feat: improve operator progress and recording accuracy

tests/unit/agents/test_operator.py

@@ -1188,140 +1188,6 @@ async def test_operator_background_tasks_prompt_injection():
         adb_registry.background.pop("task_99999", None)
 
 
-@pytest.mark.asyncio
-async def test_operator_task_plan_warning_prompt_injection():
-    from artemis.agents.operator.operator import OperatorNode
-    from artemis.context import ArtemisContext
-    from unittest.mock import AsyncMock, MagicMock, patch
-
-    mock_ctx = MagicMock(spec=ArtemisContext)
-    mock_ctx.execution_setup = None
-    mock_ctx.data_engine = None
-
-    mock_state = MagicMock()
-    mock_state.subagent_calls = []
-    mock_state.initial_goal = "Test goal"
-
-    mock_controller = MagicMock()
-    mock_controller.take_screenshot = AsyncMock(return_value="YmFzZTY0X3NjcmVlbnNob3Q=")
-    mock_device_data = MagicMock()
-    mock_device_data.elements = []
-    mock_controller.get_screen_data = AsyncMock(return_value=mock_device_data)
-
-    mock_llm = MagicMock()
-    mock_response = MagicMock()
-    mock_response.tool_calls = []
-
-    mock_llm.ainvoke = AsyncMock(return_value=mock_response)
-    mock_llm.bind_tools.return_value = mock_llm
-
-    # Case 1: task_plan updated in the last step (using the "key" argument)
-    steps_updated = [
-        {
-            "step_id": "step_1",
-            "step_number": 1,
-            "tool_calls": [{"name": "click", "args": {"target": 1}}],
-        },
-        {
-            "step_id": "step_2",
-            "step_number": 2,
-            "tool_calls": [
-                {
-                    "name": "update_note",
-                    "args": {"key": "task_plan", "target": "a", "replacement": "b"},
-                }
-            ],
-        },
-    ]
-
-    captured_messages_1 = []
-
-    async def mock_invoke_llm_loop_1(*args, **kwargs):
-        messages = await node._build_prompt(
-            state=mock_state,
-            latest_screenshot_b64="YmFzZTY0X3NjcmVlbnNob3Q=",
-            fused_xml=[],
-            minimal_list="",
-            current_step_num=3,
-            steps=steps_updated,
-            task_plan="",
-        )
-        captured_messages_1.extend(messages)
-        return [], None, None, False
-
-    with (
-        patch("artemis.agents.operator.operator.get_llm", return_value=mock_llm),
-        patch(
-            "artemis.agents.operator.operator.OperatorNode._invoke_llm_loop",
-            side_effect=mock_invoke_llm_loop_1,
-        ),
-    ):
-        node = OperatorNode(mock_ctx, transcript_config=LEGACY_TRANSCRIPT)
-        node_update = await node(mock_state)
-
-        full_content_1 = ""
-        for m in captured_messages_1:
-            if isinstance(m.content, list):
-                for part in m.content:
-                    if isinstance(part, dict) and "text" in part:
-                        full_content_1 += part["text"] + "\n"
-            elif isinstance(m.content, str):
-                full_content_1 += m.content + "\n"
-
-        assert "Reminder: your last turn executed a Turn-Ending Action" not in full_content_1
-
-    # Case 2: an action turn that did NOT update task_plan
-    steps_not_updated = [
-        {
-            "step_id": "step_1",
-            "step_number": 1,
-            "tool_calls": [{"name": "click", "args": {"target": 1}}],
-        },
-        {
-            "step_id": "step_2",
-            "step_number": 2,
-            "tool_calls": [{"name": "click", "args": {"target": 2}}],
-            "action_taken": {"action": "click", "target": 2},
-        },
-    ]
-
-    captured_messages_2 = []
-
-    async def mock_invoke_llm_loop_2(*args, **kwargs):
-        messages = await node._build_prompt(
-            state=mock_state,
-            latest_screenshot_b64="YmFzZTY0X3NjcmVlbnNob3Q=",
-            fused_xml=[],
-            minimal_list="",
-            current_step_num=3,
-            steps=steps_not_updated,
-            task_plan="",
-        )
-        captured_messages_2.extend(messages)
-        return [], None, None, False
-
-    with (
-        patch("artemis.agents.operator.operator.get_llm", return_value=mock_llm),
-        patch(
-            "artemis.agents.operator.operator.OperatorNode._invoke_llm_loop",
-            side_effect=mock_invoke_llm_loop_2,
-        ),
-    ):
-        node = OperatorNode(mock_ctx, transcript_config=LEGACY_TRANSCRIPT)
-        node_update = await node(mock_state)
-
-        full_content_2 = ""
-        for m in captured_messages_2:
-            if isinstance(m.content, list):
-                for part in m.content:
-                    if isinstance(part, dict) and "text" in part:
-                        full_content_2 += part["text"] + "\n"
-            elif isinstance(m.content, str):
-                full_content_2 += m.content + "\n"
-
-        assert "Reminder: your last turn executed a Turn-Ending Action" in full_content_2
-
-
 @pytest.mark.asyncio
 async def test_operator_tool_limit_exceeded_warning():
     from artemis.agents.operator.operator import OperatorNode