From 3e9e5689a494c81b166d41e13ce0e7c7c6f3098c Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Sat, 31 Jan 2026 11:14:38 +0000
Subject: [PATCH 1/3] Initial plan


From 2f7b3c990072963174735c5c0aa8719ded16c21f Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Sat, 31 Jan 2026 11:30:31 +0000
Subject: [PATCH 2/3] Fix flaky e2e tests: github-models-integration and
 docker-utils

Co-authored-by: gtanczyk <1281113+gtanczyk@users.noreply.github.com>
---
 e2e-tests/docker-utils.test.ts                |  4 ++++
 e2e-tests/github-models-integration.test.ts   | 14 +++++++------
 .../container-task/utils/docker-utils.ts      | 21 +++++++++++--------
 3 files changed, 24 insertions(+), 15 deletions(-)

diff --git a/e2e-tests/docker-utils.test.ts b/e2e-tests/docker-utils.test.ts
index 85ec149..cf91157 100644
--- a/e2e-tests/docker-utils.test.ts
+++ b/e2e-tests/docker-utils.test.ts
@@ -235,8 +235,12 @@ describe('Docker Utils E2E Tests', () => {
       const files = output
         .split('\n')
         .map((f) => f.trim())
+        .filter((f) => f.length > 0)
         .sort();
 
+      // Log actual files for debugging in case of flaky failures
+      console.log('Files found in container:', JSON.stringify(files, null, 2));
+
       expect(files).toContain('/data/root.txt');
       expect(files).toContain('/data/nested/nested.txt');
       // Note: Empty directories are included in the transfer
diff --git a/e2e-tests/github-models-integration.test.ts b/e2e-tests/github-models-integration.test.ts
index 953f09e..0a21b66 100644
--- a/e2e-tests/github-models-integration.test.ts
+++ b/e2e-tests/github-models-integration.test.ts
@@ -70,22 +70,24 @@ describe('GitHub Models Integration E2E Tests', () => {
     const command = `node "${CLI_PATH}" --ai-service=github-models --explicit-prompt="${prompt}" --dry-run`;
 
     try {
-      const { stderr } = await execAsync(command, {
+      const { stdout, stderr } = await execAsync(command, {
         env: {
           ...process.env,
           GITHUB_TOKEN: '', // Explicitly unset the token
         },
-        timeout: 10000,
+        timeout: 60000, // Increased timeout since the command takes time to load plugins and scan files
       });
 
       // Should show error message when no token is provided
-      // Note: The command may exit with code 0 or non-zero, but the error message should be in stderr
-      expect(stderr).toContain('GitHub Models API token not configured');
+      // Note: The error message may appear in stdout (via console output) or stderr
+      const combinedOutput = stdout + stderr;
+      expect(combinedOutput).toContain('GitHub Models API token not configured');
     } catch (error: unknown) {
       const execError = error as { code?: unknown; stdout?: string; stderr?: string };
 
-      // If the command fails, the error message should still be in stderr
-      expect(execError.stderr).toContain('GitHub Models API token not configured');
+      // If the command fails, the error message should be in either stdout or stderr
+      const combinedOutput = (execError.stdout || '') + (execError.stderr || '');
+      expect(combinedOutput).toContain('GitHub Models API token not configured');
     }
   });
 });
diff --git a/src/prompt/steps/step-iterate/handlers/container-task/utils/docker-utils.ts b/src/prompt/steps/step-iterate/handlers/container-task/utils/docker-utils.ts
index 649ae2b..6361eda 100644
--- a/src/prompt/steps/step-iterate/handlers/container-task/utils/docker-utils.ts
+++ b/src/prompt/steps/step-iterate/handlers/container-task/utils/docker-utils.ts
@@ -206,15 +206,18 @@ function addDirectoryToPack(pack: tar.Pack, directoryPath: string, relativeTo: s
   const items = fs.readdirSync(directoryPath);
   const dirStats = fs.lstatSync(directoryPath);
 
-  pack.entry(
-    {
-      name: relativeTo.endsWith('/') ? relativeTo : relativeTo + '/',
-      type: 'directory',
-      mode: dirStats.mode & 0o777,
-      mtime: dirStats.mtime,
-    },
-    () => {},
-  );
+  // Only add directory entry if not the root (to avoid creating a '/' entry which can cause issues)
+  if (relativeTo !== '') {
+    pack.entry(
+      {
+        name: relativeTo.endsWith('/') ? relativeTo : relativeTo + '/',
+        type: 'directory',
+        mode: dirStats.mode & 0o777,
+        mtime: dirStats.mtime,
+      },
+      () => {},
+    );
+  }
 
   for (const item of items) {
     const fullPath = path.join(directoryPath, item);

From 047e650279054379336d0c06ebfc06b6851ec942 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Sat, 31 Jan 2026 11:33:55 +0000
Subject: [PATCH 3/3] Remove debug logging from docker-utils test

Co-authored-by: gtanczyk <1281113+gtanczyk@users.noreply.github.com>
---
 e2e-tests/docker-utils.test.ts | 3 ---
 1 file changed, 3 deletions(-)

diff --git a/e2e-tests/docker-utils.test.ts b/e2e-tests/docker-utils.test.ts
index cf91157..5597885 100644
--- a/e2e-tests/docker-utils.test.ts
+++ b/e2e-tests/docker-utils.test.ts
@@ -238,9 +238,6 @@ describe('Docker Utils E2E Tests', () => {
         .filter((f) => f.length > 0)
         .sort();
 
-      // Log actual files for debugging in case of flaky failures
-      console.log('Files found in container:', JSON.stringify(files, null, 2));
-
       expect(files).toContain('/data/root.txt');
       expect(files).toContain('/data/nested/nested.txt');
       // Note: Empty directories are included in the transfer