fix(run): pass the startup_command as a dict to env.execute() function. (#880)

Co-authored-by: Yi Kuang <y24kuang@uwaterloo.ca>

Author: Ethan-Kuang-6666

src/minisweagent/run/benchmarks/swebench.py

@@ -88,7 +88,7 @@ def get_sb_environment(config: dict, instance: dict) -> Environment:
     env = get_environment(env_config)
     if startup_command := config.get("run", {}).get("env_startup_command"):
         startup_command = Template(startup_command, undefined=StrictUndefined).render(**instance)
-        out = env.execute(startup_command)
+        out = env.execute({"command": startup_command})
         if out["returncode"] != 0:
             raise RuntimeError(f"Error executing startup command: {out}")
     return env

tests/run/test_swebench.py

@@ -1,6 +1,6 @@
 import json
 import re
-from unittest.mock import patch
+from unittest.mock import MagicMock, patch
 
 import pytest
 from pydantic import BaseModel
@@ -9,6 +9,7 @@
 from minisweagent.models.test_models import DeterministicModel, make_output
 from minisweagent.run.benchmarks.swebench import (
     filter_instances,
+    get_sb_environment,
     get_swebench_docker_image_name,
     main,
     remove_from_preds_file,
@@ -103,6 +104,20 @@ def test_get_image_name_with_complex_instance_id():
     assert get_swebench_docker_image_name(instance) == expected
 
 
+def test_get_sb_environment_runs_startup_command_as_dict():
+    """startup_command must be passed to env.execute() as a dict, not a bare string."""
+    fake_env = MagicMock()
+    fake_env.execute.return_value = {"returncode": 0, "output": ""}
+    instance = {"instance_id": "repo1__test1", "image_name": "custom/image:tag"}
+    # The startup_command with {{instance_id}} tested the Jinja render as well.
+    config = {"run": {"env_startup_command": "echo {{instance_id}}"}}
+
+    with patch("minisweagent.run.benchmarks.swebench.get_environment", return_value=fake_env):
+        get_sb_environment(config, instance)
+
+    fake_env.execute.assert_called_once_with({"command": "echo repo1__test1"})
+
+
 def test_filter_instances_no_filters():
     """Test filter_instances with no filtering applied"""
     instances = [{"instance_id": "repo1__test1"}, {"instance_id": "repo2__test2"}, {"instance_id": "repo3__test3"}]