Files
Artemis-Eapil/tests/unit/mcp/test_input_verification.py
chenmingxuan be6e8c2abd feat: release 2.2.0 with PC-only tests and execution reliability fixes
Unify task scheduling and queue state, fix Android input and observation recovery, preserve screenshot paths, and match final review criteria accurately. Add collapsible copyable results and publish the complete 2.1.0 upgrade notes. Validation: 3189 backend tests, 326 frontend tests, 24 client tests, production build and quality gates.
2026-10-27 01:13:41 +08:00

97 lines
3.3 KiB
Python

"""Pro and Flash must judge clear versus fill against fresh input evidence."""
from types import SimpleNamespace
from unittest.mock import AsyncMock, Mock
import pytest
from artemis.agents.validator.action_execution import exec_action
from artemis.mcp.action_executor import McpActionExecutor
from artemis.mcp.action_types import ActionResult, ObserveResult
@pytest.mark.asyncio
@pytest.mark.parametrize("mode", ["pro", "flash"])
@pytest.mark.parametrize(
("text", "clear", "observed", "success"),
[
("", True, "", True),
("", True, "AR", False),
("", True, None, False),
("new", True, "new", True),
("new", True, "", False),
("new", False, "oldnew", True),
("new", False, None, False),
("", False, "", True),
("", False, "old", True),
],
)
async def test_input_verification_matches_intent(mode, text, clear, observed, success):
ctx = SimpleNamespace(
data_engine=None, device=SimpleNamespace(device_width=100, device_height=100)
)
element = {
"class": "android.widget.EditText",
"text": observed,
"focused": True,
"bounds": "[0,0][100,100]",
"center": [50, 50],
}
session = Mock()
session.call = AsyncMock(return_value=ActionResult.success("input_text", "Dispatched"))
session.ui_hierarchy = AsyncMock(return_value=[element])
if mode == "pro":
action = {
"action": "focus_and_input_text",
"coordinates": [50, 50],
"text": text,
"clear_before_input": clear,
}
ok, _ = await exec_action(ctx, session, action)
else:
executor = McpActionExecutor(ctx, actuator=Mock())
executor._session_or_start = AsyncMock(return_value=session)
executor._observe_after_action = AsyncMock(
return_value=(ObserveResult(ok=True, elements=[element]), True)
)
result = await executor._execute_device_action(
"input_text",
"input_text",
{"text": text, "clear_exist": clear},
"input",
SimpleNamespace(latest_screenshot=None, indexed_elements=[]),
)
ok = result.status == "success"
assert ok is success
session.call.assert_awaited_once()
@pytest.mark.asyncio
@pytest.mark.parametrize("observation_ok", [True, False])
async def test_flash_cannot_confirm_clear_without_fresh_hierarchy(observation_ok):
ctx = SimpleNamespace(data_engine=None, device=None)
session = Mock()
session.call = AsyncMock(return_value=ActionResult.success("input_text", "Dispatched"))
executor = McpActionExecutor(ctx, actuator=Mock())
executor._session_or_start = AsyncMock(return_value=session)
executor._observe_after_action = AsyncMock(
return_value=(
ObserveResult(
ok=observation_ok,
hierarchy_ok=False,
elements=[{"focused": True, "input_value_state": "empty", "center": [1, 1]}],
),
True,
)
)
result = await executor._execute_device_action(
"input_text",
"input_text",
{"text": "", "clear_exist": True},
"clear",
SimpleNamespace(latest_screenshot=None, indexed_elements=[]),
)
assert result.status == "error"
assert result.metadata["input_verification"] == "unknown"
session.call.assert_awaited_once()