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

Co-authored-by: matt korwel <[email protected]>

1 files changed, 34 insertions, 6 deletions

diff --git a/integration-tests/read_many_files.test.js b/integration-tests/read_many_files.test.js
index 7e770036..74d2f358 100644
--- a/integration-tests/read_many_files.test.js
+++ b/integration-tests/read_many_files.test.js
@@ -6,17 +6,45 @@
 
 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';
 
-test.skip('should be able to read multiple files', async (t) => {
+test('should be able to read multiple files', async () => {
   const rig = new TestRig();
-  rig.setup(t.name);
+  await rig.setup('should be able to read multiple files');
   rig.createFile('file1.txt', 'file 1 content');
   rig.createFile('file2.txt', 'file 2 content');
 
-  const prompt = `Read the files in this directory, list them and print them to the screen`;
+  const prompt = `Please use read_many_files to read file1.txt and file2.txt and show me what's in them`;
+
   const result = await rig.run(prompt);
 
-  assert.ok(result.includes('file 1 content'));
-  assert.ok(result.includes('file 2 content'));
+  // Check for either read_many_files or multiple read_file calls
+  const allTools = rig.readToolLogs();
+  const readManyFilesCall = await rig.waitForToolCall('read_many_files');
+  const readFileCalls = allTools.filter(
+    (t) => t.toolRequest.name === 'read_file',
+  );
+
+  // Accept either read_many_files OR at least 2 read_file calls
+  const foundValidPattern = readManyFilesCall || readFileCalls.length >= 2;
+
+  // Add debugging information
+  if (!foundValidPattern) {
+    printDebugInfo(rig, result, {
+      'read_many_files called': readManyFilesCall,
+      'read_file calls': readFileCalls.length,
+    });
+  }
+
+  assert.ok(
+    foundValidPattern,
+    'Expected to find either read_many_files or multiple read_file tool calls',
+  );
+
+  // Validate model output - will throw if no output, warn if missing expected content
+  validateModelOutput(
+    result,
+    ['file 1 content', 'file 2 content'],
+    'Read many files test',
+  );
 });