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

apps/admin_console/services/report_export.py

@@ -7,6 +7,7 @@ from pathlib import Path
 import re
 import threading
 import zipfile
+from apps.admin_console.services.request_diagnostics import diagnostics_json
 
 from docx import Document
 from docx.enum.table import WD_CELL_VERTICAL_ALIGNMENT
@@ -900,6 +901,7 @@ def export_session_media(session_id: str, kind: str) -> Path:
                     else zipfile.ZIP_DEFLATED,
                 )
             if kind == "all":
+                archive.writestr(prefix + "模型请求诊断.json", diagnostics_json(session_id))
                 archive.writestr(
                     prefix + "证据索引.json",
                     json.dumps(
@@ -980,6 +982,7 @@ def export_batch(batch_id: str, progress: Callable[[int, int, str], None] | None
                         if source.suffix.lower() in {".mp4", ".mkv", ".webm"}
                         else zipfile.ZIP_DEFLATED,
                     )
+                archive.writestr(f"{prefix}/模型请求诊断.json", diagnostics_json(report["id"]))
                 archive.writestr(
                     f"{prefix}/证据索引.json",
                     json.dumps(

apps/admin_console/services/request_diagnostics.py

@@ -0,0 +1,72 @@
+"""Read/export model diagnostics from the task's existing trace database."""
+
+from datetime import UTC, datetime
+import json
+import time
+
+from apps.admin_console.database.connection import db_session
+
+
+def request_diagnostics(session_id: str, *, db_path=None, limit: int | None = None) -> dict:
+    with db_session(db_path) as conn:
+        session = conn.execute(
+            "SELECT status FROM sessions WHERE session_id = ?", (session_id,)
+        ).fetchone()
+        sql = (
+            "SELECT payload FROM traces WHERE session_id = ? AND type = 'llm_diagnostic' "
+            "ORDER BY timestamp DESC"
+        )
+        params: tuple = (session_id,)
+        if limit is not None:
+            sql += " LIMIT ?"
+            params += (limit,)
+        rows = conn.execute(sql, params).fetchall()
+    requests = []
+    seen = set()
+    running = bool(session and session["status"] in {"running", "resuming"})
+    for row in rows:
+        try:
+            value = json.loads(row["payload"])
+        except (ValueError, TypeError):
+            continue
+        if not isinstance(value, dict):
+            continue
+        request_id = value.get("request_id")
+        value["active"] = (
+            running
+            and request_id not in seen
+            and value.get("phase")
+            in {
+                "waiting_response",
+                "waiting_output",
+                "receiving_output",
+                "generating_tool_arguments",
+                "retry_wait",
+            }
+        )
+        seen.add(request_id)
+        value["timestamps_utc"] = {
+            field: datetime.fromtimestamp(value[field], UTC).isoformat()
+            for field in (
+                "started_at",
+                "headers_at",
+                "first_output_at",
+                "last_output_at",
+                "ended_at",
+                "retry_at",
+            )
+            if isinstance(value.get(field), (int, float))
+        }
+        requests.append(value)
+    return {
+        "schema_version": 1,
+        "session_id": session_id,
+        "session_status": session["status"] if session else "unknown",
+        "collected_at": time.time(),
+        "requests": requests,
+        "note": "时间戳为 Unix 秒；timestamps_utc 为 UTC。HTTP 200 不代表完整输出。usage=null 表示服务未报告用量。历史任务没有记录的字段无法补推。",
+    }
+
+
+def diagnostics_json(session_id: str) -> str:
+    return json.dumps(request_diagnostics(session_id), ensure_ascii=False, indent=2)