test: improve coverage and organize test layout (#3085)

Author: seratch

tests/conftest.py

@@ -20,12 +20,12 @@
         [
             "test_example_workflows.py",
             "test_run_state.py",
-            "test_sandbox_memory.py",
             "sandbox/capabilities/test_filesystem_capability.py",
             "sandbox/integration_tests/test_runner_pause_resume.py",
             "sandbox/test_client_options.py",
             "sandbox/test_exposed_ports.py",
             "sandbox/test_extract.py",
+            "sandbox/test_memory.py",
             "sandbox/test_runtime.py",
             "sandbox/test_session_manager.py",
             "sandbox/test_session_sinks.py",

tests/extensions/experiemental/codex/test_payloads.py

@@ -0,0 +1,45 @@
+from __future__ import annotations
+
+import pytest
+
+from agents.extensions.experimental.codex.items import AgentMessageItem, TodoItem, TodoListItem
+
+
+def test_dict_like_supports_mapping_access_for_dataclass_fields() -> None:
+    item = AgentMessageItem(id="item-1", text="hello")
+
+    assert item["id"] == "item-1"
+    assert item["text"] == "hello"
+    assert item["type"] == "agent_message"
+    assert item.get("text") == "hello"
+    assert item.get("missing", "fallback") == "fallback"
+    assert "id" in item
+    assert "missing" not in item
+    assert object() not in item
+    assert list(item.keys()) == ["id", "text", "type"]
+
+
+def test_dict_like_raises_key_error_for_unknown_fields() -> None:
+    item = AgentMessageItem(id="item-1", text="hello")
+
+    with pytest.raises(KeyError, match="missing"):
+        _ = item["missing"]
+
+
+def test_dict_like_as_dict_recursively_converts_nested_dataclasses() -> None:
+    item = TodoListItem(
+        id="todo-list-1",
+        items=[
+            TodoItem(text="write tests", completed=True),
+            TodoItem(text="run tests", completed=False),
+        ],
+    )
+
+    assert item.as_dict() == {
+        "id": "todo-list-1",
+        "items": [
+            {"text": "write tests", "completed": True},
+            {"text": "run tests", "completed": False},
+        ],
+        "type": "todo_list",
+    }

tests/extensions/test_sandbox_blaxel.py tests/extensions/sandbox/test_blaxel.pytests/extensions/test_sandbox_blaxel.py renamed to tests/extensions/sandbox/test_blaxel.py

@@ -11,14 +11,14 @@
 
 def _load_example_module() -> Any:
     path = (
-        Path(__file__).resolve().parents[2]
+        Path(__file__).resolve().parents[3]
         / "examples"
         / "sandbox"
         / "extensions"
         / "runloop"
         / "capabilities.py"
     )
-    module_name = "tests.extensions.runloop_capabilities_example"
+    module_name = "tests.extensions.sandbox.runloop_capabilities_example"
     spec = importlib.util.spec_from_file_location(module_name, path)
     assert spec is not None
     assert spec.loader is not None