@google/gemini-cli-core 0.0.3-preview.4 → 0.0.4

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

@@ -4,16 +4,49 @@
  * SPDX-License-Identifier: Apache-2.0
  */
 import { describe, it, expect, vi, beforeEach, afterEach } from 'vitest';
-import { GeminiChat, EmptyStreamError } from './geminiChat.js';
+import { GeminiChat, EmptyStreamError, StreamEventType, } from './geminiChat.js';
 import { setSimulate429 } from '../utils/testUtils.js';
-// Mocks
-const mockModelsModule = {
-    generateContent: vi.fn(),
-    generateContentStream: vi.fn(),
-    countTokens: vi.fn(),
-    embedContent: vi.fn(),
-    batchEmbedContents: vi.fn(),
-};
+import { DEFAULT_GEMINI_FLASH_MODEL } from '../config/models.js';
+import { AuthType } from './contentGenerator.js';
+import {} from '../utils/retry.js';
+import { Kind } from '../tools/tools.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,
+    };
+});
+const { mockHandleFallback } = vi.hoisted(() => ({
+    mockHandleFallback: vi.fn(),
+}));
+// Add mock for the retry utility
+const { mockRetryWithBackoff } = vi.hoisted(() => ({
+    mockRetryWithBackoff: vi.fn(),
+}));
+vi.mock('../utils/retry.js', () => ({
+    retryWithBackoff: mockRetryWithBackoff,
+}));
+vi.mock('../fallback/handler.js', () => ({
+    handleFallback: mockHandleFallback,
+}));
 const { mockLogInvalidChunk, mockLogContentRetry, mockLogContentRetryFailure } = vi.hoisted(() => ({
     mockLogInvalidChunk: vi.fn(),
     mockLogContentRetry: vi.fn(),
@@ -24,214 +57,178 @@ vi.mock('../telemetry/loggers.js', () => ({
     logContentRetry: mockLogContentRetry,
     logContentRetryFailure: mockLogContentRetryFailure,
 }));
+vi.mock('../telemetry/uiTelemetry.js', () => ({
+    uiTelemetryService: {
+        setLastPromptTokenCount: vi.fn(),
+    },
+}));
 describe('GeminiChat', () => {
+    let mockContentGenerator;
     let chat;
     let mockConfig;
     const config = {};
     beforeEach(() => {
         vi.clearAllMocks();
+        vi.mocked(uiTelemetryService.setLastPromptTokenCount).mockClear();
+        mockContentGenerator = {
+            generateContent: vi.fn(),
+            generateContentStream: vi.fn(),
+            countTokens: vi.fn(),
+            embedContent: vi.fn(),
+            batchEmbedContents: vi.fn(),
+        };
+        mockHandleFallback.mockClear();
+        // Default mock implementation for tests that don't care about retry logic
+        mockRetryWithBackoff.mockImplementation(async (apiCall) => apiCall());
         mockConfig = {
             getSessionId: () => 'test-session-id',
             getTelemetryLogPromptsEnabled: () => true,
             getUsageStatisticsEnabled: () => true,
             getDebugMode: () => false,
-            getContentGeneratorConfig: () => ({
-                authType: 'oauth-personal',
+            getContentGeneratorConfig: vi.fn().mockReturnValue({
+                authType: 'oauth-personal', // Ensure this is set for fallback tests
                 model: 'test-model',
             }),
             getModel: vi.fn().mockReturnValue('gemini-pro'),
             setModel: vi.fn(),
+            isInFallbackMode: vi.fn().mockReturnValue(false),
             getQuotaErrorOccurred: vi.fn().mockReturnValue(false),
             setQuotaErrorOccurred: vi.fn(),
             flashFallbackHandler: undefined,
+            getProjectRoot: vi.fn().mockReturnValue('/test/project/root'),
+            storage: {
+                getProjectTempDir: vi.fn().mockReturnValue('/test/temp'),
+            },
+            getToolRegistry: vi.fn().mockReturnValue({
+                getTool: vi.fn(),
+            }),
+            getContentGenerator: vi.fn().mockReturnValue(mockContentGenerator),
         };
         // Disable 429 simulation for tests
         setSimulate429(false);
         // Reset history for each test by creating a new instance
-        chat = new GeminiChat(mockConfig, mockModelsModule, config, []);
+        chat = new GeminiChat(mockConfig, config, []);
     });
     afterEach(() => {
         vi.restoreAllMocks();
         vi.resetAllMocks();
     });
-    describe('sendMessage', () => {
-        it('should throw an error when attempting to add a user turn after another user turn', async () => {
-            // 1. Setup: Create a history that already ends with a user turn (a functionResponse).
-            const initialHistory = [
-                { role: 'user', parts: [{ text: 'Initial prompt' }] },
-                {
-                    role: 'model',
-                    parts: [{ functionCall: { name: 'test_tool', args: {} } }],
-                },
-                {
-                    role: 'user',
-                    parts: [{ functionResponse: { name: 'test_tool', response: {} } }],
-                },
-            ];
-            chat.setHistory(initialHistory);
-            // 2. Mock a valid model response so the call doesn't fail for other reasons.
-            const mockResponse = {
-                candidates: [
-                    { content: { role: 'model', parts: [{ text: 'some response' }] } },
-                ],
-            };
-            vi.mocked(mockModelsModule.generateContent).mockResolvedValue(mockResponse);
-            // 3. Action & Assert: Expect that sending another user message immediately
-            //    after a user-role turn throws the specific error.
-            await expect(chat.sendMessage({ message: 'This is an invalid consecutive user message' }, 'prompt-id-1')).rejects.toThrow('Cannot add a user turn after another user turn.');
-        });
-        it('should preserve text parts that are in the same response as a thought', async () => {
-            // 1. Mock the API to return a single response containing both a thought and visible text.
-            const mixedContentResponse = {
-                candidates: [
-                    {
-                        content: {
-                            role: 'model',
-                            parts: [
-                                { thought: 'This is a thought.' },
-                                { text: 'This is the visible text that should not be lost.' },
-                            ],
+    describe('sendMessageStream', () => {
+        it('should succeed if a tool call is followed by an empty part', async () => {
+            // 1. Mock a stream that contains a tool call, then an invalid (empty) part.
+            const streamWithToolCall = (async function* () {
+                yield {
+                    candidates: [
+                        {
+                            content: {
+                                role: 'model',
+                                parts: [{ functionCall: { name: 'test_tool', args: {} } }],
+                            },
                         },
-                    },
-                ],
-            };
-            vi.mocked(mockModelsModule.generateContent).mockResolvedValue(mixedContentResponse);
-            // 2. Action: Send a standard, non-streaming message.
-            await chat.sendMessage({ message: 'test message' }, 'prompt-id-mixed-response');
-            // 3. Assert: Check the final state of the history.
+                    ],
+                };
+                // This second chunk is invalid according to isValidResponse
+                yield {
+                    candidates: [
+                        {
+                            content: {
+                                role: 'model',
+                                parts: [{ text: '' }],
+                            },
+                        },
+                    ],
+                };
+            })();
+            vi.mocked(mockContentGenerator.generateContentStream).mockResolvedValue(streamWithToolCall);
+            // 2. Action & Assert: The stream processing should complete without throwing an error
+            // because the presence of a tool call makes the empty final chunk acceptable.
+            const stream = await chat.sendMessageStream('test-model', { message: 'test message' }, 'prompt-id-tool-call-empty-end');
+            await expect((async () => {
+                for await (const _ of stream) {
+                    /* consume stream */
+                }
+            })()).resolves.not.toThrow();
+            // 3. Verify history was recorded correctly
             const history = chat.getHistory();
-            // The history should contain two turns: the user's message and the model's response.
-            expect(history.length).toBe(2);
+            expect(history.length).toBe(2); // user turn + model turn
             const modelTurn = history[1];
-            expect(modelTurn.role).toBe('model');
-            // CRUCIAL ASSERTION:
-            // Buggy code would discard the entire response because a "thought" was present,
-            // resulting in an empty placeholder turn with 0 parts.
-            // The corrected code will pass, preserving the single visible text part.
-            expect(modelTurn?.parts?.length).toBe(1);
-            expect(modelTurn?.parts[0].text).toBe('This is the visible text that should not be lost.');
+            expect(modelTurn?.parts?.length).toBe(1); // The empty part is discarded
+            expect(modelTurn?.parts[0].functionCall).toBeDefined();
         });
-        it('should add a placeholder model turn when a tool call is followed by an empty model response', async () => {
-            // 1. Setup: A history where the model has just made a function call.
-            const initialHistory = [
-                {
-                    role: 'user',
-                    parts: [{ text: 'Find a good Italian restaurant for me.' }],
-                },
-                {
-                    role: 'model',
-                    parts: [
+        it('should fail if the stream ends with an empty part and has no finishReason', async () => {
+            // 1. Mock a stream that ends with an invalid part and has no finish reason.
+            const streamWithNoFinish = (async function* () {
+                yield {
+                    candidates: [
                         {
-                            functionCall: {
-                                name: 'find_restaurant',
-                                args: { cuisine: 'Italian' },
+                            content: {
+                                role: 'model',
+                                parts: [{ text: 'Initial content...' }],
                             },
                         },
                     ],
-                },
-            ];
-            chat.setHistory(initialHistory);
-            // 2. Mock the API to return an empty/thought-only response.
-            const emptyModelResponse = {
-                candidates: [
-                    { content: { role: 'model', parts: [{ thought: true }] } },
-                ],
-            };
-            vi.mocked(mockModelsModule.generateContent).mockResolvedValue(emptyModelResponse);
-            // 3. Action: Send the function response back to the model.
-            await chat.sendMessage({
-                message: {
-                    functionResponse: {
-                        name: 'find_restaurant',
-                        response: { name: 'Vesuvio' },
-                    },
-                },
-            }, 'prompt-id-1');
-            // 4. Assert: The history should now have four valid, alternating turns.
-            const history = chat.getHistory();
-            expect(history.length).toBe(4);
-            // The final turn must be the empty model placeholder.
-            const lastTurn = history[3];
-            expect(lastTurn.role).toBe('model');
-            expect(lastTurn?.parts?.length).toBe(0);
-            // The second-to-last turn must be the function response we sent.
-            const secondToLastTurn = history[2];
-            expect(secondToLastTurn.role).toBe('user');
-            expect(secondToLastTurn?.parts[0].functionResponse).toBeDefined();
-        });
-        it('should call generateContent with the correct parameters', async () => {
-            const response = {
-                candidates: [
-                    {
-                        content: {
-                            parts: [{ text: 'response' }],
-                            role: 'model',
+                };
+                // This second chunk is invalid and has no finishReason, so it should fail.
+                yield {
+                    candidates: [
+                        {
+                            content: {
+                                role: 'model',
+                                parts: [{ text: '' }],
+                            },
                         },
-                        finishReason: 'STOP',
-                        index: 0,
-                        safetyRatings: [],
-                    },
-                ],
-                text: () => 'response',
-            };
-            vi.mocked(mockModelsModule.generateContent).mockResolvedValue(response);
-            await chat.sendMessage({ message: 'hello' }, 'prompt-id-1');
-            expect(mockModelsModule.generateContent).toHaveBeenCalledWith({
-                model: 'gemini-pro',
-                contents: [{ role: 'user', parts: [{ text: 'hello' }] }],
-                config: {},
-            }, 'prompt-id-1');
+                    ],
+                };
+            })();
+            vi.mocked(mockContentGenerator.generateContentStream).mockResolvedValue(streamWithNoFinish);
+            // 2. Action & Assert: The stream should fail because there's no finish reason.
+            const stream = await chat.sendMessageStream('test-model', { message: 'test message' }, 'prompt-id-no-finish-empty-end');
+            await expect((async () => {
+                for await (const _ of stream) {
+                    /* consume stream */
+                }
+            })()).rejects.toThrow(EmptyStreamError);
         });
-    });
-    describe('sendMessageStream', () => {
-        it('should not consolidate text into a part that also contains a functionCall', async () => {
-            // 1. Mock the API to stream a malformed part followed by a valid text part.
-            const multiChunkStream = (async function* () {
-                // This malformed part has both text and a functionCall.
+        it('should succeed if the stream ends with an invalid part but has a finishReason and contained a valid part', async () => {
+            // 1. Mock a stream that sends a valid chunk, then an invalid one, but has a finish reason.
+            const streamWithInvalidEnd = (async function* () {
                 yield {
                     candidates: [
                         {
                             content: {
                                 role: 'model',
-                                parts: [
-                                    {
-                                        text: 'Some text',
-                                        functionCall: { name: 'do_stuff', args: {} },
-                                    },
-                                ],
+                                parts: [{ text: 'Initial valid content...' }],
                             },
                         },
                     ],
                 };
-                // This valid text part should NOT be merged into the malformed one.
+                // This second chunk is invalid, but the response has a finishReason.
                 yield {
                     candidates: [
                         {
                             content: {
                                 role: 'model',
-                                parts: [{ text: ' that should not be merged.' }],
+                                parts: [{ text: '' }], // Invalid part
                             },
+                            finishReason: 'STOP',
                         },
                     ],
                 };
             })();
-            vi.mocked(mockModelsModule.generateContentStream).mockResolvedValue(multiChunkStream);
-            // 2. Action: Send a message and consume the stream.
-            const stream = await chat.sendMessageStream({ message: 'test message' }, 'prompt-id-malformed-chunk');
-            for await (const _ of stream) {
-                // Consume the stream to trigger history recording.
-            }
-            // 3. Assert: Check that the final history was not incorrectly consolidated.
+            vi.mocked(mockContentGenerator.generateContentStream).mockResolvedValue(streamWithInvalidEnd);
+            // 2. Action & Assert: The stream should complete without throwing an error.
+            const stream = await chat.sendMessageStream('test-model', { message: 'test message' }, 'prompt-id-valid-then-invalid-end');
+            await expect((async () => {
+                for await (const _ of stream) {
+                    /* consume stream */
+                }
+            })()).resolves.not.toThrow();
+            // 3. Verify history was recorded correctly with only the valid part.
             const history = chat.getHistory();
-            expect(history.length).toBe(2);
+            expect(history.length).toBe(2); // user turn + model turn
             const modelTurn = history[1];
-            // CRUCIAL ASSERTION: There should be two separate parts.
-            // The old, non-strict logic would incorrectly merge them, resulting in one part.
-            expect(modelTurn?.parts?.length).toBe(2);
-            // Verify the contents of each part.
-            expect(modelTurn?.parts[0].text).toBe('Some text');
-            expect(modelTurn?.parts[0].functionCall).toBeDefined();
-            expect(modelTurn?.parts[1].text).toBe(' that should not be merged.');
+            expect(modelTurn?.parts?.length).toBe(1);
+            expect(modelTurn?.parts[0].text).toBe('Initial valid content...');
         });
         it('should consolidate subsequent text chunks after receiving an empty text chunk', async () => {
             // 1. Mock the API to return a stream where one chunk is just an empty text part.
@@ -246,13 +243,16 @@ describe('GeminiChat', () => {
                 // as the important part is consolidating what comes after.
                 yield {
                     candidates: [
-                        { content: { role: 'model', parts: [{ text: ' World!' }] } },
+                        {
+                            content: { role: 'model', parts: [{ text: ' World!' }] },
+                            finishReason: 'STOP',
+                        },
                     ],
                 };
             })();
-            vi.mocked(mockModelsModule.generateContentStream).mockResolvedValue(multiChunkStream);
+            vi.mocked(mockContentGenerator.generateContentStream).mockResolvedValue(multiChunkStream);
             // 2. Action: Send a message and consume the stream.
-            const stream = await chat.sendMessageStream({ message: 'test message' }, 'prompt-id-empty-chunk-consolidation');
+            const stream = await chat.sendMessageStream('test-model', { message: 'test message' }, 'prompt-id-empty-chunk-consolidation');
             for await (const _ of stream) {
                 // Consume the stream
             }
@@ -298,9 +298,9 @@ describe('GeminiChat', () => {
                     ],
                 };
             })();
-            vi.mocked(mockModelsModule.generateContentStream).mockResolvedValue(multiChunkStream);
+            vi.mocked(mockContentGenerator.generateContentStream).mockResolvedValue(multiChunkStream);
             // 2. Action: Send a message and consume the stream.
-            const stream = await chat.sendMessageStream({ message: 'test message' }, 'prompt-id-multi-chunk');
+            const stream = await chat.sendMessageStream('test-model', { message: 'test message' }, 'prompt-id-multi-chunk');
             for await (const _ of stream) {
                 // Consume the stream to trigger history recording.
             }
@@ -329,13 +329,14 @@ describe('GeminiChat', () => {
                                     { text: 'This is the visible text that should not be lost.' },
                                 ],
                             },
+                            finishReason: 'STOP',
                         },
                     ],
                 };
             })();
-            vi.mocked(mockModelsModule.generateContentStream).mockResolvedValue(mixedContentStream);
+            vi.mocked(mockContentGenerator.generateContentStream).mockResolvedValue(mixedContentStream);
             // 2. Action: Send a message and fully consume the stream to trigger history recording.
-            const stream = await chat.sendMessageStream({ message: 'test message' }, 'prompt-id-mixed-chunk');
+            const stream = await chat.sendMessageStream('test-model', { message: 'test message' }, 'prompt-id-mixed-chunk');
             for await (const _ of stream) {
                 // This loop consumes the stream.
             }
@@ -375,13 +376,16 @@ describe('GeminiChat', () => {
             const emptyStreamResponse = (async function* () {
                 yield {
                     candidates: [
-                        { content: { role: 'model', parts: [{ thought: true }] } },
+                        {
+                            content: { role: 'model', parts: [{ thought: true }] },
+                            finishReason: 'STOP',
+                        },
                     ],
                 };
             })();
-            vi.mocked(mockModelsModule.generateContentStream).mockResolvedValue(emptyStreamResponse);
+            vi.mocked(mockContentGenerator.generateContentStream).mockResolvedValue(emptyStreamResponse);
             // 3. Action: Send the function response back to the model and consume the stream.
-            const stream = await chat.sendMessageStream({
+            const stream = await chat.sendMessageStream('test-model', {
                 message: {
                     functionResponse: {
                         name: 'find_restaurant',
@@ -419,147 +423,31 @@ describe('GeminiChat', () => {
                         },
                     ],
                     text: () => 'response',
+                    usageMetadata: {
+                        promptTokenCount: 42,
+                        candidatesTokenCount: 15,
+                        totalTokenCount: 57,
+                    },
                 };
             })();
-            vi.mocked(mockModelsModule.generateContentStream).mockResolvedValue(response);
-            const stream = await chat.sendMessageStream({ message: 'hello' }, 'prompt-id-1');
+            vi.mocked(mockContentGenerator.generateContentStream).mockResolvedValue(response);
+            const stream = await chat.sendMessageStream('test-model', { message: 'hello' }, 'prompt-id-1');
             for await (const _ of stream) {
-                // consume stream to trigger internal logic
+                // consume stream
             }
-            expect(mockModelsModule.generateContentStream).toHaveBeenCalledWith({
-                model: 'gemini-pro',
-                contents: [{ role: 'user', parts: [{ text: 'hello' }] }],
+            expect(mockContentGenerator.generateContentStream).toHaveBeenCalledWith({
+                model: 'test-model',
+                contents: [
+                    {
+                        role: 'user',
+                        parts: [{ text: 'hello' }],
+                    },
+                ],
                 config: {},
             }, 'prompt-id-1');
-        });
-    });
-    describe('recordHistory', () => {
-        const userInput = {
-            role: 'user',
-            parts: [{ text: 'User input' }],
-        };
-        it('should consolidate all consecutive model turns into a single turn', () => {
-            const userInput = {
-                role: 'user',
-                parts: [{ text: 'User input' }],
-            };
-            // This simulates a multi-part model response with different part types.
-            const modelOutput = [
-                { role: 'model', parts: [{ text: 'Thinking...' }] },
-                {
-                    role: 'model',
-                    parts: [{ functionCall: { name: 'do_stuff', args: {} } }],
-                },
-            ];
-            // @ts-expect-error Accessing private method for testing
-            chat.recordHistory(userInput, modelOutput);
-            const history = chat.getHistory();
-            // The history should contain the user's turn and ONE consolidated model turn.
-            // The old code would fail here, resulting in a length of 3.
-            //expect(history).toBe([]);
-            expect(history.length).toBe(2);
-            const modelTurn = history[1];
-            expect(modelTurn.role).toBe('model');
-            // The consolidated turn should contain both the text part and the functionCall part.
-            expect(modelTurn?.parts?.length).toBe(2);
-            expect(modelTurn?.parts[0].text).toBe('Thinking...');
-            expect(modelTurn?.parts[1].functionCall).toBeDefined();
-        });
-        it('should add a placeholder model turn when a tool call is followed by an empty response', () => {
-            // 1. Setup: A history where the model has just made a function call.
-            const initialHistory = [
-                { role: 'user', parts: [{ text: 'Initial prompt' }] },
-                {
-                    role: 'model',
-                    parts: [{ functionCall: { name: 'test_tool', args: {} } }],
-                },
-            ];
-            chat.setHistory(initialHistory);
-            // 2. Action: The user provides the tool's response, and the model's
-            // final output is empty (e.g., just a thought, which gets filtered out).
-            const functionResponse = {
-                role: 'user',
-                parts: [{ functionResponse: { name: 'test_tool', response: {} } }],
-            };
-            const emptyModelOutput = [];
-            // @ts-expect-error Accessing private method for testing
-            chat.recordHistory(functionResponse, emptyModelOutput, [
-                functionResponse,
-            ]);
-            // 3. Assert: The history should now have four valid, alternating turns.
-            const history = chat.getHistory();
-            expect(history.length).toBe(4);
-            // The final turn must be the empty model placeholder.
-            const lastTurn = history[3];
-            expect(lastTurn.role).toBe('model');
-            expect(lastTurn?.parts?.length).toBe(0);
-            // The second-to-last turn must be the function response we provided.
-            const secondToLastTurn = history[2];
-            expect(secondToLastTurn.role).toBe('user');
-            expect(secondToLastTurn?.parts[0].functionResponse).toBeDefined();
-        });
-        it('should add user input and a single model output to history', () => {
-            const modelOutput = [
-                { role: 'model', parts: [{ text: 'Model output' }] },
-            ];
-            // @ts-expect-error Accessing private method for testing
-            chat.recordHistory(userInput, modelOutput);
-            const history = chat.getHistory();
-            expect(history.length).toBe(2);
-            expect(history[0]).toEqual(userInput);
-            expect(history[1]).toEqual(modelOutput[0]);
-        });
-        it('should consolidate adjacent text parts from multiple content objects', () => {
-            const modelOutput = [
-                { role: 'model', parts: [{ text: 'Part 1.' }] },
-                { role: 'model', parts: [{ text: ' Part 2.' }] },
-                { role: 'model', parts: [{ text: ' Part 3.' }] },
-            ];
-            // @ts-expect-error Accessing private method for testing
-            chat.recordHistory(userInput, modelOutput);
-            const history = chat.getHistory();
-            expect(history.length).toBe(2);
-            expect(history[1].role).toBe('model');
-            expect(history[1].parts).toEqual([{ text: 'Part 1. Part 2. Part 3.' }]);
-        });
-        it('should add an empty placeholder turn if modelOutput is empty', () => {
-            // This simulates receiving a pre-filtered, thought-only response.
-            const emptyModelOutput = [];
-            // @ts-expect-error Accessing private method for testing
-            chat.recordHistory(userInput, emptyModelOutput);
-            const history = chat.getHistory();
-            expect(history.length).toBe(2);
-            expect(history[0]).toEqual(userInput);
-            expect(history[1].role).toBe('model');
-            expect(history[1].parts).toEqual([]);
-        });
-        it('should preserve model outputs with undefined or empty parts arrays', () => {
-            const malformedOutput = [
-                { role: 'model', parts: [{ text: 'Text part' }] },
-                { role: 'model', parts: undefined },
-                { role: 'model', parts: [] },
-            ];
-            // @ts-expect-error Accessing private method for testing
-            chat.recordHistory(userInput, malformedOutput);
-            const history = chat.getHistory();
-            expect(history.length).toBe(4); // userInput + 3 model turns
-            expect(history[1].parts).toEqual([{ text: 'Text part' }]);
-            expect(history[2].parts).toBeUndefined();
-            expect(history[3].parts).toEqual([]);
-        });
-        it('should not consolidate content with different roles', () => {
-            const mixedOutput = [
-                { role: 'model', parts: [{ text: 'Model 1' }] },
-                { role: 'user', parts: [{ text: 'Unexpected User' }] },
-                { role: 'model', parts: [{ text: 'Model 2' }] },
-            ];
-            // @ts-expect-error Accessing private method for testing
-            chat.recordHistory(userInput, mixedOutput);
-            const history = chat.getHistory();
-            expect(history.length).toBe(4); // userInput, model1, unexpected_user, model2
-            expect(history[1]).toEqual(mixedOutput[0]);
-            expect(history[2]).toEqual(mixedOutput[1]);
-            expect(history[3]).toEqual(mixedOutput[2]);
+            // Verify that token counting is called when usageMetadata is present
+            expect(uiTelemetryService.setLastPromptTokenCount).toHaveBeenCalledWith(42);
+            expect(uiTelemetryService.setLastPromptTokenCount).toHaveBeenCalledTimes(1);
         });
     });
     describe('addHistory', () => {
@@ -591,9 +479,42 @@ describe('GeminiChat', () => {
         });
     });
     describe('sendMessageStream with retries', () => {
+        it('should yield a RETRY event when an invalid stream is encountered', async () => {
+            // ARRANGE: Mock the stream to fail once, then succeed.
+            vi.mocked(mockContentGenerator.generateContentStream)
+                .mockImplementationOnce(async () => 
+            // First attempt: An invalid stream with an empty text part.
+            (async function* () {
+                yield {
+                    candidates: [{ content: { parts: [{ text: '' }] } }],
+                };
+            })())
+                .mockImplementationOnce(async () => 
+            // Second attempt (the retry): A minimal valid stream.
+            (async function* () {
+                yield {
+                    candidates: [
+                        {
+                            content: { parts: [{ text: 'Success' }] },
+                            finishReason: 'STOP',
+                        },
+                    ],
+                };
+            })());
+            // ACT: Send a message and collect all events from the stream.
+            const stream = await chat.sendMessageStream('test-model', { message: 'test' }, 'prompt-id-yield-retry');
+            const events = [];
+            for await (const event of stream) {
+                events.push(event);
+            }
+            // ASSERT: Check that a RETRY event was present in the stream's output.
+            const retryEvent = events.find((e) => e.type === StreamEventType.RETRY);
+            expect(retryEvent).toBeDefined();
+            expect(retryEvent?.type).toBe(StreamEventType.RETRY);
+        });
         it('should retry on invalid content, succeed, and report metrics', async () => {
             // Use mockImplementationOnce to provide a fresh, promise-wrapped generator for each attempt.
-            vi.mocked(mockModelsModule.generateContentStream)
+            vi.mocked(mockContentGenerator.generateContentStream)
                 .mockImplementationOnce(async () => 
             // First call returns an invalid stream
             (async function* () {
@@ -606,11 +527,14 @@ describe('GeminiChat', () => {
             (async function* () {
                 yield {
                     candidates: [
-                        { content: { parts: [{ text: 'Successful response' }] } },
+                        {
+                            content: { parts: [{ text: 'Successful response' }] },
+                            finishReason: 'STOP',
+                        },
                     ],
                 };
             })());
-            const stream = await chat.sendMessageStream({ message: 'test' }, 'prompt-id-retry-success');
+            const stream = await chat.sendMessageStream('test-model', { message: 'test' }, 'prompt-id-retry-success');
             const chunks = [];
             for await (const chunk of stream) {
                 chunks.push(chunk);
@@ -619,9 +543,13 @@ describe('GeminiChat', () => {
             expect(mockLogInvalidChunk).toHaveBeenCalledTimes(1);
             expect(mockLogContentRetry).toHaveBeenCalledTimes(1);
             expect(mockLogContentRetryFailure).not.toHaveBeenCalled();
-            expect(mockModelsModule.generateContentStream).toHaveBeenCalledTimes(2);
-            expect(chunks.some((c) => c.candidates?.[0]?.content?.parts?.[0]?.text ===
-                'Successful response')).toBe(true);
+            expect(mockContentGenerator.generateContentStream).toHaveBeenCalledTimes(2);
+            // Check for a retry event
+            expect(chunks.some((c) => c.type === StreamEventType.RETRY)).toBe(true);
+            // Check for the successful content chunk
+            expect(chunks.some((c) => c.type === StreamEventType.CHUNK &&
+                c.value.candidates?.[0]?.content?.parts?.[0]?.text ===
+                    'Successful response')).toBe(true);
             // Check that history was recorded correctly once, with no duplicates.
             const history = chat.getHistory();
             expect(history.length).toBe(2);
@@ -633,9 +561,11 @@ describe('GeminiChat', () => {
                 role: 'model',
                 parts: [{ text: 'Successful response' }],
             });
+            // Verify that token counting is not called when usageMetadata is missing
+            expect(uiTelemetryService.setLastPromptTokenCount).not.toHaveBeenCalled();
         });
         it('should fail after all retries on persistent invalid content and report metrics', async () => {
-            vi.mocked(mockModelsModule.generateContentStream).mockImplementation(async () => (async function* () {
+            vi.mocked(mockContentGenerator.generateContentStream).mockImplementation(async () => (async function* () {
                 yield {
                     candidates: [
                         {
@@ -647,16 +577,14 @@ describe('GeminiChat', () => {
                     ],
                 };
             })());
-            // This helper function consumes the stream and allows us to test for rejection.
-            async function consumeStreamAndExpectError() {
-                const stream = await chat.sendMessageStream({ message: 'test' }, 'prompt-id-retry-fail');
+            const stream = await chat.sendMessageStream('test-model', { message: 'test' }, 'prompt-id-retry-fail');
+            await expect(async () => {
                 for await (const _ of stream) {
                     // Must loop to trigger the internal logic that throws.
                 }
-            }
-            await expect(consumeStreamAndExpectError()).rejects.toThrow(EmptyStreamError);
+            }).rejects.toThrow(EmptyStreamError);
             // Should be called 3 times (initial + 2 retries)
-            expect(mockModelsModule.generateContentStream).toHaveBeenCalledTimes(3);
+            expect(mockContentGenerator.generateContentStream).toHaveBeenCalledTimes(3);
             expect(mockLogInvalidChunk).toHaveBeenCalledTimes(3);
             expect(mockLogContentRetry).toHaveBeenCalledTimes(2);
             expect(mockLogContentRetryFailure).toHaveBeenCalledTimes(1);
@@ -673,7 +601,7 @@ describe('GeminiChat', () => {
         ];
         chat.setHistory(initialHistory);
         // 2. Mock the API to fail once with an empty stream, then succeed.
-        vi.mocked(mockModelsModule.generateContentStream)
+        vi.mocked(mockContentGenerator.generateContentStream)
             .mockImplementationOnce(async () => (async function* () {
             yield {
                 candidates: [{ content: { parts: [{ text: '' }] } }],
@@ -683,11 +611,16 @@ describe('GeminiChat', () => {
         // Second attempt succeeds
         (async function* () {
             yield {
-                candidates: [{ content: { parts: [{ text: 'Second answer' }] } }],
+                candidates: [
+                    {
+                        content: { parts: [{ text: 'Second answer' }] },
+                        finishReason: 'STOP',
+                    },
+                ],
             };
         })());
         // 3. Send a new message
-        const stream = await chat.sendMessageStream({ message: 'Second question' }, 'prompt-id-retry-existing');
+        const stream = await chat.sendMessageStream('test-model', { message: 'Second question' }, 'prompt-id-retry-existing');
         for await (const _ of stream) {
             // consume stream
         }
@@ -718,63 +651,9 @@ describe('GeminiChat', () => {
         }
         expect(turn4.parts[0].text).toBe('Second answer');
     });
-    describe('concurrency control', () => {
-        it('should queue a subsequent sendMessage call until the first one completes', async () => {
-            // 1. Create promises to manually control when the API calls resolve
-            let firstCallResolver;
-            const firstCallPromise = new Promise((resolve) => {
-                firstCallResolver = resolve;
-            });
-            let secondCallResolver;
-            const secondCallPromise = new Promise((resolve) => {
-                secondCallResolver = resolve;
-            });
-            // A standard response body for the mock
-            const mockResponse = {
-                candidates: [
-                    {
-                        content: { parts: [{ text: 'response' }], role: 'model' },
-                    },
-                ],
-            };
-            // 2. Mock the API to return our controllable promises in order
-            vi.mocked(mockModelsModule.generateContent)
-                .mockReturnValueOnce(firstCallPromise)
-                .mockReturnValueOnce(secondCallPromise);
-            // 3. Start the first message call. Do not await it yet.
-            const firstMessagePromise = chat.sendMessage({ message: 'first' }, 'prompt-1');
-            // Give the event loop a chance to run the async call up to the `await`
-            await new Promise(process.nextTick);
-            // 4. While the first call is "in-flight", start the second message call.
-            const secondMessagePromise = chat.sendMessage({ message: 'second' }, 'prompt-2');
-            // 5. CRUCIAL CHECK: At this point, only the first API call should have been made.
-            // The second call should be waiting on `sendPromise`.
-            expect(mockModelsModule.generateContent).toHaveBeenCalledTimes(1);
-            expect(mockModelsModule.generateContent).toHaveBeenCalledWith(expect.objectContaining({
-                contents: expect.arrayContaining([
-                    expect.objectContaining({ parts: [{ text: 'first' }] }),
-                ]),
-            }), 'prompt-1');
-            // 6. Unblock the first API call and wait for the first message to fully complete.
-            firstCallResolver(mockResponse);
-            await firstMessagePromise;
-            // Give the event loop a chance to unblock and run the second call.
-            await new Promise(process.nextTick);
-            // 7. CRUCIAL CHECK: Now, the second API call should have been made.
-            expect(mockModelsModule.generateContent).toHaveBeenCalledTimes(2);
-            expect(mockModelsModule.generateContent).toHaveBeenCalledWith(expect.objectContaining({
-                contents: expect.arrayContaining([
-                    expect.objectContaining({ parts: [{ text: 'second' }] }),
-                ]),
-            }), 'prompt-2');
-            // 8. Clean up by resolving the second call.
-            secondCallResolver(mockResponse);
-            await secondMessagePromise;
-        });
-    });
     it('should retry if the model returns a completely empty stream (no chunks)', async () => {
         // 1. Mock the API to return an empty stream first, then a valid one.
-        vi.mocked(mockModelsModule.generateContentStream)
+        vi.mocked(mockContentGenerator.generateContentStream)
             .mockImplementationOnce(
         // First call resolves to an async generator that yields nothing.
         async () => (async function* () { })())
@@ -787,20 +666,22 @@ describe('GeminiChat', () => {
                         content: {
                             parts: [{ text: 'Successful response after empty' }],
                         },
+                        finishReason: 'STOP',
                     },
                 ],
             };
         })());
         // 2. Call the method and consume the stream.
-        const stream = await chat.sendMessageStream({ message: 'test empty stream' }, 'prompt-id-empty-stream');
+        const stream = await chat.sendMessageStream('test-model', { message: 'test empty stream' }, 'prompt-id-empty-stream');
         const chunks = [];
         for await (const chunk of stream) {
             chunks.push(chunk);
         }
         // 3. Assert the results.
-        expect(mockModelsModule.generateContentStream).toHaveBeenCalledTimes(2);
-        expect(chunks.some((c) => c.candidates?.[0]?.content?.parts?.[0]?.text ===
-            'Successful response after empty')).toBe(true);
+        expect(mockContentGenerator.generateContentStream).toHaveBeenCalledTimes(2);
+        expect(chunks.some((c) => c.type === StreamEventType.CHUNK &&
+            c.value.candidates?.[0]?.content?.parts?.[0]?.text ===
+                'Successful response after empty')).toBe(true);
         const history = chat.getHistory();
         expect(history.length).toBe(2);
         // Explicitly verify the structure of each part to satisfy TypeScript
@@ -830,25 +711,35 @@ describe('GeminiChat', () => {
             };
             await firstStreamContinuePromise; // Pause the stream
             yield {
-                candidates: [{ content: { parts: [{ text: ' part 2' }] } }],
+                candidates: [
+                    {
+                        content: { parts: [{ text: ' part 2' }] },
+                        finishReason: 'STOP',
+                    },
+                ],
             };
         })();
         const secondStreamGenerator = (async function* () {
             yield {
-                candidates: [{ content: { parts: [{ text: 'second response' }] } }],
+                candidates: [
+                    {
+                        content: { parts: [{ text: 'second response' }] },
+                        finishReason: 'STOP',
+                    },
+                ],
             };
         })();
-        vi.mocked(mockModelsModule.generateContentStream)
+        vi.mocked(mockContentGenerator.generateContentStream)
             .mockResolvedValueOnce(firstStreamGenerator)
             .mockResolvedValueOnce(secondStreamGenerator);
         // 3. Start the first stream and consume only the first chunk to pause it
-        const firstStream = await chat.sendMessageStream({ message: 'first' }, 'prompt-1');
+        const firstStream = await chat.sendMessageStream('test-model', { message: 'first' }, 'prompt-1');
         const firstStreamIterator = firstStream[Symbol.asyncIterator]();
         await firstStreamIterator.next();
         // 4. While the first stream is paused, start the second call. It will block.
-        const secondStreamPromise = chat.sendMessageStream({ message: 'second' }, 'prompt-2');
+        const secondStreamPromise = chat.sendMessageStream('test-model', { message: 'second' }, 'prompt-2');
         // 5. Assert that only one API call has been made so far.
-        expect(mockModelsModule.generateContentStream).toHaveBeenCalledTimes(1);
+        expect(mockContentGenerator.generateContentStream).toHaveBeenCalledTimes(1);
         // 6. Unblock and fully consume the first stream to completion.
         continueFirstStream();
         await firstStreamIterator.next(); // Consume the rest of the stream
@@ -859,7 +750,7 @@ describe('GeminiChat', () => {
         const secondStreamIterator = secondStream[Symbol.asyncIterator]();
         await secondStreamIterator.next();
         // 9. The second API call should now have been made.
-        expect(mockModelsModule.generateContentStream).toHaveBeenCalledTimes(2);
+        expect(mockContentGenerator.generateContentStream).toHaveBeenCalledTimes(2);
         // 10. FIX: Fully consume the second stream to ensure recordHistory is called.
         await secondStreamIterator.next(); // This finishes the iterator.
         // 11. Final check on history.
@@ -871,5 +762,402 @@ describe('GeminiChat', () => {
         }
         expect(turn4.parts[0].text).toBe('second response');
     });
+    describe('stopBeforeSecondMutator', () => {
+        beforeEach(() => {
+            // Common setup for these tests: mock the tool registry.
+            const mockToolRegistry = {
+                getTool: vi.fn((toolName) => {
+                    if (toolName === 'edit') {
+                        return { kind: Kind.Edit };
+                    }
+                    return { kind: Kind.Other };
+                }),
+            };
+            vi.mocked(mockConfig.getToolRegistry).mockReturnValue(mockToolRegistry);
+        });
+        it('should stop streaming before a second mutator tool call', async () => {
+            const responses = [
+                {
+                    candidates: [
+                        { content: { role: 'model', parts: [{ text: 'First part. ' }] } },
+                    ],
+                },
+                {
+                    candidates: [
+                        {
+                            content: {
+                                role: 'model',
+                                parts: [{ functionCall: { name: 'edit', args: {} } }],
+                            },
+                        },
+                    ],
+                },
+                {
+                    candidates: [
+                        {
+                            content: {
+                                role: 'model',
+                                parts: [{ functionCall: { name: 'fetch', args: {} } }],
+                            },
+                        },
+                    ],
+                },
+                // This chunk contains the second mutator and should be clipped.
+                {
+                    candidates: [
+                        {
+                            content: {
+                                role: 'model',
+                                parts: [
+                                    { functionCall: { name: 'edit', args: {} } },
+                                    { text: 'some trailing text' },
+                                ],
+                            },
+                        },
+                    ],
+                },
+                // This chunk should never be reached.
+                {
+                    candidates: [
+                        {
+                            content: {
+                                role: 'model',
+                                parts: [{ text: 'This should not appear.' }],
+                            },
+                        },
+                    ],
+                },
+            ];
+            vi.mocked(mockContentGenerator.generateContentStream).mockResolvedValue((async function* () {
+                for (const response of responses) {
+                    yield response;
+                }
+            })());
+            const stream = await chat.sendMessageStream('test-model', { message: 'test message' }, 'prompt-id-mutator-test');
+            for await (const _ of stream) {
+                // Consume the stream to trigger history recording.
+            }
+            const history = chat.getHistory();
+            expect(history.length).toBe(2);
+            const modelTurn = history[1];
+            expect(modelTurn.role).toBe('model');
+            expect(modelTurn?.parts?.length).toBe(3);
+            expect(modelTurn?.parts[0].text).toBe('First part. ');
+            expect(modelTurn.parts[1].functionCall?.name).toBe('edit');
+            expect(modelTurn.parts[2].functionCall?.name).toBe('fetch');
+        });
+        it('should not stop streaming if only one mutator is present', async () => {
+            const responses = [
+                {
+                    candidates: [
+                        { content: { role: 'model', parts: [{ text: 'Part 1. ' }] } },
+                    ],
+                },
+                {
+                    candidates: [
+                        {
+                            content: {
+                                role: 'model',
+                                parts: [{ functionCall: { name: 'edit', args: {} } }],
+                            },
+                        },
+                    ],
+                },
+                {
+                    candidates: [
+                        {
+                            content: {
+                                role: 'model',
+                                parts: [{ text: 'Part 2.' }],
+                            },
+                            finishReason: 'STOP',
+                        },
+                    ],
+                },
+            ];
+            vi.mocked(mockContentGenerator.generateContentStream).mockResolvedValue((async function* () {
+                for (const response of responses) {
+                    yield response;
+                }
+            })());
+            const stream = await chat.sendMessageStream('test-model', { message: 'test message' }, 'prompt-id-one-mutator');
+            for await (const _ of stream) {
+                /* consume */
+            }
+            const history = chat.getHistory();
+            const modelTurn = history[1];
+            expect(modelTurn?.parts?.length).toBe(3);
+            expect(modelTurn.parts[1].functionCall?.name).toBe('edit');
+            expect(modelTurn.parts[2].text).toBe('Part 2.');
+        });
+        it('should clip the chunk containing the second mutator, preserving prior parts', async () => {
+            const responses = [
+                {
+                    candidates: [
+                        {
+                            content: {
+                                role: 'model',
+                                parts: [{ functionCall: { name: 'edit', args: {} } }],
+                            },
+                        },
+                    ],
+                },
+                // This chunk has a valid part before the second mutator.
+                // The valid part should be kept, the rest of the chunk discarded.
+                {
+                    candidates: [
+                        {
+                            content: {
+                                role: 'model',
+                                parts: [
+                                    { text: 'Keep this text. ' },
+                                    { functionCall: { name: 'edit', args: {} } },
+                                    { text: 'Discard this text.' },
+                                ],
+                            },
+                            finishReason: 'STOP',
+                        },
+                    ],
+                },
+            ];
+            const stream = (async function* () {
+                for (const response of responses) {
+                    yield response;
+                }
+            })();
+            vi.mocked(mockContentGenerator.generateContentStream).mockResolvedValue(stream);
+            const resultStream = await chat.sendMessageStream('test-model', { message: 'test' }, 'prompt-id-clip-chunk');
+            for await (const _ of resultStream) {
+                /* consume */
+            }
+            const history = chat.getHistory();
+            const modelTurn = history[1];
+            expect(modelTurn?.parts?.length).toBe(2);
+            expect(modelTurn.parts[0].functionCall?.name).toBe('edit');
+            expect(modelTurn.parts[1].text).toBe('Keep this text. ');
+        });
+        it('should handle two mutators in the same chunk (parallel call scenario)', async () => {
+            const responses = [
+                {
+                    candidates: [
+                        {
+                            content: {
+                                role: 'model',
+                                parts: [
+                                    { text: 'Some text. ' },
+                                    { functionCall: { name: 'edit', args: {} } },
+                                    { functionCall: { name: 'edit', args: {} } },
+                                ],
+                            },
+                            finishReason: 'STOP',
+                        },
+                    ],
+                },
+            ];
+            const stream = (async function* () {
+                for (const response of responses) {
+                    yield response;
+                }
+            })();
+            vi.mocked(mockContentGenerator.generateContentStream).mockResolvedValue(stream);
+            const resultStream = await chat.sendMessageStream('test-model', { message: 'test' }, 'prompt-id-parallel-mutators');
+            for await (const _ of resultStream) {
+                /* consume */
+            }
+            const history = chat.getHistory();
+            const modelTurn = history[1];
+            expect(modelTurn?.parts?.length).toBe(2);
+            expect(modelTurn.parts[0].text).toBe('Some text. ');
+            expect(modelTurn.parts[1].functionCall?.name).toBe('edit');
+        });
+    });
+    describe('Model Resolution', () => {
+        const mockResponse = {
+            candidates: [
+                {
+                    content: { parts: [{ text: 'response' }], role: 'model' },
+                    finishReason: 'STOP',
+                },
+            ],
+        };
+        it('should use the FLASH model when in fallback mode (sendMessageStream)', async () => {
+            vi.mocked(mockConfig.getModel).mockReturnValue('gemini-pro');
+            vi.mocked(mockConfig.isInFallbackMode).mockReturnValue(true);
+            vi.mocked(mockContentGenerator.generateContentStream).mockImplementation(async () => (async function* () {
+                yield mockResponse;
+            })());
+            const stream = await chat.sendMessageStream('test-model', { message: 'test' }, 'prompt-id-res3');
+            for await (const _ of stream) {
+                // consume stream
+            }
+            expect(mockContentGenerator.generateContentStream).toHaveBeenCalledWith(expect.objectContaining({
+                model: DEFAULT_GEMINI_FLASH_MODEL,
+            }), 'prompt-id-res3');
+        });
+    });
+    describe('Fallback Integration (Retries)', () => {
+        const error429 = Object.assign(new Error('API Error 429: Quota exceeded'), {
+            status: 429,
+        });
+        // Define the simulated behavior for retryWithBackoff for these tests.
+        // This simulation tries the apiCall, if it fails, it calls the callback,
+        // and then tries the apiCall again if the callback returns true.
+        const simulateRetryBehavior = async (apiCall, options) => {
+            try {
+                return await apiCall();
+            }
+            catch (error) {
+                if (options.onPersistent429) {
+                    // We simulate the "persistent" trigger here for simplicity.
+                    const shouldRetry = await options.onPersistent429(options.authType, error);
+                    if (shouldRetry) {
+                        return await apiCall();
+                    }
+                }
+                throw error; // Stop if callback returns false/null or doesn't exist
+            }
+        };
+        beforeEach(() => {
+            mockRetryWithBackoff.mockImplementation(simulateRetryBehavior);
+        });
+        afterEach(() => {
+            mockRetryWithBackoff.mockImplementation(async (apiCall) => apiCall());
+        });
+        it('should call handleFallback with the specific failed model and retry if handler returns true', async () => {
+            const authType = AuthType.LOGIN_WITH_GOOGLE;
+            vi.mocked(mockConfig.getContentGeneratorConfig).mockReturnValue({
+                authType,
+            });
+            const isInFallbackModeSpy = vi.spyOn(mockConfig, 'isInFallbackMode');
+            isInFallbackModeSpy.mockReturnValue(false);
+            vi.mocked(mockContentGenerator.generateContentStream)
+                .mockRejectedValueOnce(error429) // Attempt 1 fails
+                .mockResolvedValueOnce(
+            // Attempt 2 succeeds
+            (async function* () {
+                yield {
+                    candidates: [
+                        {
+                            content: { parts: [{ text: 'Success on retry' }] },
+                            finishReason: 'STOP',
+                        },
+                    ],
+                };
+            })());
+            mockHandleFallback.mockImplementation(async () => {
+                isInFallbackModeSpy.mockReturnValue(true);
+                return true; // Signal retry
+            });
+            const stream = await chat.sendMessageStream('test-model', { message: 'trigger 429' }, 'prompt-id-fb1');
+            // Consume stream to trigger logic
+            for await (const _ of stream) {
+                // no-op
+            }
+            expect(mockContentGenerator.generateContentStream).toHaveBeenCalledTimes(2);
+            expect(mockHandleFallback).toHaveBeenCalledTimes(1);
+            expect(mockHandleFallback).toHaveBeenCalledWith(mockConfig, 'test-model', authType, error429);
+            const history = chat.getHistory();
+            const modelTurn = history[1];
+            expect(modelTurn.parts[0].text).toBe('Success on retry');
+        });
+        it('should stop retrying if handleFallback returns false (e.g., auth intent)', async () => {
+            vi.mocked(mockConfig.getModel).mockReturnValue('gemini-pro');
+            vi.mocked(mockContentGenerator.generateContentStream).mockRejectedValue(error429);
+            mockHandleFallback.mockResolvedValue(false);
+            const stream = await chat.sendMessageStream('test-model', { message: 'test stop' }, 'prompt-id-fb2');
+            await expect((async () => {
+                for await (const _ of stream) {
+                    /* consume stream */
+                }
+            })()).rejects.toThrow(error429);
+            expect(mockContentGenerator.generateContentStream).toHaveBeenCalledTimes(1);
+            expect(mockHandleFallback).toHaveBeenCalledTimes(1);
+        });
+    });
+    it('should discard valid partial content from a failed attempt upon retry', async () => {
+        // Mock the stream to fail on the first attempt after yielding some valid content.
+        vi.mocked(mockContentGenerator.generateContentStream)
+            .mockImplementationOnce(async () => 
+        // First attempt: yields one valid chunk, then one invalid chunk
+        (async function* () {
+            yield {
+                candidates: [
+                    {
+                        content: {
+                            parts: [{ text: 'This valid part should be discarded' }],
+                        },
+                    },
+                ],
+            };
+            yield {
+                candidates: [{ content: { parts: [{ text: '' }] } }], // Invalid chunk triggers retry
+            };
+        })())
+            .mockImplementationOnce(async () => 
+        // Second attempt (the retry): succeeds
+        (async function* () {
+            yield {
+                candidates: [
+                    {
+                        content: {
+                            parts: [{ text: 'Successful final response' }],
+                        },
+                        finishReason: 'STOP',
+                    },
+                ],
+            };
+        })());
+        // Send a message and consume the stream
+        const stream = await chat.sendMessageStream('test-model', { message: 'test' }, 'prompt-id-discard-test');
+        const events = [];
+        for await (const event of stream) {
+            events.push(event);
+        }
+        // Check that a retry happened
+        expect(mockContentGenerator.generateContentStream).toHaveBeenCalledTimes(2);
+        expect(events.some((e) => e.type === StreamEventType.RETRY)).toBe(true);
+        // Check the final recorded history
+        const history = chat.getHistory();
+        expect(history.length).toBe(2); // user turn + final model turn
+        const modelTurn = history[1];
+        // The model turn should only contain the text from the successful attempt
+        expect(modelTurn.parts[0].text).toBe('Successful final response');
+        // It should NOT contain any text from the failed attempt
+        expect(modelTurn.parts[0].text).not.toContain('This valid part should be discarded');
+    });
+    describe('stripThoughtsFromHistory', () => {
+        it('should strip thought signatures', () => {
+            chat.setHistory([
+                {
+                    role: 'user',
+                    parts: [{ text: 'hello' }],
+                },
+                {
+                    role: 'model',
+                    parts: [
+                        { text: 'thinking...', thoughtSignature: 'thought-123' },
+                        {
+                            functionCall: { name: 'test', args: {} },
+                            thoughtSignature: 'thought-456',
+                        },
+                    ],
+                },
+            ]);
+            chat.stripThoughtsFromHistory();
+            expect(chat.getHistory()).toEqual([
+                {
+                    role: 'user',
+                    parts: [{ text: 'hello' }],
+                },
+                {
+                    role: 'model',
+                    parts: [
+                        { text: 'thinking...' },
+                        { functionCall: { name: 'test', args: {} } },
+                    ],
+                },
+            ]);
+        });
+    });
 });
 //# sourceMappingURL=geminiChat.test.js.map