feat: release 1.5.1 with human review and batch filters

tests/unit/admin_console/test_automation.py

@@ -780,7 +780,7 @@ def test_failed_report_exposes_logged_reason_and_formats_chinese_markdown(isolat
     )
     report = reports.session_report(job["session_id"])
     assert "连续 3 次" in report["error"]
-    assert report["verdict"] == "测试阻塞"
+    assert report["verdict"] == "测试阻塞（AI确认）"
     assert report["test_result"]["status"] == "blocked"
     assert "连续 3 次" in report["test_result"]["reason"]
     assert "abc123secret" not in reports.failure_reason(
@@ -844,8 +844,8 @@ def test_prompt_based_verdict_is_consistent_in_history_batch_and_reports(
     report = client.get(f"/api/automation/reports/{job['session_id']}").json()
     for item in (history, batch_job, report):
         assert item["status"] == "completed"
-        assert item["test_result"] == result
-    assert report["verdict"] == "测试不通过"
+        assert item["test_result"] == {**result, "source": "ai"}
+    assert report["verdict"] == "测试不通过（AI确认）"
     paths = reports.export_session(job["session_id"])
     doc = Document(next(path for path in paths if path.suffix == ".docx"))
     doc_text = "\n".join(
@@ -856,10 +856,133 @@ def test_prompt_based_verdict_is_consistent_in_history_batch_and_reports(
     assert result["evidence"][0] in doc_text
     wb = load_workbook(next(path for path in paths if path.suffix == ".xlsx"))
     values = [cell.value for row in wb["任务结果"].iter_rows() for cell in row]
-    assert "测试不通过" in values
+    assert "测试不通过（AI确认）" in values
     assert result["reason"] in values
 
 
+def test_human_review_overrides_every_report_preserves_ai_and_invalidates_batch_cache(
+    client, isolated, monkeypatch
+):
+    from artemis.runtime import case_result
+    from apps.admin_console.routers import sessions as sessions_api
+
+    monkeypatch.setattr(case_result, "TRACES_PATH", reports.TRACES_PATH)
+    batch = service.create_batch(payload())
+    jobs = isolated.jobs(batch["id"])
+    for job in jobs:
+        isolated.save_job({**job, "status": "completed", "ended_at": time.time()})
+    session_id = jobs[0]["session_id"]
+    ai = {"status": "failed", "reason": "AI识别到预期不匹配", "evidence": ["步骤 3 截图"]}
+    directory = reports.TRACES_PATH / session_id
+    directory.mkdir(parents=True)
+    outcome_file = directory / "run_outcome.json"
+    original = json.dumps({"task_status": "completed", "test_result": ai}, ensure_ascii=False)
+    outcome_file.write_text(original, encoding="utf-8")
+    archive = reports.export_batch(batch["id"])
+    isolated.put("batch", batch["id"], {**batch, "report_ready": True})
+
+    response = client.put(
+        f"/api/automation/reports/{session_id}/review",
+        json={"status": "passed", "reason": "已人工核对，符合预期"},
+    )
+    assert response.status_code == 200
+    assert response.json()["status"] == "completed"
+    assert response.json()["verdict"] == "测试通过（人工确认）"
+    assert outcome_file.read_text(encoding="utf-8") == original
+    assert not archive.exists()
+    assert not isolated.get("batch", batch["id"])["report_ready"]
+    assert len(isolated.get("case_review", session_id)["reviews"]) == 1
+    for result in (
+        response.json(),
+        client.get("/api/automation/history", params={"q": session_id}).json()["items"][0],
+        client.get(f"/api/automation/batches/{batch['id']}").json()["jobs"][0],
+    ):
+        assert result["test_result"]["source"] == "human"
+        assert result["test_result"]["status"] == "passed"
+        assert result["test_result"]["ai_result"] == {**ai, "source": "ai"}
+        assert result["test_result"]["reviewed_at"] > 0
+
+    monkeypatch.setattr(sessions_api, "store", isolated)
+    monkeypatch.setattr(sessions_api, "TRACES_PATH", reports.TRACES_PATH)
+    monkeypatch.setattr(
+        sessions_api.session_repo,
+        "get_session_by_id",
+        lambda sid: {"session_id": sid, "status": "completed"},
+    )
+    client.app.include_router(sessions_api.router)
+    detail = client.get(f"/api/sessions/{session_id}").json()
+    assert detail["test_result"]["source"] == "human"
+    assert detail["batch_id"] == batch["id"]
+
+    downloaded = client.get(f"/api/automation/batches/{batch['id']}/archive")
+    assert downloaded.status_code == 200
+    with zipfile.ZipFile(BytesIO(downloaded.content)) as bundle:
+        wb = load_workbook(BytesIO(bundle.read("批次结果.xlsx")))
+        assert wb.active["D2"].value == "测试通过（人工确认）"
+        assert "AI确认" in wb.active["D3"].value
+        doc = Document(BytesIO(bundle.read("批次汇总.docx")))
+        assert doc.tables[0].rows[1].cells[3].text == "测试通过（人工确认）"
+
+    changed = client.put(f"/api/automation/reports/{session_id}/review", json={"status": "failed"})
+    assert changed.status_code == 200
+    assert changed.json()["verdict"] == "测试不通过（人工确认）"
+    assert len(isolated.get("case_review", session_id)["reviews"]) == 2
+    assert outcome_file.read_text(encoding="utf-8") == original
+
+
+def test_review_rejects_active_missing_invalid_and_packing_tasks(client, isolated, monkeypatch):
+    job = isolated.jobs(service.create_batch(payload())["id"])[0]
+    url = f"/api/automation/reports/{job['session_id']}/review"
+    assert client.put(url, json={"status": "passed"}).status_code == 409
+    assert isolated.get("case_review", job["session_id"]) is None
+    assert client.put(url, json={"status": "inconclusive"}).status_code == 422
+    assert client.put(url, json={"status": "passed", "source": "human"}).status_code == 422
+    assert client.put(url, json={"status": "passed", "reason": "字" * 2001}).status_code == 422
+    assert (
+        client.put("/api/automation/reports/missing/review", json={"status": "passed"}).status_code
+        == 404
+    )
+    isolated.save_job({**job, "status": "completed"})
+    monkeypatch.setattr(api.archive_service, "status", lambda *args: {"status": "preparing"})
+    assert client.put(url, json={"status": "passed"}).status_code == 409
+    assert isolated.get("case_review", job["session_id"]) is None
+
+
+def test_batch_filter_searches_full_history_and_reviews_are_removed_with_batch(client, isolated):
+    first, second = [service.create_batch(payload(name=name)) for name in ("第一批", "第二批")]
+    for batch in (first, second):
+        for job in isolated.jobs(batch["id"]):
+            isolated.save_job({**job, "status": "completed"})
+    with isolated.connection() as conn:
+        conn.execute(
+            "INSERT INTO sessions(session_id,initial_goal,start_time,status) VALUES (?,?,?,?)",
+            ("single-case", "单独检查", time.time(), "completed"),
+        )
+        conn.commit()
+    filtered = client.get(
+        "/api/automation/history",
+        params={"batch_id": first["id"], "q": "用例", "size": 1, "page": 2},
+    ).json()
+    assert filtered["total"] == 2
+    assert len(filtered["items"]) == 1
+    assert filtered["items"][0]["batch_id"] == first["id"]
+    assert (
+        client.get("/api/automation/history", params={"batch_id": "__batch__"}).json()["total"] == 4
+    )
+    singles = client.get("/api/automation/history", params={"batch_id": "__single__"}).json()
+    assert [item["session_id"] for item in singles["items"]] == ["single-case"]
+    assert client.get("/api/automation/history", params={"batch_id": "bad/path"}).status_code == 400
+    reviewed_id = isolated.jobs(first["id"])[0]["session_id"]
+    isolated.save_case_review(reviewed_id, "passed", "人工核对")
+    isolated.begin_delete_batch(first["id"])
+    isolated.delete_batch(first["id"])
+    assert isolated.get("case_review", reviewed_id) is None
+    assert len(isolated.jobs(second["id"])) == 2
+    isolated.save_case_review("single-case", "failed", "不符合预期")
+    isolated.delete_history("single-case")
+    assert isolated.get("case_review", "single-case") is None
+
+
 def test_task_material_zip_contains_ordered_evidence_and_reports(
     client, isolated, tmp_path, monkeypatch
 ):