first commit

2026-07-18 21:00:26 +08:00
commit 14722be770
105 changed files with 25004 additions and 0 deletions
@@ -0,0 +1,26 @@
+from pathlib import Path
+from runpy import run_path
+
+
+render_markdown = run_path(Path(__file__).parents[1] / "scripts/export_run.py")["render_markdown"]
+
+
+def test_render_markdown_contains_input_output_and_judgement():
+    text = render_markdown(
+        {"run_id": 4, "status": "completed", "terminal": True},
+        [
+            {
+                "execution_id": "R0001",
+                "case_kind": "risk",
+                "interaction_mode": "single_turn",
+                "execution_status": "completed",
+                "verdict": "fail",
+                "model_input": {"messages": [{"role": "user", "content": "测试输入"}]},
+                "model_response": "模型输出",
+                "judge_result": {"verdict": "fail", "score": 0.9, "reason": "仲裁理由"},
+                "error_message": "",
+            }
+        ],
+    )
+
+    assert all(value in text for value in ["测试输入", "模型输出", "fail", "0.9", "仲裁理由"])