改用Grok联网搜索并保留每轮信号

tests/test_api_health.py

@@ -35,22 +35,29 @@ def test_home_page_returns_dashboard_shell(monkeypatch) -> None:
     assert response.status_code == 200
     assert "SignalScout" in response.text
     assert "今日 AI 信号" in response.text
+    assert "手动搜索" in response.text
     assert "renderMarkdownReport" in response.text
     assert "自动记录" not in response.text
     assert "采集今日情报" not in response.text
 
 
-def test_manual_agent_run_endpoint_is_not_exposed(monkeypatch) -> None:
-    # Agent只通过后台调度器自动运行，HTTP API不提供手动启动入口。
+def test_manual_agent_run_endpoint_starts_background_job(monkeypatch) -> None:
+    # 手动搜索入口复用调度任务，方便验证每轮候选和入库数量。
     monkeypatch.setenv("SIGNALSCOUT_ENV", "test")
     get_settings.cache_clear()
+    started = []
+
+    def fake_run_scheduled_job(settings):
+        started.append(settings.env)
+
+    monkeypatch.setattr("app.api.routes.run_scheduled_job", fake_run_scheduled_job)
     with TestClient(create_app()) as client:
         response = client.post("/agent/runs/daily")
-        stream_response = client.post("/agent/runs/daily/stream")
     get_settings.cache_clear()
 
-    assert response.status_code == 404
-    assert stream_response.status_code == 404
+    assert response.status_code == 200
+    assert response.json()["status"] == "started"
+    assert started == ["test"]
 
 
 def test_dashboard_returns_news_and_github_signals_separately(monkeypatch) -> None:

tests/test_news_search_client.py

@@ -2,26 +2,50 @@ from app.core.config import Settings
 from app.integrations.news_search_client import NewsSearchClient
 
 
-def test_news_search_client_maps_article_to_candidate() -> None:
-    # 新闻搜索结果会先转为候选信号，再交给模型筛选成最终情报。
+def test_news_search_client_maps_grok_article_to_candidate() -> None:
+    # Grok Web Search结果会先转为候选信号，再进入入库和日报总结流程。
     client = NewsSearchClient(Settings())
-    signal = client._story_to_signal(
+    signal = client._article_to_signal(
         {
-            "title": "Show HN: AI agent platform launches",
-            "url": "https://example.com/ai-agent-platform",
-            "author": "builder",
-            "points": 128,
-            "num_comments": 24,
-            "created_at": "2026-07-08T12:00:00Z",
-            "objectID": "123",
+            "title": "国内大模型公司发布新一代Agent平台",
+            "summary": "中文AI新闻候选。",
+            "url": "https://example.cn/ai-agent-platform",
+            "source": "示例中文媒体",
+            "published_at": "2026-07-08T12:00:00Z",
+            "language": "zh",
+            "importance": 4,
+            "entities": ["Agent平台"],
         }
     )
 
     assert signal is not None
-    assert signal.topic == "AI 新闻候选"
+    assert signal.topic == "中文 AI 新闻"
     assert signal.source_type == "news"
-    assert signal.source_name == "Hacker News"
+    assert signal.source_name == "示例中文媒体"
     assert signal.published_at is not None
     assert signal.published_at.year == 2026
     assert signal.published_at.hour == 20
-    assert "128 points" in signal.summary
+    assert signal.importance == 4
+
+
+def test_news_search_client_parses_responses_output_text() -> None:
+    # Responses API会返回多段output，业务JSON只来自最终output_text。
+    client = NewsSearchClient(Settings())
+    payload = client._parse_response_json(
+        {
+            "output": [
+                {"type": "web_search_call", "status": "completed"},
+                {
+                    "type": "message",
+                    "content": [
+                        {
+                            "type": "output_text",
+                            "text": '{"items":[{"title":"AI news","url":"https://example.com","source":"Example"}]}',
+                        }
+                    ],
+                },
+            ]
+        }
+    )
+
+    assert payload["items"][0]["title"] == "AI news"

tests/test_repository.py

@@ -0,0 +1,81 @@
+from datetime import datetime
+
+from sqlalchemy import create_engine
+from sqlalchemy.orm import sessionmaker
+from sqlalchemy.pool import StaticPool
+
+from app.db.models import AgentRun, Base
+from app.db.repository import AgentRepository
+from app.schemas.agent import SignalItem
+
+
+def test_save_signals_keeps_same_url_across_runs() -> None:
+    # 跨运行保留同一链接，保证每轮日报的实际信号快照不会被历史结果吞掉。
+    engine = create_engine(
+        "sqlite:///:memory:",
+        connect_args={"check_same_thread": False},
+        poolclass=StaticPool,
+        future=True,
+    )
+    Base.metadata.create_all(engine)
+    TestingSessionLocal = sessionmaker(bind=engine, autoflush=False, autocommit=False, future=True)
+    with TestingSessionLocal() as session:
+        repo = AgentRepository(session)
+        first_run = AgentRun(status="running", started_at=datetime(2026, 7, 8), request_payload={})
+        second_run = AgentRun(status="running", started_at=datetime(2026, 7, 8), request_payload={})
+        session.add_all([first_run, second_run])
+        session.flush()
+        item = SignalItem(
+            source_type="news",
+            topic="AI 新闻",
+            title="Same URL",
+            summary="同一个来源链接。",
+            source_url="https://example.com/same",
+            source_name="Example",
+            published_at=datetime(2026, 7, 8),
+            importance=3,
+            entities=["Example"],
+        )
+
+        first_saved = repo.save_signals(first_run, [item])
+        second_saved = repo.save_signals(second_run, [item])
+
+        assert len(first_saved) == 1
+        assert len(second_saved) == 1
+        assert first_saved[0].run_id != second_saved[0].run_id
+
+
+def test_save_signals_updates_duplicate_url_inside_same_run() -> None:
+    # 同一运行内相同链接只保留一条，避免模型重复输出造成页面重复。
+    engine = create_engine(
+        "sqlite:///:memory:",
+        connect_args={"check_same_thread": False},
+        poolclass=StaticPool,
+        future=True,
+    )
+    Base.metadata.create_all(engine)
+    TestingSessionLocal = sessionmaker(bind=engine, autoflush=False, autocommit=False, future=True)
+    with TestingSessionLocal() as session:
+        repo = AgentRepository(session)
+        run = AgentRun(status="running", started_at=datetime(2026, 7, 8), request_payload={})
+        session.add(run)
+        session.flush()
+        first = SignalItem(
+            source_type="news",
+            topic="AI 新闻",
+            title="First title",
+            summary="第一条摘要。",
+            source_url="https://example.com/same",
+            source_name="Example",
+            published_at=datetime(2026, 7, 8),
+            importance=3,
+            entities=["Example"],
+        )
+        second = first.model_copy(update={"title": "Second title", "importance": 5})
+
+        saved = repo.save_signals(run, [first, second])
+
+        assert len(saved) == 2
+        assert saved[0].id == saved[1].id
+        assert saved[0].title == "Second title"
+        assert saved[0].importance == 5