@google/gemini-cli-core 0.6.0-nightly.20250910.a31830a3 → 0.6.0-preview.0

package/dist/src/core/client.test.js

@@ -4,6 +4,7 @@
  * SPDX-License-Identifier: Apache-2.0
  */
 import { describe, it, expect, vi, beforeEach, afterEach, } from 'vitest';
+import { createUserContent } from '@google/genai';
 import { findIndexAfterFraction, isThinkingDefault, isThinkingSupported, GeminiClient, } from './client.js';
 import { AuthType, } from './contentGenerator.js';
 import {} from './geminiChat.js';
@@ -13,7 +14,7 @@ import { DEFAULT_GEMINI_FLASH_MODEL } from '../config/models.js';
 import { FileDiscoveryService } from '../services/fileDiscoveryService.js';
 import { setSimulate429 } from '../utils/testUtils.js';
 import { tokenLimit } from './tokenLimits.js';
-import { ideContext } from '../ide/ideContext.js';
+import { ideContextStore } from '../ide/ideContext.js';
 import { ClearcutLogger } from '../telemetry/clearcut-logger/clearcut-logger.js';
 // Mock fs module to prevent actual file system operations during tests
 const mockFileSystem = new Map();
@@ -108,22 +109,22 @@ describe('findIndexAfterFraction', () => {
         // 0: 66
         // 1: 66 + 68 = 134
         // 2: 134 + 66 = 200
-        // 200 >= 166.5, so index is 2
-        expect(findIndexAfterFraction(history, 0.5)).toBe(2);
+        // 200 >= 166.5, so index is 3
+        expect(findIndexAfterFraction(history, 0.5)).toBe(3);
     });
     it('should handle a fraction that results in the last index', () => {
         // 333 * 0.9 = 299.7
         // ...
         // 3: 200 + 68 = 268
         // 4: 268 + 65 = 333
-        // 333 >= 299.7, so index is 4
-        expect(findIndexAfterFraction(history, 0.9)).toBe(4);
+        // 333 >= 299.7, so index is 5
+        expect(findIndexAfterFraction(history, 0.9)).toBe(5);
     });
     it('should handle an empty history', () => {
         expect(findIndexAfterFraction([], 0.5)).toBe(0);
     });
     it('should handle a history with only one item', () => {
-        expect(findIndexAfterFraction(history.slice(0, 1), 0.5)).toBe(0);
+        expect(findIndexAfterFraction(history.slice(0, 1), 0.5)).toBe(1);
     });
     it('should handle history with weird parts', () => {
         const historyWithEmptyParts = [
@@ -131,7 +132,7 @@ describe('findIndexAfterFraction', () => {
             { role: 'model', parts: [{ fileData: { fileUri: 'derp' } }] },
             { role: 'user', parts: [{ text: 'Message 2' }] },
         ];
-        expect(findIndexAfterFraction(historyWithEmptyParts, 0.5)).toBe(1);
+        expect(findIndexAfterFraction(historyWithEmptyParts, 0.5)).toBe(2);
     });
 });
 describe('isThinkingSupported', () => {
@@ -176,8 +177,7 @@ describe('Gemini Client (client.ts)', () => {
         mockContentGenerator = {
             generateContent: mockGenerateContentFn,
             generateContentStream: vi.fn(),
-            countTokens: vi.fn(),
-            embedContent: vi.fn(),
+            countTokens: vi.fn().mockResolvedValue({ totalTokens: 100 }),
             batchEmbedContents: vi.fn(),
         };
         // Because the GeminiClient constructor kicks off an async process (startChat)
@@ -189,7 +189,6 @@ describe('Gemini Client (client.ts)', () => {
         };
         const fileService = new FileDiscoveryService('/test/dir');
         const contentGeneratorConfig = {
-            model: 'test-model',
             apiKey: 'test-key',
             vertexai: false,
             authType: AuthType.USE_GEMINI,
@@ -222,16 +221,26 @@ describe('Gemini Client (client.ts)', () => {
                 getDirectories: vi.fn().mockReturnValue(['/test/dir']),
             }),
             getGeminiClient: vi.fn(),
+            getModelRouterService: vi.fn().mockReturnValue({
+                route: vi.fn().mockResolvedValue({ model: 'default-routed-model' }),
+            }),
             isInFallbackMode: vi.fn().mockReturnValue(false),
             setFallbackMode: vi.fn(),
             getChatCompression: vi.fn().mockReturnValue(undefined),
             getSkipNextSpeakerCheck: vi.fn().mockReturnValue(false),
             getUseSmartEdit: vi.fn().mockReturnValue(false),
+            getUseModelRouter: vi.fn().mockReturnValue(false),
             getProjectRoot: vi.fn().mockReturnValue('/test/project/root'),
             storage: {
                 getProjectTempDir: vi.fn().mockReturnValue('/test/temp'),
             },
             getContentGenerator: vi.fn().mockReturnValue(mockContentGenerator),
+            getBaseLlmClient: vi.fn().mockReturnValue({
+                generateJson: vi.fn().mockResolvedValue({
+                    next_speaker: 'user',
+                    reasoning: 'test',
+                }),
+            }),
         };
         client = new GeminiClient(mockConfig);
         await client.initialize();
@@ -240,129 +249,6 @@ describe('Gemini Client (client.ts)', () => {
     afterEach(() => {
         vi.restoreAllMocks();
     });
-    describe('generateEmbedding', () => {
-        const texts = ['hello world', 'goodbye world'];
-        const testEmbeddingModel = 'test-embedding-model';
-        it('should call embedContent with correct parameters and return embeddings', async () => {
-            const mockEmbeddings = [
-                [0.1, 0.2, 0.3],
-                [0.4, 0.5, 0.6],
-            ];
-            vi.mocked(mockContentGenerator.embedContent).mockResolvedValue({
-                embeddings: [
-                    { values: mockEmbeddings[0] },
-                    { values: mockEmbeddings[1] },
-                ],
-            });
-            const result = await client.generateEmbedding(texts);
-            expect(mockContentGenerator.embedContent).toHaveBeenCalledTimes(1);
-            expect(mockContentGenerator.embedContent).toHaveBeenCalledWith({
-                model: testEmbeddingModel,
-                contents: texts,
-            });
-            expect(result).toEqual(mockEmbeddings);
-        });
-        it('should return an empty array if an empty array is passed', async () => {
-            const result = await client.generateEmbedding([]);
-            expect(result).toEqual([]);
-            expect(mockContentGenerator.embedContent).not.toHaveBeenCalled();
-        });
-        it('should throw an error if API response has no embeddings array', async () => {
-            vi.mocked(mockContentGenerator.embedContent).mockResolvedValue({});
-            await expect(client.generateEmbedding(texts)).rejects.toThrow('No embeddings found in API response.');
-        });
-        it('should throw an error if API response has an empty embeddings array', async () => {
-            vi.mocked(mockContentGenerator.embedContent).mockResolvedValue({
-                embeddings: [],
-            });
-            await expect(client.generateEmbedding(texts)).rejects.toThrow('No embeddings found in API response.');
-        });
-        it('should throw an error if API returns a mismatched number of embeddings', async () => {
-            vi.mocked(mockContentGenerator.embedContent).mockResolvedValue({
-                embeddings: [{ values: [1, 2, 3] }], // Only one for two texts
-            });
-            await expect(client.generateEmbedding(texts)).rejects.toThrow('API returned a mismatched number of embeddings. Expected 2, got 1.');
-        });
-        it('should throw an error if any embedding has nullish values', async () => {
-            vi.mocked(mockContentGenerator.embedContent).mockResolvedValue({
-                embeddings: [{ values: [1, 2, 3] }, { values: undefined }], // Second one is bad
-            });
-            await expect(client.generateEmbedding(texts)).rejects.toThrow('API returned an empty embedding for input text at index 1: "goodbye world"');
-        });
-        it('should throw an error if any embedding has an empty values array', async () => {
-            vi.mocked(mockContentGenerator.embedContent).mockResolvedValue({
-                embeddings: [{ values: [] }, { values: [1, 2, 3] }], // First one is bad
-            });
-            await expect(client.generateEmbedding(texts)).rejects.toThrow('API returned an empty embedding for input text at index 0: "hello world"');
-        });
-        it('should propagate errors from the API call', async () => {
-            vi.mocked(mockContentGenerator.embedContent).mockRejectedValue(new Error('API Failure'));
-            await expect(client.generateEmbedding(texts)).rejects.toThrow('API Failure');
-        });
-    });
-    describe('generateJson', () => {
-        it('should call generateContent with the correct parameters', async () => {
-            const contents = [{ role: 'user', parts: [{ text: 'hello' }] }];
-            const schema = { type: 'string' };
-            const abortSignal = new AbortController().signal;
-            vi.mocked(mockContentGenerator.countTokens).mockResolvedValue({
-                totalTokens: 1,
-            });
-            await client.generateJson(contents, schema, abortSignal, DEFAULT_GEMINI_FLASH_MODEL);
-            expect(mockContentGenerator.generateContent).toHaveBeenCalledWith({
-                model: DEFAULT_GEMINI_FLASH_MODEL,
-                config: {
-                    abortSignal,
-                    systemInstruction: getCoreSystemPrompt(''),
-                    temperature: 0,
-                    topP: 1,
-                    responseJsonSchema: schema,
-                    responseMimeType: 'application/json',
-                },
-                contents,
-            }, 'test-session-id');
-        });
-        it('should allow overriding model and config', async () => {
-            const contents = [
-                { role: 'user', parts: [{ text: 'hello' }] },
-            ];
-            const schema = { type: 'string' };
-            const abortSignal = new AbortController().signal;
-            const customModel = 'custom-json-model';
-            const customConfig = { temperature: 0.9, topK: 20 };
-            vi.mocked(mockContentGenerator.countTokens).mockResolvedValue({
-                totalTokens: 1,
-            });
-            await client.generateJson(contents, schema, abortSignal, customModel, customConfig);
-            expect(mockContentGenerator.generateContent).toHaveBeenCalledWith({
-                model: customModel,
-                config: {
-                    abortSignal,
-                    systemInstruction: getCoreSystemPrompt(''),
-                    temperature: 0.9,
-                    topP: 1, // from default
-                    topK: 20,
-                    responseJsonSchema: schema,
-                    responseMimeType: 'application/json',
-                },
-                contents,
-            }, 'test-session-id');
-        });
-        it('should use the Flash model when fallback mode is active', async () => {
-            const contents = [{ role: 'user', parts: [{ text: 'hello' }] }];
-            const schema = { type: 'string' };
-            const abortSignal = new AbortController().signal;
-            const requestedModel = 'gemini-2.5-pro'; // A non-flash model
-            // Mock config to be in fallback mode
-            // We access the mock via the client instance which holds the mocked config
-            vi.spyOn(client['config'], 'isInFallbackMode').mockReturnValue(true);
-            await client.generateJson(contents, schema, abortSignal, requestedModel);
-            // Assert that the Flash model was used, not the requested model
-            expect(mockContentGenerator.generateContent).toHaveBeenCalledWith(expect.objectContaining({
-                model: DEFAULT_GEMINI_FLASH_MODEL,
-            }), 'test-session-id');
-        });
-    });
     describe('addHistory', () => {
         it('should call chat.addHistory with the provided content', async () => {
             const mockChat = {
@@ -400,7 +286,6 @@ describe('Gemini Client (client.ts)', () => {
         });
     });
     describe('tryCompressChat', () => {
-        const mockSendMessage = vi.fn();
         const mockGetHistory = vi.fn();
         beforeEach(() => {
             vi.mock('./tokenLimits', () => ({
@@ -410,7 +295,6 @@ describe('Gemini Client (client.ts)', () => {
                 getHistory: mockGetHistory,
                 addHistory: vi.fn(),
                 setHistory: vi.fn(),
-                sendMessage: mockSendMessage,
             };
         });
         function setup({ chatHistory = [
@@ -420,7 +304,6 @@ describe('Gemini Client (client.ts)', () => {
             const mockChat = {
                 getHistory: vi.fn().mockReturnValue(chatHistory),
                 setHistory: vi.fn(),
-                sendMessage: vi.fn().mockResolvedValue({ text: 'Summary' }),
             };
             vi.mocked(mockContentGenerator.countTokens)
                 .mockResolvedValueOnce({ totalTokens: 1000 })
@@ -435,8 +318,12 @@ describe('Gemini Client (client.ts)', () => {
                 vi.mocked(mockContentGenerator.countTokens).mockResolvedValue({
                     totalTokens: 1000,
                 });
-                await client.tryCompressChat('prompt-id-4'); // Fails
-                const result = await client.tryCompressChat('prompt-id-4', true);
+                await client.tryCompressChat('prompt-id-4', false, [
+                    { text: 'request' },
+                ]); // Fails
+                const result = await client.tryCompressChat('prompt-id-4', true, [
+                    { text: 'request' },
+                ]);
                 expect(result).toEqual({
                     compressionStatus: CompressionStatus.COMPRESSED,
                     newTokenCount: 1000,
@@ -448,7 +335,9 @@ describe('Gemini Client (client.ts)', () => {
                 vi.mocked(mockContentGenerator.countTokens).mockResolvedValue({
                     totalTokens: 1000,
                 });
-                const result = await client.tryCompressChat('prompt-id-4', true);
+                const result = await client.tryCompressChat('prompt-id-4', false, [
+                    { text: 'request' },
+                ]);
                 expect(result).toEqual({
                     compressionStatus: CompressionStatus.COMPRESSION_FAILED_INFLATED_TOKEN_COUNT,
                     newTokenCount: 5000,
@@ -457,7 +346,9 @@ describe('Gemini Client (client.ts)', () => {
             });
             it('does not manipulate the source chat', async () => {
                 const { client, mockChat } = setup();
-                await client.tryCompressChat('prompt-id-4', true);
+                await client.tryCompressChat('prompt-id-4', false, [
+                    { text: 'request' },
+                ]);
                 expect(client['chat']).toBe(mockChat); // a new chat session was not created
             });
             it('restores the history back to the original', async () => {
@@ -473,14 +364,18 @@ describe('Gemini Client (client.ts)', () => {
                 const { client } = setup({
                     chatHistory: originalHistory,
                 });
-                const { compressionStatus } = await client.tryCompressChat('prompt-id-4');
+                const { compressionStatus } = await client.tryCompressChat('prompt-id-4', false, [{ text: 'what is your wisdom?' }]);
                 expect(compressionStatus).toBe(CompressionStatus.COMPRESSION_FAILED_INFLATED_TOKEN_COUNT);
                 expect(client['chat']?.setHistory).toHaveBeenCalledWith(originalHistory);
             });
             it('will not attempt to compress context after a failure', async () => {
                 const { client } = setup();
-                await client.tryCompressChat('prompt-id-4');
-                const result = await client.tryCompressChat('prompt-id-5');
+                await client.tryCompressChat('prompt-id-4', false, [
+                    { text: 'request' },
+                ]);
+                const result = await client.tryCompressChat('prompt-id-5', false, [
+                    { text: 'request' },
+                ]);
                 // it counts tokens for {original, compressed} and then never again
                 expect(mockContentGenerator.countTokens).toHaveBeenCalledTimes(2);
                 expect(result).toEqual({
@@ -500,7 +395,9 @@ describe('Gemini Client (client.ts)', () => {
                 totalTokens: MOCKED_TOKEN_LIMIT * 0.699, // TOKEN_THRESHOLD_FOR_SUMMARIZATION = 0.7
             });
             const initialChat = client.getChat();
-            const result = await client.tryCompressChat('prompt-id-2');
+            const result = await client.tryCompressChat('prompt-id-2', false, [
+                { text: '...history...' },
+            ]);
             const newChat = client.getChat();
             expect(tokenLimit).toHaveBeenCalled();
             expect(result).toEqual({
@@ -527,11 +424,19 @@ describe('Gemini Client (client.ts)', () => {
                 .mockResolvedValueOnce({ totalTokens: originalTokenCount }) // First call for the check
                 .mockResolvedValueOnce({ totalTokens: newTokenCount }); // Second call for the new history
             // Mock the summary response from the chat
-            mockSendMessage.mockResolvedValue({
-                role: 'model',
-                parts: [{ text: 'This is a summary.' }],
+            mockGenerateContentFn.mockResolvedValue({
+                candidates: [
+                    {
+                        content: {
+                            role: 'model',
+                            parts: [{ text: 'This is a summary.' }],
+                        },
+                    },
+                ],
             });
-            await client.tryCompressChat('prompt-id-3');
+            await client.tryCompressChat('prompt-id-3', false, [
+                { text: '...history...' },
+            ]);
             expect(ClearcutLogger.prototype.logChatCompressionEvent).toHaveBeenCalledWith(expect.objectContaining({
                 tokens_before: originalTokenCount,
                 tokens_after: newTokenCount,
@@ -553,15 +458,23 @@ describe('Gemini Client (client.ts)', () => {
                 .mockResolvedValueOnce({ totalTokens: originalTokenCount }) // First call for the check
                 .mockResolvedValueOnce({ totalTokens: newTokenCount }); // Second call for the new history
             // Mock the summary response from the chat
-            mockSendMessage.mockResolvedValue({
-                role: 'model',
-                parts: [{ text: 'This is a summary.' }],
+            mockGenerateContentFn.mockResolvedValue({
+                candidates: [
+                    {
+                        content: {
+                            role: 'model',
+                            parts: [{ text: 'This is a summary.' }],
+                        },
+                    },
+                ],
             });
             const initialChat = client.getChat();
-            const result = await client.tryCompressChat('prompt-id-3');
+            const result = await client.tryCompressChat('prompt-id-3', false, [
+                { text: '...history...' },
+            ]);
             const newChat = client.getChat();
             expect(tokenLimit).toHaveBeenCalled();
-            expect(mockSendMessage).toHaveBeenCalled();
+            expect(mockGenerateContentFn).toHaveBeenCalled();
             // Assert that summarization happened and returned the correct stats
             expect(result).toEqual({
                 compressionStatus: CompressionStatus.COMPRESSED,
@@ -598,15 +511,23 @@ describe('Gemini Client (client.ts)', () => {
                 .mockResolvedValueOnce({ totalTokens: originalTokenCount }) // First call for the check
                 .mockResolvedValueOnce({ totalTokens: newTokenCount }); // Second call for the new history
             // Mock the summary response from the chat
-            mockSendMessage.mockResolvedValue({
-                role: 'model',
-                parts: [{ text: 'This is a summary.' }],
+            mockGenerateContentFn.mockResolvedValue({
+                candidates: [
+                    {
+                        content: {
+                            role: 'model',
+                            parts: [{ text: 'This is a summary.' }],
+                        },
+                    },
+                ],
             });
             const initialChat = client.getChat();
-            const result = await client.tryCompressChat('prompt-id-3');
+            const result = await client.tryCompressChat('prompt-id-3', false, [
+                { text: '...history...' },
+            ]);
             const newChat = client.getChat();
             expect(tokenLimit).toHaveBeenCalled();
-            expect(mockSendMessage).toHaveBeenCalled();
+            expect(mockGenerateContentFn).toHaveBeenCalled();
             // Assert that summarization happened and returned the correct stats
             expect(result).toEqual({
                 compressionStatus: CompressionStatus.COMPRESSED,
@@ -620,7 +541,7 @@ describe('Gemini Client (client.ts)', () => {
             // 3. compressed summary message
             // 4. standard canned user summary message
             // 5. The last user message (not the last 3 because that would start with a function response)
-            expect(newChat.getHistory().length).toEqual(5);
+            expect(newChat.getHistory().length).toEqual(6);
         });
         it('should always trigger summarization when force is true, regardless of token count', async () => {
             mockGetHistory.mockReturnValue([
@@ -632,14 +553,22 @@ describe('Gemini Client (client.ts)', () => {
                 .mockResolvedValueOnce({ totalTokens: originalTokenCount })
                 .mockResolvedValueOnce({ totalTokens: newTokenCount });
             // Mock the summary response from the chat
-            mockSendMessage.mockResolvedValue({
-                role: 'model',
-                parts: [{ text: 'This is a summary.' }],
+            mockGenerateContentFn.mockResolvedValue({
+                candidates: [
+                    {
+                        content: {
+                            role: 'model',
+                            parts: [{ text: 'This is a summary.' }],
+                        },
+                    },
+                ],
             });
             const initialChat = client.getChat();
-            const result = await client.tryCompressChat('prompt-id-1', true); // force = true
+            const result = await client.tryCompressChat('prompt-id-1', false, [
+                { text: '...history...' },
+            ]); // force = true
             const newChat = client.getChat();
-            expect(mockSendMessage).toHaveBeenCalled();
+            expect(mockGenerateContentFn).toHaveBeenCalled();
             expect(result).toEqual({
                 compressionStatus: CompressionStatus.COMPRESSED,
                 originalTokenCount,
@@ -665,17 +594,18 @@ describe('Gemini Client (client.ts)', () => {
                 { role: 'model', parts: [{ text: 'Long response' }] },
             ];
             const mockChat = {
-                getHistory: vi.fn().mockReturnValue(mockChatHistory),
+                getHistory: vi.fn().mockImplementation(() => [...mockChatHistory]),
                 setHistory: vi.fn(),
                 sendMessage: mockSendMessage,
             };
             client['chat'] = mockChat;
             client['startChat'] = vi.fn().mockResolvedValue(mockChat);
-            const result = await client.tryCompressChat('prompt-id-4', true);
+            const request = [{ text: 'Long conversation' }];
+            const result = await client.tryCompressChat('prompt-id-4', false, request);
             expect(mockContentGenerator.countTokens).toHaveBeenCalledTimes(2);
             expect(mockContentGenerator.countTokens).toHaveBeenNthCalledWith(1, {
                 model: firstCurrentModel,
-                contents: mockChatHistory,
+                contents: [...mockChatHistory, createUserContent(request)],
             });
             expect(mockContentGenerator.countTokens).toHaveBeenNthCalledWith(2, {
                 model: secondCurrentModel,
@@ -740,7 +670,7 @@ describe('Gemini Client (client.ts)', () => {
         });
         it('should include editor context when ideMode is enabled', async () => {
             // Arrange
-            vi.mocked(ideContext.getIdeContext).mockReturnValue({
+            vi.mocked(ideContextStore.get).mockReturnValue({
                 workspaceState: {
                     openFiles: [
                         {
@@ -762,6 +692,11 @@ describe('Gemini Client (client.ts)', () => {
                 },
             });
             vi.mocked(mockConfig.getIdeMode).mockReturnValue(true);
+            vi.spyOn(client, 'tryCompressChat').mockResolvedValue({
+                originalTokenCount: 0,
+                newTokenCount: 0,
+                compressionStatus: CompressionStatus.COMPRESSED,
+            });
             mockTurnRunFn.mockReturnValue((async function* () {
                 yield { type: 'content', value: 'Hello' };
             })());
@@ -777,7 +712,7 @@ describe('Gemini Client (client.ts)', () => {
                 // consume stream
             }
             // Assert
-            expect(ideContext.getIdeContext).toHaveBeenCalled();
+            expect(ideContextStore.get).toHaveBeenCalled();
             const expectedContext = `
 Here is the user's editor context as a JSON object. This is for your information only.
 \`\`\`json
@@ -802,7 +737,7 @@ ${JSON.stringify({
         });
         it('should not add context if ideMode is enabled but no open files', async () => {
             // Arrange
-            vi.mocked(ideContext.getIdeContext).mockReturnValue({
+            vi.mocked(ideContextStore.get).mockReturnValue({
                 workspaceState: {
                     openFiles: [],
                 },
@@ -824,12 +759,16 @@ ${JSON.stringify({
                 // consume stream
             }
             // Assert
-            expect(ideContext.getIdeContext).toHaveBeenCalled();
-            expect(mockTurnRunFn).toHaveBeenCalledWith(initialRequest, expect.any(Object));
+            expect(ideContextStore.get).toHaveBeenCalled();
+            // The `turn.run` method is now called with the model name as the first
+            // argument. We use `expect.any(String)` because this test is
+            // concerned with the IDE context logic, not the model routing,
+            // which is tested in its own dedicated suite.
+            expect(mockTurnRunFn).toHaveBeenCalledWith(expect.any(String), initialRequest, expect.any(Object));
         });
         it('should add context if ideMode is enabled and there is one active file', async () => {
             // Arrange
-            vi.mocked(ideContext.getIdeContext).mockReturnValue({
+            vi.mocked(ideContextStore.get).mockReturnValue({
                 workspaceState: {
                     openFiles: [
                         {
@@ -843,6 +782,11 @@ ${JSON.stringify({
                 },
             });
             vi.spyOn(client['config'], 'getIdeMode').mockReturnValue(true);
+            vi.spyOn(client, 'tryCompressChat').mockResolvedValue({
+                originalTokenCount: 0,
+                newTokenCount: 0,
+                compressionStatus: CompressionStatus.COMPRESSED,
+            });
             const mockStream = (async function* () {
                 yield { type: 'content', value: 'Hello' };
             })();
@@ -859,7 +803,7 @@ ${JSON.stringify({
                 // consume stream
             }
             // Assert
-            expect(ideContext.getIdeContext).toHaveBeenCalled();
+            expect(ideContextStore.get).toHaveBeenCalled();
             const expectedContext = `
 Here is the user's editor context as a JSON object. This is for your information only.
 \`\`\`json
@@ -883,7 +827,7 @@ ${JSON.stringify({
         });
         it('should add context if ideMode is enabled and there are open files but no active file', async () => {
             // Arrange
-            vi.mocked(ideContext.getIdeContext).mockReturnValue({
+            vi.mocked(ideContextStore.get).mockReturnValue({
                 workspaceState: {
                     openFiles: [
                         {
@@ -898,6 +842,11 @@ ${JSON.stringify({
                 },
             });
             vi.spyOn(client['config'], 'getIdeMode').mockReturnValue(true);
+            vi.spyOn(client, 'tryCompressChat').mockResolvedValue({
+                originalTokenCount: 0,
+                newTokenCount: 0,
+                compressionStatus: CompressionStatus.COMPRESSED,
+            });
             const mockStream = (async function* () {
                 yield { type: 'content', value: 'Hello' };
             })();
@@ -914,7 +863,7 @@ ${JSON.stringify({
                 // consume stream
             }
             // Assert
-            expect(ideContext.getIdeContext).toHaveBeenCalled();
+            expect(ideContextStore.get).toHaveBeenCalled();
             const expectedContext = `
 Here is the user's editor context as a JSON object. This is for your information only.
 \`\`\`json
@@ -1110,6 +1059,91 @@ ${JSON.stringify({
             console.log(`Infinite loop protection working: checkNextSpeaker called ${callCount} times, ` +
                 `${eventCount} events generated (properly bounded by MAX_TURNS)`);
         });
+        describe('Model Routing', () => {
+            let mockRouterService;
+            beforeEach(() => {
+                mockRouterService = {
+                    route: vi
+                        .fn()
+                        .mockResolvedValue({ model: 'routed-model', reason: 'test' }),
+                };
+                vi.mocked(mockConfig.getModelRouterService).mockReturnValue(mockRouterService);
+                mockTurnRunFn.mockReturnValue((async function* () {
+                    yield { type: 'content', value: 'Hello' };
+                })());
+            });
+            it('should use the model router service to select a model on the first turn', async () => {
+                const stream = client.sendMessageStream([{ text: 'Hi' }], new AbortController().signal, 'prompt-1');
+                await fromAsync(stream); // consume stream
+                expect(mockConfig.getModelRouterService).toHaveBeenCalled();
+                expect(mockRouterService.route).toHaveBeenCalled();
+                expect(mockTurnRunFn).toHaveBeenCalledWith('routed-model', // The model from the router
+                [{ text: 'Hi' }], expect.any(Object));
+            });
+            it('should use the same model for subsequent turns in the same prompt (stickiness)', async () => {
+                // First turn
+                let stream = client.sendMessageStream([{ text: 'Hi' }], new AbortController().signal, 'prompt-1');
+                await fromAsync(stream);
+                expect(mockRouterService.route).toHaveBeenCalledTimes(1);
+                expect(mockTurnRunFn).toHaveBeenCalledWith('routed-model', [{ text: 'Hi' }], expect.any(Object));
+                // Second turn
+                stream = client.sendMessageStream([{ text: 'Continue' }], new AbortController().signal, 'prompt-1');
+                await fromAsync(stream);
+                // Router should not be called again
+                expect(mockRouterService.route).toHaveBeenCalledTimes(1);
+                // Should stick to the first model
+                expect(mockTurnRunFn).toHaveBeenCalledWith('routed-model', [{ text: 'Continue' }], expect.any(Object));
+            });
+            it('should reset the sticky model and re-route when the prompt_id changes', async () => {
+                // First prompt
+                let stream = client.sendMessageStream([{ text: 'Hi' }], new AbortController().signal, 'prompt-1');
+                await fromAsync(stream);
+                expect(mockRouterService.route).toHaveBeenCalledTimes(1);
+                expect(mockTurnRunFn).toHaveBeenCalledWith('routed-model', [{ text: 'Hi' }], expect.any(Object));
+                // New prompt
+                mockRouterService.route.mockResolvedValue({
+                    model: 'new-routed-model',
+                    reason: 'test',
+                });
+                stream = client.sendMessageStream([{ text: 'A new topic' }], new AbortController().signal, 'prompt-2');
+                await fromAsync(stream);
+                // Router should be called again for the new prompt
+                expect(mockRouterService.route).toHaveBeenCalledTimes(2);
+                // Should use the newly routed model
+                expect(mockTurnRunFn).toHaveBeenCalledWith('new-routed-model', [{ text: 'A new topic' }], expect.any(Object));
+            });
+            it('should use the fallback model and bypass routing when in fallback mode', async () => {
+                vi.mocked(mockConfig.isInFallbackMode).mockReturnValue(true);
+                mockRouterService.route.mockResolvedValue({
+                    model: DEFAULT_GEMINI_FLASH_MODEL,
+                    reason: 'fallback',
+                });
+                const stream = client.sendMessageStream([{ text: 'Hi' }], new AbortController().signal, 'prompt-1');
+                await fromAsync(stream);
+                expect(mockTurnRunFn).toHaveBeenCalledWith(DEFAULT_GEMINI_FLASH_MODEL, [{ text: 'Hi' }], expect.any(Object));
+            });
+            it('should stick to the fallback model for the entire sequence even if fallback mode ends', async () => {
+                // Start the sequence in fallback mode
+                vi.mocked(mockConfig.isInFallbackMode).mockReturnValue(true);
+                mockRouterService.route.mockResolvedValue({
+                    model: DEFAULT_GEMINI_FLASH_MODEL,
+                    reason: 'fallback',
+                });
+                let stream = client.sendMessageStream([{ text: 'Hi' }], new AbortController().signal, 'prompt-fallback-stickiness');
+                await fromAsync(stream);
+                // First call should use fallback model
+                expect(mockTurnRunFn).toHaveBeenCalledWith(DEFAULT_GEMINI_FLASH_MODEL, [{ text: 'Hi' }], expect.any(Object));
+                // End fallback mode
+                vi.mocked(mockConfig.isInFallbackMode).mockReturnValue(false);
+                // Second call in the same sequence
+                stream = client.sendMessageStream([{ text: 'Continue' }], new AbortController().signal, 'prompt-fallback-stickiness');
+                await fromAsync(stream);
+                // Router should still not be called, and it should stick to the fallback model
+                expect(mockTurnRunFn).toHaveBeenCalledTimes(2); // Ensure it was called again
+                expect(mockTurnRunFn).toHaveBeenLastCalledWith(DEFAULT_GEMINI_FLASH_MODEL, // Still the fallback model
+                [{ text: 'Continue' }], expect.any(Object));
+            });
+        });
         describe('Editor context delta', () => {
             const mockStream = (async function* () {
                 yield { type: 'content', value: 'Hello' };
@@ -1126,7 +1160,6 @@ ${JSON.stringify({
                 const mockChat = {
                     addHistory: vi.fn(),
                     setHistory: vi.fn(),
-                    sendMessage: vi.fn().mockResolvedValue({ text: 'summary' }),
                     // Assume history is not empty for delta checks
                     getHistory: vi
                         .fn()
@@ -1250,7 +1283,7 @@ ${JSON.stringify({
                     },
                 };
                 // Setup current context
-                vi.mocked(ideContext.getIdeContext).mockReturnValue({
+                vi.mocked(ideContextStore.get).mockReturnValue({
                     workspaceState: {
                         openFiles: [
                             { ...currentActiveFile, isActive: true, timestamp: Date.now() },
@@ -1296,7 +1329,7 @@ ${JSON.stringify({
                     },
                 };
                 // Setup current context (same as previous)
-                vi.mocked(ideContext.getIdeContext).mockReturnValue({
+                vi.mocked(ideContextStore.get).mockReturnValue({
                     workspaceState: {
                         openFiles: [
                             { ...activeFile, isActive: true, timestamp: Date.now() },
@@ -1341,11 +1374,10 @@ ${JSON.stringify({
                     addHistory: vi.fn(),
                     getHistory: vi.fn().mockReturnValue([]), // Default empty history
                     setHistory: vi.fn(),
-                    sendMessage: vi.fn().mockResolvedValue({ text: 'summary' }),
                 };
                 client['chat'] = mockChat;
                 vi.spyOn(client['config'], 'getIdeMode').mockReturnValue(true);
-                vi.mocked(ideContext.getIdeContext).mockReturnValue({
+                vi.mocked(ideContextStore.get).mockReturnValue({
                     workspaceState: {
                         openFiles: [{ path: '/path/to/file.ts', timestamp: Date.now() }],
                     },
@@ -1421,7 +1453,7 @@ ${JSON.stringify({
                         openFiles: [{ path: '/path/to/fileA.ts', timestamp: Date.now() }],
                     },
                 };
-                vi.mocked(ideContext.getIdeContext).mockReturnValue(initialIdeContext);
+                vi.mocked(ideContextStore.get).mockReturnValue(initialIdeContext);
                 // Act: Send the tool response
                 let stream = client.sendMessageStream([
                     {
@@ -1467,7 +1499,7 @@ ${JSON.stringify({
                         openFiles: [{ path: '/path/to/fileB.ts', timestamp: Date.now() }],
                     },
                 };
-                vi.mocked(ideContext.getIdeContext).mockReturnValue(newIdeContext);
+                vi.mocked(ideContextStore.get).mockReturnValue(newIdeContext);
                 // Act: Send a new, regular user message
                 stream = client.sendMessageStream([{ text: 'Thanks!' }], new AbortController().signal, 'prompt-id-final');
                 for await (const _ of stream) {
@@ -1497,7 +1529,7 @@ ${JSON.stringify({
                         ],
                     },
                 };
-                vi.mocked(ideContext.getIdeContext).mockReturnValue(contextA);
+                vi.mocked(ideContextStore.get).mockReturnValue(contextA);
                 // Act: Send a regular message to establish the initial context
                 let stream = client.sendMessageStream([{ text: 'Initial message' }], new AbortController().signal, 'prompt-id-initial');
                 for await (const _ of stream) {
@@ -1530,7 +1562,7 @@ ${JSON.stringify({
                         ],
                     },
                 };
-                vi.mocked(ideContext.getIdeContext).mockReturnValue(contextB);
+                vi.mocked(ideContextStore.get).mockReturnValue(contextB);
                 // Act: Send the tool response
                 stream = client.sendMessageStream([
                     {
@@ -1575,7 +1607,7 @@ ${JSON.stringify({
                         ],
                     },
                 };
-                vi.mocked(ideContext.getIdeContext).mockReturnValue(contextC);
+                vi.mocked(ideContextStore.get).mockReturnValue(contextC);
                 // Act: Send a new, regular user message
                 stream = client.sendMessageStream([{ text: 'Thanks!' }], new AbortController().signal, 'prompt-id-final');
                 for await (const _ of stream) {
@@ -1640,6 +1672,35 @@ ${JSON.stringify({
             // Assert
             expect(mockCheckNextSpeaker).not.toHaveBeenCalled();
         });
+        it('should abort linked signal when loop is detected', async () => {
+            // Arrange
+            vi.spyOn(client['loopDetector'], 'turnStarted').mockResolvedValue(false);
+            vi.spyOn(client['loopDetector'], 'addAndCheck')
+                .mockReturnValueOnce(false)
+                .mockReturnValueOnce(true);
+            let capturedSignal;
+            mockTurnRunFn.mockImplementation((model, request, signal) => {
+                capturedSignal = signal;
+                return (async function* () {
+                    yield { type: 'content', value: 'First event' };
+                    yield { type: 'content', value: 'Second event' };
+                })();
+            });
+            const mockChat = {
+                addHistory: vi.fn(),
+                getHistory: vi.fn().mockReturnValue([]),
+            };
+            client['chat'] = mockChat;
+            // Act
+            const stream = client.sendMessageStream([{ text: 'Hi' }], new AbortController().signal, 'prompt-id-loop');
+            const events = [];
+            for await (const event of stream) {
+                events.push(event);
+            }
+            // Assert
+            expect(events).toContainEqual({ type: GeminiEventType.LoopDetected });
+            expect(capturedSignal.aborted).toBe(true);
+        });
     });
     describe('generateContent', () => {
         it('should call generateContent with the correct parameters', async () => {