Artemis-Eapil/tests/unit/agents/test_operator_recovery.py

"""Regression coverage for quoted click targets and prose-only recovery loops."""

from types import SimpleNamespace
from unittest.mock import AsyncMock, Mock

from langchain_core.messages import AIMessage, SystemMessage, ToolMessage
import pytest

from artemis.agents.operator.operator import OperatorNode, OperatorNoProgressError
from artemis.config.settings import settings


@pytest.fixture
def operator(monkeypatch):
    node = object.__new__(OperatorNode)
    node.ctx = SimpleNamespace(transcript_ledger=None)
    node._available_device_actions = lambda: {"click", "long_press", "input_text"}
    node._plan_changed_this_turn = lambda: True
    node._max_burst_actions = lambda: 4
    state = SimpleNamespace(
        operator_raw_data={"width": 1280, "height": 2856},
        indexed_elements=[{"center": [310, 2610], "text": "model menu", "resource_id": "model"}]
        * 13,
        open_incident=None,
    )
    monkeypatch.setattr(settings, "ARTEMIS_OPERATOR_MAX_NO_PROGRESS_ATTEMPTS", 3)
    return node, state


@pytest.mark.parametrize("name", ["click", "long_press", "input_text"])
@pytest.mark.parametrize("target", ["13", "[240, 914]", "240, 914"])
def test_quoted_targets_keep_same_action_and_grounding(operator, name, target):
    node, state = operator
    expected_target = 13 if target == "13" else [240, 914]
    args = {"target": target, "target_description": "model menu", "text": "test"}
    actions, error = node._translate_and_validate_tool({"name": name, "args": args}, state)
    expected, _ = node._translate_and_validate_tool(
        {"name": name, "args": {**args, "target": expected_target}}, state
    )
    assert error is None
    assert actions == expected
    if target == "13":
        assert actions[0]["target_resource_id"] == "model"
    else:
        assert actions[0]["target_description"] == "model menu"


@pytest.mark.parametrize("target", ["0", "14", "model menu", True, "true"])
def test_invalid_index_is_not_guessed_or_dispatched(operator, target):
    node, state = operator
    actions, error = node._translate_and_validate_tool(
        {"name": "click", "args": {"target": target}}, state
    )
    assert not actions
    assert error


def test_quoted_coordinates_still_need_description(operator):
    node, state = operator
    actions, error = node._translate_and_validate_tool(
        {"name": "click", "args": {"target": "[240, 914]"}}, state
    )
    assert not actions
    assert "target_description" in error


def click(target="13"):
    return AIMessage(
        content="Open the model menu.",
        tool_calls=[{"name": "click", "args": {"target": target}, "id": "call_click"}],
    )


async def run_loop(operator, monkeypatch, responses):
    node, state = operator
    invoke = AsyncMock(side_effect=responses)
    monkeypatch.setattr("artemis.agents.operator.operator.acomplete", invoke)
    messages = []
    result = await node._invoke_llm_loop(Mock(), messages, [], [], state)
    return result, messages, invoke


@pytest.mark.asyncio
async def test_prose_retry_keeps_context_and_recovers_once(operator, monkeypatch):
    prose = AIMessage(content="I will click the model menu.")
    result, messages, invoke = await run_loop(operator, monkeypatch, [prose, click()])
    assert invoke.await_count == 2
    assert len(result[0]) == 1
    assert result[0][0]["coordinates"] == [310, 2610]
    assert prose in messages
    assert any(isinstance(m, SystemMessage) and "No tool call" in m.content for m in messages)


@pytest.mark.asyncio
async def test_invalid_target_can_be_corrected_before_limit(operator, monkeypatch):
    result, messages, invoke = await run_loop(operator, monkeypatch, [click("14"), click("13")])
    assert invoke.await_count == 2
    assert len(result[0]) == 1
    assert result[0][0]["coordinates"] == [310, 2610]
    outputs = [m for m in messages if isinstance(m, ToolMessage)]
    assert [m.status for m in outputs] == ["error", "success"]


@pytest.mark.asyncio
@pytest.mark.parametrize("limit", [1, 3])
async def test_empty_responses_fail_at_configured_limit(operator, monkeypatch, limit):
    monkeypatch.setattr(settings, "ARTEMIS_OPERATOR_MAX_NO_PROGRESS_ATTEMPTS", limit)
    invoke = AsyncMock(return_value=AIMessage(content="I will click now."))
    monkeypatch.setattr("artemis.agents.operator.operator.acomplete", invoke)
    node, state = operator
    with pytest.raises(OperatorNoProgressError, match="未提供可执行的工具调用"):
        await node._invoke_llm_loop(Mock(), [], [], [], state)
    assert invoke.await_count == limit


@pytest.mark.asyncio
async def test_invalid_calls_and_prose_share_bounded_recovery(operator, monkeypatch):
    invoke = AsyncMock(side_effect=[click("missing"), AIMessage(content="Retry."), click("14")])
    monkeypatch.setattr("artemis.agents.operator.operator.acomplete", invoke)
    node, state = operator
    messages = []
    with pytest.raises(OperatorNoProgressError, match="校验仍失败"):
        await node._invoke_llm_loop(Mock(), messages, [], [], state)
    assert invoke.await_count == 3
    assert any(isinstance(m, ToolMessage) and m.status == "error" for m in messages)
    assert not any(isinstance(m, ToolMessage) and m.content == "Action Recorded" for m in messages)


@pytest.mark.asyncio
async def test_iteration_cap_cannot_restart_no_progress_next_turn(operator, monkeypatch):
    monkeypatch.setattr("artemis.agents.operator.operator.OPERATOR_MAX_TOOL_ITERATIONS", 2)
    invoke = AsyncMock(return_value=AIMessage(content="I will click now."))
    monkeypatch.setattr("artemis.agents.operator.operator.acomplete", invoke)
    node, state = operator
    with pytest.raises(OperatorNoProgressError, match="预算内未恢复"):
        await node._invoke_llm_loop(Mock(), [], [], [], state)
    assert invoke.await_count == 2