRequest = PermissionHandler.ApproveAll
 });
 
@@ -168,6 +175,46 @@ var session = await client.CreateSessionAsync(new SessionConfig
 // client.StopAsync() is automatically called when exiting scope
 ```
 
+## Tagging message provenance
+
+Since v1.0.14, `MessageOptions` has a `Source` property so you can tag *why* a message was sent —
+useful when errors or unexpected turns show up in transcripts and you need to tell human input
+apart from messages an automated system (a webhook handler, a scheduled job, another agent) sent
+on the user's behalf.
+
+```csharp
+using GitHub.Copilot;
+
+// Ordinary human input (also the default when Source is omitted).
+await session.SendAsync(new MessageOptions
+{
+    Prompt = "What changed in the last release?",
+    Source = MessageSource.User
+});
+
+// A message injected by your own system rather than typed by a person,
+// e.g. a scheduled health check or automated retry.
+await session.SendAsync(new MessageOptions
+{
+    Prompt = "Re-run the failed step and report the result.",
+    Source = MessageSource.System
+});
+
+// A message sent by another agent or automation acting on the user's behalf,
+// tagged with a caller-supplied identifier.
+await session.SendAsync(new MessageOptions
+{
+    Prompt = "Summarize the open incidents.",
+    Source = MessageSource.Agent("incident-bot")
+});
+```
+
+`MessageSource` is a closed record with `User`, `System`, and a `MessageSource.Agent(string id)`
+factory (serialized as `"agent-<id>"`). Tagging a source only records provenance on the message —
+it does not change how the message is delivered, replace the session's system prompt, or grant
+extra permissions. When `Source` is omitted, the field is left unset and the runtime treats the
+message as ordinary user input.
+
 ## Best practices
 
 Permission handling is opt-in. If a session may need tool, file, or system access, set `OnPermissionRequest` explicitly when creating it.

cookbook/copilot-sdk/dotnet/in-process-runtime.md

@@ -0,0 +1,88 @@
+# In-Process Runtime
+
+Run the Copilot runtime inside your own process instead of launching a separate Copilot CLI process.
+
+> **Runnable example:** [recipe/in-process-runtime.cs](recipe/in-process-runtime.cs)
+>
+> ```bash
+> dotnet run recipe/in-process-runtime.cs
+> ```
+
+## Example scenario
+
+By default, `CopilotClient` launches and manages a separate Copilot CLI child process and talks to
+it over stdio or TCP. Some deployments — for example, tightly sandboxed hosts, single-binary
+services, or environments where spawning child processes is restricted — need the runtime loaded
+directly into the application process instead.
+
+## Using the in-process connection
+
+```csharp
+using GitHub.Copilot;
+
+// RuntimeConnection.ForInProcess() is an experimental API (diagnostic GHCP001).
+#pragma warning disable GHCP001
+
+var client = new CopilotClient(new CopilotClientOptions
+{
+    Connection = RuntimeConnection.ForInProcess()
+});
+
+await client.StartAsync();
+
+var session = await client.CreateSessionAsync(new SessionConfig
+{
+    Model = "auto",
+    OnPermissionRequest = PermissionHandler.ApproveAll
+});
+
+var response = await session.SendAndWaitAsync(
+    new MessageOptions { Prompt = "Hello from the in-process runtime!" });
+Console.WriteLine(response?.Data.Content);
+
+await client.StopAsync();
+```
+
+`RuntimeConnection.ForInProcess()` is marked `[Experimental("GHCP001")]`, so any code that calls it
+must either suppress the diagnostic with `#pragma warning disable GHCP001` (as above) or opt in at
+the project level. Everything past client construction — sessions, streaming events, tools, hooks,
+and permissions — behaves the same as with the default CLI-launching connection.
+
+## How it works
+
+Instead of spawning a child process, the SDK loads the native Copilot runtime library directly into
+your application and communicates with it over an in-memory connection using the same
+`Content-Length`-framed JSON-RPC protocol the CLI transport uses. The runtime can invoke SDK
+callbacks from native worker threads; the SDK handles marshalling those calls back onto managed
+threads for you.
+
+You can also select the in-process transport without changing application code by setting
+`COPILOT_SDK_DEFAULT_CONNECTION=inprocess` before startup — the SDK only falls back to this
+environment variable when the client doesn't specify a connection explicitly.
+
+## Limitations
+
+The in-process transport is experimental and comes with a few constraints to be aware of:
+
+- **Shared process state**: every in-process client shares the host process's environment,
+  current working directory, and native runtime library — there's no per-client working directory
+  or environment override.
+- **Rejected process options**: `CopilotClientOptions.Environment`, telemetry configuration, and
+  similar options that assume a separate child process are not supported and throw if set.
+- **One runtime version per process**: once a native runtime library is loaded, starting another
+  client with a different runtime library path or version fails. Starting additional clients that
+  reuse the same already-loaded library is fine.
+- **Persistent load**: the native library stays loaded for the lifetime of the process even after
+  `StopAsync()` gracefully shuts down the client's sessions and connection — don't rely on being
+  able to unload and swap in a different runtime build later.
+
+## Best practices
+
+1. **Test per platform**: validate startup, model turns, and shutdown on every OS/architecture
+   combination you deploy, since runtime library support varies.
+2. **Prefer the default CLI transport** unless you specifically need to avoid a child process —
+   it's the most established and isolated deployment path.
+3. **Still call `StopAsync()`**: graceful shutdown still closes sessions and the JSON-RPC
+   connection cleanly, even though the native library itself remains loaded.
+4. **Set process-wide values early**: configure environment variables and working directory before
+   creating the first in-process client, since later clients can't override them per-instance.

cookbook/copilot-sdk/dotnet/managing-local-files.md

@@ -25,14 +25,14 @@ await client.StartAsync();
 // Define tools for file operations
 var session = await client.CreateSessionAsync(new SessionConfig
 {
-    Model = "gpt-5",
+    Model = "auto",
     OnPermissionRequest = PermissionHandler.ApproveAll
 });
 
 // Wait for completion
 var done = new TaskCompletionSource();
 
-session.On(evt =>
+session.On<SessionEvent>(evt =>
 {
     switch (evt)
     {

cookbook/copilot-sdk/dotnet/multiple-sessions.md

@@ -20,20 +20,22 @@ using GitHub.Copilot;
 await using var client = new CopilotClient();
 await client.StartAsync();
 
-// Create multiple independent sessions
+// Create multiple independent sessions. Most sessions should let Copilot pick the
+// best model automatically; pin an explicit model only when you have a deliberate
+// reason to (e.g. an A/B comparison, as with session3 here).
 var session1 = await client.CreateSessionAsync(new SessionConfig
 {
-    Model = "gpt-5",
+    Model = "auto",
     OnPermissionRequest = PermissionHandler.ApproveAll
 });
 var session2 = await client.CreateSessionAsync(new SessionConfig
 {
-    Model = "gpt-5",
+    Model = "auto",
     OnPermissionRequest = PermissionHandler.ApproveAll
 });
 var session3 = await client.CreateSessionAsync(new SessionConfig
 {
-    Model = "claude-sonnet-4.5",
+    Model = "claude-sonnet-5",
     OnPermissionRequest = PermissionHandler.ApproveAll
 });
 
@@ -61,7 +63,7 @@ Use custom IDs for easier tracking:
 var session = await client.CreateSessionAsync(new SessionConfig
 {
     SessionId = "user-123-chat",
-    Model = "gpt-5",
+    Model = "auto",
     OnPermissionRequest = PermissionHandler.ApproveAll
 });
 
@@ -89,4 +91,5 @@ await client.DeleteSessionAsync("user-123-chat");
 
 - **Multi-user applications**: One session per user
 - **Multi-task workflows**: Separate sessions for different tasks
-- **A/B testing**: Compare responses from different models
+- **A/B testing**: Compare responses from different models by pinning an explicit
+  `Model` per session (as `session3` does above) instead of the default `"auto"`

cookbook/copilot-sdk/dotnet/persisting-sessions.md

@@ -25,7 +25,7 @@ await client.StartAsync();
 var session = await client.CreateSessionAsync(new SessionConfig
 {
     SessionId = "user-123-conversation",
-    Model = "gpt-5",
+    Model = "auto",
     OnPermissionRequest = PermissionHandler.ApproveAll
 });
 

cookbook/copilot-sdk/dotnet/pr-visualization.md

@@ -164,7 +164,7 @@ await client.StartAsync();
 
 var session = await client.CreateSessionAsync(new SessionConfig
 {
-    Model = "gpt-5",
+    Model = "auto",
     OnPermissionRequest = PermissionHandler.ApproveAll,
     SystemMessage = new SystemMessageConfig
     {
@@ -185,7 +185,7 @@ The current working directory is: {Environment.CurrentDirectory}
 });
 
 // Set up event handling
-session.On(evt =>
+session.On<SessionEvent>(evt =>
 {
     switch (evt)
     {
@@ -256,3 +256,37 @@ while (true)
 | Flexibility     | Fixed logic       | **AI decides best approach**      |
 | Chart types     | What you coded    | **Any type Copilot can generate** |
 | Data grouping   | Hardcoded buckets | **Intelligent grouping**          |
+
+## Deferring tools with tool search
+
+The GitHub MCP Server alone exposes dozens of tools, and this recipe also has the file and code
+execution tools available — well past the point where stuffing every tool description into the
+model's context on every turn helps more than it costs. `SessionConfig.ToolSearch` controls when
+the SDK defers less-frequently-needed tools behind a search step instead of listing them all
+up front:
+
+```csharp
+var session = await client.CreateSessionAsync(new SessionConfig
+{
+    Model = "auto",
+    OnPermissionRequest = PermissionHandler.ApproveAll,
+    ToolSearch = new ToolSearchConfig
+    {
+        Enabled = true,
+        DeferThreshold = 20
+    },
+    SystemMessage = new SystemMessageConfig { Content = "..." }
+});
+```
+
+- `Enabled` turns tool search on or off explicitly; leave it `null` to use the runtime default.
+- `DeferThreshold` is the tool count above which MCP/external tools are deferred behind a
+  `tool_search_tool` call instead of being listed directly (the runtime default is 30 when
+  unset). Lowering it — as above — is useful once a recipe pulls in a large MCP server like the
+  GitHub MCP Server alongside file and code-execution tools, since it keeps the per-turn tool
+  list small while still letting Copilot search for the exact tool it needs.
+- To customize how the search itself behaves, register your own tool named `"tool_search_tool"`
+  with `OverridesBuiltInTool = true` to replace the built-in implementation.
+
+This is a per-session setting, not a global one, so you can tune it independently for
+tool-heavy recipes like this one versus lighter sessions elsewhere in your app.

cookbook/copilot-sdk/dotnet/ralph-loop.md

@@ -60,13 +60,13 @@ try
         var session = await client.CreateSessionAsync(
             new SessionConfig
             {
-                Model = "gpt-5.1-codex-mini",
+                Model = "gpt-5.3-codex",
                 OnPermissionRequest = PermissionHandler.ApproveAll
             });
         try
         {
             var done = new TaskCompletionSource<string>();
-            session.On(evt =>
+            session.On<SessionEvent>(evt =>
             {
                 if (evt is AssistantMessageEvent msg)
                     done.TrySetResult(msg.Data.Content);
@@ -125,7 +125,7 @@ try
         var session = await client.CreateSessionAsync(
             new SessionConfig
             {
-                Model = "gpt-5.1-codex-mini",
+                Model = "gpt-5.3-codex",
                 // Pin the agent to the project directory
                 WorkingDirectory = Environment.CurrentDirectory,
                 // Auto-approve tool calls for unattended operation
@@ -134,7 +134,7 @@ try
         try
         {
             var done = new TaskCompletionSource<string>();
-            session.On(evt =>
+            session.On<SessionEvent>(evt =>
             {
                 // Log tool usage for visibility
                 if (evt is ToolExecutionStartEvent toolStart)

cookbook/copilot-sdk/dotnet/recipe/README.md

@@ -26,6 +26,7 @@ dotnet run <filename>.cs
 | Persisting Sessions  | `dotnet run persisting-sessions.cs`  | Save and resume sessions across restarts   |
 | Accessibility Report ℹ️ | `dotnet run accessibility-report.cs` | Analyzes web page accessibility            |
 | Ralph Loop ⚠️         | `dotnet run ralph-loop.cs`           | Autonomous development loop                |
+| In-Process Runtime   | `dotnet run in-process-runtime.cs`  | Runs the Copilot runtime in-process         |
 
 ### Examples with Arguments
 

cookbook/copilot-sdk/dotnet/recipe/accessibility-report.cs

@@ -31,7 +31,7 @@ Console.WriteLine("Please wait...\n");
 // Create a session with Playwright MCP server
 await using var session = await client.CreateSessionAsync(new SessionConfig
 {
-  Model = "claude-opus-4.6",
+  Model = "auto",
   Streaming = true,
   OnPermissionRequest = PermissionHandler.ApproveAll,
   McpServers = new Dictionary<string, McpServerConfig>()
@@ -49,7 +49,7 @@ await using var session = await client.CreateSessionAsync(new SessionConfig
 // Wait for response using session.idle event
 var done = new TaskCompletionSource();
 
-session.On(evt =>
+session.On<SessionEvent>(evt =>
 {
   switch (evt)
   {

cookbook/copilot-sdk/dotnet/recipe/error-handling.cs

@@ -11,12 +11,12 @@ try
     await client.StartAsync();
     var session = await client.CreateSessionAsync(new SessionConfig
     {
-        Model = "gpt-5",
+        Model = "auto",
         OnPermissionRequest = PermissionHandler.ApproveAll
     });
 
     var done = new TaskCompletionSource<string>();
-    session.On(evt =>
+    session.On<SessionEvent>(evt =>
     {
         if (evt is AssistantMessageEvent msg)
         {

cookbook/copilot-sdk/dotnet/recipe/in-process-runtime.cs

@@ -0,0 +1,38 @@
+#:package GitHub.Copilot.SDK@*
+#:property PublishAot=false
+
+// The GitHub.Copilot.SDK package exposes the GitHub.Copilot namespace.
+using GitHub.Copilot;
+
+// RuntimeConnection.ForInProcess() is an experimental API (diagnostic GHCP001):
+// it loads the native Copilot runtime directly into this process instead of
+// launching a separate Copilot CLI process.
+#pragma warning disable GHCP001
+
+var client = new CopilotClient(new CopilotClientOptions
+{
+    Connection = RuntimeConnection.ForInProcess()
+});
+
+try
+{
+    await client.StartAsync();
+
+    var session = await client.CreateSessionAsync(new SessionConfig
+    {
+        Model = "auto",
+        OnPermissionRequest = PermissionHandler.ApproveAll
+    });
+
+    var response = await session.SendAndWaitAsync(
+        new MessageOptions { Prompt = "Hello from the in-process runtime!" });
+    Console.WriteLine(response?.Data.Content);
+
+    await session.DisposeAsync();
+}
+finally
+{
+    // Gracefully stop; the native runtime library itself stays loaded for the
+    // lifetime of the process (see the "Lifecycle behavior" notes in the docs).
+    await client.StopAsync();
+}

cookbook/copilot-sdk/dotnet/recipe/managing-local-files.cs

@@ -11,14 +11,14 @@ await client.StartAsync();
 // Define tools for file operations
 var session = await client.CreateSessionAsync(new SessionConfig
 {
-    Model = "gpt-5",
+    Model = "auto",
     OnPermissionRequest = PermissionHandler.ApproveAll
 });
 
 // Wait for completion
 var done = new TaskCompletionSource();
 
-session.On(evt =>
+session.On<SessionEvent>(evt =>
 {
     switch (evt)
     {

cookbook/copilot-sdk/dotnet/recipe/multiple-sessions.cs

@@ -7,20 +7,22 @@ using GitHub.Copilot;
 await using var client = new CopilotClient();
 await client.StartAsync();
 
-// Create multiple independent sessions
+// Create multiple independent sessions. Most sessions should let Copilot pick the
+// best model automatically; pin an explicit model only when you have a deliberate
+// reason to (e.g. an A/B comparison, as with session3 here).
 var session1 = await client.CreateSessionAsync(new SessionConfig
 {
-    Model = "gpt-5",
+    Model = "auto",
     OnPermissionRequest = PermissionHandler.ApproveAll
 });
 var session2 = await client.CreateSessionAsync(new SessionConfig
 {
-    Model = "gpt-5",
+    Model = "auto",
     OnPermissionRequest = PermissionHandler.ApproveAll
 });
 var session3 = await client.CreateSessionAsync(new SessionConfig
 {
-    Model = "claude-sonnet-4.5",
+    Model = "claude-sonnet-5",
     OnPermissionRequest = PermissionHandler.ApproveAll
 });
 

cookbook/copilot-sdk/dotnet/recipe/persisting-sessions.cs

@@ -11,7 +11,7 @@ await client.StartAsync();
 var session = await client.CreateSessionAsync(new SessionConfig
 {
     SessionId = "user-123-conversation",
-    Model = "gpt-5",
+    Model = "auto",
     OnPermissionRequest = PermissionHandler.ApproveAll
 });
 

cookbook/copilot-sdk/dotnet/recipe/pr-visualization.cs

@@ -132,7 +132,7 @@ await client.StartAsync();
 
 var session = await client.CreateSessionAsync(new SessionConfig
 {
-    Model = "gpt-5",
+    Model = "auto",
     OnPermissionRequest = PermissionHandler.ApproveAll,
     SystemMessage = new SystemMessageConfig
     {

cookbook/copilot-sdk/dotnet/recipe/ralph-loop.cs

@@ -45,7 +45,7 @@ try
         var session = await client.CreateSessionAsync(
             new SessionConfig
             {
-                Model = "gpt-5.1-codex-mini",
+                Model = "gpt-5.3-codex",
                 // Pin the agent to the project directory
                 WorkingDirectory = Environment.CurrentDirectory,
                 // Auto-approve tool calls for unattended operation
@@ -55,7 +55,7 @@ try
         try
         {
             var done = new TaskCompletionSource<string>();
-            session.On(evt =>
+            session.On<SessionEvent>(evt =>
             {
                 // Log tool usage for visibility
                 if (evt is ToolExecutionStartEvent toolStart)