feat(cursor): harden daemon integration and model discovery

tests/unit/cursor/cursor-daemon.test.ts

@@ -153,6 +153,13 @@ describe('startDaemon', () => {
       const running = await isDaemonRunning(port);
       expect(running).toBe(true);
 
+      // Verify models endpoint exists and is OpenAI-compatible list shape
+      const modelsResponse = await fetch(`http://127.0.0.1:${port}/v1/models`);
+      expect(modelsResponse.status).toBe(200);
+      const modelsJson = (await modelsResponse.json()) as { object?: string; data?: unknown[] };
+      expect(modelsJson.object).toBe('list');
+      expect(Array.isArray(modelsJson.data)).toBe(true);
+
       // Verify chat endpoint exists (requires auth, should not be 404)
       const chatResponse = await fetch(`http://127.0.0.1:${port}/v1/chat/completions`, {
         method: 'POST',

tests/unit/cursor/cursor-models.test.ts

@@ -2,7 +2,8 @@
  * Unit tests for Cursor models module
  */
 
-import { describe, it, expect } from 'bun:test';
+import { describe, it, expect, beforeEach } from 'bun:test';
+import * as http from 'http';
 import {
   DEFAULT_CURSOR_MODELS,
   DEFAULT_CURSOR_PORT,
@@ -11,6 +12,9 @@ import {
   detectProvider,
   formatModelName,
   fetchModelsFromDaemon,
+  fetchModelsFromCursorApi,
+  getModelsForDaemon,
+  clearCursorModelsCache,
 } from '../../../src/cursor/cursor-models';
 
 describe('DEFAULT_CURSOR_MODELS', () => {
@@ -102,3 +106,127 @@ describe('fetchModelsFromDaemon', () => {
     expect(models).toEqual(DEFAULT_CURSOR_MODELS);
   });
 });
+
+describe('fetchModelsFromCursorApi', () => {
+  it('parses model list from API response', async () => {
+    const server = http.createServer((_req, res) => {
+      res.writeHead(200, { 'Content-Type': 'application/json' });
+      res.end(
+        JSON.stringify({
+          data: [
+            { id: 'gpt-5.3-codex', name: 'GPT-5.3 Codex', provider: 'openai' },
+            { id: 'claude-4.6-opus', name: 'Claude 4.6 Opus', provider: 'anthropic' },
+          ],
+        })
+      );
+    });
+
+    await new Promise<void>((resolve) => server.listen(0, '127.0.0.1', resolve));
+    const address = server.address();
+    if (!address || typeof address === 'string') {
+      throw new Error('Unable to resolve test server port');
+    }
+
+    try {
+      const models = await fetchModelsFromCursorApi(
+        {
+          accessToken: 'test-token-123',
+          machineId: '1234567890abcdef1234567890abcdef',
+        },
+        {
+          endpoint: `http://127.0.0.1:${address.port}/v1/models`,
+          timeoutMs: 2000,
+        }
+      );
+
+      expect(models).not.toBeNull();
+      expect(models?.[0].id).toBe('gpt-5.3-codex');
+      expect(models?.[1].id).toBe('claude-4.6-opus');
+    } finally {
+      await new Promise<void>((resolve) => server.close(() => resolve()));
+    }
+  });
+
+  it('returns null for non-200 responses', async () => {
+    const server = http.createServer((_req, res) => {
+      res.writeHead(403, { 'Content-Type': 'application/json' });
+      res.end(JSON.stringify({ error: 'forbidden' }));
+    });
+
+    await new Promise<void>((resolve) => server.listen(0, '127.0.0.1', resolve));
+    const address = server.address();
+    if (!address || typeof address === 'string') {
+      throw new Error('Unable to resolve test server port');
+    }
+
+    try {
+      const models = await fetchModelsFromCursorApi(
+        {
+          accessToken: 'test-token-123',
+          machineId: '1234567890abcdef1234567890abcdef',
+        },
+        {
+          endpoint: `http://127.0.0.1:${address.port}/v1/models`,
+          timeoutMs: 2000,
+        }
+      );
+
+      expect(models).toBeNull();
+    } finally {
+      await new Promise<void>((resolve) => server.close(() => resolve()));
+    }
+  });
+});
+
+describe('getModelsForDaemon', () => {
+  beforeEach(() => {
+    clearCursorModelsCache();
+  });
+
+  it('falls back to defaults without credentials', async () => {
+    const models = await getModelsForDaemon();
+    expect(models).toEqual(DEFAULT_CURSOR_MODELS);
+  });
+
+  it('uses cached live models when endpoint becomes unavailable', async () => {
+    const liveModelId = 'test-live-model';
+    const server = http.createServer((_req, res) => {
+      res.writeHead(200, { 'Content-Type': 'application/json' });
+      res.end(
+        JSON.stringify({
+          data: [{ id: liveModelId, name: 'Live Model', provider: 'openai' }],
+        })
+      );
+    });
+
+    await new Promise<void>((resolve) => server.listen(0, '127.0.0.1', resolve));
+    const address = server.address();
+    if (!address || typeof address === 'string') {
+      throw new Error('Unable to resolve test server port');
+    }
+
+    const endpoint = `http://127.0.0.1:${address.port}/v1/models`;
+
+    try {
+      const first = await getModelsForDaemon({
+        credentials: {
+          accessToken: 'test-token-123',
+          machineId: '1234567890abcdef1234567890abcdef',
+        },
+        endpoint,
+        timeoutMs: 2000,
+      });
+
+      expect(first[0]?.id).toBe(liveModelId);
+    } finally {
+      await new Promise<void>((resolve) => server.close(() => resolve()));
+    }
+
+    const second = await getModelsForDaemon({
+      endpoint: 'http://127.0.0.1:9/v1/models',
+      timeoutMs: 250,
+    });
+
+    expect(second[0]?.id).toBe(liveModelId);
+  });
+});

tests/unit/cursor/cursor-protobuf.test.ts

@@ -531,6 +531,28 @@ describe('CursorExecutor', () => {
       const body = JSON.parse(bodyText);
       expect(body.error.type).toBe('rate_limit_error');
     });
+
+    it('should surface reasoning_content when thinking payload is present', async () => {
+      const textContent = 'Final answer';
+      const thinkingContent = 'Internal reasoning trail';
+      const thinkingField = encodeField(FIELD.Thinking.TEXT, WIRE_TYPE.LEN, thinkingContent);
+      const chatResponse = concatArrays(
+        encodeField(FIELD.ChatResponse.TEXT, WIRE_TYPE.LEN, textContent),
+        encodeField(FIELD.ChatResponse.THINKING, WIRE_TYPE.LEN, thinkingField)
+      );
+      const responseMsg = encodeField(FIELD.Response.RESPONSE, WIRE_TYPE.LEN, chatResponse);
+      const frame = wrapConnectRPCFrame(responseMsg, false);
+
+      const result = executor.transformProtobufToJSON(Buffer.from(frame), 'gpt-4', {
+        messages: [],
+      });
+
+      expect(result.status).toBe(200);
+      const bodyText = await result.text();
+      const body = JSON.parse(bodyText);
+      expect(body.choices[0].message.content).toBe(textContent);
+      expect(body.choices[0].message.reasoning_content).toBe(thinkingContent);
+    });
   });
 
   describe('transformProtobufToSSE', () => {
@@ -572,6 +594,23 @@ describe('CursorExecutor', () => {
       const body = JSON.parse(bodyText);
       expect(body.error.type).toBe('rate_limit_error');
     });
+
+    it('should emit reasoning_content deltas for thinking payloads', async () => {
+      const thinkingContent = 'Deliberate reasoning';
+      const thinkingField = encodeField(FIELD.Thinking.TEXT, WIRE_TYPE.LEN, thinkingContent);
+      const chatResponse = encodeField(FIELD.ChatResponse.THINKING, WIRE_TYPE.LEN, thinkingField);
+      const responseMsg = encodeField(FIELD.Response.RESPONSE, WIRE_TYPE.LEN, chatResponse);
+      const frame = wrapConnectRPCFrame(responseMsg, false);
+
+      const result = executor.transformProtobufToSSE(Buffer.from(frame), 'gpt-4', {
+        messages: [],
+      });
+
+      expect(result.status).toBe(200);
+      const bodyText = await result.text();
+      expect(bodyText).toContain('reasoning_content');
+      expect(bodyText).toContain(thinkingContent);
+    });
   });
 
   describe('decompressPayload error handling', () => {
@@ -676,6 +715,16 @@ function buildTextFrame(text: string): Buffer {
   return buildFrame(responseMsg);
 }
 
+/**
+ * Helper: build a protobuf thinking response frame
+ */
+function buildThinkingFrame(thinking: string): Buffer {
+  const thinkingField = encodeField(FIELD.Thinking.TEXT, WIRE_TYPE.LEN, thinking);
+  const responseField = encodeField(FIELD.ChatResponse.THINKING, WIRE_TYPE.LEN, thinkingField);
+  const responseMsg = encodeField(FIELD.Response.RESPONSE, WIRE_TYPE.LEN, responseField);
+  return buildFrame(responseMsg);
+}
+
 describe('StreamingFrameParser', () => {
   it('should parse a complete single frame', () => {
     const parser = new StreamingFrameParser();
@@ -784,6 +833,18 @@ describe('StreamingFrameParser', () => {
     }
   });
 
+  it('should parse thinking frames', () => {
+    const parser = new StreamingFrameParser();
+    const frame = buildThinkingFrame('Think step by step');
+    const results = parser.push(frame);
+
+    expect(results.length).toBe(1);
+    expect(results[0].type).toBe('thinking');
+    if (results[0].type === 'thinking') {
+      expect(results[0].text).toBe('Think step by step');
+    }
+  });
+
   it('should report hasPartial() correctly', () => {
     const parser = new StreamingFrameParser();
     expect(parser.hasPartial()).toBe(false);