feat: add tool-call fallback and resilient batch exports

tests/unit/agents/test_flash_runner_ledger.py

@@ -384,7 +384,7 @@ def _make_runner(mock_context, responses, max_turns=None):
     pending = list(responses)
 
     async def _invoke(llm, tools, messages):
-        runner.calls.append({"tools": list(tools), "messages": list(messages)})
+        runner.calls.append({"llm": llm, "tools": list(tools), "messages": list(messages)})
         return pending.pop(0)
 
     runner._invoke_model = AsyncMock(side_effect=_invoke)
@@ -647,9 +647,11 @@ async def test_flash_stops_repeated_no_tool_responses(mock_context, monkeypatch,
     from artemis.config.settings import settings
 
     monkeypatch.setattr(settings, "ARTEMIS_FLASH_MAX_NO_PROGRESS_ATTEMPTS", attempts)
-    responses = [AIMessage(content="I will click now.", tool_calls=[]) for _ in range(attempts)]
+    monkeypatch.setattr(settings, "ARTEMIS_TOOL_FALLBACK_MAX_NO_PROGRESS_ATTEMPTS", 3)
+    responses = [AIMessage(content="I will click now.", tool_calls=[]) for _ in range(attempts + 3)]
     with (
         patch("artemis.controllers.unified_controller.get_driver"),
+        patch("artemis.agents.flash.runner.get_llm", return_value=Mock()) as get_fallback,
         patch(
             "artemis.agents.flash.runner.capture_screenshot_and_parse_ui",
             AsyncMock(return_value=_INITIAL_OBSERVATION),
@@ -659,11 +661,55 @@ async def test_flash_stops_repeated_no_tool_responses(mock_context, monkeypatch,
         result = await runner.run(State(initial_goal="Open Settings"))
 
     assert result["status"] == "failed"
-    assert f"连续 {attempts} 次未调用工具" in result["explanation"]
-    assert runner._invoke_model.await_count == attempts
+    assert f"连续 {attempts} 次" in result["explanation"]
+    assert "兜底模型连续 3 次" in result["explanation"]
+    assert runner._invoke_model.await_count == attempts + 3
+    get_fallback.assert_called_once_with(mock_context, name="operator", use_fallback=True)
     runner.executor.execute.assert_not_awaited()
 
 
+@pytest.mark.asyncio
+async def test_flash_uses_fallback_for_one_tool_turn_then_primary(mock_context):
+    no_call = AIMessage(content="I will click now.")
+    observe = AIMessage(
+        content="Check the target.",
+        tool_calls=[
+            {
+                "name": "ask_explorer",
+                "args": {"question": "Find Settings"},
+                "id": "fallback-observe",
+            }
+        ],
+    )
+    click = AIMessage(
+        content="Tap Settings.",
+        tool_calls=[{"name": "click", "args": {"target": 1}, "id": "fallback-click"}],
+    )
+    fallback = Mock()
+    responses = [no_call, no_call, no_call, observe, click, _report()]
+    with (
+        patch("artemis.controllers.unified_controller.get_driver"),
+        patch("artemis.agents.flash.runner.get_llm", return_value=fallback),
+        patch(
+            "artemis.agents.flash.runner.capture_screenshot_and_parse_ui",
+            AsyncMock(return_value=_INITIAL_OBSERVATION),
+        ),
+    ):
+        runner = _make_runner(mock_context, responses, max_turns=0)
+        result = await runner.run(State(initial_goal="Open Settings"))
+
+    primary = runner.calls[0]["llm"]
+    assert [call["llm"] for call in runner.calls] == [
+        primary,
+        primary,
+        primary,
+        fallback,
+        fallback,
+        primary,
+    ]
+    assert result["status"] == "completed"
+
+
 # ---------------------------------------------------------------------------
 # _TurnRecord.result: the turn's execution result in the validator-report shape
 # ---------------------------------------------------------------------------

tests/unit/agents/test_operator_recovery.py

@@ -107,23 +107,56 @@ async def test_invalid_target_can_be_corrected_before_limit(operator, monkeypatc
 @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)
+    monkeypatch.setattr(settings, "ARTEMIS_TOOL_FALLBACK_MAX_NO_PROGRESS_ATTEMPTS", 3)
+    fallback = Mock()
+    get_fallback = Mock(return_value=fallback)
+    monkeypatch.setattr("artemis.agents.operator.operator.get_llm", get_fallback)
     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
+    assert invoke.await_count == limit + 3
+    get_fallback.assert_called_once_with(ctx=node.ctx, name="operator", use_fallback=True)
+    assert invoke.call_args_list[-1].args[0] is fallback.bind_tools.return_value
+
+
+@pytest.mark.asyncio
+async def test_operator_fallback_is_scoped_to_current_step(operator, monkeypatch):
+    node, state = operator
+    primary = Mock()
+    fallback = Mock()
+    get_fallback = Mock(return_value=fallback)
+    monkeypatch.setattr("artemis.agents.operator.operator.get_llm", get_fallback)
+    invoke = AsyncMock(side_effect=[AIMessage(content="no tool")] * 3 + [click(), click()])
+    monkeypatch.setattr("artemis.agents.operator.operator.acomplete", invoke)
+
+    first = await node._invoke_llm_loop(primary, [], [], [], state)
+    second = await node._invoke_llm_loop(primary, [], [], [], state)
+
+    assert first[0] and second[0]
+    assert get_fallback.call_count == 1
+    assert [call.args[0] for call in invoke.call_args_list] == [
+        primary.bind_tools.return_value,
+        primary.bind_tools.return_value,
+        primary.bind_tools.return_value,
+        fallback.bind_tools.return_value,
+        primary.bind_tools.return_value,
+    ]
 
 
 @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")])
+    invoke = AsyncMock(
+        side_effect=[click("missing"), AIMessage(content="Retry."), click("14")] + [click("14")] * 3
+    )
+    monkeypatch.setattr("artemis.agents.operator.operator.get_llm", Mock(return_value=Mock()))
     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 invoke.await_count == 6
     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)