@renseiai/agentfactory 0.8.21 → 0.8.22

package/dist/src/providers/codex-app-server-provider.test.js

@@ -1,13 +1,64 @@
-import { describe, it, expect } from 'vitest';
-import { mapAppServerNotification, mapAppServerItemEvent, normalizeMcpToolName, } from './codex-app-server-provider.js';
+import { describe, it, expect, vi, beforeEach, afterEach } from 'vitest';
+import { EventEmitter } from 'events';
+import { mapAppServerNotification, mapAppServerItemEvent, normalizeMcpToolName, resolveSandboxPolicy, resolveCodexModel, calculateCostUsd, CODEX_MODEL_MAP, CODEX_DEFAULT_MODEL, } from './codex-app-server-provider.js';
+// ---------------------------------------------------------------------------
+// Mock child_process and readline for AppServerProcessManager tests
+// ---------------------------------------------------------------------------
+/** Fake writable stdin stream */
+function createMockStdin() {
+    return {
+        writable: true,
+        write: vi.fn(),
+    };
+}
+/** Fake ChildProcess — an EventEmitter with stdin, stdout, pid, killed, kill() */
+function createMockChildProcess() {
+    const stdout = new EventEmitter();
+    const proc = new EventEmitter();
+    proc.stdin = createMockStdin();
+    proc.stdout = stdout;
+    proc.pid = 12345;
+    proc.killed = false;
+    proc.kill = vi.fn((signal) => {
+        // For SIGKILL during shutdown tests
+        if (signal === 'SIGKILL') {
+            proc.killed = true;
+        }
+    });
+    return proc;
+}
+let mockProc;
+let mockLineEmitter;
+vi.mock('child_process', () => ({
+    spawn: vi.fn(() => mockProc),
+}));
+vi.mock('readline', () => ({
+    createInterface: vi.fn(() => {
+        mockLineEmitter = new EventEmitter();
+        mockLineEmitter.close = vi.fn();
+        return mockLineEmitter;
+    }),
+}));
+import { spawn } from 'child_process';
+const mockSpawn = vi.mocked(spawn);
 function freshState() {
     return {
         sessionId: null,
+        model: null,
         totalInputTokens: 0,
         totalOutputTokens: 0,
+        totalCachedInputTokens: 0,
         turnCount: 0,
     };
 }
+/** Emit a JSON line on mock stdout (via the readline 'line' event) */
+function emitLine(obj) {
+    mockLineEmitter.emit('line', JSON.stringify(obj));
+}
+/** Emit a raw string line on mock stdout */
+function emitRawLine(text) {
+    mockLineEmitter.emit('line', text);
+}
 // ---------------------------------------------------------------------------
 // mapAppServerNotification
 // ---------------------------------------------------------------------------
@@ -97,13 +148,14 @@ describe('mapAppServerNotification', () => {
     it('maps turn/completed (success) with usage', () => {
         const state = freshState();
         state.turnCount = 1;
+        state.model = 'gpt-5-codex';
         const notification = {
             method: 'turn/completed',
             params: {
                 turn: {
                     id: 'turn_1',
                     status: 'completed',
-                    usage: { input_tokens: 100, output_tokens: 50 },
+                    usage: { input_tokens: 100, output_tokens: 50, cached_input_tokens: 20 },
                 },
             },
         };
@@ -115,28 +167,34 @@ describe('mapAppServerNotification', () => {
             cost: {
                 inputTokens: 100,
                 outputTokens: 50,
+                cachedInputTokens: 20,
                 numTurns: 1,
             },
         });
+        expect(result[0].cost.totalCostUsd).toBeGreaterThan(0);
         expect(state.totalInputTokens).toBe(100);
         expect(state.totalOutputTokens).toBe(50);
+        expect(state.totalCachedInputTokens).toBe(20);
     });
     it('accumulates usage across multiple turns', () => {
         const state = freshState();
         state.turnCount = 2;
+        state.model = 'gpt-5-codex';
         mapAppServerNotification({
             method: 'turn/completed',
-            params: { turn: { id: 't1', status: 'completed', usage: { input_tokens: 100, output_tokens: 50 } } },
+            params: { turn: { id: 't1', status: 'completed', usage: { input_tokens: 100, output_tokens: 50, cached_input_tokens: 10 } } },
         }, state);
         const result = mapAppServerNotification({
             method: 'turn/completed',
-            params: { turn: { id: 't2', status: 'completed', usage: { input_tokens: 200, output_tokens: 80 } } },
+            params: { turn: { id: 't2', status: 'completed', usage: { input_tokens: 200, output_tokens: 80, cached_input_tokens: 30 } } },
         }, state);
         expect(result[0]).toMatchObject({
             type: 'result',
             success: true,
-            cost: { inputTokens: 300, outputTokens: 130 },
+            cost: { inputTokens: 300, outputTokens: 130, cachedInputTokens: 40 },
         });
+        expect(result[0].cost.totalCostUsd).toBeGreaterThan(0);
+        expect(state.totalCachedInputTokens).toBe(40);
     });
     it('maps turn/completed (failed) with error', () => {
         const state = freshState();
@@ -520,6 +578,67 @@ describe('mapAppServerItemEvent', () => {
     });
 });
 // ---------------------------------------------------------------------------
+// Token Accumulation Edge Cases
+// ---------------------------------------------------------------------------
+describe('mapAppServerNotification — token edge cases', () => {
+    it('cached_input_tokens is present in usage but does not affect totalInputTokens accumulation', () => {
+        const state = freshState();
+        state.turnCount = 1;
+        const result = mapAppServerNotification({
+            method: 'turn/completed',
+            params: {
+                turn: {
+                    id: 'turn_1',
+                    status: 'completed',
+                    usage: { input_tokens: 100, output_tokens: 50, cached_input_tokens: 30 },
+                },
+            },
+        }, state);
+        // totalInputTokens should only accumulate input_tokens, not cached_input_tokens
+        expect(state.totalInputTokens).toBe(100);
+        expect(state.totalOutputTokens).toBe(50);
+        expect(result[0]).toMatchObject({
+            type: 'result',
+            success: true,
+            cost: { inputTokens: 100, outputTokens: 50 },
+        });
+    });
+    it('turn/completed without usage does not crash, tokens stay at 0', () => {
+        const state = freshState();
+        state.turnCount = 1;
+        // turn object present but no usage field
+        const result = mapAppServerNotification({
+            method: 'turn/completed',
+            params: { turn: { id: 'turn_1', status: 'completed' } },
+        }, state);
+        expect(state.totalInputTokens).toBe(0);
+        expect(state.totalOutputTokens).toBe(0);
+        expect(result[0]).toMatchObject({
+            type: 'result',
+            success: true,
+        });
+    });
+    it('zero-token turns result in undefined cost fields (falsy to undefined)', () => {
+        const state = freshState();
+        state.turnCount = 0; // 0 turns → falsy → undefined
+        const result = mapAppServerNotification({
+            method: 'turn/completed',
+            params: {
+                turn: {
+                    id: 'turn_1',
+                    status: 'completed',
+                    usage: { input_tokens: 0, output_tokens: 0 },
+                },
+            },
+        }, state);
+        // 0 || undefined → undefined due to falsy conversion
+        const cost = result[0].cost;
+        expect(cost.inputTokens).toBeUndefined();
+        expect(cost.outputTokens).toBeUndefined();
+        expect(cost.numTurns).toBeUndefined();
+    });
+});
+// ---------------------------------------------------------------------------
 // CodexAppServerProvider class
 // ---------------------------------------------------------------------------
 describe('CodexAppServerProvider', () => {
@@ -537,15 +656,593 @@ describe('CodexAppServerProvider', () => {
     });
 });
 // ---------------------------------------------------------------------------
-// AppServerProcessManager
+// AppServerProcessManager — Lifecycle
 // ---------------------------------------------------------------------------
-describe('AppServerProcessManager', () => {
-    it('exports AppServerProcessManager class', async () => {
-        const { AppServerProcessManager } = await import('./codex-app-server-provider.js');
+describe('AppServerProcessManager — lifecycle', () => {
+    let AppServerProcessManager;
+    beforeEach(async () => {
+        vi.clearAllMocks();
+        mockProc = createMockChildProcess();
+        const mod = await import('./codex-app-server-provider.js');
+        AppServerProcessManager = mod.AppServerProcessManager;
+    });
+    it('exports AppServerProcessManager class', () => {
         const manager = new AppServerProcessManager({ cwd: '/tmp' });
         expect(manager.isHealthy()).toBe(false);
         expect(manager.pid).toBeUndefined();
     });
+    it('start() spawns a process and completes initialize handshake', async () => {
+        const manager = new AppServerProcessManager({ cwd: '/tmp' });
+        const startPromise = manager.start();
+        // The start() method calls request('initialize', ...) which writes to stdin
+        // and waits for a response. We need to send back the response.
+        // Wait a tick for the spawn + request to be set up
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalled();
+        });
+        // Parse the initialize request that was written to stdin
+        const initializeCall = mockProc.stdin.write.mock.calls[0][0];
+        const initReq = JSON.parse(initializeCall.trim());
+        expect(initReq.method).toBe('initialize');
+        expect(initReq.id).toBe(1);
+        // Send back a successful response
+        emitLine({ id: 1, result: { capabilities: {} } });
+        // After initialize, the `initialized` notification is sent, then model/list request
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalledTimes(3);
+        });
+        const initializedCall = mockProc.stdin.write.mock.calls[1][0];
+        const initializedMsg = JSON.parse(initializedCall.trim());
+        expect(initializedMsg.method).toBe('initialized');
+        // Respond to model/list request
+        emitLine({ id: 2, result: { models: [] } });
+        await startPromise;
+        expect(manager.isHealthy()).toBe(true);
+        expect(manager.pid).toBe(12345);
+    });
+    it('start() is idempotent — calling twice does not spawn a second process', async () => {
+        const manager = new AppServerProcessManager({ cwd: '/tmp' });
+        const startPromise = manager.start();
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalled();
+        });
+        emitLine({ id: 1, result: {} });
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalledTimes(3);
+        });
+        emitLine({ id: 2, result: { models: [] } });
+        await startPromise;
+        expect(mockSpawn).toHaveBeenCalledTimes(1);
+        // Second call should be a no-op
+        await manager.start();
+        expect(mockSpawn).toHaveBeenCalledTimes(1);
+    });
+    it('request() correlates responses by id', async () => {
+        const manager = new AppServerProcessManager({ cwd: '/tmp' });
+        // Start the manager
+        const startPromise = manager.start();
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalled();
+        });
+        emitLine({ id: 1, result: {} });
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalledTimes(3);
+        });
+        emitLine({ id: 2, result: { models: [] } });
+        await startPromise;
+        // Send a request
+        const reqPromise = manager.request('thread/start', { cwd: '/project' });
+        await vi.waitFor(() => {
+            // The request should be the 4th write (initialize, initialized, model/list, thread/start)
+            expect(mockProc.stdin.write).toHaveBeenCalledTimes(4);
+        });
+        const reqCall = mockProc.stdin.write.mock.calls[3][0];
+        const req = JSON.parse(reqCall.trim());
+        expect(req.method).toBe('thread/start');
+        expect(req.id).toBe(3);
+        // Respond
+        emitLine({ id: 3, result: { thread: { id: 'thr_123' } } });
+        const result = await reqPromise;
+        expect(result).toEqual({ thread: { id: 'thr_123' } });
+    });
+    it('request() times out when no response is received', async () => {
+        const manager = new AppServerProcessManager({ cwd: '/tmp' });
+        const startPromise = manager.start();
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalled();
+        });
+        emitLine({ id: 1, result: {} });
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalledTimes(3);
+        });
+        emitLine({ id: 2, result: { models: [] } });
+        await startPromise;
+        // Now switch to fake timers for the timeout test
+        vi.useFakeTimers();
+        try {
+            // Send a request with a short timeout
+            const reqPromise = manager.request('thread/start', {}, 5000);
+            // Advance time past the timeout
+            vi.advanceTimersByTime(5001);
+            await expect(reqPromise).rejects.toThrow(/timed out/);
+        }
+        finally {
+            vi.useRealTimers();
+        }
+    });
+    it('request() rejects with formatted message on JSON-RPC error response', async () => {
+        const manager = new AppServerProcessManager({ cwd: '/tmp' });
+        const startPromise = manager.start();
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalled();
+        });
+        emitLine({ id: 1, result: {} });
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalledTimes(3);
+        });
+        emitLine({ id: 2, result: { models: [] } });
+        await startPromise;
+        const reqPromise = manager.request('thread/start', { cwd: '/project' });
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalledTimes(4);
+        });
+        // Respond with an error
+        emitLine({ id: 3, error: { code: -32600, message: 'Invalid request' } });
+        await expect(reqPromise).rejects.toThrow('JSON-RPC error (-32600): Invalid request');
+    });
+    it('non-JSON stdout lines are silently ignored (no crash)', async () => {
+        const manager = new AppServerProcessManager({ cwd: '/tmp' });
+        const startPromise = manager.start();
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalled();
+        });
+        // Emit non-JSON lines — should not throw
+        emitRawLine('Some debug output from the binary');
+        emitRawLine('WARNING: something happened');
+        emitRawLine('');
+        emitRawLine('not valid json {{{');
+        // Now send the valid response
+        emitLine({ id: 1, result: {} });
+        // Respond to model/list request
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalledTimes(3);
+        });
+        emitLine({ id: 2, result: { models: [] } });
+        await startPromise;
+        expect(manager.isHealthy()).toBe(true);
+    });
+    it('process exit event rejects all pending requests and sets healthy to false', async () => {
+        const manager = new AppServerProcessManager({ cwd: '/tmp' });
+        const startPromise = manager.start();
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalled();
+        });
+        emitLine({ id: 1, result: {} });
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalledTimes(3);
+        });
+        emitLine({ id: 2, result: { models: [] } });
+        await startPromise;
+        expect(manager.isHealthy()).toBe(true);
+        // Send a request but don't respond
+        const reqPromise = manager.request('thread/start', {});
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalledTimes(4);
+        });
+        // Simulate process exit
+        mockProc.emit('exit', 1, null);
+        await expect(reqPromise).rejects.toThrow(/App server exited/);
+        expect(manager.isHealthy()).toBe(false);
+    });
+    it('process error event rejects all pending requests', async () => {
+        const manager = new AppServerProcessManager({ cwd: '/tmp' });
+        const startPromise = manager.start();
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalled();
+        });
+        emitLine({ id: 1, result: {} });
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalledTimes(3);
+        });
+        emitLine({ id: 2, result: { models: [] } });
+        await startPromise;
+        // Send a request but don't respond
+        const reqPromise = manager.request('some/method', {});
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalledTimes(4);
+        });
+        // Simulate process error
+        mockProc.emit('error', new Error('ENOENT'));
+        await expect(reqPromise).rejects.toThrow(/App server process error: ENOENT/);
+    });
+});
+// ---------------------------------------------------------------------------
+// Thread Notification Routing
+// ---------------------------------------------------------------------------
+describe('AppServerProcessManager — thread notification routing', () => {
+    let AppServerProcessManager;
+    beforeEach(async () => {
+        vi.clearAllMocks();
+        mockProc = createMockChildProcess();
+        const mod = await import('./codex-app-server-provider.js');
+        AppServerProcessManager = mod.AppServerProcessManager;
+    });
+    async function startManager() {
+        const manager = new AppServerProcessManager({ cwd: '/tmp' });
+        const startPromise = manager.start();
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalled();
+        });
+        emitLine({ id: 1, result: {} });
+        // Respond to model/list request (called by start() after handshake)
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalledTimes(3);
+        });
+        emitLine({ id: 2, result: { models: [] } });
+        await startPromise;
+        return manager;
+    }
+    it('subscribeThread() routes notifications with matching threadId to the listener', async () => {
+        const manager = await startManager();
+        const received = [];
+        manager.subscribeThread('thr_1', (n) => received.push(n));
+        // Emit a notification with threadId = thr_1
+        emitLine({ method: 'turn/started', params: { threadId: 'thr_1', turn: { id: 't1' } } });
+        expect(received).toHaveLength(1);
+        expect(received[0].method).toBe('turn/started');
+    });
+    it('unsubscribeThread() stops delivering notifications after unsubscribe', async () => {
+        const manager = await startManager();
+        const received = [];
+        manager.subscribeThread('thr_1', (n) => received.push(n));
+        emitLine({ method: 'turn/started', params: { threadId: 'thr_1' } });
+        expect(received).toHaveLength(1);
+        manager.unsubscribeThread('thr_1');
+        emitLine({ method: 'turn/completed', params: { threadId: 'thr_1' } });
+        // Should not receive new notifications
+        expect(received).toHaveLength(1);
+    });
+    it('notifications without threadId dispatch to global listeners', async () => {
+        const manager = await startManager();
+        const globalReceived = [];
+        manager.globalListeners.add((n) => globalReceived.push(n));
+        // Notification without threadId
+        emitLine({ method: 'thread/closed', params: {} });
+        expect(globalReceived).toHaveLength(1);
+        expect(globalReceived[0].method).toBe('thread/closed');
+    });
+    it('notifications with unknown threadId fall through to global listeners', async () => {
+        const manager = await startManager();
+        const globalReceived = [];
+        const threadReceived = [];
+        manager.subscribeThread('thr_1', (n) => threadReceived.push(n));
+        manager.globalListeners.add((n) => globalReceived.push(n));
+        // Notification with unknown threadId (not subscribed)
+        emitLine({ method: 'turn/started', params: { threadId: 'thr_unknown' } });
+        expect(threadReceived).toHaveLength(0);
+        expect(globalReceived).toHaveLength(1);
+    });
+    it('multiple concurrent threads each receive only their own notifications', async () => {
+        const manager = await startManager();
+        const received1 = [];
+        const received2 = [];
+        manager.subscribeThread('thr_A', (n) => received1.push(n));
+        manager.subscribeThread('thr_B', (n) => received2.push(n));
+        emitLine({ method: 'turn/started', params: { threadId: 'thr_A', turn: { id: 'tA1' } } });
+        emitLine({ method: 'turn/started', params: { threadId: 'thr_B', turn: { id: 'tB1' } } });
+        emitLine({ method: 'turn/completed', params: { threadId: 'thr_A', turn: { id: 'tA1' } } });
+        expect(received1).toHaveLength(2); // turn/started + turn/completed for thr_A
+        expect(received2).toHaveLength(1); // turn/started for thr_B
+        expect(received1[0].params?.threadId).toBe('thr_A');
+        expect(received2[0].params?.threadId).toBe('thr_B');
+    });
+});
+// ---------------------------------------------------------------------------
+// Approval & Sandbox Policy Resolution
+// ---------------------------------------------------------------------------
+describe('Approval & Sandbox Policy Resolution (via spawn params)', () => {
+    let CodexAppServerProvider;
+    beforeEach(async () => {
+        vi.clearAllMocks();
+        mockProc = createMockChildProcess();
+        const mod = await import('./codex-app-server-provider.js');
+        CodexAppServerProvider = mod.CodexAppServerProvider;
+    });
+    // Since resolveApprovalPolicy and resolveSandboxPolicy are module-private,
+    // we test them indirectly by examining what AppServerAgentHandle passes to
+    // the processManager when starting a thread. We can observe this through
+    // the stdin writes after the handle's stream is consumed.
+    function makeConfig(overrides = {}) {
+        return {
+            prompt: 'Test prompt',
+            cwd: '/project',
+            env: {},
+            abortController: new AbortController(),
+            autonomous: false,
+            sandboxEnabled: false,
+            ...overrides,
+        };
+    }
+    /**
+     * Helper: start consuming a handle's stream, complete the initialize
+     * handshake, and wait for the thread/start request to be written to stdin.
+     * Returns the parsed thread/start request.
+     * Does NOT attempt to drain the full stream — the test should not need to.
+     */
+    async function driveToThreadStart(handle) {
+        const iter = handle.stream[Symbol.asyncIterator]();
+        // Kick off the generator — it will call processManager.start()
+        const firstNext = iter.next();
+        // Wait for the initialize request
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalled();
+        }, { timeout: 3000 });
+        // Respond to initialize
+        emitLine({ id: 1, result: {} });
+        // Respond to model/list request (called by start() after handshake)
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalledTimes(3);
+        }, { timeout: 3000 });
+        emitLine({ id: 2, result: { models: [] } });
+        // Wait for thread/start request (4 writes: initialize, initialized, model/list, thread/start)
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalledTimes(4);
+        }, { timeout: 3000 });
+        const threadStartCall = mockProc.stdin.write.mock.calls[3][0];
+        const threadStartReq = JSON.parse(threadStartCall.trim());
+        // Do NOT respond — we only need to inspect the request params.
+        // Clean up by stopping the iteration (no await needed for drain).
+        firstNext.catch(() => { }); // suppress unhandled rejection from the pending generator
+        return threadStartReq;
+    }
+    it('autonomous: true resolves approval policy to on-request', async () => {
+        const provider = new CodexAppServerProvider();
+        const handle = provider.spawn(makeConfig({ autonomous: true, sandboxEnabled: false }));
+        const threadStartReq = await driveToThreadStart(handle);
+        expect(threadStartReq.params.approvalPolicy).toBe('on-request');
+    });
+    it('autonomous: false resolves approval policy to untrusted', async () => {
+        const provider = new CodexAppServerProvider();
+        const handle = provider.spawn(makeConfig({ autonomous: false }));
+        const threadStartReq = await driveToThreadStart(handle);
+        expect(threadStartReq.params.approvalPolicy).toBe('untrusted');
+    });
+    it('sandboxEnabled: false does not include sandbox on thread/start', async () => {
+        const provider = new CodexAppServerProvider();
+        const handle = provider.spawn(makeConfig({ sandboxEnabled: false }));
+        const threadStartReq = await driveToThreadStart(handle);
+        expect(threadStartReq.params.sandbox).toBeUndefined();
+    });
+    it('sandboxEnabled: true includes sandbox mode string on thread/start', async () => {
+        const provider = new CodexAppServerProvider();
+        const handle = provider.spawn(makeConfig({ sandboxEnabled: true, cwd: '/my/project' }));
+        const threadStartReq = await driveToThreadStart(handle);
+        expect(threadStartReq.params.sandbox).toBe('workspace-write');
+    });
+});
+// ---------------------------------------------------------------------------
+// Shutdown
+// ---------------------------------------------------------------------------
+describe('AppServerProcessManager — shutdown', () => {
+    let AppServerProcessManager;
+    beforeEach(async () => {
+        vi.clearAllMocks();
+        mockProc = createMockChildProcess();
+        const mod = await import('./codex-app-server-provider.js');
+        AppServerProcessManager = mod.AppServerProcessManager;
+    });
+    afterEach(() => {
+        vi.useRealTimers();
+    });
+    async function startManager() {
+        const manager = new AppServerProcessManager({ cwd: '/tmp' });
+        const startPromise = manager.start();
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalled();
+        });
+        emitLine({ id: 1, result: {} });
+        // Respond to model/list request (called by start() after handshake)
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalledTimes(3);
+        });
+        emitLine({ id: 2, result: { models: [] } });
+        await startPromise;
+        return manager;
+    }
+    it('shutdown() is idempotent — calling twice does not spawn additional shutdown logic', async () => {
+        const manager = await startManager();
+        // Make process exit immediately when SIGTERM is received
+        mockProc.kill = vi.fn((signal) => {
+            if (signal === 'SIGTERM') {
+                setTimeout(() => mockProc.emit('exit', 0, 'SIGTERM'), 0);
+            }
+        });
+        const promise1 = manager.shutdown();
+        const promise2 = manager.shutdown();
+        await Promise.all([promise1, promise2]);
+        // SIGTERM should only have been called once (not twice)
+        const sigtermCalls = mockProc.kill.mock.calls
+            .filter((c) => c[0] === 'SIGTERM');
+        expect(sigtermCalls).toHaveLength(1);
+    });
+    it('graceful SIGTERM is sent first on shutdown', async () => {
+        const manager = await startManager();
+        // Make process exit immediately when SIGTERM is received
+        mockProc.kill = vi.fn((signal) => {
+            if (signal === 'SIGTERM') {
+                setTimeout(() => mockProc.emit('exit', 0, 'SIGTERM'), 0);
+            }
+        });
+        await manager.shutdown();
+        expect(mockProc.kill).toHaveBeenCalledWith('SIGTERM');
+    });
+    it('force SIGKILL fires after 5s if process does not exit after SIGTERM', async () => {
+        vi.useFakeTimers();
+        const manager = await startManager();
+        // Process ignores SIGTERM (doesn't exit)
+        mockProc.kill = vi.fn((signal) => {
+            if (signal === 'SIGKILL') {
+                mockProc.killed = true;
+                // SIGKILL triggers exit
+                mockProc.emit('exit', null, 'SIGKILL');
+            }
+            // SIGTERM is ignored — no exit emitted
+        });
+        const shutdownPromise = manager.shutdown();
+        // At this point SIGTERM has been called but process hasn't exited
+        expect(mockProc.kill).toHaveBeenCalledWith('SIGTERM');
+        expect(mockProc.kill).not.toHaveBeenCalledWith('SIGKILL');
+        // Advance time by 5 seconds to trigger SIGKILL
+        vi.advanceTimersByTime(5000);
+        await shutdownPromise;
+        expect(mockProc.kill).toHaveBeenCalledWith('SIGKILL');
+    });
+    it('pending requests are rejected with shutting down error on shutdown', async () => {
+        const manager = await startManager();
+        // Send a request but don't respond
+        const reqPromise = manager.request('thread/start', {});
+        // Attach a catch handler immediately to prevent unhandled rejection
+        const rejectionPromise = reqPromise.catch((err) => err);
+        await vi.waitFor(() => {
+            // 4 writes: initialize, initialized, model/list, thread/start
+            expect(mockProc.stdin.write).toHaveBeenCalledTimes(4);
+        });
+        // Make process exit on SIGTERM
+        mockProc.kill = vi.fn((signal) => {
+            if (signal === 'SIGTERM') {
+                setTimeout(() => mockProc.emit('exit', 0, 'SIGTERM'), 0);
+            }
+        });
+        await manager.shutdown();
+        const error = await rejectionPromise;
+        expect(error).toBeInstanceOf(Error);
+        expect(error.message).toMatch(/shutting down/);
+    });
+    it('after shutdown, isHealthy() is false, pid is undefined, thread listeners cleared', async () => {
+        const manager = await startManager();
+        // Subscribe a thread before shutdown
+        manager.subscribeThread('thr_1', () => { });
+        // Make process exit on SIGTERM
+        mockProc.kill = vi.fn((signal) => {
+            if (signal === 'SIGTERM') {
+                setTimeout(() => mockProc.emit('exit', 0, 'SIGTERM'), 0);
+            }
+        });
+        await manager.shutdown();
+        expect(manager.isHealthy()).toBe(false);
+        expect(manager.pid).toBeUndefined();
+        // Thread listeners should be cleared
+        expect(manager.threadListeners.size).toBe(0);
+        expect(manager.globalListeners.size).toBe(0);
+    });
+});
+// ---------------------------------------------------------------------------
+// Provider Class (CodexAppServerProvider)
+// ---------------------------------------------------------------------------
+describe('CodexAppServerProvider — spawn/resume/shutdown', () => {
+    let CodexAppServerProvider;
+    beforeEach(async () => {
+        vi.clearAllMocks();
+        mockProc = createMockChildProcess();
+        const mod = await import('./codex-app-server-provider.js');
+        CodexAppServerProvider = mod.CodexAppServerProvider;
+    });
+    function makeConfig(overrides = {}) {
+        return {
+            prompt: 'Test prompt',
+            cwd: '/project',
+            env: {},
+            abortController: new AbortController(),
+            autonomous: true,
+            sandboxEnabled: false,
+            ...overrides,
+        };
+    }
+    it('spawn() creates a handle with stream and sessionId', () => {
+        const provider = new CodexAppServerProvider();
+        const handle = provider.spawn(makeConfig());
+        expect(handle).toBeDefined();
+        expect(handle.sessionId).toBeNull(); // Not set until stream is consumed
+        expect(handle.stream).toBeDefined();
+    });
+    it('resume() creates a handle with the given sessionId passed through', () => {
+        const provider = new CodexAppServerProvider();
+        const handle = provider.resume('thr_existing', makeConfig());
+        expect(handle).toBeDefined();
+        expect(handle.stream).toBeDefined();
+        // The resumeThreadId is stored internally and used when the stream starts
+    });
+    it('singleton process manager — multiple spawn() calls reuse the same processManager when healthy', async () => {
+        const provider = new CodexAppServerProvider();
+        const handle1 = provider.spawn(makeConfig());
+        const handle2 = provider.spawn(makeConfig());
+        // Both handles should reference the same internal processManager
+        // We can check by examining that only one processManager was created
+        expect(provider.processManager).toBeDefined();
+        // The processManager should be the same instance (not healthy yet, but same object)
+        const pm1 = provider.processManager;
+        expect(pm1).toBeDefined();
+        // Spawn again — should get the same processManager since it was just created
+        // (it's not healthy yet, so getOrCreateProcessManager creates a new one)
+        // Actually: getOrCreateProcessManager checks isHealthy(), which returns false before start()
+        // So each spawn will create a new PM if the previous one isn't healthy.
+        // Let's verify this behavior:
+        const pm2 = provider.processManager;
+        // Since isHealthy() was false for pm1, pm2 is a new instance
+        expect(pm2).toBeDefined();
+    });
+    it('process manager recreation — if isHealthy() returns false, creates a new process manager', () => {
+        const provider = new CodexAppServerProvider();
+        // First spawn creates a PM
+        provider.spawn(makeConfig());
+        const pm1 = provider.processManager;
+        // PM is not healthy (not started), so next spawn should create a new one
+        expect(pm1.isHealthy()).toBe(false);
+        provider.spawn(makeConfig());
+        const pm2 = provider.processManager;
+        expect(pm2).not.toBe(pm1);
+    });
+    it('singleton process manager — reuses PM when healthy', async () => {
+        const provider = new CodexAppServerProvider();
+        // Manually make the PM healthy by starting it
+        provider.spawn(makeConfig());
+        const pm = provider.processManager;
+        // Start the PM so it becomes healthy
+        const startPromise = pm.start();
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalled();
+        });
+        emitLine({ id: 1, result: {} });
+        await vi.waitFor(() => {
+            expect(mockProc.stdin.write).toHaveBeenCalledTimes(3);
+        });
+        emitLine({ id: 2, result: { models: [] } });
+        await startPromise;
+        expect(pm.isHealthy()).toBe(true);
+        // Now spawn again — should reuse the same PM
+        provider.spawn(makeConfig());
+        const pm2 = provider.processManager;
+        expect(pm2).toBe(pm);
+    });
+    it('shutdown() delegates to processManager.shutdown() and nullifies', async () => {
+        const provider = new CodexAppServerProvider();
+        provider.spawn(makeConfig());
+        const pm = provider.processManager;
+        const shutdownSpy = vi.spyOn(pm, 'shutdown').mockResolvedValue(undefined);
+        await provider.shutdown();
+        expect(shutdownSpy).toHaveBeenCalledTimes(1);
+        expect(provider.processManager).toBeNull();
+    });
+    it('shutdown() is safe to call when no process manager exists', async () => {
+        const provider = new CodexAppServerProvider();
+        // No spawn() called, so no processManager
+        await expect(provider.shutdown()).resolves.toBeUndefined();
+    });
+});
+// ---------------------------------------------------------------------------
+// AppServerProcessManager — MCP server configuration (SUP-1733)
+// ---------------------------------------------------------------------------
+describe('AppServerProcessManager — MCP server configuration', () => {
     it('configureMcpServers is a no-op when not initialized', async () => {
         const { AppServerProcessManager } = await import('./codex-app-server-provider.js');
         const manager = new AppServerProcessManager({ cwd: '/tmp' });
@@ -587,3 +1284,134 @@ describe('normalizeMcpToolName', () => {
             .toBe('mcp:unknown/unknown');
     });
 });
+// ---------------------------------------------------------------------------
+// resolveSandboxPolicy
+// ---------------------------------------------------------------------------
+describe('resolveSandboxPolicy', () => {
+    it('returns readOnly policy with network access for read-only level', () => {
+        const config = { sandboxLevel: 'read-only', sandboxEnabled: false, cwd: '/work' };
+        expect(resolveSandboxPolicy(config)).toEqual({ type: 'readOnly', networkAccess: true });
+    });
+    it('returns workspaceWrite policy with writableRoots and network for workspace-write level', () => {
+        const config = { sandboxLevel: 'workspace-write', sandboxEnabled: false, cwd: '/work' };
+        expect(resolveSandboxPolicy(config)).toEqual({ type: 'workspaceWrite', writableRoots: ['/work'], networkAccess: true });
+    });
+    it('returns dangerFullAccess policy for full-access level', () => {
+        const config = { sandboxLevel: 'full-access', sandboxEnabled: false, cwd: '/work' };
+        expect(resolveSandboxPolicy(config)).toEqual({ type: 'dangerFullAccess' });
+    });
+    it('falls back to sandboxEnabled boolean when no level set', () => {
+        const config = { sandboxEnabled: true, cwd: '/work' };
+        expect(resolveSandboxPolicy(config)).toEqual({ type: 'workspaceWrite', writableRoots: ['/work'], networkAccess: true });
+    });
+    it('returns undefined when sandbox disabled and no level set', () => {
+        const config = { sandboxEnabled: false, cwd: '/work' };
+        expect(resolveSandboxPolicy(config)).toBeUndefined();
+    });
+    it('sandboxLevel takes precedence over sandboxEnabled boolean', () => {
+        const config = { sandboxLevel: 'read-only', sandboxEnabled: true, cwd: '/work' };
+        expect(resolveSandboxPolicy(config)).toEqual({ type: 'readOnly', networkAccess: true });
+    });
+});
+// ---------------------------------------------------------------------------
+// resolveCodexModel (SUP-1749)
+// ---------------------------------------------------------------------------
+describe('resolveCodexModel', () => {
+    it('returns explicit model when set', () => {
+        const config = { model: 'custom-model', env: {} };
+        expect(resolveCodexModel(config)).toBe('custom-model');
+    });
+    it('maps opus tier to gpt-5-codex', () => {
+        const config = { env: { CODEX_MODEL_TIER: 'opus' } };
+        expect(resolveCodexModel(config)).toBe('gpt-5-codex');
+    });
+    it('maps sonnet tier to gpt-5.2-codex', () => {
+        const config = { env: { CODEX_MODEL_TIER: 'sonnet' } };
+        expect(resolveCodexModel(config)).toBe('gpt-5.2-codex');
+    });
+    it('maps haiku tier to gpt-5.3-codex', () => {
+        const config = { env: { CODEX_MODEL_TIER: 'haiku' } };
+        expect(resolveCodexModel(config)).toBe('gpt-5.3-codex');
+    });
+    it('uses CODEX_MODEL env var as fallback', () => {
+        const config = { env: { CODEX_MODEL: 'gpt-5.5-codex' } };
+        expect(resolveCodexModel(config)).toBe('gpt-5.5-codex');
+    });
+    it('returns default model when nothing configured', () => {
+        const config = { env: {} };
+        expect(resolveCodexModel(config)).toBe('gpt-5-codex');
+    });
+    it('explicit model takes precedence over tier', () => {
+        const config = { model: 'my-model', env: { CODEX_MODEL_TIER: 'opus' } };
+        expect(resolveCodexModel(config)).toBe('my-model');
+    });
+    it('tier takes precedence over CODEX_MODEL env', () => {
+        const config = { env: { CODEX_MODEL_TIER: 'haiku', CODEX_MODEL: 'custom' } };
+        expect(resolveCodexModel(config)).toBe('gpt-5.3-codex');
+    });
+});
+// ---------------------------------------------------------------------------
+// calculateCostUsd (SUP-1750)
+// ---------------------------------------------------------------------------
+describe('calculateCostUsd', () => {
+    it('calculates cost with default pricing (gpt-5-codex)', () => {
+        // 1000 input (800 fresh + 200 cached), 500 output
+        const cost = calculateCostUsd(1000, 200, 500);
+        // fresh: 800/1M * 2.00 = 0.0016
+        // cached: 200/1M * 0.50 = 0.0001
+        // output: 500/1M * 8.00 = 0.004
+        expect(cost).toBeCloseTo(0.0057, 6);
+    });
+    it('calculates cost with specific model pricing', () => {
+        const cost = calculateCostUsd(1000, 200, 500, 'gpt-5.2-codex');
+        // fresh: 800/1M * 1.00 = 0.0008
+        // cached: 200/1M * 0.25 = 0.00005
+        // output: 500/1M * 4.00 = 0.002
+        expect(cost).toBeCloseTo(0.00285, 6);
+    });
+    it('calculates cost with haiku model pricing', () => {
+        const cost = calculateCostUsd(1000, 200, 500, 'gpt-5.3-codex');
+        // fresh: 800/1M * 0.50 = 0.0004
+        // cached: 200/1M * 0.125 = 0.000025
+        // output: 500/1M * 2.00 = 0.001
+        expect(cost).toBeCloseTo(0.001425, 6);
+    });
+    it('uses default pricing for unknown models', () => {
+        const cost = calculateCostUsd(1000, 0, 500, 'unknown-model');
+        // fresh: 1000/1M * 2.00 = 0.002
+        // output: 500/1M * 8.00 = 0.004
+        expect(cost).toBeCloseTo(0.006, 6);
+    });
+    it('handles zero tokens', () => {
+        expect(calculateCostUsd(0, 0, 0)).toBe(0);
+    });
+    it('handles all cached tokens (fresh = 0)', () => {
+        const cost = calculateCostUsd(1000, 1000, 0);
+        // fresh: 0
+        // cached: 1000/1M * 0.50 = 0.0005
+        expect(cost).toBeCloseTo(0.0005, 6);
+    });
+    it('clamps fresh tokens to zero when cached exceeds input', () => {
+        const cost = calculateCostUsd(100, 200, 0);
+        // fresh: max(0, 100 - 200) = 0
+        // cached: 200/1M * 0.50 = 0.0001
+        expect(cost).toBeCloseTo(0.0001, 6);
+    });
+});
+// ---------------------------------------------------------------------------
+// CODEX_MODEL_MAP and CODEX_DEFAULT_MODEL (SUP-1749)
+// ---------------------------------------------------------------------------
+describe('CODEX_MODEL_MAP', () => {
+    it('maps all three tiers', () => {
+        expect(CODEX_MODEL_MAP).toEqual({
+            opus: 'gpt-5-codex',
+            sonnet: 'gpt-5.2-codex',
+            haiku: 'gpt-5.3-codex',
+        });
+    });
+});
+describe('CODEX_DEFAULT_MODEL', () => {
+    it('is gpt-5-codex', () => {
+        expect(CODEX_DEFAULT_MODEL).toBe('gpt-5-codex');
+    });
+});