fix(tests): refactor integration tests to be less flaky (#4890)

Co-authored-by: matt korwel <matt.korwel@gmail.com>

integration-tests/write_file.test.js

@@ -6,16 +6,63 @@
 
 import { test } from 'node:test';
 import { strict as assert } from 'assert';
-import { TestRig } from './test-helper.js';
+import {
+  TestRig,
+  createToolCallErrorMessage,
+  printDebugInfo,
+  validateModelOutput,
+} from './test-helper.js';
 
-test('should be able to write a file', async (t) => {
+test('should be able to write a file', async () => {
   const rig = new TestRig();
-  rig.setup(t.name);
+  await rig.setup('should be able to write a file');
   const prompt = `show me an example of using the write tool. put a dad joke in dad.txt`;
 
-  await rig.run(prompt);
+  const result = await rig.run(prompt);
+
+  const foundToolCall = await rig.waitForToolCall('write_file');
+
+  // Add debugging information
+  if (!foundToolCall) {
+    printDebugInfo(rig, result);
+  }
+
+  const allTools = rig.readToolLogs();
+  assert.ok(
+    foundToolCall,
+    createToolCallErrorMessage(
+      'write_file',
+      allTools.map((t) => t.toolRequest.name),
+      result,
+    ),
+  );
+
+  // Validate model output - will throw if no output, warn if missing expected content
+  validateModelOutput(result, 'dad.txt', 'Write file test');
+
   const newFilePath = 'dad.txt';
 
   const newFileContent = rig.readFile(newFilePath);
-  assert.notEqual(newFileContent, '');
+
+  // Add debugging for file content
+  if (newFileContent === '') {
+    console.error('File was created but is empty');
+    console.error(
+      'Tool calls:',
+      rig.readToolLogs().map((t) => ({
+        name: t.toolRequest.name,
+        args: t.toolRequest.args,
+      })),
+    );
+  }
+
+  assert.notEqual(newFileContent, '', 'Expected file to have content');
+
+  // Log success info if verbose
+  if (process.env.VERBOSE === 'true') {
+    console.log(
+      'File created successfully with content:',
+      newFileContent.substring(0, 100) + '...',
+    );
+  }
 });