代码修改

offline-backend-20260512/backend/scripts/inspect_latest_roundtable_run.py

@@ -0,0 +1,114 @@
+"""一次性巡检：把最新圆桌草稿里「后台作业 run」的关键字段 dump 到文件，
+用于验证维度3 存储 + 报告 HTML + 席位 md 文件是否都对齐了。
+
+用法（在 backend 目录）：
+    PYTHONPATH=. uv run python scripts/inspect_latest_roundtable_run.py
+
+输出写到 scripts/_roundtable_run_report.txt（UTF-8），直接打开看，避免 PowerShell GBK 乱码。
+"""
+
+from __future__ import annotations
+
+import json
+import sqlite3
+from pathlib import Path
+
+DB = Path(".deer-flow/data/deerflow.db")
+OUT = Path("scripts/_roundtable_run_report.txt")
+
+
+def _loads(v):
+    if isinstance(v, (bytes, bytearray)):
+        v = v.decode("utf-8", "replace")
+    if isinstance(v, str):
+        try:
+            return json.loads(v)
+        except Exception:
+            return v
+    return v
+
+
+def main() -> None:
+    con = sqlite3.connect(str(DB))
+    con.row_factory = sqlite3.Row
+    lines: list[str] = []
+
+    # 最新一条有 step2.runs 的草稿
+    rows = con.execute(
+        "SELECT id, step2, step3 FROM roundtable_drafts ORDER BY updated_at DESC LIMIT 5"
+    ).fetchall()
+
+    draft = None
+    top_step3 = None
+    for r in rows:
+        s2 = _loads(r["step2"]) or {}
+        if isinstance(s2, dict) and s2.get("runs"):
+            draft = (r["id"], s2)
+            top_step3 = _loads(r["step3"])
+            break
+
+    if not draft:
+        lines.append("没找到带 step2.runs 的草稿。")
+        OUT.write_text("\n".join(lines), encoding="utf-8")
+        print(f"written: {OUT}")
+        return
+
+    draft_id, s2 = draft
+    runs = s2.get("runs") or []
+    job_run = next((x for x in runs if str(x.get("id", "")).startswith("job-")), runs[-1])
+
+    lines.append(f"draft_id = {draft_id}")
+    lines.append(f"run.id = {job_run.get('id')}  jobStatus = {job_run.get('jobStatus')}  hasConsensus = {job_run.get('hasConsensus')}")
+    lines.append("")
+
+    # ── 维度3：threadIds ──
+    lines.append("【threadIds】(应为真实 map，不是 null)")
+    lines.append(json.dumps(job_run.get("threadIds"), ensure_ascii=False, indent=2))
+    lines.append("")
+
+    # ── 维度3：各对话的 steps / agentThreadId / time ──
+    lines.append("【step2RoundtableDialogues 摘要】")
+    for d in job_run.get("step2RoundtableDialogues") or []:
+        steps = d.get("steps") or []
+        step_keys = [s.get("key") for s in steps]
+        lines.append(
+            f"- [{d.get('sender')}] time={d.get('time')!r} thread={d.get('agentThreadId')!r} "
+            f"steps={step_keys} content_len={len(d.get('content') or '')}"
+        )
+    lines.append("")
+
+    # ── 席位 md 文件：seatStubMessages 里有没有 write_file ──
+    lines.append("【seatStubMessages】(席位写的文件 → 文件卡片数据源)")
+    stubs = job_run.get("seatStubMessages") or {}
+    if not stubs:
+        lines.append("  （空 —— 席位没有写文件 / 没采集到）")
+    for tid, msgs in stubs.items():
+        for m in msgs:
+            for tc in m.get("tool_calls") or []:
+                args = tc.get("args") or {}
+                lines.append(f"  thread={tid} {tc.get('name')} path={args.get('path')!r} content_len={len(str(args.get('content') or ''))}")
+    lines.append("")
+
+    # ── 报告 HTML（P0）──
+    lines.append("【step3 报告】")
+    step3 = job_run.get("step3") or {}
+    html = step3.get("html") or ""
+    lines.append(f"  html_len = {len(html)}")
+    lines.append(f"  含 <html 标签 = {'<html' in html.lower()}  (True=真 HTML / False=纯文字回退=P0未解决)")
+    lines.append(f"  summary = {(step3.get('summary') or '')[:120]!r}")
+    lines.append("")
+
+    # ── 顶层 draft.step3（结果绘制聊天记录靠它重建报告气泡）──
+    lines.append("【顶层 draft.step3】(前端 hydrateStep3 用它重建报告聊天记录气泡)")
+    if not top_step3:
+        lines.append("  ❌ 空 —— 报告聊天记录加载后会丢失（本次修复点：应非空）")
+    else:
+        th = (top_step3.get("html") or "") if isinstance(top_step3, dict) else ""
+        lines.append(f"  ✅ 已存  html_len={len(th)}  summary={(top_step3.get('summary') or '')[:80]!r}" if isinstance(top_step3, dict) else f"  {top_step3!r}")
+
+    OUT.write_text("\n".join(lines), encoding="utf-8")
+    print(f"written: {OUT.resolve()}")
+
+
+if __name__ == "__main__":
+    main()