Fix remaining Medium issues from second subagent review pass

- strictModeEnabled: false for MCP tool registration — external MCP
  schemas are not guaranteed to satisfy OpenAI strict-mode constraints
  (all properties required, additionalProperties: false everywhere).
  A single non-conforming schema with strictModeEnabled: true would
  cause a 400 at registration time for ALL tools in the request.

- Guard ParseToolArguments in non-streaming path — JsonException on
  malformed model-generated arguments would previously bubble up as
  a 500. Now caught identically to the streaming path: logs warning
  and sends an error FunctionCallOutputResponseItem so the model can
  recover without aborting the loop.

- Guard currentLegResponseId null before tool-call continuation —
  if the API never emits StreamingResponseCreatedUpdate, proceeding
  with PreviousResponseId = null causes a confusing 400 from the API.
  Now throws InvalidOperationException with a diagnostic message.

- Log original ConversationContextLimitExceededException in mid-stream
  handler — previously wrapped in a new InvalidOperationException,
  losing the stack trace, PreviousResponseId, and inner exception.

Author: BenjaminMichaelis

EssentialCSharp.Chat.Shared/Services/AIChatService.cs

@@ -232,6 +232,13 @@ private static string SanitizeForXmlContext(string? input) =>
         // GetChatCompletionCore and avoids conversation branching.
         if (pendingFunctionCalls is { Count: > 0 } && mcpClient != null)
         {
+            // Guard: if the API never sent StreamingResponseCreatedUpdate, currentLegResponseId
+            // is null. Continuing would send FunctionCallOutputResponseItems referencing CallIds
+            // the server has no record of (PreviousResponseId = null), causing a 400.
+            if (currentLegResponseId is null)
+                throw new InvalidOperationException(
+                    "Cannot continue tool-call chain: the streaming leg completed with tool calls but emitted no response ID.");
+
             var continuationOptions = CloneOptionsWithPreviousResponseId(responseOptions, currentLegResponseId);
             var outputItems = new List<ResponseItem>(pendingFunctionCalls.Count);
 
@@ -387,7 +394,11 @@ private async Task<ResponseCreationOptions> CreateResponseOptionsAsync(
                 }
 
 #pragma warning disable OPENAI001 // Type is for evaluation purposes only and is subject to change or removal in future updates. Suppress this diagnostic to proceed.
-                options.Tools.Add(ResponseTool.CreateFunctionTool(tool.Name, functionDescription: tool.Description, strictModeEnabled: true, functionParameters: BinaryData.FromString(tool.JsonSchema.GetRawText())));
+                // strictModeEnabled: false — MCP tool schemas come from external servers and are not
+                // guaranteed to satisfy OpenAI strict-mode constraints (all properties required,
+                // additionalProperties: false everywhere). A single non-conforming schema with
+                // strict mode enabled would cause a 400 at registration time for ALL tools.
+                options.Tools.Add(ResponseTool.CreateFunctionTool(tool.Name, functionDescription: tool.Description, strictModeEnabled: false, functionParameters: BinaryData.FromString(tool.JsonSchema.GetRawText())));
 #pragma warning restore OPENAI001 // Type is for evaluation purposes only and is subject to change or removal in future updates. Suppress this diagnostic to proceed.
             }
         }
@@ -466,7 +477,20 @@ private async Task<ResponseCreationOptions> CreateResponseOptionsAsync(
                     }
 
                     LogMcpToolCallInvoked(_Logger, functionCallItem.FunctionName, iteration, endUserId);
-                    var arguments = ParseToolArguments(functionCallItem.FunctionArguments);
+
+                    Dictionary<string, object?> arguments;
+                    try
+                    {
+                        arguments = ParseToolArguments(functionCallItem.FunctionArguments);
+                    }
+                    catch (Exception ex)
+                    {
+                        LogMcpToolArgumentParseError(_Logger, functionCallItem.FunctionName, ex, endUserId);
+                        responseItems.Add(new FunctionCallOutputResponseItem(
+                            functionCallItem.CallId,
+                            $"Error parsing arguments for '{functionCallItem.FunctionName}': invalid JSON."));
+                        continue;
+                    }
 
                     var toolResult = await mcpClient.CallToolAsync(
                         functionCallItem.FunctionName,

EssentialCSharp.Web/Controllers/ChatController.cs

@@ -143,9 +143,9 @@ public async Task StreamMessage([FromBody] ChatMessageRequest request, Cancellat
             Response.ContentType = "application/json";
             await Response.WriteAsJsonAsync(new { error = "This conversation has grown too long. Please start a new one.", errorCode = "context_limit_exceeded" }, CancellationToken.None);
         }
-        catch (ConversationContextLimitExceededException)
+        catch (ConversationContextLimitExceededException ex)
         {
-            LogChatStreamErrorMidStream(_Logger, new InvalidOperationException("Context limit exceeded mid-stream"), User.Identity?.Name);
+            LogChatStreamErrorMidStream(_Logger, ex, User.Identity?.Name);
             try
             {
                 await Response.WriteAsync("data: {\"type\":\"error\",\"message\":\"This conversation has grown too long. Please start a new one.\",\"errorCode\":\"context_limit_exceeded\"}\n\n", CancellationToken.None);