fix: bound task failures and release stalled workers

tests/unit/agents/test_flash_runner.py

@@ -15,6 +15,9 @@
 """Unit tests for Universal FlashRunner."""
 
 import base64
+from io import BytesIO
+from PIL import Image
+from artemis.mcp.observation import DeviceObservationError
 from types import SimpleNamespace
 from unittest.mock import AsyncMock, Mock, patch
 import pytest
@@ -72,9 +75,18 @@ def test_flash_runner_tools_initialization(mock_context):
 # --- Post-action screenshot fallback (failed action / unknown tool) ----------------
 
 
+def _post_image():
+    buffer = BytesIO()
+    Image.new("RGB", (64, 64), "white").save(buffer, format="PNG")
+    return buffer.getvalue()
+
+
+_POST_IMG = _post_image()
+
+
 def _screen(elements):
     return SimpleNamespace(
-        base64=base64.b64encode(b"POST-IMG").decode("ascii"),
+        base64=base64.b64encode(_POST_IMG).decode("ascii"),
         elements=elements,
         width=1080,
         height=2400,
@@ -120,7 +132,7 @@ async def test_fallback_observation_renders_the_indexed_ui_list(mock_context, tm
             exec_result, "click", frozenset({"click"}), state
         )
 
-    assert post == b"POST-IMG"
+    assert post == _POST_IMG
     assert isinstance(exec_result.ui_elements_text, str)
     assert exec_result.ui_elements_text == "[1] Text: 'Settings' | Bounds: [37,50][962,112]"
     assert "{'text'" not in exec_result.ui_elements_text
@@ -154,8 +166,8 @@ async def test_fallback_observation_is_skipped_for_non_action_tools(mock_context
 
 
 @pytest.mark.asyncio
-async def test_fallback_observation_keeps_the_executors_ui_list(mock_context, tmp_path):
-    """An action that reported a UI list but no screenshot keeps its own list."""
+async def test_fallback_observation_replaces_stale_executor_ui_list(mock_context, tmp_path):
+    """Fresh screenshots and element lists must describe the same observation."""
     with patch("artemis.controllers.unified_controller.get_driver"):
         runner = FlashRunner(mock_context, goal="g")
     runner.controller = Mock()
@@ -175,12 +187,13 @@ async def test_fallback_observation_keeps_the_executors_ui_list(mock_context, tm
             exec_result, "click", frozenset({"click"}), None
         )
 
-    assert post == b"POST-IMG"
-    assert exec_result.ui_elements_text == "[1] Text: 'Mine' | Bounds: [0,0][10,10]"
+    assert post == _POST_IMG
+    assert "Other" in exec_result.ui_elements_text
+    assert "Mine" not in exec_result.ui_elements_text
 
 
 @pytest.mark.asyncio
-async def test_fallback_observation_failure_is_contained(mock_context):
+async def test_fallback_observation_failure_stops_after_bounded_recovery(mock_context):
     with patch("artemis.controllers.unified_controller.get_driver"):
         runner = FlashRunner(mock_context, goal="g")
     runner.controller = Mock()
@@ -189,11 +202,13 @@ async def test_fallback_observation_failure_is_contained(mock_context):
     state.latest_screenshot = "old.jpg"
     exec_result = _failed_click()
 
-    post = await runner._capture_post_screenshot(exec_result, "click", frozenset({"click"}), state)
-
-    assert post is None
-    assert exec_result.ui_elements_text is None
-    assert state.latest_screenshot == "old.jpg"
+    with patch("artemis.mcp.observation.settings.ARTEMIS_OBSERVATION_MAX_ATTEMPTS", 3):
+        with pytest.raises(DeviceObservationError, match="adb gone"):
+            await runner._capture_post_screenshot(exec_result, "click", frozenset({"click"}), state)
+    assert runner.controller.get_screen_data.await_count == 3
+    assert state.latest_screenshot is None
+    assert state.indexed_elements == []
+    assert state.indexed_points == []
 
 
 def test_flash_runner_executor_follows_the_flash_explorer_tier(mock_context):

tests/unit/agents/test_flash_runner_ledger.py

@@ -213,7 +213,8 @@ def test_observation_tail_has_pro_shape(mock_context):
         texts = [b["text"] for b in later.content]
         assert OBSERVATION_HEADER_RE.match(texts[0])
         assert "objective" not in "".join(texts).lower()
-        assert texts[1:] == [
+        assert "当前没有可用的元素索引" in texts[1]
+        assert texts[2:] == [
             "nudge",
             "[INJECTED]",
             'User instruction: "stop"',
@@ -408,6 +409,98 @@ def _report(status="completed", tc_id="tc-report", **extra):
 _INITIAL_OBSERVATION = ("shot0.png", b"IMG0", "[1] Settings")
 
 
+@pytest.mark.asyncio
+@pytest.mark.parametrize("fallback_succeeds", [True, False])
+async def test_failed_actions_have_finite_fallback_and_restore_primary(
+    mock_context, monkeypatch, fallback_succeeds
+):
+    monkeypatch.setattr("artemis.agents.flash.runner.settings.ARTEMIS_ACTION_FAILURE_ATTEMPTS", 3)
+    monkeypatch.setattr(
+        "artemis.agents.flash.runner.settings.ARTEMIS_TOOL_FALLBACK_MAX_NO_PROGRESS_ATTEMPTS", 3
+    )
+    action_count = 4 if fallback_succeeds else 6
+    responses = [
+        AIMessage(
+            content="try", tool_calls=[{"name": "click", "args": {"index": 1}, "id": f"tc-{i}"}]
+        )
+        for i in range(action_count)
+    ]
+    if fallback_succeeds:
+        responses.append(_report())
+    with patch("artemis.controllers.unified_controller.get_driver"):
+        runner = _make_runner(mock_context, responses)
+    fallback = Mock()
+    count = 0
+
+    def execute(name, args, tc_id, state, **kwargs):
+        nonlocal count
+        count += 1
+        result = _exec_result(tc_id, name, b"current-screen")
+        if count <= 3 or not fallback_succeeds:
+            result.status = "error"
+            result.text_summary = "目标不存在"
+        return result
+
+    runner.executor.execute = AsyncMock(side_effect=execute)
+    with (
+        patch(
+            "artemis.agents.flash.runner.capture_screenshot_and_parse_ui",
+            AsyncMock(return_value=_INITIAL_OBSERVATION),
+        ) as observe,
+        patch("artemis.agents.flash.runner.get_llm", return_value=fallback),
+    ):
+        outcome = await runner.run(State(initial_goal="Open Settings"))
+    primary = runner._init_llm.return_value
+    assert all(call["llm"] is primary for call in runner.calls[:3])
+    assert runner.calls[3]["llm"] is fallback
+    assert observe.await_count == 2  # initial + fresh observation before fallback
+    if fallback_succeeds:
+        assert outcome["status"] == "completed"
+        assert runner.calls[-1]["llm"] is primary
+    else:
+        assert outcome["status"] == "failed"
+        assert "兜底模型连续 3" in outcome["explanation"]
+        assert "目标不存在" in outcome["explanation"]
+        assert len(runner.calls) == 6
+
+
+@pytest.mark.asyncio
+async def test_empty_current_ui_clears_previous_turn_element_list(mock_context):
+    response = AIMessage(
+        content="tap", tool_calls=[{"name": "click", "args": {"index": 1}, "id": "tc"}]
+    )
+    with patch("artemis.controllers.unified_controller.get_driver"):
+        runner = _make_runner(mock_context, [response, _report()])
+    runner.executor.execute = AsyncMock(
+        return_value=_exec_result("tc", "click", b"fresh-screen", ui="")
+    )
+    with patch(
+        "artemis.agents.flash.runner.capture_screenshot_and_parse_ui",
+        AsyncMock(return_value=_INITIAL_OBSERVATION),
+    ):
+        await runner.run(State(initial_goal="Open Settings"))
+    tail = runner.calls[-1]["messages"][-1]
+    text = "\n".join(block.get("text", "") for block in tail.content)
+    assert "当前没有可用的元素索引" in text
+    assert "[1] Settings" not in text
+
+
+@pytest.mark.asyncio
+async def test_unavailable_screen_ends_before_any_model_call(mock_context):
+    from artemis.mcp.observation import DeviceObservationError
+
+    with patch("artemis.controllers.unified_controller.get_driver"):
+        runner = _make_runner(mock_context, [])
+    with patch(
+        "artemis.agents.flash.runner.capture_screenshot_and_parse_ui",
+        AsyncMock(side_effect=DeviceObservationError("设备断连")),
+    ):
+        outcome = await runner.run(State(initial_goal="Open Settings"))
+    assert outcome["code"] == "device_observation_failed"
+    assert outcome["error"] == "设备断连"
+    runner._invoke_model.assert_not_awaited()
+
+
 @pytest.mark.asyncio
 async def test_run_builds_prompt_from_ledger_with_session_offsets(mock_context):
     """Turn 2 must see: the static system prefix, the committed turn 1 (its

tests/unit/agents/test_planner_node.py

@@ -33,6 +33,17 @@ class DummyState:
         self.operator_replan_reason = operator_replan_reason
 
 
+import base64
+from io import BytesIO
+from PIL import Image
+
+
+def _valid_screen_base64():
+    buffer = BytesIO()
+    Image.new("RGB", (64, 64), "white").save(buffer, format="PNG")
+    return base64.b64encode(buffer.getvalue()).decode("ascii")
+
+
 @pytest.fixture
 def mock_context():
     ctx = Mock(spec=ArtemisContext)
@@ -41,7 +52,7 @@ def mock_context():
     ctx.adb_client = Mock()
     ctx.ui_adb_client = Mock()
     mock_screen_data = Mock()
-    mock_screen_data.base64 = "YmFzZTY0"
+    mock_screen_data.base64 = _valid_screen_base64()
     mock_screen_data.elements = []
     mock_screen_data.width = 1080
     mock_screen_data.height = 2400