feat: add official Codex skill and harden task execution

Install and update artemis-use with task diagnostics and plugin guidance. Preserve cross-device file receipts, review canonical reports, guard stage transitions and reduce small history summaries. Improve exploration lifecycle, request recovery and browser live connections.

tests/unit/agents/test_checker_structured_report.py

@@ -87,6 +87,58 @@ async def test_complete_checker_json_does_not_request_another_verdict(status):
     llm.with_structured_output.assert_not_called()
 
 
+@pytest.mark.asyncio
+@pytest.mark.parametrize("status", ["passed", "failed", "inconclusive"])
+async def test_complete_verdict_alias_and_evidence_list_are_not_sent_for_reformatting(status):
+    payload = {
+        "verdicts": [
+            {
+                "kind": "assert",
+                "item_text": "检查本轮页面",
+                "verdict": status,
+                "evidence": ["步骤 2 的实际页面", "步骤 3 的对照截图"],
+            }
+        ],
+        "test_result": _result(status),
+    }
+    llm, structured = _llm(CheckReport())
+    report = await _structured_report(
+        llm,
+        [AIMessage(content=json.dumps(payload))],
+        [SimpleNamespace(kind="assert", text="检查本轮页面")],
+    )
+    assert report.verdicts[0].status == status
+    assert report.verdicts[0].evidence == "步骤 2 的实际页面\n步骤 3 的对照截图"
+    assert report.test_result.model_dump() == payload["test_result"]
+    structured.ainvoke.assert_not_awaited()
+    llm.with_structured_output.assert_not_called()
+
+
+@pytest.mark.parametrize(
+    "fields",
+    [
+        {"status": "failed", "verdict": "passed", "evidence": ["图 1"]},
+        {"verdict": "unknown", "evidence": ["图 1"]},
+        {"verdict": "passed", "evidence": [{"text": "图 1"}]},
+        {"verdict": "passed", "evidence": [1]},
+    ],
+)
+def test_conflicting_or_invalid_verdict_shapes_are_not_silently_accepted(fields):
+    with pytest.raises(ValidationError):
+        CheckReport.model_validate(
+            {"verdicts": [{"kind": "assert", "item_text": "页面", **fields}]}
+        )
+
+
+def test_empty_evidence_list_does_not_turn_an_unsupported_failure_into_evidence():
+    from artemis.agents.checker.checker import _normalize_report
+
+    report = CheckReport.model_validate(
+        {"verdicts": [{"kind": "assert", "item_text": "页面", "verdict": "failed", "evidence": []}]}
+    )
+    assert _normalize_report(report, []).verdicts[0].status == "inconclusive"
+
+
 def test_nested_json_string_result_is_decoded_and_still_validated():
     result = _result()
     assert (