fix: correct telemetry response for Responses API output

src/platform/endpoint/node/responsesApi.ts

@@ -18,7 +18,8 @@ import { ConfigKey, IConfigurationService } from '../../configuration/common/con
 import { ILogService } from '../../log/common/logService';
 import { FinishedCallback, IResponseDelta, OpenAiResponsesFunctionTool } from '../../networking/common/fetch';
 import { IChatEndpoint, ICreateEndpointBodyOptions, IEndpointBody } from '../../networking/common/networking';
-import { ChatCompletion, FinishedCompletionReason, modelsWithoutResponsesContextManagement, openAIContextManagementCompactionType, OpenAIContextManagementResponse, TokenLogProb } from '../../networking/common/openai';
+import { ChatCompletion, FinishedCompletionReason, modelsWithoutResponsesContextManagement, openAIContextManagementCompactionType, OpenAIContextManagementResponse, TokenLogProb, rawMessageToCAPI } from '../../networking/common/openai';
+import { sendEngineMessagesTelemetry } from '../../networking/node/chatStream';
 import { IExperimentationService } from '../../telemetry/common/nullExperimentationService';
 import { ITelemetryService } from '../../telemetry/common/telemetry';
 import { TelemetryData } from '../../telemetry/common/telemetryData';
@@ -427,6 +428,17 @@ export async function processResponseFromChatEndpoint(instantiationService: IIns
 				logService.trace(`SSE: ${ev.data}`);
 				const completion = processor.push({ type: ev.type, ...JSON.parse(ev.data) }, finishCallback);
 				if (completion) {
+					const telemetryMessage = rawMessageToCAPI(completion.message);
+					let telemetryDataWithUsage = telemetryData;
+					if (completion.usage) {
+						telemetryDataWithUsage = telemetryData.extendedBy({}, {
+							promptTokens: completion.usage.prompt_tokens,
+							completionTokens: completion.usage.completion_tokens,
+							totalTokens: completion.usage.total_tokens,
+						});
+					}
+
+					sendEngineMessagesTelemetry(telemetryService, [telemetryMessage], telemetryDataWithUsage, true, logService);
 					feed.emitOne(completion);
 				}
 			} catch (e) {

src/platform/endpoint/node/test/responsesApi.spec.ts

@@ -6,7 +6,13 @@
 import { Raw } from '@vscode/prompt-tsx';
 import type { OpenAI } from 'openai';
 import { describe, expect, it } from 'vitest';
-import { responseApiInputToRawMessagesForLogging } from '../responsesApi';
+import { IInstantiationService } from '../../../../util/vs/platform/instantiation/common/instantiation';
+import { ILogService } from '../../../log/common/logService';
+import { TelemetryData } from '../../../telemetry/common/telemetryData';
+import { SpyingTelemetryService } from '../../../telemetry/node/spyingTelemetryService';
+import { createFakeStreamResponse } from '../../../test/node/fetcher';
+import { createPlatformServices } from '../../../test/node/services';
+import { processResponseFromChatEndpoint, responseApiInputToRawMessagesForLogging } from '../responsesApi';
 
 describe('responseApiInputToRawMessagesForLogging', () => {
 
@@ -207,3 +213,64 @@ describe('responseApiInputToRawMessagesForLogging', () => {
 		expect((result[0] as Raw.AssistantChatMessage).toolCalls).toHaveLength(2);
 	});
 });
+
+describe('processResponseFromChatEndpoint telemetry', () => {
+	it('emits engine.messages for Responses API assistant output', async () => {
+		const services = createPlatformServices();
+		const accessor = services.createTestingAccessor();
+		const instantiationService = accessor.get(IInstantiationService);
+		const logService = accessor.get(ILogService);
+		const telemetryService = new SpyingTelemetryService();
+
+		const completedEvent = {
+			type: 'response.completed',
+			response: {
+				id: 'resp_123',
+				model: 'gpt-5-mini',
+				created_at: 123,
+				usage: {
+					input_tokens: 11,
+					output_tokens: 7,
+					total_tokens: 18,
+					input_tokens_details: { cached_tokens: 0 },
+					output_tokens_details: { reasoning_tokens: 0 },
+				},
+				output: [
+					{
+						type: 'message',
+						content: [{ type: 'output_text', text: 'final assistant reply' }],
+					}
+				],
+			}
+		};
+
+		const response = createFakeStreamResponse(`data: ${JSON.stringify(completedEvent)}\n\n`);
+		const telemetryData = TelemetryData.createAndMarkAsIssued({ modelCallId: 'model-call-1' }, {});
+
+		const stream = await processResponseFromChatEndpoint(
+			instantiationService,
+			telemetryService,
+			logService,
+			response,
+			1,
+			async () => undefined,
+			telemetryData
+		);
+
+		for await (const _ of stream) {
+			// consume all completions to flush telemetry side effects
+		}
+
+		const events = telemetryService.getEvents().telemetryServiceEvents.filter(e => e.eventName === 'engine.messages');
+		expect(events.length).toBeGreaterThan(0);
+
+		const outputEvent = events[events.length - 1];
+		const messagesJson = JSON.parse(String((outputEvent.properties as Record<string, string>)?.messagesJson));
+		expect(messagesJson).toHaveLength(1);
+		expect(messagesJson[0].role).toBe('assistant');
+		expect(messagesJson[0].content).toBe('final assistant reply');
+
+		accessor.dispose();
+		services.dispose();
+	});
+});