fix for sandbox start_background_job

packages/prime-sandboxes/src/prime_sandboxes/sandbox.py

@@ -435,14 +435,14 @@ def start_background_job(
         stderr_log_file_quoted = shlex.quote(stderr_log_file)
         # Wrap command in subshell so 'exit' terminates the subshell, not the outer shell.
         # This ensures 'echo $?' always runs to capture the exit code.
-        sh_command = f"({command_body}); echo $? > {exit_file_quoted}"
+        sh_command = (
+            f"({command_body}) > {stdout_log_file_quoted} 2> {stderr_log_file_quoted}; "
+            f"echo $? > {exit_file_quoted}"
+        )
         quoted_sh_command = shlex.quote(sh_command)
 
-        # Start detached process with separate stdout and stderr log files
-        bg_cmd = (
-            f"nohup sh -c {quoted_sh_command} "
-            f"> {stdout_log_file_quoted} 2> {stderr_log_file_quoted} &"
-        )
+        # Outer nohup redirects to /dev/null since output goes to log files inside sh -c
+        bg_cmd = f"nohup sh -c {quoted_sh_command} < /dev/null > /dev/null 2>&1 &"
         self.execute_command(sandbox_id, bg_cmd, timeout=10)
 
         return BackgroundJob(
@@ -979,14 +979,14 @@ async def start_background_job(
         stderr_log_file_quoted = shlex.quote(stderr_log_file)
         # Wrap command in subshell so 'exit' terminates the subshell, not the outer shell.
         # This ensures 'echo $?' always runs to capture the exit code.
-        sh_command = f"({command_body}); echo $? > {exit_file_quoted}"
+        sh_command = (
+            f"({command_body}) > {stdout_log_file_quoted} 2> {stderr_log_file_quoted}; "
+            f"echo $? > {exit_file_quoted}"
+        )
         quoted_sh_command = shlex.quote(sh_command)
 
-        # Start detached process with separate stdout and stderr log files
-        bg_cmd = (
-            f"nohup sh -c {quoted_sh_command} "
-            f"> {stdout_log_file_quoted} 2> {stderr_log_file_quoted} &"
-        )
+        # Outer nohup redirects to /dev/null since output goes to log files inside sh -c
+        bg_cmd = f"nohup sh -c {quoted_sh_command} < /dev/null > /dev/null 2>&1 &"
         await self.execute_command(sandbox_id, bg_cmd, timeout=10)
 
         return BackgroundJob(

packages/prime-sandboxes/tests/test_command_execution.py

@@ -226,6 +226,28 @@ def test_start_background_job(sandbox_client, shared_sandbox):
     print(f"✓ Background execution completed: {status.stdout.strip()}")
 
 
+def test_start_background_job_returns_immediately(sandbox_client, shared_sandbox):
+    """Test that start_background_job returns immediately without waiting for the job."""
+    print("\nTesting start_background_job returns immediately...")
+
+    start_time = time.time()
+    job = sandbox_client.start_background_job(
+        shared_sandbox.id,
+        "sleep 30 && echo done",
+    )
+    elapsed = time.time() - start_time
+
+    assert job.job_id is not None
+    # Should return in under 5 seconds, not 30
+    assert elapsed < 5, f"start_background_job took {elapsed:.1f}s, expected < 5s"
+    print(f"✓ Job started in {elapsed:.2f}s (sleep 30 is running in background)")
+
+    # Verify job is still running (not completed yet)
+    status = sandbox_client.get_background_job(shared_sandbox.id, job)
+    assert not status.completed, "Job should still be running"
+    print("✓ Job correctly running in background")
+
+
 def test_start_background_job_with_working_dir(sandbox_client, shared_sandbox):
     """Test start_background_job with working directory"""
     sandbox_client.execute_command(shared_sandbox.id, "mkdir -p /tmp/bgtest")