@google/gemini-cli-core 0.0.3-preview.4 → 0.0.3

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

@@ -4,25 +4,43 @@
  * SPDX-License-Identifier: Apache-2.0
  */
 import { describe, it, expect, vi, beforeEach, afterEach, } from 'vitest';
-import { GoogleGenAI } from '@google/genai';
-import { findIndexAfterFraction, GeminiClient } from './client.js';
+import { findCompressSplitPoint, isThinkingDefault, isThinkingSupported, GeminiClient, } from './client.js';
 import { AuthType, } from './contentGenerator.js';
 import {} from './geminiChat.js';
-import { Config } from '../config/config.js';
 import { CompressionStatus, GeminiEventType, Turn, } from './turn.js';
 import { getCoreSystemPrompt } from './prompts.js';
 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';
+import { uiTelemetryService } from '../telemetry/uiTelemetry.js';
+// Mock fs module to prevent actual file system operations during tests
+const mockFileSystem = new Map();
+vi.mock('node:fs', () => {
+    const fsModule = {
+        mkdirSync: vi.fn(),
+        writeFileSync: vi.fn((path, data) => {
+            mockFileSystem.set(path, data);
+        }),
+        readFileSync: vi.fn((path) => {
+            if (mockFileSystem.has(path)) {
+                return mockFileSystem.get(path);
+            }
+            throw Object.assign(new Error('ENOENT: no such file or directory'), {
+                code: 'ENOENT',
+            });
+        }),
+        existsSync: vi.fn((path) => mockFileSystem.has(path)),
+    };
+    return {
+        default: fsModule,
+        ...fsModule,
+    };
+});
 // --- Mocks ---
-const mockChatCreateFn = vi.fn();
-const mockGenerateContentFn = vi.fn();
-const mockEmbedContentFn = vi.fn();
 const mockTurnRunFn = vi.fn();
-vi.mock('@google/genai');
 vi.mock('./turn', async (importOriginal) => {
     const actual = await importOriginal();
     // Define a mock class that has the same shape as the real Turn
@@ -59,6 +77,11 @@ vi.mock('../telemetry/index.js', () => ({
     logApiError: vi.fn(),
 }));
 vi.mock('../ide/ideContext.js');
+vi.mock('../telemetry/uiTelemetry.js', () => ({
+    uiTelemetryService: {
+        setLastPromptTokenCount: vi.fn(),
+    },
+}));
 /**
  * Array.fromAsync ponyfill, which will be available in es 2024.
  *
@@ -72,41 +95,59 @@ async function fromAsync(promise) {
     return results;
 }
 describe('findIndexAfterFraction', () => {
-    const history = [
-        { role: 'user', parts: [{ text: 'This is the first message.' }] }, // JSON length: 66
-        { role: 'model', parts: [{ text: 'This is the second message.' }] }, // JSON length: 68
-        { role: 'user', parts: [{ text: 'This is the third message.' }] }, // JSON length: 66
-        { role: 'model', parts: [{ text: 'This is the fourth message.' }] }, // JSON length: 68
-        { role: 'user', parts: [{ text: 'This is the fifth message.' }] }, // JSON length: 65
-    ];
-    // Total length: 333
     it('should throw an error for non-positive numbers', () => {
-        expect(() => findIndexAfterFraction(history, 0)).toThrow('Fraction must be between 0 and 1');
+        expect(() => findCompressSplitPoint([], 0)).toThrow('Fraction must be between 0 and 1');
     });
     it('should throw an error for a fraction greater than or equal to 1', () => {
-        expect(() => findIndexAfterFraction(history, 1)).toThrow('Fraction must be between 0 and 1');
+        expect(() => findCompressSplitPoint([], 1)).toThrow('Fraction must be between 0 and 1');
+    });
+    it('should handle an empty history', () => {
+        expect(findCompressSplitPoint([], 0.5)).toBe(0);
     });
     it('should handle a fraction in the middle', () => {
-        // 333 * 0.5 = 166.5
-        // 0: 66
-        // 1: 66 + 68 = 134
-        // 2: 134 + 66 = 200
-        // 200 >= 166.5, so index is 2
-        expect(findIndexAfterFraction(history, 0.5)).toBe(2);
+        const history = [
+            { role: 'user', parts: [{ text: 'This is the first message.' }] }, // JSON length: 66 (19%)
+            { role: 'model', parts: [{ text: 'This is the second message.' }] }, // JSON length: 68 (40%)
+            { role: 'user', parts: [{ text: 'This is the third message.' }] }, // JSON length: 66 (60%)
+            { role: 'model', parts: [{ text: 'This is the fourth message.' }] }, // JSON length: 68 (80%)
+            { role: 'user', parts: [{ text: 'This is the fifth message.' }] }, // JSON length: 65 (100%)
+        ];
+        expect(findCompressSplitPoint(history, 0.5)).toBe(2);
     });
-    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);
+    it('should handle a fraction of last index', () => {
+        const history = [
+            { role: 'user', parts: [{ text: 'This is the first message.' }] }, // JSON length: 66 (19%)
+            { role: 'model', parts: [{ text: 'This is the second message.' }] }, // JSON length: 68 (40%)
+            { role: 'user', parts: [{ text: 'This is the third message.' }] }, // JSON length: 66 (60%)
+            { role: 'model', parts: [{ text: 'This is the fourth message.' }] }, // JSON length: 68 (80%)
+            { role: 'user', parts: [{ text: 'This is the fifth message.' }] }, // JSON length: 65 (100%)
+        ];
+        expect(findCompressSplitPoint(history, 0.9)).toBe(4);
     });
-    it('should handle an empty history', () => {
-        expect(findIndexAfterFraction([], 0.5)).toBe(0);
+    it('should handle a fraction of after last index', () => {
+        const history = [
+            { role: 'user', parts: [{ text: 'This is the first message.' }] }, // JSON length: 66 (24%%)
+            { role: 'model', parts: [{ text: 'This is the second message.' }] }, // JSON length: 68 (50%)
+            { role: 'user', parts: [{ text: 'This is the third message.' }] }, // JSON length: 66 (74%)
+            { role: 'model', parts: [{ text: 'This is the fourth message.' }] }, // JSON length: 68 (100%)
+        ];
+        expect(findCompressSplitPoint(history, 0.8)).toBe(4);
+    });
+    it('should return earlier splitpoint if no valid ones are after threshhold', () => {
+        const history = [
+            { role: 'user', parts: [{ text: 'This is the first message.' }] },
+            { role: 'model', parts: [{ text: 'This is the second message.' }] },
+            { role: 'user', parts: [{ text: 'This is the third message.' }] },
+            { role: 'model', parts: [{ functionCall: {} }] },
+        ];
+        // Can't return 4 because the previous item has a function call.
+        expect(findCompressSplitPoint(history, 0.99)).toBe(2);
     });
     it('should handle a history with only one item', () => {
-        expect(findIndexAfterFraction(history.slice(0, 1), 0.5)).toBe(0);
+        const historyWithEmptyParts = [
+            { role: 'user', parts: [{ text: 'Message 1' }] },
+        ];
+        expect(findCompressSplitPoint(historyWithEmptyParts, 0.5)).toBe(0);
     });
     it('should handle history with weird parts', () => {
         const historyWithEmptyParts = [
@@ -114,37 +155,55 @@ describe('findIndexAfterFraction', () => {
             { role: 'model', parts: [{ fileData: { fileUri: 'derp' } }] },
             { role: 'user', parts: [{ text: 'Message 2' }] },
         ];
-        expect(findIndexAfterFraction(historyWithEmptyParts, 0.5)).toBe(1);
+        expect(findCompressSplitPoint(historyWithEmptyParts, 0.5)).toBe(2);
+    });
+});
+describe('isThinkingSupported', () => {
+    it('should return true for gemini-2.5', () => {
+        expect(isThinkingSupported('gemini-2.5')).toBe(true);
+    });
+    it('should return true for gemini-2.5-pro', () => {
+        expect(isThinkingSupported('gemini-2.5-pro')).toBe(true);
+    });
+    it('should return false for other models', () => {
+        expect(isThinkingSupported('gemini-1.5-flash')).toBe(false);
+        expect(isThinkingSupported('some-other-model')).toBe(false);
+    });
+});
+describe('isThinkingDefault', () => {
+    it('should return false for gemini-2.5-flash-lite', () => {
+        expect(isThinkingDefault('gemini-2.5-flash-lite')).toBe(false);
+    });
+    it('should return true for gemini-2.5', () => {
+        expect(isThinkingDefault('gemini-2.5')).toBe(true);
+    });
+    it('should return true for gemini-2.5-pro', () => {
+        expect(isThinkingDefault('gemini-2.5-pro')).toBe(true);
+    });
+    it('should return false for other models', () => {
+        expect(isThinkingDefault('gemini-1.5-flash')).toBe(false);
+        expect(isThinkingDefault('some-other-model')).toBe(false);
     });
 });
 describe('Gemini Client (client.ts)', () => {
+    let mockContentGenerator;
+    let mockConfig;
     let client;
+    let mockGenerateContentFn;
     beforeEach(async () => {
         vi.resetAllMocks();
+        vi.mocked(uiTelemetryService.setLastPromptTokenCount).mockClear();
+        mockGenerateContentFn = vi.fn().mockResolvedValue({
+            candidates: [{ content: { parts: [{ text: '{"key": "value"}' }] } }],
+        });
         // Disable 429 simulation for tests
         setSimulate429(false);
-        // Set up the mock for GoogleGenAI constructor and its methods
-        const MockedGoogleGenAI = vi.mocked(GoogleGenAI);
-        MockedGoogleGenAI.mockImplementation(() => {
-            const mock = {
-                chats: { create: mockChatCreateFn },
-                models: {
-                    generateContent: mockGenerateContentFn,
-                    embedContent: mockEmbedContentFn,
-                },
-            };
-            return mock;
-        });
-        mockChatCreateFn.mockResolvedValue({});
-        mockGenerateContentFn.mockResolvedValue({
-            candidates: [
-                {
-                    content: {
-                        parts: [{ text: '{"key": "value"}' }],
-                    },
-                },
-            ],
-        });
+        mockContentGenerator = {
+            generateContent: mockGenerateContentFn,
+            generateContentStream: vi.fn(),
+            countTokens: vi.fn().mockResolvedValue({ totalTokens: 100 }),
+            batchEmbedContents: vi.fn(),
+        };
         // Because the GeminiClient constructor kicks off an async process (startChat)
         // that depends on a fully-formed Config object, we need to mock the
         // entire implementation of Config for these tests.
@@ -154,12 +213,11 @@ 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,
         };
-        const mockConfigObject = {
+        mockConfig = {
             getContentGeneratorConfig: vi
                 .fn()
                 .mockReturnValue(contentGeneratorConfig),
@@ -187,159 +245,34 @@ 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',
+                }),
+            }),
         };
-        const MockedConfig = vi.mocked(Config, true);
-        MockedConfig.mockImplementation(() => mockConfigObject);
-        // We can instantiate the client here since Config is mocked
-        // and the constructor will use the mocked GoogleGenAI
-        client = new GeminiClient(new Config({ sessionId: 'test-session-id' }));
-        mockConfigObject.getGeminiClient.mockReturnValue(client);
-        await client.initialize(contentGeneratorConfig);
+        client = new GeminiClient(mockConfig);
+        await client.initialize();
+        vi.mocked(mockConfig.getGeminiClient).mockReturnValue(client);
     });
     afterEach(() => {
         vi.restoreAllMocks();
     });
-    // NOTE: The following tests for startChat were removed due to persistent issues with
-    // the @google/genai mock. Specifically, the mockChatCreateFn (representing instance.chats.create)
-    // was not being detected as called by the GeminiClient instance.
-    // This likely points to a subtle issue in how the GoogleGenerativeAI class constructor
-    // and its instance methods are mocked and then used by the class under test.
-    // For future debugging, ensure that the `this.client` in `GeminiClient` (which is an
-    // instance of the mocked GoogleGenerativeAI) correctly has its `chats.create` method
-    // pointing to `mockChatCreateFn`.
-    // it('startChat should call getCoreSystemPrompt with userMemory and pass to chats.create', async () => { ... });
-    // it('startChat should call getCoreSystemPrompt with empty string if userMemory is empty', async () => { ... });
-    // NOTE: The following tests for generateJson were removed due to persistent issues with
-    // the @google/genai mock, similar to the startChat tests. The mockGenerateContentFn
-    // (representing instance.models.generateContent) was not being detected as called, or the mock
-    // was not preventing an actual API call (leading to API key errors).
-    // For future debugging, ensure `this.client.models.generateContent` in `GeminiClient` correctly
-    // uses the `mockGenerateContentFn`.
-    // it('generateJson should call getCoreSystemPrompt with userMemory and pass to generateContent', async () => { ... });
-    // it('generateJson should call getCoreSystemPrompt with empty string if userMemory is empty', async () => { ... });
-    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],
-            ];
-            const mockResponse = {
-                embeddings: [
-                    { values: mockEmbeddings[0] },
-                    { values: mockEmbeddings[1] },
-                ],
-            };
-            mockEmbedContentFn.mockResolvedValue(mockResponse);
-            const result = await client.generateEmbedding(texts);
-            expect(mockEmbedContentFn).toHaveBeenCalledTimes(1);
-            expect(mockEmbedContentFn).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(mockEmbedContentFn).not.toHaveBeenCalled();
-        });
-        it('should throw an error if API response has no embeddings array', async () => {
-            mockEmbedContentFn.mockResolvedValue({}); // No `embeddings` key
-            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 () => {
-            const mockResponse = {
-                embeddings: [],
-            };
-            mockEmbedContentFn.mockResolvedValue(mockResponse);
-            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 () => {
-            const mockResponse = {
-                embeddings: [{ values: [1, 2, 3] }], // Only one for two texts
-            };
-            mockEmbedContentFn.mockResolvedValue(mockResponse);
-            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 () => {
-            const mockResponse = {
-                embeddings: [{ values: [1, 2, 3] }, { values: undefined }], // Second one is bad
-            };
-            mockEmbedContentFn.mockResolvedValue(mockResponse);
-            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 () => {
-            const mockResponse = {
-                embeddings: [{ values: [] }, { values: [1, 2, 3] }], // First one is bad
-            };
-            mockEmbedContentFn.mockResolvedValue(mockResponse);
-            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 () => {
-            const apiError = new Error('API Failure');
-            mockEmbedContentFn.mockRejectedValue(apiError);
-            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;
-            // Mock countTokens
-            const mockGenerator = {
-                countTokens: vi.fn().mockResolvedValue({ totalTokens: 1 }),
-                generateContent: mockGenerateContentFn,
-            };
-            client['contentGenerator'] = mockGenerator;
-            await client.generateJson(contents, schema, abortSignal);
-            expect(mockGenerateContentFn).toHaveBeenCalledWith({
-                model: 'test-model', // Should use current model from config
-                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 };
-            const mockGenerator = {
-                countTokens: vi.fn().mockResolvedValue({ totalTokens: 1 }),
-                generateContent: mockGenerateContentFn,
-            };
-            client['contentGenerator'] = mockGenerator;
-            await client.generateJson(contents, schema, abortSignal, customModel, customConfig);
-            expect(mockGenerateContentFn).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');
-        });
-    });
     describe('addHistory', () => {
         it('should call chat.addHistory with the provided content', async () => {
             const mockChat = {
@@ -377,21 +310,15 @@ describe('Gemini Client (client.ts)', () => {
         });
     });
     describe('tryCompressChat', () => {
-        const mockCountTokens = vi.fn();
-        const mockSendMessage = vi.fn();
         const mockGetHistory = vi.fn();
         beforeEach(() => {
             vi.mock('./tokenLimits', () => ({
                 tokenLimit: vi.fn(),
             }));
-            client['contentGenerator'] = {
-                countTokens: mockCountTokens,
-            };
             client['chat'] = {
                 getHistory: mockGetHistory,
                 addHistory: vi.fn(),
                 setHistory: vi.fn(),
-                sendMessage: mockSendMessage,
             };
         });
         function setup({ chatHistory = [
@@ -401,28 +328,21 @@ describe('Gemini Client (client.ts)', () => {
             const mockChat = {
                 getHistory: vi.fn().mockReturnValue(chatHistory),
                 setHistory: vi.fn(),
-                sendMessage: vi.fn().mockResolvedValue({ text: 'Summary' }),
             };
-            const mockCountTokens = vi
-                .fn()
+            vi.mocked(mockContentGenerator.countTokens)
                 .mockResolvedValueOnce({ totalTokens: 1000 })
                 .mockResolvedValueOnce({ totalTokens: 5000 });
-            const mockGenerator = {
-                countTokens: mockCountTokens,
-            };
             client['chat'] = mockChat;
-            client['contentGenerator'] = mockGenerator;
             client['startChat'] = vi.fn().mockResolvedValue({ ...mockChat });
-            return { client, mockChat, mockGenerator };
+            return { client, mockChat };
         }
         describe('when compression inflates the token count', () => {
-            it('uses the truncated history for compression');
             it('allows compression to be forced/manual after a failure', async () => {
-                const { client, mockGenerator } = setup();
-                mockGenerator.countTokens?.mockResolvedValue({
+                const { client } = setup();
+                vi.mocked(mockContentGenerator.countTokens).mockResolvedValue({
                     totalTokens: 1000,
                 });
-                await client.tryCompressChat('prompt-id-4'); // Fails
+                await client.tryCompressChat('prompt-id-4', false); // Fails
                 const result = await client.tryCompressChat('prompt-id-4', true);
                 expect(result).toEqual({
                     compressionStatus: CompressionStatus.COMPRESSED,
@@ -432,21 +352,26 @@ describe('Gemini Client (client.ts)', () => {
             });
             it('yields the result even if the compression inflated the tokens', async () => {
                 const { client } = setup();
-                const result = await client.tryCompressChat('prompt-id-4', true);
+                vi.mocked(mockContentGenerator.countTokens).mockResolvedValue({
+                    totalTokens: 1000,
+                });
+                const result = await client.tryCompressChat('prompt-id-4', false);
                 expect(result).toEqual({
                     compressionStatus: CompressionStatus.COMPRESSION_FAILED_INFLATED_TOKEN_COUNT,
                     newTokenCount: 5000,
                     originalTokenCount: 1000,
                 });
+                expect(uiTelemetryService.setLastPromptTokenCount).toHaveBeenCalledWith(5000);
+                expect(uiTelemetryService.setLastPromptTokenCount).toHaveBeenCalledTimes(1);
             });
             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);
                 expect(client['chat']).toBe(mockChat); // a new chat session was not created
             });
             it('restores the history back to the original', async () => {
                 vi.mocked(tokenLimit).mockReturnValue(1000);
-                mockCountTokens.mockResolvedValue({
+                vi.mocked(mockContentGenerator.countTokens).mockResolvedValue({
                     totalTokens: 999,
                 });
                 const originalHistory = [
@@ -457,16 +382,16 @@ 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);
                 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, mockGenerator } = setup();
-                await client.tryCompressChat('prompt-id-4');
-                const result = await client.tryCompressChat('prompt-id-5');
+                const { client } = setup();
+                await client.tryCompressChat('prompt-id-4', false);
+                const result = await client.tryCompressChat('prompt-id-5', false);
                 // it counts tokens for {original, compressed} and then never again
-                expect(mockGenerator.countTokens).toHaveBeenCalledTimes(2);
+                expect(mockContentGenerator.countTokens).toHaveBeenCalledTimes(2);
                 expect(result).toEqual({
                     compressionStatus: CompressionStatus.NOOP,
                     newTokenCount: 0,
@@ -474,37 +399,17 @@ describe('Gemini Client (client.ts)', () => {
                 });
             });
         });
-        it('attempts to compress with a maxOutputTokens set to the original token count', async () => {
-            vi.mocked(tokenLimit).mockReturnValue(1000);
-            mockCountTokens.mockResolvedValue({
-                totalTokens: 999,
-            });
-            mockGetHistory.mockReturnValue([
-                { role: 'user', parts: [{ text: '...history...' }] },
-            ]);
-            // Mock the summary response from the chat
-            mockSendMessage.mockResolvedValue({
-                role: 'model',
-                parts: [{ text: 'This is a summary.' }],
-            });
-            await client.tryCompressChat('prompt-id-2', true);
-            expect(mockSendMessage).toHaveBeenCalledWith(expect.objectContaining({
-                config: expect.objectContaining({
-                    maxOutputTokens: 999,
-                }),
-            }), 'prompt-id-2');
-        });
         it('should not trigger summarization if token count is below threshold', async () => {
             const MOCKED_TOKEN_LIMIT = 1000;
             vi.mocked(tokenLimit).mockReturnValue(MOCKED_TOKEN_LIMIT);
             mockGetHistory.mockReturnValue([
                 { role: 'user', parts: [{ text: '...history...' }] },
             ]);
-            mockCountTokens.mockResolvedValue({
+            vi.mocked(mockContentGenerator.countTokens).mockResolvedValue({
                 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);
             const newChat = client.getChat();
             expect(tokenLimit).toHaveBeenCalled();
             expect(result).toEqual({
@@ -527,19 +432,27 @@ describe('Gemini Client (client.ts)', () => {
             ]);
             const originalTokenCount = MOCKED_TOKEN_LIMIT * MOCKED_CONTEXT_PERCENTAGE_THRESHOLD;
             const newTokenCount = 100;
-            mockCountTokens
+            vi.mocked(mockContentGenerator.countTokens)
                 .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);
             expect(ClearcutLogger.prototype.logChatCompressionEvent).toHaveBeenCalledWith(expect.objectContaining({
                 tokens_before: originalTokenCount,
                 tokens_after: newTokenCount,
             }));
+            expect(uiTelemetryService.setLastPromptTokenCount).toHaveBeenCalledWith(newTokenCount);
+            expect(uiTelemetryService.setLastPromptTokenCount).toHaveBeenCalledTimes(1);
         });
         it('should trigger summarization if token count is at threshold with contextPercentageThreshold setting', async () => {
             const MOCKED_TOKEN_LIMIT = 1000;
@@ -553,19 +466,25 @@ describe('Gemini Client (client.ts)', () => {
             ]);
             const originalTokenCount = MOCKED_TOKEN_LIMIT * MOCKED_CONTEXT_PERCENTAGE_THRESHOLD;
             const newTokenCount = 100;
-            mockCountTokens
+            vi.mocked(mockContentGenerator.countTokens)
                 .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);
             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,
@@ -598,19 +517,25 @@ describe('Gemini Client (client.ts)', () => {
             ]);
             const originalTokenCount = 1000 * 0.7;
             const newTokenCount = 100;
-            mockCountTokens
+            vi.mocked(mockContentGenerator.countTokens)
                 .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);
             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,
@@ -632,18 +557,24 @@ describe('Gemini Client (client.ts)', () => {
             ]);
             const originalTokenCount = 10; // Well below threshold
             const newTokenCount = 5;
-            mockCountTokens
+            vi.mocked(mockContentGenerator.countTokens)
                 .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); // force = true
             const newChat = client.getChat();
-            expect(mockSendMessage).toHaveBeenCalled();
+            expect(mockGenerateContentFn).toHaveBeenCalled();
             expect(result).toEqual({
                 compressionStatus: CompressionStatus.COMPRESSED,
                 originalTokenCount,
@@ -653,9 +584,14 @@ describe('Gemini Client (client.ts)', () => {
             expect(newChat).not.toBe(initialChat);
         });
         it('should use current model from config for token counting after sendMessage', async () => {
-            const initialModel = client['config'].getModel();
-            const mockCountTokens = vi
-                .fn()
+            const initialModel = mockConfig.getModel();
+            // mock the model has been changed between calls of `countTokens`
+            const firstCurrentModel = initialModel + '-changed-1';
+            const secondCurrentModel = initialModel + '-changed-2';
+            vi.mocked(mockConfig.getModel)
+                .mockReturnValueOnce(firstCurrentModel)
+                .mockReturnValueOnce(secondCurrentModel);
+            vi.mocked(mockContentGenerator.countTokens)
                 .mockResolvedValueOnce({ totalTokens: 100000 })
                 .mockResolvedValueOnce({ totalTokens: 5000 });
             const mockSendMessage = vi.fn().mockResolvedValue({ text: 'Summary' });
@@ -664,29 +600,19 @@ 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,
             };
-            const mockGenerator = {
-                countTokens: mockCountTokens,
-            };
-            // mock the model has been changed between calls of `countTokens`
-            const firstCurrentModel = initialModel + '-changed-1';
-            const secondCurrentModel = initialModel + '-changed-2';
-            vi.spyOn(client['config'], 'getModel')
-                .mockReturnValueOnce(firstCurrentModel)
-                .mockReturnValueOnce(secondCurrentModel);
             client['chat'] = mockChat;
-            client['contentGenerator'] = mockGenerator;
             client['startChat'] = vi.fn().mockResolvedValue(mockChat);
-            const result = await client.tryCompressChat('prompt-id-4', true);
-            expect(mockCountTokens).toHaveBeenCalledTimes(2);
-            expect(mockCountTokens).toHaveBeenNthCalledWith(1, {
+            const result = await client.tryCompressChat('prompt-id-4', false);
+            expect(mockContentGenerator.countTokens).toHaveBeenCalledTimes(2);
+            expect(mockContentGenerator.countTokens).toHaveBeenNthCalledWith(1, {
                 model: firstCurrentModel,
-                contents: mockChatHistory,
+                contents: [...mockChatHistory],
             });
-            expect(mockCountTokens).toHaveBeenNthCalledWith(2, {
+            expect(mockContentGenerator.countTokens).toHaveBeenNthCalledWith(2, {
                 model: secondCurrentModel,
                 contents: expect.any(Array),
             });
@@ -700,20 +626,9 @@ describe('Gemini Client (client.ts)', () => {
     describe('sendMessageStream', () => {
         it('emits a compression event when the context was automatically compressed', async () => {
             // Arrange
-            const mockStream = (async function* () {
+            mockTurnRunFn.mockReturnValue((async function* () {
                 yield { type: 'content', value: 'Hello' };
-            })();
-            mockTurnRunFn.mockReturnValue(mockStream);
-            const mockChat = {
-                addHistory: vi.fn(),
-                getHistory: vi.fn().mockReturnValue([]),
-            };
-            client['chat'] = mockChat;
-            const mockGenerator = {
-                countTokens: vi.fn().mockResolvedValue({ totalTokens: 0 }),
-                generateContent: mockGenerateContentFn,
-            };
-            client['contentGenerator'] = mockGenerator;
+            })());
             const compressionInfo = {
                 compressionStatus: CompressionStatus.COMPRESSED,
                 originalTokenCount: 1000,
@@ -743,16 +658,6 @@ describe('Gemini Client (client.ts)', () => {
                 yield { type: 'content', value: 'Hello' };
             })();
             mockTurnRunFn.mockReturnValue(mockStream);
-            const mockChat = {
-                addHistory: vi.fn(),
-                getHistory: vi.fn().mockReturnValue([]),
-            };
-            client['chat'] = mockChat;
-            const mockGenerator = {
-                countTokens: vi.fn().mockResolvedValue({ totalTokens: 0 }),
-                generateContent: mockGenerateContentFn,
-            };
-            client['contentGenerator'] = mockGenerator;
             const compressionInfo = {
                 compressionStatus,
                 originalTokenCount: 1000,
@@ -770,7 +675,7 @@ 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: [
                         {
@@ -791,21 +696,20 @@ describe('Gemini Client (client.ts)', () => {
                     ],
                 },
             });
-            vi.spyOn(client['config'], 'getIdeMode').mockReturnValue(true);
-            const mockStream = (async function* () {
+            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' };
-            })();
-            mockTurnRunFn.mockReturnValue(mockStream);
+            })());
             const mockChat = {
                 addHistory: vi.fn(),
                 getHistory: vi.fn().mockReturnValue([]),
             };
             client['chat'] = mockChat;
-            const mockGenerator = {
-                countTokens: vi.fn().mockResolvedValue({ totalTokens: 0 }),
-                generateContent: mockGenerateContentFn,
-            };
-            client['contentGenerator'] = mockGenerator;
             const initialRequest = [{ text: 'Hi' }];
             // Act
             const stream = client.sendMessageStream(initialRequest, new AbortController().signal, 'prompt-id-ide');
@@ -813,7 +717,7 @@ 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
@@ -838,7 +742,7 @@ ${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: [],
                 },
@@ -853,11 +757,6 @@ ${JSON.stringify({
                 getHistory: vi.fn().mockReturnValue([]),
             };
             client['chat'] = mockChat;
-            const mockGenerator = {
-                countTokens: vi.fn().mockResolvedValue({ totalTokens: 0 }),
-                generateContent: mockGenerateContentFn,
-            };
-            client['contentGenerator'] = mockGenerator;
             const initialRequest = [{ text: 'Hi' }];
             // Act
             const stream = client.sendMessageStream(initialRequest, new AbortController().signal, 'prompt-id-ide');
@@ -865,12 +764,16 @@ ${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: [
                         {
@@ -884,6 +787,11 @@ ${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' };
             })();
@@ -893,11 +801,6 @@ ${JSON.stringify({
                 getHistory: vi.fn().mockReturnValue([]),
             };
             client['chat'] = mockChat;
-            const mockGenerator = {
-                countTokens: vi.fn().mockResolvedValue({ totalTokens: 0 }),
-                generateContent: mockGenerateContentFn,
-            };
-            client['contentGenerator'] = mockGenerator;
             const initialRequest = [{ text: 'Hi' }];
             // Act
             const stream = client.sendMessageStream(initialRequest, new AbortController().signal, 'prompt-id-ide');
@@ -905,7 +808,7 @@ ${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
@@ -929,7 +832,7 @@ ${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: [
                         {
@@ -944,6 +847,11 @@ ${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' };
             })();
@@ -953,11 +861,6 @@ ${JSON.stringify({
                 getHistory: vi.fn().mockReturnValue([]),
             };
             client['chat'] = mockChat;
-            const mockGenerator = {
-                countTokens: vi.fn().mockResolvedValue({ totalTokens: 0 }),
-                generateContent: mockGenerateContentFn,
-            };
-            client['contentGenerator'] = mockGenerator;
             const initialRequest = [{ text: 'Hi' }];
             // Act
             const stream = client.sendMessageStream(initialRequest, new AbortController().signal, 'prompt-id-ide');
@@ -965,7 +868,7 @@ ${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
@@ -991,11 +894,6 @@ ${JSON.stringify({
                 getHistory: vi.fn().mockReturnValue([]),
             };
             client['chat'] = mockChat;
-            const mockGenerator = {
-                countTokens: vi.fn().mockResolvedValue({ totalTokens: 0 }),
-                generateContent: mockGenerateContentFn,
-            };
-            client['contentGenerator'] = mockGenerator;
             // Act
             const stream = client.sendMessageStream([{ text: 'Hi' }], new AbortController().signal, 'prompt-id-1');
             // Consume the stream manually to get the final return value.
@@ -1028,11 +926,6 @@ ${JSON.stringify({
                 getHistory: vi.fn().mockReturnValue([]),
             };
             client['chat'] = mockChat;
-            const mockGenerator = {
-                countTokens: vi.fn().mockResolvedValue({ totalTokens: 0 }),
-                generateContent: mockGenerateContentFn,
-            };
-            client['contentGenerator'] = mockGenerator;
             // Use a signal that never gets aborted
             const abortController = new AbortController();
             const signal = abortController.signal;
@@ -1095,11 +988,6 @@ ${JSON.stringify({
                 getHistory: vi.fn().mockReturnValue([]),
             };
             client['chat'] = mockChat;
-            const mockGenerator = {
-                countTokens: vi.fn().mockResolvedValue({ totalTokens: 0 }),
-                generateContent: mockGenerateContentFn,
-            };
-            client['contentGenerator'] = mockGenerator;
             // Act & Assert
             // Run up to the limit
             for (let i = 0; i < MAX_SESSION_TURNS; i++) {
@@ -1138,11 +1026,6 @@ ${JSON.stringify({
                 getHistory: vi.fn().mockReturnValue([]),
             };
             client['chat'] = mockChat;
-            const mockGenerator = {
-                countTokens: vi.fn().mockResolvedValue({ totalTokens: 0 }),
-                generateContent: mockGenerateContentFn,
-            };
-            client['contentGenerator'] = mockGenerator;
             // Use a signal that never gets aborted
             const abortController = new AbortController();
             const signal = abortController.signal;
@@ -1181,6 +1064,91 @@ ${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' };
@@ -1197,7 +1165,6 @@ ${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()
@@ -1206,11 +1173,6 @@ ${JSON.stringify({
                     ]),
                 };
                 client['chat'] = mockChat;
-                const mockGenerator = {
-                    countTokens: vi.fn().mockResolvedValue({ totalTokens: 0 }),
-                    generateContent: mockGenerateContentFn,
-                };
-                client['contentGenerator'] = mockGenerator;
             });
             const testCases = [
                 {
@@ -1326,7 +1288,7 @@ ${JSON.stringify({
                     },
                 };
                 // Setup current context
-                vi.mocked(ideContext.getIdeContext).mockReturnValue({
+                vi.mocked(ideContextStore.get).mockReturnValue({
                     workspaceState: {
                         openFiles: [
                             { ...currentActiveFile, isActive: true, timestamp: Date.now() },
@@ -1372,7 +1334,7 @@ ${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() },
@@ -1417,15 +1379,10 @@ ${JSON.stringify({
                     addHistory: vi.fn(),
                     getHistory: vi.fn().mockReturnValue([]), // Default empty history
                     setHistory: vi.fn(),
-                    sendMessage: vi.fn().mockResolvedValue({ text: 'summary' }),
                 };
                 client['chat'] = mockChat;
-                const mockGenerator = {
-                    countTokens: vi.fn().mockResolvedValue({ totalTokens: 0 }),
-                };
-                client['contentGenerator'] = mockGenerator;
                 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() }],
                     },
@@ -1501,7 +1458,7 @@ ${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([
                     {
@@ -1547,7 +1504,7 @@ ${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) {
@@ -1577,7 +1534,7 @@ ${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) {
@@ -1610,7 +1567,7 @@ ${JSON.stringify({
                         ],
                     },
                 };
-                vi.mocked(ideContext.getIdeContext).mockReturnValue(contextB);
+                vi.mocked(ideContextStore.get).mockReturnValue(contextB);
                 // Act: Send the tool response
                 stream = client.sendMessageStream([
                     {
@@ -1655,7 +1612,7 @@ ${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) {
@@ -1687,11 +1644,6 @@ ${JSON.stringify({
                 getHistory: vi.fn().mockReturnValue([]),
             };
             client['chat'] = mockChat;
-            const mockGenerator = {
-                countTokens: vi.fn().mockResolvedValue({ totalTokens: 0 }),
-                generateContent: mockGenerateContentFn,
-            };
-            client['contentGenerator'] = mockGenerator;
             // Act
             const stream = client.sendMessageStream([{ text: 'Hi' }], new AbortController().signal, 'prompt-id-error');
             for await (const _ of stream) {
@@ -1717,11 +1669,6 @@ ${JSON.stringify({
                 getHistory: vi.fn().mockReturnValue([]),
             };
             client['chat'] = mockChat;
-            const mockGenerator = {
-                countTokens: vi.fn().mockResolvedValue({ totalTokens: 0 }),
-                generateContent: mockGenerateContentFn,
-            };
-            client['contentGenerator'] = mockGenerator;
             // Act
             const stream = client.sendMessageStream([{ text: 'Hi' }], new AbortController().signal, 'prompt-id-error');
             for await (const _ of stream) {
@@ -1730,21 +1677,44 @@ ${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 () => {
             const contents = [{ role: 'user', parts: [{ text: 'hello' }] }];
             const generationConfig = { temperature: 0.5 };
             const abortSignal = new AbortController().signal;
-            // Mock countTokens
-            const mockGenerator = {
-                countTokens: vi.fn().mockResolvedValue({ totalTokens: 1 }),
-                generateContent: mockGenerateContentFn,
-            };
-            client['contentGenerator'] = mockGenerator;
-            await client.generateContent(contents, generationConfig, abortSignal);
-            expect(mockGenerateContentFn).toHaveBeenCalledWith({
-                model: 'test-model',
+            await client.generateContent(contents, generationConfig, abortSignal, DEFAULT_GEMINI_FLASH_MODEL);
+            expect(mockContentGenerator.generateContent).toHaveBeenCalledWith({
+                model: DEFAULT_GEMINI_FLASH_MODEL,
                 config: {
                     abortSignal,
                     systemInstruction: getCoreSystemPrompt(''),
@@ -1759,98 +1729,29 @@ ${JSON.stringify({
             const contents = [{ role: 'user', parts: [{ text: 'test' }] }];
             const currentModel = initialModel + '-changed';
             vi.spyOn(client['config'], 'getModel').mockReturnValueOnce(currentModel);
-            const mockGenerator = {
-                countTokens: vi.fn().mockResolvedValue({ totalTokens: 1 }),
-                generateContent: mockGenerateContentFn,
-            };
-            client['contentGenerator'] = mockGenerator;
-            await client.generateContent(contents, {}, new AbortController().signal);
-            expect(mockGenerateContentFn).not.toHaveBeenCalledWith({
+            await client.generateContent(contents, {}, new AbortController().signal, DEFAULT_GEMINI_FLASH_MODEL);
+            expect(mockContentGenerator.generateContent).not.toHaveBeenCalledWith({
                 model: initialModel,
                 config: expect.any(Object),
                 contents,
             });
-            expect(mockGenerateContentFn).toHaveBeenCalledWith({
-                model: currentModel,
+            expect(mockContentGenerator.generateContent).toHaveBeenCalledWith({
+                model: DEFAULT_GEMINI_FLASH_MODEL,
                 config: expect.any(Object),
                 contents,
             }, 'test-session-id');
         });
-    });
-    describe('handleFlashFallback', () => {
-        it('should use current model from config when checking for fallback', async () => {
-            const initialModel = client['config'].getModel();
-            const fallbackModel = DEFAULT_GEMINI_FLASH_MODEL;
-            // mock config been changed
-            const currentModel = initialModel + '-changed';
-            const getModelSpy = vi.spyOn(client['config'], 'getModel');
-            getModelSpy.mockReturnValue(currentModel);
-            const mockFallbackHandler = vi.fn().mockResolvedValue(true);
-            client['config'].flashFallbackHandler = mockFallbackHandler;
-            client['config'].setModel = vi.fn();
-            const result = await client['handleFlashFallback'](AuthType.LOGIN_WITH_GOOGLE);
-            expect(result).toBe(fallbackModel);
-            expect(mockFallbackHandler).toHaveBeenCalledWith(currentModel, fallbackModel, undefined);
-        });
-    });
-    describe('setHistory', () => {
-        it('should strip thought signatures when stripThoughts is true', () => {
-            const mockChat = {
-                setHistory: vi.fn(),
-            };
-            client['chat'] = mockChat;
-            const historyWithThoughts = [
-                {
-                    role: 'user',
-                    parts: [{ text: 'hello' }],
-                },
-                {
-                    role: 'model',
-                    parts: [
-                        { text: 'thinking...', thoughtSignature: 'thought-123' },
-                        {
-                            functionCall: { name: 'test', args: {} },
-                            thoughtSignature: 'thought-456',
-                        },
-                    ],
-                },
-            ];
-            client.setHistory(historyWithThoughts, { stripThoughts: true });
-            const expectedHistory = [
-                {
-                    role: 'user',
-                    parts: [{ text: 'hello' }],
-                },
-                {
-                    role: 'model',
-                    parts: [
-                        { text: 'thinking...' },
-                        { functionCall: { name: 'test', args: {} } },
-                    ],
-                },
-            ];
-            expect(mockChat.setHistory).toHaveBeenCalledWith(expectedHistory);
-        });
-        it('should not strip thought signatures when stripThoughts is false', () => {
-            const mockChat = {
-                setHistory: vi.fn(),
-            };
-            client['chat'] = mockChat;
-            const historyWithThoughts = [
-                {
-                    role: 'user',
-                    parts: [{ text: 'hello' }],
-                },
-                {
-                    role: 'model',
-                    parts: [
-                        { text: 'thinking...', thoughtSignature: 'thought-123' },
-                        { text: 'ok', thoughtSignature: 'thought-456' },
-                    ],
-                },
-            ];
-            client.setHistory(historyWithThoughts, { stripThoughts: false });
-            expect(mockChat.setHistory).toHaveBeenCalledWith(historyWithThoughts);
+        it('should use the Flash model when fallback mode is active', async () => {
+            const contents = [{ role: 'user', parts: [{ text: 'hello' }] }];
+            const generationConfig = { temperature: 0.5 };
+            const abortSignal = new AbortController().signal;
+            const requestedModel = 'gemini-2.5-pro'; // A non-flash model
+            // Mock config to be in fallback mode
+            vi.spyOn(client['config'], 'isInFallbackMode').mockReturnValue(true);
+            await client.generateContent(contents, generationConfig, abortSignal, requestedModel);
+            expect(mockGenerateContentFn).toHaveBeenCalledWith(expect.objectContaining({
+                model: DEFAULT_GEMINI_FLASH_MODEL,
+            }), 'test-session-id');
         });
     });
 });