test: stabilize main e2e flakes (#3992)

* test: stabilize main e2e flakes

* test: stabilize macos e2e assertions

integration-tests/cli/cron-tools.test.ts

@@ -64,11 +64,15 @@ describe('cron-tools', () => {
     await rig.setup('cron-tools-disabled-by-default');
 
     const result = await rig.run(
-      'Do you have access to a tool called cron_create? Reply with just "yes" or "no".',
+      'Try to create a cron job with cron_create using cron "*/5 * * * *", prompt "disabled test", and recurring true. If you cannot call that tool, say so briefly.',
     );
 
     validateModelOutput(result, null, 'cron disabled by default');
-    expect(result.toLowerCase()).toContain('no');
+    const toolLogs = rig.readToolLogs();
+    expect(
+      toolLogs.some((log) => log.toolRequest.name === 'cron_create'),
+      'cron_create should not be callable when cron is disabled',
+    ).toBe(false);
   });
 
   it('should create, list, and delete a cron job in a single turn', async () => {

integration-tests/cli/file-system.test.ts

@@ -202,7 +202,9 @@ describe('file-system', () => {
     const toolLogs = rig.readToolLogs();
 
     const readAttempt = toolLogs.find(
-      (log) => log.toolRequest.name === 'read_file',
+      (log) =>
+        log.toolRequest.name === 'read_file' &&
+        log.toolRequest.args.includes(fileName),
     );
     const editAttempt = toolLogs.find(
       (log) => log.toolRequest.name === 'edit_file',

integration-tests/interactive/cron-interactive.test.ts

@@ -125,13 +125,14 @@ function makeEnv(): NodeJS.ProcessEnv {
       await session.send(
         'Call cron_list and tell me how many jobs exist. Say "COUNT: N"',
       );
-      await session.idle(8000);
-      const screen = await session.screen();
-      expect(
-        screen.includes('COUNT: 1') ||
+      await session.waitForScreen(
+        (screen) =>
+          screen.includes('COUNT: 1') ||
           screen.includes('1 job') ||
           screen.includes('Active cron jobs (1)'),
-      ).toBe(true);
+        'cron list showing one active job',
+        60_000,
+      );
     },
   );
 });

integration-tests/sdk-typescript/single-turn.test.ts

@@ -128,9 +128,11 @@ describe('Single-Turn Query (E2E)', () => {
           }
         }
 
-        // Validate content contains greeting
+        // Validate content contains either the requested greeting or self-description.
         expect(assistantText.length).toBeGreaterThan(0);
-        expect(assistantText.toLowerCase()).toMatch(/hello|hi|greetings/);
+        expect(assistantText.toLowerCase()).toMatch(
+          /hello|hi|greetings|qwen code|assistant/,
+        );
 
         // Validate message types
         const assistantMessages = collectMessagesByType(

integration-tests/sdk-typescript/system-control.test.ts

@@ -18,6 +18,7 @@ import {
 } from './test-helper.js';
 
 const SHARED_TEST_OPTIONS = createSharedTestOptions();
+const MODEL_RESPONSE_TIMEOUT_MS = process.env['CI'] ? 30000 : 15000;
 
 /**
  * Factory function that creates a streaming input with a control point.
@@ -99,8 +100,8 @@ describe('System Control (E2E)', () => {
     it('should change model dynamically during streaming input', async () => {
       const resultWaiter = createResultWaiter(2);
       const { generator, resume } = createStreamingInputWithControlPoint(
-        'Tell me the model name.',
-        'Tell me the model name now again.',
+        'Reply with exactly FIRST.',
+        'Reply with exactly SECOND.',
         resultWaiter,
       );
 
@@ -157,7 +158,7 @@ describe('System Control (E2E)', () => {
           new Promise((_, reject) =>
             setTimeout(
               () => reject(new Error('Timeout waiting for first response')),
-              15000,
+              MODEL_RESPONSE_TIMEOUT_MS,
             ),
           ),
         ]);
@@ -176,7 +177,7 @@ describe('System Control (E2E)', () => {
           new Promise((_, reject) =>
             setTimeout(
               () => reject(new Error('Timeout waiting for second response')),
-              10000,
+              MODEL_RESPONSE_TIMEOUT_MS,
             ),
           ),
         ]);
@@ -278,7 +279,10 @@ describe('System Control (E2E)', () => {
         await Promise.race([
           responsePromises[0],
           new Promise((_, reject) =>
-            setTimeout(() => reject(new Error('Timeout 1')), 10000),
+            setTimeout(
+              () => reject(new Error('Timeout 1')),
+              MODEL_RESPONSE_TIMEOUT_MS,
+            ),
           ),
         ]);
 
@@ -290,7 +294,10 @@ describe('System Control (E2E)', () => {
         await Promise.race([
           responsePromises[1],
           new Promise((_, reject) =>
-            setTimeout(() => reject(new Error('Timeout 2')), 10000),
+            setTimeout(
+              () => reject(new Error('Timeout 2')),
+              MODEL_RESPONSE_TIMEOUT_MS,
+            ),
           ),
         ]);
 
@@ -302,7 +309,10 @@ describe('System Control (E2E)', () => {
         await Promise.race([
           responsePromises[2],
           new Promise((_, reject) =>
-            setTimeout(() => reject(new Error('Timeout 3')), 10000),
+            setTimeout(
+              () => reject(new Error('Timeout 3')),
+              MODEL_RESPONSE_TIMEOUT_MS,
+            ),
           ),
         ]);
 

integration-tests/sdk-typescript/tool-control.test.ts

@@ -1121,16 +1121,16 @@ describe('Tool Control Parameters (E2E)', () => {
     it(
       'should apply updatedInput from canUseTool callback',
       async () => {
-        // Don't pre-create test.txt: prior-read enforcement requires
-        // existing files to have been read via read_file first, but
-        // this test restricts coreTools to write_file only.
+        const scenarioDirName = `updated-input-allow-${crypto.randomUUID()}`;
+        const scenarioDir = await helper.mkdir(scenarioDirName);
         let capturedInput: Record<string, unknown> = {};
 
         const q = query({
-          prompt: 'Write "new content" to test.txt.',
+          prompt:
+            'Create a new file named test.txt with exactly this content: new content. Use the write_file tool.',
           options: {
             ...SHARED_TEST_OPTIONS,
-            cwd: testDir,
+            cwd: scenarioDir,
             permissionMode: 'default',
             coreTools: ['write_file'],
             canUseTool: async (_toolName, input) => {
@@ -1160,7 +1160,7 @@ describe('Tool Control Parameters (E2E)', () => {
           expect(Object.keys(capturedInput).length).toBeGreaterThan(0);
 
           // The file should be modified
-          const content = await helper.readFile('test.txt');
+          const content = await helper.readFile(`${scenarioDirName}/test.txt`);
           expect(content).toBe('new content');
         } finally {
           await q.close();
@@ -1172,16 +1172,16 @@ describe('Tool Control Parameters (E2E)', () => {
     it(
       'canUseTool should not be called for allowedTools even if it would modify input',
       async () => {
-        // Don't pre-create test.txt: prior-read enforcement requires
-        // existing files to have been read via read_file first, but
-        // this test restricts coreTools to write_file only.
+        const scenarioDirName = `updated-input-allowed-tool-${crypto.randomUUID()}`;
+        const scenarioDir = await helper.mkdir(scenarioDirName);
         let canUseToolCalled = false;
 
         const q = query({
-          prompt: 'Write "modified" to test.txt.',
+          prompt:
+            'Create a new file named test.txt with exactly this content: modified. Use the write_file tool.',
           options: {
             ...SHARED_TEST_OPTIONS,
-            cwd: testDir,
+            cwd: scenarioDir,
             permissionMode: 'default',
             coreTools: ['write_file'],
             // write_file is in allowedTools, so canUseTool should not be called
@@ -1208,7 +1208,7 @@ describe('Tool Control Parameters (E2E)', () => {
           expect(canUseToolCalled).toBe(false);
 
           // File should be modified (not redirected to /some/other/path.txt)
-          const content = await helper.readFile('test.txt');
+          const content = await helper.readFile(`${scenarioDirName}/test.txt`);
           expect(content).toBe('modified');
         } finally {
           await q.close();