@qduc/term2 0.7.2 → 0.8.0

package/dist/services/conversation/conversation-service.test.js

@@ -0,0 +1,1138 @@
+import { it, expect, beforeAll, beforeEach } from 'vitest';
+import { ConversationService } from './conversation-service.js';
+import { MockStream } from '../test-helpers/mock-stream.js';
+import { clearApprovalRejectionMarkers, markToolCallAsApprovalRejection, } from '../../utils/streaming/extract-command-messages.js';
+import { registerToolFormatters } from '../../tools/command-message-formatters.js';
+import { formatShellCommandMessage } from '../../tools/system/shell.js';
+const mockLogger = {
+    info: () => { },
+    warn: () => { },
+    error: () => { },
+    debug: () => { },
+    security: () => { },
+    setCorrelationId: () => { },
+    getCorrelationId: () => undefined,
+    clearCorrelationId: () => { },
+};
+const sessionContextService = {
+    runWithContext: (_context, fn) => fn(),
+    getContext: () => null,
+};
+/**
+ * Creates a partial ConversationAgentClient with safe no-op defaults.
+ * Pass only the methods the test actually exercises.
+ */
+function partialClient(methods = {}) {
+    return {
+        chat: async () => '',
+        abort: () => { },
+        setModel: () => { },
+        addToolInterceptor: () => () => { },
+        startStream: async () => new MockStream([]),
+        continueRunStream: async () => new MockStream([]),
+        ...methods,
+    };
+}
+function asFinal(result) {
+    if (result.type !== 'response')
+        throw new Error('Expected FinalTerminal');
+    return result;
+}
+function asApproval(result) {
+    if (result.type !== 'approval_required')
+        throw new Error('Expected ApprovalRequiredTerminal');
+    return result;
+}
+beforeAll(() => {
+    registerToolFormatters([{ name: 'shell', formatCommandMessage: formatShellCommandMessage }]);
+});
+beforeEach(() => {
+    clearApprovalRejectionMarkers();
+});
+it('emits live text chunks for response.output_text.delta events', async () => {
+    expect.assertions(3);
+    const events = [
+        { type: 'response.output_text.delta', delta: 'Hello' },
+        { type: 'response.output_text.delta', delta: ' world' },
+    ];
+    const mockClient = partialClient({
+        async startStream() {
+            return new MockStream(events);
+        },
+    });
+    const chunks = [];
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    const result = await service.sendMessage('hi', {
+        onTextChunk(full, chunk) {
+            chunks.push({ full, chunk });
+        },
+    });
+    expect(chunks).toEqual([
+        { full: 'Hello', chunk: 'Hello' },
+        { full: 'Hello world', chunk: ' world' },
+    ]);
+    expect(result.type).toBe('response');
+    expect(asFinal(result).finalText).toBe('Hello world');
+});
+it('emits ConversationEvents (text_delta → final) in order', async () => {
+    const events = [
+        { type: 'response.output_text.delta', delta: 'Hello' },
+        { type: 'response.output_text.delta', delta: ' world' },
+    ];
+    const mockClient = partialClient({
+        async startStream() {
+            return new MockStream(events);
+        },
+    });
+    const emitted = [];
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    const result = await service.sendMessage('hi', {
+        onEvent(event) {
+            emitted.push(event);
+        },
+    });
+    expect(result.type).toBe('response');
+    expect(asFinal(result).finalText).toBe('Hello world');
+    expect(emitted.map((e) => e.type)).toEqual(['text_delta', 'text_delta', 'final']);
+});
+it('emits approval_required ConversationEvent for interruptions', async () => {
+    const interruption = {
+        name: 'bash',
+        agent: { name: 'CLI Agent' },
+        arguments: JSON.stringify({ command: 'echo hi' }),
+        callId: 'call-xyz',
+    };
+    const initialStream = new MockStream([]);
+    initialStream.interruptions = [interruption];
+    const mockClient = partialClient({
+        async startStream() {
+            return initialStream;
+        },
+    });
+    const emitted = [];
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    const result = await service.sendMessage('run command', {
+        onEvent(event) {
+            emitted.push(event);
+        },
+    });
+    expect(result.type).toBe('approval_required');
+    expect(asApproval(result).approval.callId).toBe('call-xyz');
+    expect(emitted.length).toBe(1);
+    expect(emitted[0].type).toBe('approval_required');
+    expect(emitted[0].approval.toolName).toBe('bash');
+    expect(emitted[0].approval.argumentsText).toBe('echo hi');
+    expect(emitted[0].approval.callId).toBe('call-xyz');
+});
+it('compacts whitespace-heavy JSON arguments for approvals', async () => {
+    const interruption = {
+        name: 'apply_patch',
+        agent: { name: 'CLI Agent' },
+        arguments: '{"path":"a.txt","timeout_ms":   \n\t\t    null,"max_output_length":\n\t null}',
+        callId: 'call-trim',
+    };
+    const initialStream = new MockStream([]);
+    initialStream.interruptions = [interruption];
+    const mockClient = partialClient({
+        async startStream() {
+            return initialStream;
+        },
+    });
+    const emitted = [];
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    const result = await service.sendMessage('run tool', {
+        onEvent(event) {
+            emitted.push(event);
+        },
+    });
+    expect(result.type).toBe('approval_required');
+    expect(emitted.length).toBe(1);
+    expect(emitted[0].type).toBe('approval_required');
+    expect(emitted[0].approval.argumentsText).toBe('{"path":"a.txt","timeout_ms":null,"max_output_length":null}');
+});
+it('emits events when resolving aborted approval on next message', async () => {
+    const interruption = {
+        name: 'bash',
+        agent: { name: 'CLI Agent' },
+        arguments: JSON.stringify({ command: 'echo hi' }),
+        callId: 'call-abort',
+    };
+    const initialStream = new MockStream([]);
+    initialStream.interruptions = [interruption];
+    initialStream.state = {
+        approveCalls: [],
+        rejectCalls: [],
+        approve(arg) {
+            this.approveCalls.push(arg);
+        },
+        reject(arg) {
+            this.rejectCalls.push(arg);
+        },
+    };
+    const continuationStream = new MockStream([{ type: 'response.output_text.delta', delta: 'After abort' }]);
+    continuationStream.finalOutput = 'After abort';
+    let interceptorCount = 0;
+    const mockClient = partialClient({
+        abort() { },
+        addToolInterceptor() {
+            interceptorCount++;
+            return () => {
+                interceptorCount--;
+            };
+        },
+        async startStream() {
+            return initialStream;
+        },
+        async continueRunStream(state, options) {
+            expect(state).toBe(initialStream.state);
+            expect(options).toEqual({
+                previousResponseId: 'resp_test',
+                sessionId: 'default',
+                toolResultCallIds: ['call-abort'],
+            });
+            return continuationStream;
+        },
+    });
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    const approvalEvents = [];
+    const approvalResult = await service.sendMessage('run command', {
+        onEvent(event) {
+            approvalEvents.push(event);
+        },
+    });
+    expect(approvalResult.type).toBe('approval_required');
+    expect(approvalEvents[0].type).toBe('approval_required');
+    service.abort();
+    const resolvedEvents = [];
+    const resolvedResult = await service.sendMessage('new input', {
+        onEvent(event) {
+            resolvedEvents.push(event);
+        },
+    });
+    expect(resolvedResult.type).toBe('response');
+    expect(asFinal(resolvedResult).finalText).toBe('After abort');
+    expect(resolvedEvents.some((e) => e.type === 'text_delta')).toBe(true);
+    expect(resolvedEvents.some((e) => e.type === 'final')).toBe(true);
+    expect(interceptorCount).toBe(0);
+    expect(initialStream.state.approveCalls).toEqual([]);
+    expect(initialStream.state.rejectCalls).toEqual([interruption]);
+});
+it('reject with reason and abort+new input yield the same history', async () => {
+    const baseHistory = [
+        {
+            role: 'user',
+            type: 'message',
+            content: 'run command',
+        },
+        {
+            type: 'function_call',
+            name: 'shell',
+            arguments: JSON.stringify({ command: 'echo hi' }),
+            callId: 'call-1',
+        },
+    ];
+    const rejectionHistory = [
+        ...baseHistory,
+        {
+            type: 'tool_call_output_item',
+            name: 'shell',
+            output: 'Tool execution was not approved.',
+            callId: 'call-1',
+        },
+    ];
+    const runFlow = async (mode) => {
+        const interruption = {
+            name: 'shell',
+            agent: { name: 'CLI Agent' },
+            arguments: JSON.stringify({ command: 'echo hi' }),
+            callId: 'call-1',
+        };
+        const initialStream = new MockStream([]);
+        initialStream.interruptions = [interruption];
+        initialStream.state = {
+            approveCalls: [],
+            rejectCalls: [],
+            approve(arg) {
+                this.approveCalls.push(arg);
+            },
+            reject(arg) {
+                this.rejectCalls.push(arg);
+            },
+        };
+        initialStream.history = baseHistory;
+        const continuationStream = new MockStream([{ type: 'response.output_text.delta', delta: 'Rejected' }]);
+        continuationStream.finalOutput = 'Rejected';
+        continuationStream.history = rejectionHistory;
+        const nextStream = new MockStream([]);
+        nextStream.finalOutput = 'Next';
+        const startCalls = [];
+        const mockClient = partialClient({
+            getProvider() {
+                return 'openrouter';
+            },
+            addToolInterceptor() {
+                return () => { };
+            },
+            abort() { },
+            async startStream(input, options) {
+                startCalls.push({ input, options });
+                if (startCalls.length === 1)
+                    return initialStream;
+                return nextStream;
+            },
+            async continueRunStream() {
+                return continuationStream;
+            },
+        });
+        const service = new ConversationService({
+            agentClient: mockClient,
+            deps: { logger: mockLogger, sessionContextService },
+        });
+        const approvalResult = await service.sendMessage('run command');
+        expect(approvalResult.type).toBe('approval_required');
+        if (mode === 'reject') {
+            await service.handleApprovalDecision('n', 'no thanks');
+        }
+        else {
+            service.abort();
+            await service.sendMessage('new input');
+        }
+        await service.sendMessage('next');
+        expect(startCalls.length).toBe(2);
+        return startCalls[1].input;
+    };
+    const rejectionHistoryAfter = await runFlow('reject');
+    const abortHistoryAfter = await runFlow('abort');
+    expect(abortHistoryAfter).toEqual(rejectionHistoryAfter);
+});
+it('passes previous response ids into subsequent runs', async () => {
+    const streams = [new MockStream([]), new MockStream([])];
+    streams[0].lastResponseId = 'resp-1';
+    streams[0].finalOutput = 'First run done.';
+    streams[1].lastResponseId = 'resp-2';
+    streams[1].finalOutput = 'Second run done.';
+    const startCalls = [];
+    const mockClient = partialClient({
+        async startStream(text, options) {
+            const index = startCalls.length;
+            startCalls.push({ text, options });
+            return streams[index];
+        },
+    });
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    await service.sendMessage('first');
+    const secondResult = await service.sendMessage('second');
+    expect(startCalls).toEqual([
+        { text: 'first', options: { previousResponseId: null, sessionId: 'default' } },
+        { text: 'second', options: { previousResponseId: 'resp-1', sessionId: 'default' } },
+    ]);
+    expect(secondResult.type).toBe('response');
+    expect(asFinal(secondResult).finalText).toBe('Second run done.');
+});
+it('emits approval interruptions and resumes after approval', async () => {
+    const interruption = {
+        name: 'bash',
+        agent: { name: 'CLI Agent' },
+        arguments: JSON.stringify({ command: 'echo hi' }),
+    };
+    const initialStream = new MockStream([]);
+    initialStream.interruptions = [interruption];
+    initialStream.state = {
+        approveCalls: [],
+        rejectCalls: [],
+        approve(arg) {
+            this.approveCalls.push(arg);
+        },
+        reject(arg) {
+            this.rejectCalls.push(arg);
+        },
+    };
+    const continuationStream = new MockStream([{ type: 'response.output_text.delta', delta: 'Approved run' }]);
+    continuationStream.finalOutput = 'Approved run';
+    const mockClient = partialClient({
+        async startStream() {
+            return initialStream;
+        },
+        async continueRunStream(state, options) {
+            expect(state).toBe(initialStream.state);
+            expect(options).toEqual({
+                previousResponseId: 'resp_test',
+                sessionId: 'default',
+                toolResultCallIds: [],
+            });
+            return continuationStream;
+        },
+    });
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    const approvalResult = await service.sendMessage('run command');
+    expect(approvalResult.type).toBe('approval_required');
+    expect(asApproval(approvalResult).approval.toolName).toBe('bash');
+    expect(asApproval(approvalResult).approval.argumentsText).toBe('echo hi');
+    const finalResult = await service.handleApprovalDecision('y');
+    expect(finalResult).toBeTruthy();
+    expect(finalResult.type).toBe('response');
+    expect(asFinal(finalResult).finalText).toBe('Approved run');
+    expect(initialStream.state.approveCalls).toEqual([interruption]);
+    expect(initialStream.state.rejectCalls).toEqual([]);
+});
+it('dedupes command messages emitted live from run events', async () => {
+    const commandPayload = 'exit 0\nfile.txt';
+    const rawItem = {
+        id: 'call-123',
+        type: 'function_call_result',
+        name: 'shell',
+        arguments: JSON.stringify({ commands: 'ls' }),
+    };
+    const commandItem = {
+        type: 'tool_call_output_item',
+        name: 'shell',
+        output: commandPayload,
+        rawItem,
+    };
+    const events = [{ type: 'run_item_stream_event', item: commandItem }];
+    const stream = new MockStream(events);
+    stream.newItems = [commandItem];
+    const emitted = [];
+    const mockClient = partialClient({
+        async startStream() {
+            return stream;
+        },
+    });
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    const result = await service.sendMessage('run shell', {
+        onCommandMessage(message) {
+            emitted.push(message);
+        },
+    });
+    expect(emitted).toEqual([
+        {
+            id: 'call-123-0',
+            callId: 'call-123',
+            sender: 'command',
+            status: 'completed',
+            command: 'ls',
+            output: 'file.txt',
+            success: true,
+            isApprovalRejection: false,
+            failureReason: undefined,
+            toolName: 'shell',
+        },
+    ]);
+    expect(asFinal(result).commandMessages).toEqual([]);
+});
+it('attaches cached shell args when output uses call_id', async () => {
+    const functionCallItem = {
+        rawItem: {
+            type: 'function_call',
+            id: 'call-abc',
+            name: 'shell',
+            arguments: JSON.stringify({ command: 'npm run lint' }),
+        },
+    };
+    const outputItem = {
+        type: 'tool_call_output_item',
+        name: 'shell',
+        output: 'exit 0\n> md-preview@0.0.0 lint\n> eslint .',
+        rawItem: {
+            type: 'function_call_result',
+            name: 'shell',
+            id: 'result-1',
+            call_id: 'call-abc',
+        },
+    };
+    const events = [
+        { type: 'run_item_stream_event', item: functionCallItem },
+        { type: 'run_item_stream_event', item: outputItem },
+    ];
+    const stream = new MockStream(events);
+    stream.newItems = [functionCallItem, outputItem];
+    const emitted = [];
+    const mockClient = partialClient({
+        async startStream() {
+            return stream;
+        },
+    });
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    const result = await service.sendMessage('run shell', {
+        onCommandMessage(message) {
+            emitted.push(message);
+        },
+    });
+    expect(emitted).toEqual([
+        {
+            id: 'result-1-0',
+            callId: 'call-abc',
+            sender: 'command',
+            status: 'completed',
+            command: 'npm run lint',
+            output: '> md-preview@0.0.0 lint\n> eslint .',
+            success: true,
+            isApprovalRejection: false,
+            failureReason: undefined,
+            toolName: 'shell',
+        },
+    ]);
+    expect(asFinal(result).commandMessages).toEqual([]);
+});
+it('preserves approval rejection command messages', async () => {
+    const rejectionPayload = JSON.stringify({
+        output: [
+            {
+                command: 'should-not-show',
+                stdout: 'fake output',
+                stderr: '',
+                outcome: { type: 'exit', exitCode: 1 },
+            },
+        ],
+    });
+    const rawItem = {
+        id: 'rejection-call',
+        callId: 'rejection-call',
+        type: 'function_call_result',
+        name: 'shell',
+        arguments: JSON.stringify({ commands: ['should-not-show'] }),
+    };
+    markToolCallAsApprovalRejection(rawItem.callId);
+    const commandItem = {
+        type: 'tool_call_output_item',
+        name: 'shell',
+        output: rejectionPayload,
+        rawItem,
+    };
+    const events = [{ type: 'run_item_stream_event', item: commandItem }];
+    const stream = new MockStream(events);
+    stream.newItems = [commandItem];
+    const emitted = [];
+    const mockClient = partialClient({
+        async startStream() {
+            return stream;
+        },
+    });
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    const result = await service.sendMessage('run shell', {
+        onCommandMessage(message) {
+            emitted.push(message);
+        },
+    });
+    expect(emitted.length).toBe(1);
+    expect(emitted[0].isApprovalRejection).toBe(true);
+    expect(emitted[0].command).toBe('should-not-show');
+    expect(emitted[0].output).toBe(rejectionPayload);
+    expect(asFinal(result).commandMessages).toEqual([]);
+});
+it('dedupes commands from initial stream when continuation history contains them', async () => {
+    // This test simulates the scenario where:
+    // 1. Initial stream emits command 'ls' before hitting an approval interruption
+    // 2. User approves, continuation stream runs
+    // 3. Continuation stream's history/newItems contains the 'ls' command again
+    // The 'ls' command should NOT be duplicated in the final result
+    const lsCommandPayload = 'exit 0\nfile.txt';
+    const lsRawItem = {
+        id: 'call-ls-123',
+        type: 'function_call_result',
+        name: 'shell',
+        arguments: JSON.stringify({ commands: 'ls' }),
+    };
+    const lsCommandItem = {
+        type: 'tool_call_output_item',
+        name: 'shell',
+        output: lsCommandPayload,
+        rawItem: lsRawItem,
+    };
+    const sedCommandPayload = 'exit 0\ncontent';
+    const sedRawItem = {
+        id: 'call-sed-456',
+        type: 'function_call_result',
+        name: 'shell',
+        arguments: JSON.stringify({ commands: 'sed -n "1,10p" file.txt' }),
+    };
+    const sedCommandItem = {
+        type: 'tool_call_output_item',
+        name: 'shell',
+        output: sedCommandPayload,
+        rawItem: sedRawItem,
+    };
+    const interruption = {
+        name: 'shell',
+        agent: { name: 'CLI Agent' },
+        arguments: JSON.stringify({ commands: 'sed -n "1,10p" file.txt' }),
+    };
+    // Initial stream: emits 'ls' command, then hits approval for 'sed'
+    const initialEvents = [{ type: 'run_item_stream_event', item: lsCommandItem }];
+    const initialStream = new MockStream(initialEvents);
+    initialStream.interruptions = [interruption];
+    initialStream.state = {
+        approveCalls: [],
+        approve(arg) {
+            this.approveCalls.push(arg);
+        },
+        reject() { },
+    };
+    // Continuation stream: emits 'sed' command, history contains BOTH 'ls' and 'sed'
+    const continuationEvents = [{ type: 'run_item_stream_event', item: sedCommandItem }];
+    const continuationStream = new MockStream(continuationEvents);
+    continuationStream.finalOutput = 'Done';
+    // Simulate that the continuation stream's history contains both commands
+    continuationStream.history = [lsCommandItem, sedCommandItem];
+    const mockClient = partialClient({
+        async startStream() {
+            return initialStream;
+        },
+        async continueRunStream() {
+            return continuationStream;
+        },
+    });
+    const emittedCommands = [];
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    // Send initial message - should emit 'ls' and return approval_required
+    const approvalResult = await service.sendMessage('run commands', {
+        onCommandMessage(message) {
+            emittedCommands.push(message);
+        },
+    });
+    expect(approvalResult.type).toBe('approval_required');
+    expect(emittedCommands.length).toBe(1);
+    expect(emittedCommands[0].id).toBe('call-ls-123-0');
+    // Handle approval - should emit 'sed' but NOT duplicate 'ls'
+    const finalResult = await service.handleApprovalDecision('y', undefined, {
+        onCommandMessage(message) {
+            emittedCommands.push(message);
+        },
+    });
+    expect(finalResult.type).toBe('response');
+    // Only 'sed' should be emitted during continuation
+    expect(emittedCommands.length).toBe(2);
+    expect(emittedCommands[1].id).toBe('call-sed-456-0');
+    // Final result should have no additional commands since both were emitted live
+    expect(asFinal(finalResult).commandMessages).toEqual([]);
+});
+it('continuation replays the just-emitted tool in newItems without re-emitting it in the final result', async () => {
+    // Regression: when the SDK continuation stream replays the tool that the
+    // current turn has already emitted via command_message (in stream.newItems
+    // and/or stream.history), the final response must not include that tool in
+    // commandMessages. Otherwise applyServiceResult appends it to the messages
+    // state and the user sees the finished tool output printed again after the
+    // model's final answer.
+    const commandPayload = 'exit 0\ncontent';
+    const rawItem = {
+        id: 'call-replay-1',
+        type: 'function_call_result',
+        name: 'shell',
+        arguments: JSON.stringify({ commands: 'sed -n "1,10p" file.txt' }),
+    };
+    const commandItem = {
+        type: 'tool_call_output_item',
+        name: 'shell',
+        output: commandPayload,
+        rawItem,
+    };
+    const interruption = {
+        name: 'shell',
+        agent: { name: 'CLI Agent' },
+        arguments: JSON.stringify({ commands: 'sed -n "1,10p" file.txt' }),
+    };
+    const initialStream = new MockStream([]);
+    initialStream.interruptions = [interruption];
+    initialStream.state = {
+        approveCalls: [],
+        approve(arg) {
+            this.approveCalls.push(arg);
+        },
+        reject() { },
+    };
+    const continuationEvents = [{ type: 'run_item_stream_event', item: commandItem }];
+    const continuationStream = new MockStream(continuationEvents);
+    continuationStream.finalOutput = 'Done';
+    continuationStream.newItems = [commandItem];
+    continuationStream.history = [commandItem];
+    const mockClient = partialClient({
+        async startStream() {
+            return initialStream;
+        },
+        async continueRunStream() {
+            return continuationStream;
+        },
+    });
+    const emittedCommands = [];
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    const approvalResult = await service.sendMessage('run command', {
+        onCommandMessage(message) {
+            emittedCommands.push(message);
+        },
+    });
+    expect(approvalResult.type).toBe('approval_required');
+    const finalResult = await service.handleApprovalDecision('y', undefined, {
+        onCommandMessage(message) {
+            emittedCommands.push(message);
+        },
+    });
+    expect(finalResult.type).toBe('response');
+    expect(emittedCommands.length).toBe(1);
+    expect(emittedCommands[0].id).toBe('call-replay-1-0');
+    expect(asFinal(finalResult).commandMessages).toEqual([]);
+});
+it('resetWithNewId() clears conversation state', async () => {
+    const streams = [new MockStream([]), new MockStream([])];
+    streams[0].lastResponseId = 'resp-1';
+    const startCalls = [];
+    const mockClient = partialClient({
+        async startStream(text, options) {
+            startCalls.push({ text, options });
+            return streams[startCalls.length - 1];
+        },
+    });
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    await service.sendMessage('first');
+    service.resetWithNewId('new-id');
+    await service.sendMessage('second');
+    expect(startCalls[1].options).toMatchObject({ previousResponseId: null });
+});
+it('resetWithNewId() updates sessionId', () => {
+    const service = new ConversationService({
+        agentClient: partialClient(),
+        sessionId: 'old-id',
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    expect(service.sessionId).toBe('old-id');
+    service.resetWithNewId('new-id');
+    expect(service.sessionId).toBe('new-id');
+});
+it('resetWithNewId() clears provider conversations when supported', async () => {
+    let clearCalls = 0;
+    const mockClient = partialClient({
+        clearConversations() {
+            clearCalls++;
+        },
+        async startStream() {
+            return new MockStream([]);
+        },
+    });
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    await service.sendMessage('first message');
+    service.resetWithNewId('new-id');
+    expect(clearCalls).toBe(1);
+});
+it('resetWithNewId() rediscover skills when skills service is available', async () => {
+    let discoverCalls = 0;
+    const mockSkillsService = {
+        discoverSkills() {
+            discoverCalls++;
+        },
+    };
+    const service = new ConversationService({
+        agentClient: partialClient(),
+        deps: { logger: mockLogger, sessionContextService, skillsService: mockSkillsService },
+    });
+    service.resetWithNewId('new-id');
+    expect(discoverCalls).toBe(1);
+});
+it('setModel() delegates to agent client', () => {
+    let setModelCalledWith = null;
+    const mockClient = partialClient({
+        setModel(model) {
+            setModelCalledWith = model;
+        },
+    });
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    service.setModel('gpt-4');
+    expect(setModelCalledWith).toBe('gpt-4');
+});
+it('setTemperature() delegates to agent client when supported', () => {
+    let setTemperatureCalledWith = null;
+    const mockClient = partialClient({
+        setTemperature(value) {
+            setTemperatureCalledWith = value;
+        },
+    });
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    service.setTemperature(0.7);
+    expect(setTemperatureCalledWith).toBe(0.7);
+});
+it('abort() delegates to agent client and clears pending approval', async () => {
+    let abortCalled = false;
+    const mockClient = partialClient({
+        abort() {
+            abortCalled = true;
+        },
+        async startStream() {
+            const stream = new MockStream([]);
+            stream.interruptions = [
+                {
+                    name: 'bash',
+                    agent: { name: 'CLI Agent' },
+                    arguments: JSON.stringify({ command: 'echo hi' }),
+                },
+            ];
+            return stream;
+        },
+    });
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    // Trigger a pending approval
+    await service.sendMessage('run command');
+    service.abort();
+    expect(abortCalled).toBe(true);
+    const result = await service.handleApprovalDecision('y');
+    expect(result).toBe(null);
+});
+it('abort() preserves the aborted tool turn in exported state and snapshot', async () => {
+    const interruption = {
+        name: 'shell',
+        agent: { name: 'CLI Agent' },
+        arguments: JSON.stringify({ command: 'echo hi' }),
+        callId: 'call-abort',
+    };
+    const initialStream = new MockStream([]);
+    initialStream.interruptions = [interruption];
+    initialStream.state = {
+        approve() { },
+        reject() { },
+    };
+    const service = new ConversationService({
+        agentClient: partialClient({
+            abort() { },
+            async startStream() {
+                return initialStream;
+            },
+        }),
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    const first = await service.sendMessage('run command');
+    expect(first.type).toBe('approval_required');
+    service.abort();
+    const exported = service.exportState();
+    const snapshot = service.getCurrentSnapshot();
+    for (const state of [exported, snapshot]) {
+        expect(state.history.map((item) => item.type)).toEqual(['message', 'function_call', 'tool_call_output_item']);
+        expect(state.history[1].callId).toBe('call-abort');
+        expect(state.history[2].callId).toBe('call-abort');
+        expect(state.history[2].output).toBe('Tool execution was not approved.');
+    }
+});
+it('switchProvider() after abort reuses the preserved tool turn in the next full-history request', async () => {
+    const interruption = {
+        name: 'shell',
+        agent: { name: 'CLI Agent' },
+        arguments: JSON.stringify({ command: 'echo hi' }),
+        callId: 'call-abort',
+    };
+    const initialStream = new MockStream([]);
+    initialStream.interruptions = [interruption];
+    initialStream.state = {
+        approve() { },
+        reject() { },
+    };
+    const followupStream = new MockStream([]);
+    followupStream.finalOutput = 'Next';
+    const startCalls = [];
+    const mockClient = partialClient({
+        getProvider() {
+            return 'openrouter';
+        },
+        abort() { },
+        setProvider() { },
+        clearConversations() { },
+        async startStream(input) {
+            startCalls.push(input);
+            return startCalls.length === 1 ? initialStream : followupStream;
+        },
+    });
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    const first = await service.sendMessage('run command');
+    expect(first.type).toBe('approval_required');
+    service.abort();
+    service.switchProvider('openrouter');
+    const second = await service.sendMessage('next');
+    expect(second.type).toBe('response');
+    expect(startCalls.length).toBe(2);
+    expect(startCalls[1].map((item) => item.type)).toEqual([
+        'message',
+        'function_call',
+        'tool_call_output_item',
+        'message',
+    ]);
+    expect(startCalls[1][1].callId).toBe('call-abort');
+    expect(startCalls[1][2].callId).toBe('call-abort');
+    expect(startCalls[1][2].output).toBe('Tool execution was not approved.');
+    expect(startCalls[1][3].content).toBe('next');
+});
+it('handleApprovalDecision() rejects interruption when answer is n', async () => {
+    const interruption = {
+        name: 'bash',
+        agent: { name: 'CLI Agent' },
+        arguments: JSON.stringify({ command: 'echo hi' }),
+    };
+    const initialStream = new MockStream([]);
+    initialStream.interruptions = [interruption];
+    initialStream.state = {
+        approveCalls: [],
+        rejectCalls: [],
+        approve(arg) {
+            this.approveCalls.push(arg);
+        },
+        reject(arg) {
+            this.rejectCalls.push(arg);
+        },
+    };
+    const continuationStream = new MockStream([{ type: 'response.output_text.delta', delta: 'Rejected run' }]);
+    continuationStream.finalOutput = 'Rejected run';
+    const mockClient = partialClient({
+        async startStream() {
+            return initialStream;
+        },
+        async continueRunStream(_state, options) {
+            expect(options).toEqual({
+                previousResponseId: 'resp_test',
+                sessionId: 'default',
+                toolResultCallIds: [],
+            });
+            return continuationStream;
+        },
+        addToolInterceptor(_fn) {
+            return () => { };
+        },
+    });
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    await service.sendMessage('run command');
+    const finalResult = await service.handleApprovalDecision('n');
+    expect(finalResult.type).toBe('response');
+    expect(initialStream.state.approveCalls).toEqual([]);
+    expect(initialStream.state.rejectCalls).toEqual([interruption]);
+});
+it('handleApprovalDecision() returns null when no pending approval', async () => {
+    const service = new ConversationService({
+        agentClient: partialClient(),
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    const result = await service.handleApprovalDecision('y');
+    expect(result).toBe(null);
+});
+it('emits live reasoning chunks', async () => {
+    expect.assertions(3);
+    // Reasoning is now streamed via model events carrying reasoning_summary_text deltas
+    const events = [
+        {
+            type: 'response.output_text.delta',
+            data: {
+                type: 'model',
+                event: {
+                    type: 'response.reasoning_summary_text.delta',
+                    delta: 'Thinking...',
+                },
+            },
+        },
+        {
+            type: 'response.output_text.delta',
+            data: {
+                type: 'model',
+                event: {
+                    type: 'response.reasoning_summary_text.delta',
+                    delta: ' Still thinking.',
+                },
+            },
+        },
+    ];
+    const mockClient = partialClient({
+        async startStream() {
+            return new MockStream(events);
+        },
+    });
+    const chunks = [];
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    const result = await service.sendMessage('hi', {
+        onReasoningChunk(full, chunk) {
+            chunks.push({ full, chunk });
+        },
+    });
+    expect(chunks).toEqual([
+        { full: 'Thinking...', chunk: 'Thinking...' },
+        { full: 'Thinking... Still thinking.', chunk: ' Still thinking.' },
+    ]);
+    expect(result.type).toBe('response');
+    expect(asFinal(result).reasoningText).toBe('Thinking... Still thinking.');
+});
+it('retries on tool hallucination error (ModelBehaviorError)', async () => {
+    // Import ModelBehaviorError dynamically
+    const { ModelBehaviorError } = await import('@openai/agents');
+    let callCount = 0;
+    const mockClient = partialClient({
+        async startStream() {
+            callCount++;
+            if (callCount === 1) {
+                // First call: model hallucinates a non-existent tool
+                throw new ModelBehaviorError('Tool open_file not found in agent Terminal Assistant.');
+            }
+            // Second call: succeeds
+            const stream = new MockStream([
+                {
+                    type: 'response.output_text.delta',
+                    delta: 'Retried successfully',
+                },
+            ]);
+            stream.finalOutput = 'Retried successfully';
+            return stream;
+        },
+    });
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    const result = await service.sendMessage('explain this file');
+    // Should have retried and succeeded on second attempt
+    expect(callCount).toBe(2);
+    expect(result.type).toBe('response');
+    expect(asFinal(result).finalText).toBe('Retried successfully');
+});
+it('stops retrying after max hallucination retries', async () => {
+    const { ModelBehaviorError } = await import('@openai/agents');
+    let callCount = 0;
+    const mockClient = partialClient({
+        async startStream() {
+            callCount++;
+            // Always throw hallucination error
+            throw new ModelBehaviorError('Tool fake_tool not found in agent Test Agent.');
+        },
+    });
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    try {
+        await service.sendMessage('test message');
+        throw new Error('Should have thrown error after max retries');
+    }
+    catch (error) {
+        // collectTerminalResult re-throws the error event message as a plain Error,
+        // so the original ModelBehaviorError is not preserved.
+        expect(error.message).toBe('Tool fake_tool not found in agent Test Agent.');
+        // Should have tried: initial + 2 retries = 3 total attempts
+        expect(callCount).toBe(3);
+    }
+});
+it('does not retry on non-hallucination ModelBehaviorError', async () => {
+    const { ModelBehaviorError } = await import('@openai/agents');
+    let callCount = 0;
+    const mockClient = partialClient({
+        async startStream() {
+            callCount++;
+            // Throw a different kind of ModelBehaviorError
+            throw new ModelBehaviorError('Model violated safety guidelines');
+        },
+    });
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    try {
+        await service.sendMessage('test message');
+        throw new Error('Should have thrown error');
+    }
+    catch (error) {
+        // collectTerminalResult re-throws the error event message as a plain Error,
+        // so the original ModelBehaviorError is not preserved.
+        expect(error.message).toBe('Model violated safety guidelines');
+        // Should NOT retry - only 1 call
+        expect(callCount).toBe(1);
+    }
+});
+it('failed user turn is dropped from history after non-retryable provider error', async () => {
+    const startCalls = [];
+    const mockClient = partialClient({
+        getProvider() {
+            return 'openrouter';
+        },
+        async startStream(input) {
+            startCalls.push(input);
+            if (startCalls.length === 1) {
+                throw new Error('400 Error from provider: bad request');
+            }
+            const stream = new MockStream([
+                {
+                    type: 'response.output_text.delta',
+                    delta: 'ok',
+                },
+            ]);
+            stream.finalOutput = 'ok';
+            return stream;
+        },
+    });
+    const service = new ConversationService({
+        agentClient: mockClient,
+        deps: { logger: mockLogger, sessionContextService },
+    });
+    await expect(service.sendMessage('first failed message')).rejects.toThrow('400 Error from provider: bad request');
+    const result = await service.sendMessage('second message');
+    expect(result.type).toBe('response');
+    expect(startCalls.length).toBe(2);
+    // The session removes the failed user turn from the store before yielding the
+    // error event (so the drop happens regardless of generator-cleanup semantics),
+    // and surfaces the dropped text on the error event for UI restoration. The
+    // next request must not include the stranded first turn.
+    expect(startCalls[1]).toEqual([
+        {
+            role: 'user',
+            type: 'message',
+            content: 'second message',
+        },
+    ]);
+});
+//# sourceMappingURL=conversation-service.test.js.map