重写为 AI 情报 Agent 并接入 Jenkins 流水线

2026-07-08 21:41:46 +08:00
commit f471c2a08a
45 changed files with 2163 additions and 0 deletions
@@ -0,0 +1,136 @@
+from __future__ import annotations
+
+import json
+import logging
+from datetime import date
+from typing import Iterator, Optional
+
+from app.agent.report_writer import ReportWriter
+from app.core.config import Settings
+from app.db.models import AgentRun, Signal
+from app.db.repository import AgentRepository
+from app.db.session import SessionLocal
+from app.integrations.github_client import GitHubHotProjectClient
+from app.integrations.news_search_client import NewsSearchClient
+from app.llm.grok_client import GrokIntelligenceClient
+
+logger = logging.getLogger(__name__)
+
+
+def stream_daily_agent_events(settings: Settings, run_date: Optional[date] = None) -> Iterator[str]:
+    """以SSE方式运行每日Agent，并在模型完成后逐条发送结构化情报。"""
+    db = SessionLocal()
+    repo = AgentRepository(db)
+    current_date = run_date or date.today()
+    run: Optional[AgentRun] = None
+
+    try:
+        request_payload = {
+            "run_date": current_date.isoformat(),
+            "mode": "grok_agent_stream",
+        }
+        run = repo.create_run(request_payload=request_payload)
+        db.commit()
+        logger.info("agent_stream_started run_id=%s run_date=%s mode=grok_agent", run.id, current_date)
+        yield _sse("started", {"run_id": run.id, "status": run.status})
+
+        news = NewsSearchClient(settings)
+        github = GitHubHotProjectClient(settings)
+        grok = GrokIntelligenceClient(settings)
+        topics = repo.list_enabled_topics()
+        yield _sse("stage", {"run_id": run.id, "stage": "news"})
+        news_candidates = news.collect_ai_news(run_date=current_date)
+        yield _sse("stage", {"run_id": run.id, "stage": "github"})
+        github_projects = github.collect_hot_projects(run_date=current_date)
+        yield _sse("stage", {"run_id": run.id, "stage": "grok"})
+        result = grok.collect_daily_intelligence(
+            topics=topics,
+            news_candidates=news_candidates,
+            github_projects=github_projects,
+            run_date=current_date,
+        )
+
+        saved_count = 0
+        final_signals = []
+        for item in result.signals:
+            # 模型一次性完成筛选和摘要，SSE只负责把最终情报逐条落库并推给前端。
+            signal = repo.save_or_update_signal(run=run, item=item)
+            if signal is None:
+                continue
+            db.commit()
+            db.refresh(signal)
+            final_signals.append(signal)
+            saved_count += 1
+            yield _sse("candidate", {"signal": _signal_payload(signal), "saved_count": saved_count})
+
+        writer = ReportWriter(settings.report_dir)
+        report_path = writer.write_daily_report(current_date, result.report_markdown)
+        repo.save_report(
+            run=run,
+            title=result.title,
+            content_md=result.report_markdown,
+            file_path=str(report_path),
+        )
+        repo.mark_run_complete(
+            run=run,
+            raw_response={
+                "mode": "grok_agent_stream",
+                "topic_count": len(topics),
+                "news_candidate_count": len(news_candidates),
+                "github_project_count": len(github_projects),
+                "model": settings.llm_model,
+                "model_response": result.raw_response,
+            },
+        )
+        db.commit()
+        logger.info(
+            "agent_stream_completed run_id=%s candidates=%s final_signals=%s report_path=%s",
+            run.id,
+            saved_count,
+            len(final_signals),
+            report_path,
+        )
+        yield _sse(
+            "completed",
+            {
+                "run_id": run.id,
+                "status": "completed",
+                "candidate_count": saved_count,
+                "final_signal_count": len(final_signals),
+                "report_path": str(report_path),
+            },
+        )
+    except Exception as exc:
+        if run is not None:
+            repo.mark_run_failed(run=run, error=str(exc))
+            db.commit()
+            logger.exception("agent_stream_failed run_id=%s error=%s", run.id, exc)
+            yield _sse("failed", {"run_id": run.id, "status": "failed", "error": str(exc)})
+        else:
+            logger.exception("agent_stream_failed_before_run error=%s", exc)
+            yield _sse("failed", {"run_id": None, "status": "failed", "error": str(exc)})
+    finally:
+        db.close()
+
+
+def _signal_payload(signal: Signal) -> dict:
+    # SSE载荷只使用JSON安全字段，方便前端直接追加到列表。
+    return {
+        "id": signal.id,
+        "run_id": signal.run_id,
+        "topic": signal.topic,
+        "title": signal.title,
+        "summary": signal.summary,
+        "source_url": signal.source_url,
+        "source_name": signal.source_name,
+        "published_at": signal.published_at.isoformat() if signal.published_at else None,
+        "importance": signal.importance,
+        "entities": signal.entities,
+        "created_at": signal.created_at.isoformat() if signal.created_at else None,
+    }
+
+
+def _sse(event: str, data: dict) -> str:
+    # 标准SSE帧让前端可以通过流式请求消费事件。
+    payload = json.dumps(data, ensure_ascii=False)
+    return f"event: {event}\ndata: {payload}\n\n"