1 files changed, 45 insertions, 7 deletions
diff --git a/integration-tests/list_directory.test.js b/integration-tests/list_directory.test.js
index af7aae78..16f49f4b 100644
--- a/integration-tests/list_directory.test.js
+++ b/integration-tests/list_directory.test.js
@@ -6,19 +6,57 @@
 
 import { test } from 'node:test';
 import { strict as assert } from 'assert';
-import { TestRig } from './test-helper.js';
+import { TestRig, printDebugInfo, validateModelOutput } from './test-helper.js';
+import { existsSync } from 'fs';
+import { join } from 'path';
 
-test('should be able to list a directory', async (t) => {
+test('should be able to list a directory', async () => {
   const rig = new TestRig();
-  rig.setup(t.name);
+  await rig.setup('should be able to list a directory');
   rig.createFile('file1.txt', 'file 1 content');
   rig.mkdir('subdir');
   rig.sync();
 
+  // Poll for filesystem changes to propagate in containers
+  await rig.poll(
+    () => {
+      // Check if the files exist in the test directory
+      const file1Path = join(rig.testDir, 'file1.txt');
+      const subdirPath = join(rig.testDir, 'subdir');
+      return existsSync(file1Path) && existsSync(subdirPath);
+    },
+    1000, // 1 second max wait
+    50, // check every 50ms
+  );
+
   const prompt = `Can you list the files in the current directory. Display them in the style of 'ls'`;
-  const result = rig.run(prompt);
 
-  const lines = result.split('\n').filter((line) => line.trim() !== '');
-  assert.ok(lines.some((line) => line.includes('file1.txt')));
-  assert.ok(lines.some((line) => line.includes('subdir')));
+  const result = await rig.run(prompt);
+
+  const foundToolCall = await rig.waitForToolCall('list_directory');
+
+  // Add debugging information
+  if (
+    !foundToolCall ||
+    !result.includes('file1.txt') ||
+    !result.includes('subdir')
+  ) {
+    const allTools = printDebugInfo(rig, result, {
+      'Found tool call': foundToolCall,
+      'Contains file1.txt': result.includes('file1.txt'),
+      'Contains subdir': result.includes('subdir'),
+    });
+
+    console.error(
+      'List directory calls:',
+      allTools
+        .filter((t) => t.toolRequest.name === 'list_directory')
+        .map((t) => t.toolRequest.args),
+    );
+  }
+
+  assert.ok(foundToolCall, 'Expected to find a list_directory tool call');
+
+  // Validate model output - will throw if no output, warn if missing expected content
+  validateModelOutput(result, ['file1.txt', 'subdir'], 'List directory test');
 });