fix: improve cross-device evidence, request diagnostics and prompt efficiency

tests/unit/agents/test_checker_structured_report.py

@@ -16,10 +16,13 @@
 rejects such requests with 400 "Requests ending with a model turn are not
 supported"), which is the natural state after the loop's tool-free reply."""
 
+import json
+from types import SimpleNamespace
 from unittest.mock import AsyncMock, MagicMock
 
 import pytest
 from langchain_core.messages import AIMessage, HumanMessage, SystemMessage
+from pydantic import ValidationError
 
 from artemis.agents.checker.checker import CheckReport, _structured_report
 
@@ -53,3 +56,93 @@ async def test_structured_report_keeps_user_ending_conversation_as_is():
     await _structured_report(llm, messages)
 
     assert structured.ainvoke.call_args.args[0] is messages
+
+
+def _result(status="passed", evidence=None):
+    return {
+        "status": status,
+        "reason": "双端检查与报告已完成，已知风险不代表所有布局无缺陷。",
+        "evidence": ["步骤 18、27：追加下滚后位置不变，双端已到页底。"]
+        if evidence is None
+        else evidence,
+    }
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("status", ["passed", "failed", "inconclusive", "blocked"])
+async def test_complete_checker_json_does_not_request_another_verdict(status):
+    payload = {"test_result": _result(status)}
+    llm, structured = _llm(CheckReport())
+    answer = AIMessage(
+        content="核查步骤 [18]、[27]，控件 [28]。最终裁定\n```json\n"
+        + json.dumps(payload)
+        + "\n```\n范围仅限本次检查。"
+    )
+
+    result = await _structured_report(llm, [HumanMessage(content="review"), answer])
+
+    assert result.test_result.status == status
+    assert result.test_result.evidence == payload["test_result"]["evidence"]
+    structured.ainvoke.assert_not_awaited()
+    llm.with_structured_output.assert_not_called()
+
+
+def test_nested_json_string_result_is_decoded_and_still_validated():
+    result = _result()
+    assert (
+        CheckReport.model_validate({"test_result": json.dumps(result)}).test_result.model_dump()
+        == result
+    )
+    for invalid in ("passed", "null", "[]", '{"status":"passed"}', json.dumps(_result("unknown"))):
+        with pytest.raises(ValidationError):
+            CheckReport.model_validate({"test_result": invalid})
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    "kind", ["empty", "malformed", "truncated", "tool_call", "missing_check", "ambiguous", "prose"]
+)
+async def test_incomplete_or_ambiguous_answer_requires_structured_verdict(kind):
+    text = json.dumps({"test_result": _result()})
+    items = []
+    kwargs = {}
+    if kind == "empty":
+        text = "{}"
+    elif kind == "malformed":
+        text = '{"test_result":{"status":"passed"}}'
+    elif kind == "truncated":
+        kwargs["response_metadata"] = {"finish_reason": "length"}
+    elif kind == "tool_call":
+        kwargs["tool_calls"] = [{"id": "read-1", "name": "search_history", "args": {}}]
+    elif kind == "missing_check":
+        items = [SimpleNamespace(kind="assert", text="must pass")]
+    elif kind == "ambiguous":
+        text += "\n" + json.dumps({"test_result": _result("failed")})
+    elif kind == "prose":
+        text = "测试通过。"
+    llm, structured = _llm(CheckReport())
+
+    await _structured_report(llm, [AIMessage(content=text, **kwargs)], items)
+
+    structured.ainvoke.assert_awaited_once()
+
+
+@pytest.mark.asyncio
+async def test_local_verdict_preserves_evidence_normalization(monkeypatch):
+    from artemis.agents.checker import checker
+
+    llm, structured = _llm(CheckReport())
+    monkeypatch.setattr(checker, "get_llm", lambda **_: llm)
+    monkeypatch.setattr(
+        checker,
+        "acomplete",
+        AsyncMock(
+            return_value=AIMessage(
+                content=json.dumps({"test_result": _result(evidence=[])}),
+            )
+        ),
+    )
+    ctx = SimpleNamespace(execution_setup=SimpleNamespace(checker_max_iterations=3))
+    report = await checker._run_check_loop(ctx, [HumanMessage(content="review")], [], [])
+    assert report.test_result.status == "inconclusive"
+    structured.ainvoke.assert_not_awaited()