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

tests/unit/admin_console/test_automation.py

@@ -583,7 +583,11 @@ async def test_fail_fast_includes_skipped_reports(isolated, monkeypatch):
     assert [j["status"] for j in isolated.jobs(batch["id"])] == ["failed", "skipped"]
     archive = reports.export_batch(batch["id"])
     with zipfile.ZipFile(archive) as z:
-        assert len(z.namelist()) == 8  # summaries + two reports and evidence index per task
+        assert len(z.namelist()) == 10  # summaries + reports, diagnostics and evidence per task
+        diagnostics = [name for name in z.namelist() if name.endswith("模型请求诊断.json")]
+        assert len(diagnostics) == 2
+        for name in diagnostics:
+            assert json.loads(z.read(name))["schema_version"] == 1
         assert len([name for name in z.namelist() if name.endswith("证据索引.json")]) == 2
         assert sum(n.endswith(".docx") for n in z.namelist()) == 3