From 7d32b41e90eaa098b471f6fe825327ee4bfc5cbe Mon Sep 17 00:00:00 2001
From: fengxiangdebaozi <10845145+fengxiangdebaozi@user.noreply.gitee.com>
Date: Mon, 22 Jun 2026 10:46:57 +0800
Subject: [PATCH 01/17] =?UTF-8?q?=E5=8A=A0=E8=BD=BD=E6=A1=86=E6=A0=B7?=
 =?UTF-8?q?=E5=BC=8F?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 .../pages/RoundtablePlanningPage.tsx          | 59 ++++++++++++++++++-
 1 file changed, 58 insertions(+), 1 deletion(-)

diff --git a/frontend-web/src/roundtable-planning/pages/RoundtablePlanningPage.tsx b/frontend-web/src/roundtable-planning/pages/RoundtablePlanningPage.tsx
index 4b9cf882..bb7b0733 100644
--- a/frontend-web/src/roundtable-planning/pages/RoundtablePlanningPage.tsx
+++ b/frontend-web/src/roundtable-planning/pages/RoundtablePlanningPage.tsx
@@ -296,6 +296,10 @@ export default function RoundtablePlanningPage() {
   const [taskRestoreResolved, setTaskRestoreResolved] = useState<boolean>(
     !prefillRef.current?.taskId,
   );
+  // 深链/预填自动会话：进入页面后要先建意图线程、（带 taskId 时）回显历史草稿，这段时间
+  // 页面只有一句欢迎语、问答还没真正开始，线程创建慢时像卡住。下面这个标志用来在超时后
+  // 强制撤掉「启动加载遮罩」，避免万一初始化卡死把界面永久锁住。
+  const [bootLoadingExpired, setBootLoadingExpired] = useState<boolean>(false);
 
   const { models: availableModels, isLoading: modelsLoading } = useModels();
 
@@ -1143,6 +1147,35 @@ export default function RoundtablePlanningPage() {
     };
   }, []);
 
+  // ── 启动加载遮罩（深链/预填自动会话）──────────────────────────────
+  // 是否是「自动会话」：深链 -A、taskId、businessCode/chainId 任一存在即自动进入问答，
+  // 用户没有手动操作，进入页面到问答之间必须给出明确加载反馈。
+  const isAutoSession = Boolean(
+    prefillRef.current?.autoStart ||
+      prefillRef.current?.taskId ||
+      prefillRef.current?.businessCode ||
+      prefillRef.current?.chainId,
+  );
+  // 「问答已开始」：自动发出的首条用户消息已入列 / 正在流式 / 已得到意图结果，任一为真即认为
+  // 已经动起来，可以撤掉遮罩交给 Step1Panel 自身的流式占位。
+  const conversationStarted =
+    step1.isIntentSending ||
+    step1.intentReady != null ||
+    step1.interactiveMessages.some((m) => m.sender === "user");
+  // 遮罩展示条件：自动会话 + 仍在 Step1 + 问答还没开始 + 没有初始化报错 + 没到安全超时。
+  const showBootLoading =
+    isAutoSession &&
+    currentStep === 1 &&
+    !conversationStarted &&
+    !step1.intentInitError &&
+    !bootLoadingExpired;
+  // 安全超时：无论如何最多遮 30s，避免初始化卡死锁住界面（用户仍可手动操作底层页面）。
+  useEffect(() => {
+    if (!showBootLoading) return;
+    const timer = window.setTimeout(() => setBootLoadingExpired(true), 30_000);
+    return () => window.clearTimeout(timer);
+  }, [showBootLoading]);
+
   /** 组装无界回传载荷：总控共识 + 各智能体最终交付（剔除总控席位）。 */
   const buildWujieConsensusPayload = useCallback(
     () => ({
@@ -1780,10 +1813,34 @@ export default function RoundtablePlanningPage() {
   return (
     <div
       className={cn(
-        "roundtable-planning theme-transition flex h-full min-h-0 flex-col bg-[#fef6f6] font-sans text-slate-800 selection:bg-red-200",
+        "roundtable-planning theme-transition relative flex h-full min-h-0 flex-col bg-[#fef6f6] font-sans text-slate-800 selection:bg-red-200",
         isDark && "roundtable-planning--dark",
       )}
     >
+      {/* 启动加载遮罩：深链/预填自动会话从进入页面到问答真正开始（建意图线程 / 回显历史草稿）
+          期间全屏覆盖，避免线程创建慢时页面像卡住。问答一旦开始或超时即撤掉。 */}
+      {showBootLoading && (
+        <div
+          className="absolute inset-0 z-50 flex items-center justify-center bg-white"
+          role="status"
+          aria-live="polite"
+        >
+          <div className="flex flex-col items-center gap-7 px-8">
+            <div className="flex flex-col items-center gap-2 text-center">
+              <p className="text-lg font-semibold tracking-wide text-slate-800">
+                正在准备多智能体会商
+              </p>
+              <p className="text-sm text-slate-400">正在初始化任务理解智能体，请稍候…</p>
+            </div>
+            <span className="inline-flex items-center gap-1.5">
+              <span className="h-2 w-2 animate-bounce rounded-full bg-red-400 [animation-delay:-0.3s]" />
+              <span className="h-2 w-2 animate-bounce rounded-full bg-red-400 [animation-delay:-0.15s]" />
+              <span className="h-2 w-2 animate-bounce rounded-full bg-red-400" />
+            </span>
+          </div>
+        </div>
+      )}
+
       <RecommendAgentsDialog
         open={isRecommendDialogOpen}
         intent={step1.intentReady}

From 79f1a004a0616f834914429dbf48df625fa33d20 Mon Sep 17 00:00:00 2001
From: fengxiangdebaozi <10845145+fengxiangdebaozi@user.noreply.gitee.com>
Date: Mon, 22 Jun 2026 11:02:51 +0800
Subject: [PATCH 02/17] =?UTF-8?q?feat(layout):=20=E7=AE=80=E6=B4=81?=
 =?UTF-8?q?=E6=A8=A1=E5=BC=8F=E5=8F=B3=E4=B8=8A=E8=A7=92=E5=8A=A0=E6=8D=A2?=
 =?UTF-8?q?=E8=82=A4=20+=20=E7=95=8C=E9=9D=A2=E6=A8=A1=E5=BC=8F=E4=B8=8B?=
 =?UTF-8?q?=E6=8B=89=E5=88=87=E6=8D=A2?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

抽出共享 appearance-controls（ThemeSwitcher 按 tone 适配 banner/content
配色，新增图标+下拉的 LayoutModeSwitcher）；Header 复用换肤组件，简洁模式
ChatPage 顶栏右上角补上换肤与界面模式切换，与全量模式保持一致。

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 .../workspace/appearance-controls.tsx         | 178 ++++++++++++++++++
 frontend-web/src/pages/ChatPage.tsx           |   8 +
 .../src/strategy-components/Header.tsx        |  96 +---------
 3 files changed, 191 insertions(+), 91 deletions(-)
 create mode 100644 frontend-web/src/components/workspace/appearance-controls.tsx

diff --git a/frontend-web/src/components/workspace/appearance-controls.tsx b/frontend-web/src/components/workspace/appearance-controls.tsx
new file mode 100644
index 00000000..8decbe91
--- /dev/null
+++ b/frontend-web/src/components/workspace/appearance-controls.tsx
@@ -0,0 +1,178 @@
+"use client";
+
+import { Check, LayoutIcon, PanelLeftIcon, Shirt } from "lucide-react";
+import { useEffect, useRef, useState } from "react";
+
+import { Button } from "@/components/ui/button";
+import {
+  DropdownMenu,
+  DropdownMenuContent,
+  DropdownMenuItem,
+  DropdownMenuLabel,
+  DropdownMenuSeparator,
+  DropdownMenuTrigger,
+} from "@/components/ui/dropdown-menu";
+import { useTheme, type ThemeType } from "@/contexts/ThemeContext";
+import { useLayoutMode } from "@/core/system-settings/hooks";
+import { cn } from "@/lib/utils";
+
+/**
+ * 换肤 / 界面模式 顶栏控件。深色顶栏（Header）与简洁模式页面级顶栏（ChatPage 等）共用，
+ * 通过 `tone` 适配不同底色：
+ *   - "banner-dark" / "banner-light"：Header 落在主题 banner 上（白字 / 浅色 hover）。
+ *   - "content"：简洁模式聊天页浅底顶栏，走 foreground / muted-foreground。
+ */
+export type AppearanceControlTone = "banner-dark" | "banner-light" | "content";
+
+// 换肤色板：红色=红白主题(light)，蓝色=浅蓝主题(lightblue)，深蓝色=深色主题(dark)
+export const THEME_SWATCHES: { id: ThemeType; label: string; color: string }[] = [
+  { id: "light", label: "红色", color: "#cf1322" },
+  { id: "lightblue", label: "蓝色", color: "#0062d9" },
+  { id: "dark", label: "深蓝色", color: "#002466" },
+];
+
+const triggerToneClass = (tone: AppearanceControlTone) =>
+  tone === "banner-dark"
+    ? "text-slate-400 hover:bg-slate-800/50 hover:text-slate-400"
+    : tone === "banner-light"
+      ? "text-white hover:bg-[var(--brand-950)] hover:text-white active:bg-[#8a0618]"
+      : "text-muted-foreground hover:bg-muted hover:text-foreground";
+
+export function ThemeSwitcher({ tone }: { tone: AppearanceControlTone }) {
+  const { theme, setTheme } = useTheme();
+  const [open, setOpen] = useState(false);
+  const closeTimer = useRef<ReturnType<typeof setTimeout> | null>(null);
+
+  const cancelClose = () => {
+    if (closeTimer.current) {
+      clearTimeout(closeTimer.current);
+      closeTimer.current = null;
+    }
+  };
+  const scheduleClose = () => {
+    cancelClose();
+    closeTimer.current = setTimeout(() => setOpen(false), 120);
+  };
+
+  useEffect(() => () => cancelClose(), []);
+
+  return (
+    <div
+      className="relative"
+      onMouseEnter={() => {
+        cancelClose();
+        setOpen(true);
+      }}
+      onMouseLeave={scheduleClose}
+    >
+      <Button
+        type="button"
+        variant="ghost"
+        size="icon-sm"
+        className={cn("size-8 shrink-0 rounded-sm", triggerToneClass(tone))}
+        title="换肤"
+        aria-haspopup="true"
+        aria-expanded={open}
+      >
+        <Shirt className="size-4" />
+      </Button>
+      {open ? (
+        <div className="absolute top-full left-1/2 z-50 -translate-x-1/2 pt-2">
+          <div className="relative flex items-center gap-2 rounded-lg border border-black/5 bg-white px-2.5 py-1.5 shadow-lg">
+            <div className="absolute -top-1 left-1/2 size-2 -translate-x-1/2 rotate-45 rounded-[1px] border-t border-l border-black/5 bg-white" />
+            {THEME_SWATCHES.map((swatch) => {
+              const active = theme === swatch.id;
+              return (
+                <button
+                  key={swatch.id}
+                  type="button"
+                  title={swatch.label}
+                  aria-label={swatch.label}
+                  aria-pressed={active}
+                  onClick={() => {
+                    setTheme(swatch.id);
+                    setOpen(false);
+                  }}
+                  className="size-5 rounded-full transition-transform hover:scale-110"
+                  style={{
+                    backgroundColor: swatch.color,
+                    boxShadow: active
+                      ? `0 0 0 1.5px #ffffff, 0 0 0 3px ${swatch.color}`
+                      : undefined,
+                  }}
+                />
+              );
+            })}
+          </div>
+        </div>
+      ) : null}
+    </div>
+  );
+}
+
+const LAYOUT_OPTIONS = [
+  {
+    id: "full" as const,
+    label: "全量模式",
+    description: "展示侧边栏与顶栏导航",
+    icon: PanelLeftIcon,
+  },
+  {
+    id: "compact" as const,
+    label: "简洁模式",
+    description: "隐藏侧边栏与顶栏，获得更大工作区",
+    icon: LayoutIcon,
+  },
+];
+
+/**
+ * 界面模式（全量 / 简洁）下拉切换。简洁模式下全局顶栏被隐藏，故把外观设置里的
+ * 「界面模式」切换也搬到页面级顶栏右上角，图标 + 下拉，与换肤并排。
+ */
+export function LayoutModeSwitcher({ tone }: { tone: AppearanceControlTone }) {
+  const [mode, setLayoutMode] = useLayoutMode();
+  // 未显式设置过时简洁模式才会渲染本控件（全量模式有全局顶栏），故当前态按 compact 兜底。
+  const current = mode === "full" ? "full" : "compact";
+  const TriggerIcon = current === "full" ? PanelLeftIcon : LayoutIcon;
+
+  return (
+    <DropdownMenu>
+      <DropdownMenuTrigger asChild>
+        <Button
+          type="button"
+          variant="ghost"
+          size="icon-sm"
+          className={cn("size-8 shrink-0 rounded-sm", triggerToneClass(tone))}
+          title="界面模式"
+          aria-label="界面模式"
+        >
+          <TriggerIcon className="size-4" />
+        </Button>
+      </DropdownMenuTrigger>
+      <DropdownMenuContent align="end" sideOffset={8} className="min-w-56 rounded-lg">
+        <DropdownMenuLabel>界面模式</DropdownMenuLabel>
+        <DropdownMenuSeparator />
+        {LAYOUT_OPTIONS.map((option) => {
+          const Icon = option.icon;
+          const active = current === option.id;
+          return (
+            <DropdownMenuItem
+              key={option.id}
+              onClick={() => setLayoutMode(option.id)}
+              className="gap-2"
+            >
+              <Icon className="size-4 shrink-0" />
+              <div className="flex min-w-0 flex-col">
+                <span className="text-sm leading-tight">{option.label}</span>
+                <span className="text-muted-foreground text-xs leading-tight">
+                  {option.description}
+                </span>
+              </div>
+              {active && <Check className="ml-auto size-4 shrink-0" />}
+            </DropdownMenuItem>
+          );
+        })}
+      </DropdownMenuContent>
+    </DropdownMenu>
+  );
+}
diff --git a/frontend-web/src/pages/ChatPage.tsx b/frontend-web/src/pages/ChatPage.tsx
index b1c24df3..325a8739 100644
--- a/frontend-web/src/pages/ChatPage.tsx
+++ b/frontend-web/src/pages/ChatPage.tsx
@@ -26,6 +26,10 @@ import { ThreadContext } from "@/components/workspace/messages/context";
 import { QuestionNavigator } from "@/components/workspace/question-navigator";
 import { ThreadTitle } from "@/components/workspace/thread-title";
 import { TodoList } from "@/components/workspace/todo-list";
+import {
+  LayoutModeSwitcher,
+  ThemeSwitcher,
+} from "@/components/workspace/appearance-controls";
 import { TokenUsageIndicator } from "@/components/workspace/token-usage-indicator";
 import { UserIdentityBadge } from "@/components/workspace/user-identity-badge";
 import { Welcome } from "@/components/workspace/welcome";
@@ -484,8 +488,12 @@ export default function ChatPage() {
               <ArtifactTrigger />
               {/* 不带全局顶栏（紧凑模式）时，名称+岗位回落到聊天页顶栏右侧，
                   与全量模式 Header 右上角保持一致。全量模式已有顶栏则不重复展示。 */}
+              {/* 简洁模式无全局顶栏：把换肤 + 界面模式切换搬到页面级顶栏右上角，
+                  与全量模式 Header 右上角保持一致（content 配色贴浅底顶栏）。 */}
               {!compactChrome && (
                 <>
+                  <ThemeSwitcher tone="content" />
+                  <LayoutModeSwitcher tone="content" />
                   <div
                     className="bg-border/70 mx-1 h-6 w-px shrink-0"
                     aria-hidden
diff --git a/frontend-web/src/strategy-components/Header.tsx b/frontend-web/src/strategy-components/Header.tsx
index 4d75ccbd..57d7480e 100644
--- a/frontend-web/src/strategy-components/Header.tsx
+++ b/frontend-web/src/strategy-components/Header.tsx
@@ -1,9 +1,10 @@
-import { Bell, ChevronDown, FileText, Shirt, User } from "lucide-react";
-import { useEffect, useRef, useState } from "react";
+import { Bell, ChevronDown, FileText, User } from "lucide-react";
+import { useState } from "react";
 import { useLocation, useNavigate } from "react-router-dom";
 
 import { Button } from "@/components/ui/button";
 import "@/components/page-sdebar.css";
+import { ThemeSwitcher } from "@/components/workspace/appearance-controls";
 import { NotificationDialog } from "@/components/workspace/notification-dialog";
 import { getStoredAuth, getAccountDisplayName } from "@/core/auth";
 import { isWujieEnv } from "@/core/embed/wujie";
@@ -18,7 +19,7 @@ import {
 import { useSidebarMenuItems } from "@/core/page-layout/use-sidebar-menu";
 import { cn } from "@/lib/utils";
 
-import { useTheme, type ThemeType } from "../contexts/ThemeContext";
+import { useTheme } from "../contexts/ThemeContext";
 
 const iconButtonClass = (isDark: boolean) =>
   cn(
@@ -32,93 +33,6 @@ const iconButtonClass = (isDark: boolean) =>
 const LEFT_MENU_IDS = ["qa-center", "task-management", "light-app-center"];
 const RIGHT_MENU_IDS = ["comprehensive-management", "resource-management"];
 
-// 换肤色板：红色=红白主题(light)，蓝色=浅蓝主题(lightblue)，深蓝色=深色主题(dark)
-const THEME_SWATCHES: { id: ThemeType; label: string; color: string }[] = [
-  { id: "light", label: "红色", color: "#cf1322" },
-  { id: "lightblue", label: "蓝色", color: "#0062d9" },
-  { id: "dark", label: "深蓝色", color: "#002466" },
-];
-
-function ThemeSwitcher({ isDark }: { isDark: boolean }) {
-  const { theme, setTheme } = useTheme();
-  const [open, setOpen] = useState(false);
-  const closeTimer = useRef<ReturnType<typeof setTimeout> | null>(null);
-
-  const cancelClose = () => {
-    if (closeTimer.current) {
-      clearTimeout(closeTimer.current);
-      closeTimer.current = null;
-    }
-  };
-  const scheduleClose = () => {
-    cancelClose();
-    closeTimer.current = setTimeout(() => setOpen(false), 120);
-  };
-
-  useEffect(() => () => cancelClose(), []);
-
-  // 切换按钮：悬浮只改背景，不改图标颜色（显式锁定 hover 文字色等于常态色）
-  const triggerClass = cn(
-    "size-8 shrink-0 rounded-sm",
-    isDark
-      ? "text-slate-400 hover:bg-slate-800/50 hover:text-slate-400"
-      : "text-white hover:bg-[var(--brand-950)] hover:text-white active:bg-[#8a0618]",
-  );
-
-  return (
-    <div
-      className="relative"
-      onMouseEnter={() => {
-        cancelClose();
-        setOpen(true);
-      }}
-      onMouseLeave={scheduleClose}
-    >
-      <Button
-        type="button"
-        variant="ghost"
-        size="icon-sm"
-        className={triggerClass}
-        title="换肤"
-        aria-haspopup="true"
-        aria-expanded={open}
-      >
-        <Shirt className="size-4" />
-      </Button>
-      {open ? (
-        <div className="absolute top-full left-1/2 z-50 -translate-x-1/2 pt-2">
-          <div className="relative flex items-center gap-2 rounded-lg border border-black/5 bg-white px-2.5 py-1.5 shadow-lg">
-            <div className="absolute -top-1 left-1/2 size-2 -translate-x-1/2 rotate-45 rounded-[1px] border-t border-l border-black/5 bg-white" />
-            {THEME_SWATCHES.map((swatch) => {
-              const active = theme === swatch.id;
-              return (
-                <button
-                  key={swatch.id}
-                  type="button"
-                  title={swatch.label}
-                  aria-label={swatch.label}
-                  aria-pressed={active}
-                  onClick={() => {
-                    setTheme(swatch.id);
-                    setOpen(false);
-                  }}
-                  className="size-5 rounded-full transition-transform hover:scale-110"
-                  style={{
-                    backgroundColor: swatch.color,
-                    boxShadow: active
-                      ? `0 0 0 1.5px #ffffff, 0 0 0 3px ${swatch.color}`
-                      : undefined,
-                  }}
-                />
-              );
-            })}
-          </div>
-        </div>
-      ) : null}
-    </div>
-  );
-}
-
 type HeaderProps = {
   collapsed?: boolean;
 };
@@ -210,7 +124,7 @@ const Header = ({ collapsed }: HeaderProps) => {
   // 右侧操作区（换肤/通知/用户）—— 深浅两套布局共用
   const actionCluster = (
     <>
-      <ThemeSwitcher isDark={isDark} />
+      <ThemeSwitcher tone={isDark ? "banner-dark" : "banner-light"} />
 
       <Button
         type="button"

From 2b6424b3fffad12b2caa9b54a08cf0ba08a0a3e3 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E5=90=89=E9=91=AB=E5=8D=87?= <xsji@kland.com.cn>
Date: Mon, 22 Jun 2026 11:20:35 +0800
Subject: [PATCH 03/17] =?UTF-8?q?feat(open-chat):=20=E6=96=B0=E5=A2=9E=20/?=
 =?UTF-8?q?api/open/chat=20=E5=BC=80=E6=94=BE=E9=97=AE=E7=AD=94=E6=8E=A5?=
 =?UTF-8?q?=E5=8F=A3?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- 无需登录(/api/open/ 加入 auth 白名单 + CSRF 豁免),供外部系统直接调用本项目智能体做问答
- 可指定模型(model_name)与智能体(agent_name 中文名称,扫描 agents/*/config.yaml 解析为运行 id);GET /api/open/agents 列出可选名称
- 只返回最终结果(answer = 最后一段 markdown),不含中间过程/工具调用
- 可靠关闭思考:thinking_enabled=false 时下发 thinking_force_disabled,经 create_chat_model(force_disable_thinking=True) 强制注入 enable_thinking=false(及 GLM thinking.type=disabled),覆盖只声明 supports_thinking 的内网 vLLM/Qwen 模型(与生成标题关思考一致)
- 一次性问答设 is_scheduled_run 跳过自动起标题

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 offline-backend-20260512/backend/CLAUDE.md    |   2 +
 .../backend/app/gateway/app.py                |   4 +
 .../backend/app/gateway/auth_middleware.py    |   3 +
 .../backend/app/gateway/csrf_middleware.py    |   4 +
 .../backend/app/gateway/routers/open_chat.py  | 219 ++++++++++++++++++
 .../backend/app/gateway/services.py           |   4 +
 .../deerflow/agents/lead_agent/agent.py       |  11 +-
 .../harness/deerflow/models/factory.py        |  22 +-
 test_open_chat_api.py                         |  72 ++++++
 9 files changed, 337 insertions(+), 4 deletions(-)
 create mode 100644 offline-backend-20260512/backend/app/gateway/routers/open_chat.py
 create mode 100644 test_open_chat_api.py

diff --git a/offline-backend-20260512/backend/CLAUDE.md b/offline-backend-20260512/backend/CLAUDE.md
index 17eb0b9b..efe0577d 100644
--- a/offline-backend-20260512/backend/CLAUDE.md
+++ b/offline-backend-20260512/backend/CLAUDE.md
@@ -243,6 +243,7 @@ FastAPI application on port 8001 with health check at `GET /health`. Set `GATEWA
 | **Thread Runs** (`/api/threads/{id}/runs`) | `POST /` - create background run; `POST /stream` - create + SSE stream; `POST /wait` - create + block; `GET /` - list runs; `GET /{rid}` - run details; `POST /{rid}/cancel` - cancel; `GET /{rid}/join` - join SSE; `GET /{rid}/messages` - paginated messages `{data, has_more}`; `GET /{rid}/events` - full event stream; `GET /../messages` - thread messages with feedback; `GET /../token-usage` - aggregate tokens |
 | **Feedback** (`/api/threads/{id}/runs/{rid}/feedback`) | `PUT /` - upsert feedback; `DELETE /` - delete user feedback; `POST /` - create feedback; `GET /` - list feedback; `GET /stats` - aggregate stats; `DELETE /{fid}` - delete specific |
 | **Runs** (`/api/runs`) | `POST /stream` - stateless run + SSE; `POST /wait` - stateless run + block; `GET /{rid}/messages` - paginated messages by run_id `{data, has_more}` (cursor: `after_seq`/`before_seq`); `GET /{rid}/feedback` - list feedback by run_id |
+| **Open Chat** (`/api/open/chat`) | **Unauthenticated** open Q&A entry for external systems (the `/api/open/` prefix is whitelisted in `auth_middleware._PUBLIC_PATH_PREFIXES` + CSRF-exempt in `csrf_middleware.should_check_csrf`). `POST /chat` `{message \| messages[], model_name?, agent_name?, agent_id?, thinking_enabled=false, strip_think=true, thread_id?, recursion_limit=100}` → resolves the agent by **Chinese display name** (`agent_name`, matched against each `.deer-flow/agents/*/config.yaml` `name` via `_resolve_agent_id`; explicit `agent_id` wins if given), builds the lead agent via `make_lead_agent` (so the chosen **agent's** SOUL/tools/skills all apply), `ainvoke`s it once (stateless — no checkpointer, no thread persistence; runs under the `"default"` user bucket), and returns **only the final** `{answer, model_name, agent_name, agent_id, thinking_enabled, thread_id}` (no intermediate tool-call/process). `GET /agents` lists `{name, id}` so callers know what `agent_name` to send. **Thinking off is reliable**: when `thinking_enabled=false` the route also sets `thinking_force_disabled=true` so the lead agent calls `create_chat_model(force_disable_thinking=True)`, which injects `extra_body.chat_template_kwargs.enable_thinking=false` (+ GLM `thinking.type="disabled"`) even for internal vLLM/Qwen models that only declare `supports_thinking: true` — the same trick `TitleMiddleware._force_disable_thinking` uses for title generation. Sets `is_scheduled_run=true` internally to skip the auto-title LLM call. Router `app/gateway/routers/open_chat.py`; standalone caller `test_open_chat_api.py` (repo root); tests `tests/test_open_chat.py`. |
 | **Thread Shares** (`/api/thread-shares`) | Per-conversation sharing with two modes (`thread_shares.mode`). **`import`**: `POST /` `{thread_id, mode}` - create (or reuse) a revocable share code; `GET /{code}/preview` + `POST /{code}/import` - copy that conversation into the caller's account (fresh `thread_id` via `app/gateway/thread_copy.py` + new `threads_meta` row + `user-data` copy; dedup on `metadata.shared_origin`). **`view`**: a public, read-only static page ? see Public Shares below. Both modes: `GET /` - list my codes; `DELETE /{code}` - revoke. |
 | **Public Shares** (`/api/public/shares`) | **Unauthenticated** (the `/api/public/` prefix is whitelisted in `auth_middleware._PUBLIC_PATH_PREFIXES`). `GET /{code}` - read-only conversation snapshot (title + serialized messages + artifacts) for a `view`-mode share; messages are read straight from the latest checkpoint. `GET /{code}/files/{path}` - serves a file referenced by that conversation (images, Markdown, text, PDF, ?) so the shared page's links are openable; path resolution is confined to the owner's thread directory. Active content (`text/html`, `application/xhtml+xml`, `image/svg+xml`) is forced to download (`_public_share_disposition`); everything else is served inline. Missing/revoked/non-`view` codes all return 404. |
 | **Roundtable Drafts** (`/api/roundtable-drafts`) | **Strictly per-user** persistence for the roundtable-planning page (replaces the old browser `localStorage`). `GET /` - list drafts (lightweight meta); `POST /` - create; `GET/PUT/DELETE /{id}` - full read / partial update / delete (all owner-scoped). A draft = one full session (Step 1 intent + Step 2 roundtable), step snapshots stored as JSON in `PortableLongText`. Recommendation history bound to a draft: `GET /{id}/recommendations` - list newest-first; `POST /{id}/recommendations` - append one run (objective/status/rationale/picks/candidates). taskId 深链会商记录 are **not** here — they live in the separate `/api/roundtable-task-drafts` store (see below) so personal history never mixes with task records. Backed by `deerflow.persistence.roundtable_drafts` (tables `roundtable_drafts`, `roundtable_recommend_history`); store wired in `deps.py` as `app.state.roundtable_draft_store`. Tests: `tests/test_roundtable_task_drafts.py`. |
@@ -614,6 +615,7 @@ scheduler runs once per deployment. Per run:
 
 - `create_chat_model(name, thinking_enabled)` instantiates LLM from config via reflection
 - Supports `thinking_enabled` flag with per-model `when_thinking_enabled` overrides
+- `force_disable_thinking=True` (kwarg) hard-disables thinking even when the model config declares **no** `when_thinking_*`/`thinking` shape — many internal vLLM/Qwen models only set `supports_thinking: true`, so `thinking_enabled=False` alone injects nothing and they keep reasoning on. The flag injects both OpenAI-compatible disable shapes (`extra_body.chat_template_kwargs.enable_thinking=false` for vLLM/Qwen + `extra_body.thinking.type="disabled"` for Zhipu GLM), only for classes that accept `extra_body`. Mirrors `TitleMiddleware._force_disable_thinking`. The lead agent reads the runtime flag `thinking_force_disabled` and forwards it (used by `/api/open/chat`)
 - Supports vLLM-style thinking toggles via `when_thinking_enabled.extra_body.chat_template_kwargs.enable_thinking` for Qwen reasoning models, while normalizing legacy `thinking` configs for backward compatibility
 - Supports `supports_vision` flag for image understanding models
 - Config values starting with `$` resolved as environment variables
diff --git a/offline-backend-20260512/backend/app/gateway/app.py b/offline-backend-20260512/backend/app/gateway/app.py
index a32575ef..995f76b3 100644
--- a/offline-backend-20260512/backend/app/gateway/app.py
+++ b/offline-backend-20260512/backend/app/gateway/app.py
@@ -50,6 +50,7 @@ from app.gateway.routers import (
     models,
     multi_agent,
     notifications,
+    open_chat,
     parallel_agent_tasks,
     parallel_agents,
     positions,
@@ -1454,6 +1455,9 @@ This gateway provides custom endpoints for models, MCP configuration, skills, an
     # Stateless Runs API (stream/wait without a pre-existing thread)
     app.include_router(runs.router)
 
+    # 开放问答接口 /api/open/chat —— 无需登录,指定模型 + 智能体做问答,可关闭思考模式
+    app.include_router(open_chat.router)
+
     # --- ?????????? /page/workspace/roundtable/planning?---
     # ??? app/gateway/routers/{multi_agent,intent,recommend}.py?
     # ??? frontend-web/docs/multi-agent-backend-dev.md
diff --git a/offline-backend-20260512/backend/app/gateway/auth_middleware.py b/offline-backend-20260512/backend/app/gateway/auth_middleware.py
index ae124dfb..e6c84b4d 100644
--- a/offline-backend-20260512/backend/app/gateway/auth_middleware.py
+++ b/offline-backend-20260512/backend/app/gateway/auth_middleware.py
@@ -35,6 +35,9 @@ _PUBLIC_PATH_PREFIXES: tuple[str, ...] = (
     "/redoc",
     "/openapi.json",
     "/api/public/",
+    # 开放问答接口:无需登录即可指定模型 + 智能体做问答(见 routers/open_chat.py)。
+    # 以 "default" 用户桶运行,只暴露受控的问答能力。
+    "/api/open/",
 )
 
 # Exact auth paths that are public (login/register/status check).
diff --git a/offline-backend-20260512/backend/app/gateway/csrf_middleware.py b/offline-backend-20260512/backend/app/gateway/csrf_middleware.py
index c5494277..afc9554a 100644
--- a/offline-backend-20260512/backend/app/gateway/csrf_middleware.py
+++ b/offline-backend-20260512/backend/app/gateway/csrf_middleware.py
@@ -52,6 +52,10 @@ def should_check_csrf(request: Request) -> bool:
     # PUT/POST returns 403 even though auth was bypassed earlier in the chain.
     if path.startswith("/api/public/"):
         return False
+    # 开放问答接口同样是匿名、无会话的(见 auth_middleware._PUBLIC_PATH_PREFIXES),
+    # 双提交 Cookie 既无 token 可比、也无会话可被伪造,豁免之。
+    if path.startswith("/api/open/"):
+        return False
     return True
 
 
diff --git a/offline-backend-20260512/backend/app/gateway/routers/open_chat.py b/offline-backend-20260512/backend/app/gateway/routers/open_chat.py
new file mode 100644
index 00000000..1b8f0b78
--- /dev/null
+++ b/offline-backend-20260512/backend/app/gateway/routers/open_chat.py
@@ -0,0 +1,219 @@
+"""开放问答接口(无需登录)。
+
+给外部系统一个最简单的 HTTP 入口来「调用本项目的智能体做问答」:
+
+- 可指定 **模型**(``model_name``,对应 ``config.yaml`` 中 ``models[]`` 的 name)。
+- 可指定 **智能体**(``agent_id``,对应 ``.deer-flow/agents/<id>/``;留空 = 默认 lead agent)。
+- 可关闭 **大模型思考模式**(``thinking_enabled=false``,默认即关闭)。内网 vLLM/Qwen
+  模型仅靠 ``thinking_enabled=False`` 往往关不掉思考(其 config 只声明
+  ``supports_thinking: true``),本接口在关闭思考时会把
+  ``thinking_force_disabled`` 一并下发,经
+  ``create_chat_model(force_disable_thinking=True)`` 强制注入
+  ``extra_body.chat_template_kwargs.enable_thinking=false``(及 GLM 的
+  ``thinking.type=disabled``)——与「生成标题」时关闭思考的做法一致。
+
+该路由前缀 ``/api/open/`` 已在 ``auth_middleware`` 白名单里,**绕过登录鉴权**,
+以 ``"default"`` 用户桶运行,只暴露受控的问答能力(无文件/线程持久化)。
+
+请求示例::
+
+    POST /api/open/chat
+    {
+        "message": "用一句话介绍一下你自己",
+        "model_name": "deepseek-ai/DeepSeek-V4-Flash",
+        "agent_id": null,
+        "thinking_enabled": false
+    }
+
+返回::
+
+    {"answer": "...", "model_name": "...", "agent_id": null,
+     "thinking_enabled": false, "thread_id": "open-...."}
+"""
+
+from __future__ import annotations
+
+import logging
+import re
+import uuid
+from typing import Any, Literal
+
+from fastapi import APIRouter, HTTPException
+from langchain_core.messages import AIMessage, BaseMessage, HumanMessage, SystemMessage
+from pydantic import BaseModel, Field
+
+from deerflow.agents.lead_agent.agent import make_lead_agent
+from deerflow.config.agents_config import list_custom_agents, validate_agent_id
+
+logger = logging.getLogger(__name__)
+
+router = APIRouter(prefix="/api/open", tags=["open"])
+
+_THINK_TAG_RE = re.compile(r"<think>[\s\S]*?</think>", re.IGNORECASE)
+
+
+class OpenChatMessage(BaseModel):
+    """多轮历史中的一条消息。"""
+
+    role: Literal["user", "assistant", "system"] = "user"
+    content: str
+
+
+class OpenChatRequest(BaseModel):
+    """开放问答请求体。
+
+    ``message`` 与 ``messages`` 二选一:单轮直接传 ``message``;需要带上下文时传
+    ``messages``(完整历史,本接口本身无状态、不落库)。
+    """
+
+    message: str | None = Field(default=None, description="单轮问题文本;与 messages 二选一")
+    messages: list[OpenChatMessage] | None = Field(default=None, description="多轮历史;与 message 二选一")
+    model_name: str | None = Field(default=None, description="models[] 中的模型名;留空用默认模型")
+    agent_name: str | None = Field(default=None, description="智能体名称(中文显示名);留空用默认 lead agent")
+    agent_id: str | None = Field(default=None, description="智能体 id;一般不用,优先级高于 agent_name")
+    thinking_enabled: bool = Field(default=False, description="是否开启思考;默认 False = 关闭思考(内网模型也会被强制关闭)")
+    strip_think: bool = Field(default=True, description="关闭思考时,从答案里剥离残留的 <think>…</think> 块")
+    thread_id: str | None = Field(default=None, description="可选;仅用于文件隔离,留空自动生成")
+    recursion_limit: int = Field(default=100, ge=1, le=300, description="智能体单次问答的最大步数")
+
+
+class OpenChatResponse(BaseModel):
+    answer: str
+    model_name: str | None
+    agent_name: str | None
+    agent_id: str | None
+    thinking_enabled: bool
+    thread_id: str
+
+
+def _resolve_agent_id(agent_id: str | None, agent_name: str | None) -> str | None:
+    """把 agent_id / 中文名称 解析成运行 id。
+
+    优先级:显式 agent_id > agent_name(按 .deer-flow/agents/*/config.yaml 的 name 匹配)。
+    都没传 = None(默认 lead agent)。名称匹配先精确、再忽略大小写/首尾空格。
+    """
+    if agent_id and agent_id.strip():
+        try:
+            return validate_agent_id(agent_id)
+        except ValueError as exc:
+            raise HTTPException(status_code=422, detail=f"非法的 agent_id:{exc}") from exc
+
+    if not agent_name or not agent_name.strip():
+        return None
+
+    target = agent_name.strip()
+    agents = list_custom_agents()
+    for a in agents:
+        if (a.name or "").strip() == target:
+            return a.id
+    lowered = target.lower()
+    for a in agents:
+        if (a.name or "").strip().lower() == lowered:
+            return a.id
+    raise HTTPException(status_code=404, detail=f"未找到名称为「{agent_name}」的智能体。")
+
+
+def _build_lc_messages(body: OpenChatRequest) -> list[BaseMessage]:
+    """把请求体转成 LangChain 消息列表。"""
+    if body.messages:
+        out: list[BaseMessage] = []
+        for m in body.messages:
+            text = (m.content or "").strip()
+            if not text:
+                continue
+            if m.role == "assistant":
+                out.append(AIMessage(content=m.content))
+            elif m.role == "system":
+                out.append(SystemMessage(content=m.content))
+            else:
+                out.append(HumanMessage(content=m.content))
+        if out:
+            return out
+    if body.message and body.message.strip():
+        return [HumanMessage(content=body.message)]
+    raise HTTPException(status_code=422, detail="必须提供非空的 'message' 或 'messages'。")
+
+
+def _extract_text(content: Any) -> str:
+    """从 AIMessage.content(str 或 block 列表)抽取纯文本。"""
+    if isinstance(content, str):
+        return content
+    if isinstance(content, list):
+        parts: list[str] = []
+        for block in content:
+            if isinstance(block, str):
+                parts.append(block)
+            elif isinstance(block, dict):
+                text_value = block.get("text")
+                if isinstance(text_value, str):
+                    parts.append(text_value)
+        return "".join(parts) if parts else ""
+    return ""
+
+
+def _extract_answer(result: Any, *, strip_think: bool) -> str:
+    """从 ainvoke 的最终状态里取最后一条 AI 文本。"""
+    messages = (result or {}).get("messages") if isinstance(result, dict) else None
+    answer = ""
+    if isinstance(messages, list):
+        for msg in reversed(messages):
+            if isinstance(msg, AIMessage):
+                text = _extract_text(msg.content).strip()
+                if text:
+                    answer = text
+                    break
+    if strip_think and answer:
+        answer = _THINK_TAG_RE.sub("", answer).strip()
+    return answer
+
+
+@router.post("/chat", response_model=OpenChatResponse)
+async def open_chat(body: OpenChatRequest) -> OpenChatResponse:
+    """指定模型 + 智能体做一次问答(无状态、可关闭思考)。"""
+    lc_messages = _build_lc_messages(body)
+
+    agent_id = _resolve_agent_id(body.agent_id, body.agent_name)
+
+    thread_id = body.thread_id or f"open-{uuid.uuid4()}"
+
+    configurable: dict[str, Any] = {
+        "thread_id": thread_id,
+        "model_name": body.model_name,
+        "thinking_enabled": body.thinking_enabled,
+        # 关闭思考时强制下发,确保内网模型也真的关掉(见模块 docstring)。
+        "thinking_force_disabled": not body.thinking_enabled,
+        # 一次性问答:跳过自动起标题(省一次 LLM 调用)、不挂定时任务工具。
+        "is_scheduled_run": True,
+    }
+    if agent_id:
+        configurable["agent_id"] = agent_id
+
+    config: dict[str, Any] = {"configurable": configurable, "recursion_limit": body.recursion_limit}
+
+    try:
+        agent = make_lead_agent(config)
+        result = await agent.ainvoke({"messages": lc_messages}, config=config)
+    except Exception as exc:  # noqa: BLE001 — 开放接口需把内部错误归一成 502
+        logger.exception("open_chat 调用智能体失败 (model=%s, agent=%s)", body.model_name, agent_id)
+        raise HTTPException(status_code=502, detail=f"智能体调用失败:{exc}") from exc
+
+    answer = _extract_answer(result, strip_think=body.strip_think and not body.thinking_enabled)
+    if not answer:
+        raise HTTPException(status_code=502, detail="智能体未返回任何文本答案。")
+
+    return OpenChatResponse(
+        answer=answer,
+        model_name=body.model_name,
+        agent_name=body.agent_name,
+        agent_id=agent_id,
+        thinking_enabled=body.thinking_enabled,
+        thread_id=thread_id,
+    )
+
+
+@router.get("/agents")
+async def open_list_agents() -> dict[str, Any]:
+    """列出可用智能体的名称(方便知道 agent_name 该填什么)。"""
+    agents = list_custom_agents()
+    items = [{"name": a.name, "id": a.id} for a in agents if (a.name or "").strip()]
+    return {"agents": items, "count": len(items)}
diff --git a/offline-backend-20260512/backend/app/gateway/services.py b/offline-backend-20260512/backend/app/gateway/services.py
index 362c5b9c..d16243a3 100644
--- a/offline-backend-20260512/backend/app/gateway/services.py
+++ b/offline-backend-20260512/backend/app/gateway/services.py
@@ -222,6 +222,10 @@ _CONTEXT_CONFIGURABLE_KEYS: frozenset[str] = frozenset(
         "model_name",
         "mode",
         "thinking_enabled",
+        # 强制关闭思考(内网 vLLM/Qwen 模型只声明 supports_thinking 时,单靠
+        # thinking_enabled=False 不会注入 enable_thinking=false)。透传到 run config
+        # 后由 lead agent 交给 create_chat_model(force_disable_thinking=...)。
+        "thinking_force_disabled",
         "reasoning_effort",
         "rag_mode_enabled",
         "writing_mode",
diff --git a/offline-backend-20260512/backend/packages/harness/deerflow/agents/lead_agent/agent.py b/offline-backend-20260512/backend/packages/harness/deerflow/agents/lead_agent/agent.py
index 00a2d1cb..e2d27c2f 100644
--- a/offline-backend-20260512/backend/packages/harness/deerflow/agents/lead_agent/agent.py
+++ b/offline-backend-20260512/backend/packages/harness/deerflow/agents/lead_agent/agent.py
@@ -412,6 +412,11 @@ def _make_lead_agent(config: RunnableConfig, *, app_config: AppConfig):
     resolved_app_config = app_config
 
     thinking_enabled = cfg.get("thinking_enabled", True)
+    # Hard "thinking off" switch for internal vLLM/Qwen models whose config only
+    # declares `supports_thinking: true` (no disable shape) — passes through to
+    # create_chat_model(force_disable_thinking=...) so enable_thinking=false is
+    # actually injected. Default off → no behavior change for normal chats.
+    thinking_force_disabled = bool(cfg.get("thinking_force_disabled", False))
     reasoning_effort = cfg.get("reasoning_effort", None)
     requested_model_name: str | None = cfg.get("model_name") or cfg.get("model")
     is_plan_mode = cfg.get("is_plan_mode", False)
@@ -501,7 +506,7 @@ def _make_lead_agent(config: RunnableConfig, *, app_config: AppConfig):
         from deerflow.agents.middlewares.writing_setup_round_cap_middleware import WritingSetupRoundCapMiddleware
 
         return create_agent(
-            model=create_chat_model(name=model_name, thinking_enabled=thinking_enabled, app_config=resolved_app_config),
+            model=create_chat_model(name=model_name, thinking_enabled=thinking_enabled, force_disable_thinking=thinking_force_disabled, app_config=resolved_app_config),
             tools=[setup_writing, ask_clarification_tool],
             middleware=[
                 *_build_middlewares(config, model_name=model_name, app_config=resolved_app_config),
@@ -514,7 +519,7 @@ def _make_lead_agent(config: RunnableConfig, *, app_config: AppConfig):
     if is_bootstrap:
         # Special bootstrap agent with minimal prompt for initial custom agent creation flow
         return create_agent(
-            model=create_chat_model(name=model_name, thinking_enabled=thinking_enabled, app_config=resolved_app_config),
+            model=create_chat_model(name=model_name, thinking_enabled=thinking_enabled, force_disable_thinking=thinking_force_disabled, app_config=resolved_app_config),
             tools=get_available_tools(
                 model_name=model_name,
                 subagent_enabled=subagent_enabled,
@@ -593,7 +598,7 @@ def _make_lead_agent(config: RunnableConfig, *, app_config: AppConfig):
         )
 
     return create_agent(
-        model=create_chat_model(name=model_name, thinking_enabled=thinking_enabled, reasoning_effort=reasoning_effort, app_config=resolved_app_config),
+        model=create_chat_model(name=model_name, thinking_enabled=thinking_enabled, reasoning_effort=reasoning_effort, force_disable_thinking=thinking_force_disabled, app_config=resolved_app_config),
         tools=default_tools,
         middleware=_build_middlewares(config, model_name=model_name, agent_name=agent_id, app_config=resolved_app_config),
         system_prompt=system_prompt,
diff --git a/offline-backend-20260512/backend/packages/harness/deerflow/models/factory.py b/offline-backend-20260512/backend/packages/harness/deerflow/models/factory.py
index 58c33562..50096990 100644
--- a/offline-backend-20260512/backend/packages/harness/deerflow/models/factory.py
+++ b/offline-backend-20260512/backend/packages/harness/deerflow/models/factory.py
@@ -47,11 +47,22 @@ def _enable_stream_usage_by_default(model_use_path: str, model_settings_from_con
         model_settings_from_config["stream_usage"] = True
 
 
-def create_chat_model(name: str | None = None, thinking_enabled: bool = False, *, app_config: AppConfig | None = None, **kwargs) -> BaseChatModel:
+def create_chat_model(name: str | None = None, thinking_enabled: bool = False, *, app_config: AppConfig | None = None, force_disable_thinking: bool = False, **kwargs) -> BaseChatModel:
     """Create a chat model instance from the config.
 
     Args:
         name: The name of the model to create. If None, the first model in the config will be used.
+        thinking_enabled: Turn the model's extended thinking / reasoning on.
+        force_disable_thinking: Hard override that disables thinking even when the
+            per-model config declares no ``when_thinking_*`` shape. Many internal
+            vLLM / Qwen models only set ``supports_thinking: true`` (no disable
+            shape), so passing ``thinking_enabled=False`` alone injects nothing and
+            the model keeps reasoning on. When this flag is set (and thinking is
+            off) we additionally inject both common OpenAI-compatible disable shapes
+            — vLLM/Qwen ``chat_template_kwargs.enable_thinking=False`` and Zhipu GLM
+            ``thinking.type="disabled"`` — mirroring
+            ``TitleMiddleware._force_disable_thinking``. No effect when thinking is
+            enabled, and skipped for model classes that don't accept ``extra_body``.
 
     Returns:
         A chat model instance.
@@ -120,6 +131,15 @@ def create_chat_model(name: str | None = None, thinking_enabled: bool = False, *
         kwargs.pop("reasoning_effort", None)
         model_settings_from_config.pop("reasoning_effort", None)
 
+    # Hard override: force-disable thinking regardless of the per-model config shape.
+    # Injected only for OpenAI-compatible classes (those declaring `extra_body`);
+    # other providers would reject the kwarg, so they are skipped (they ignore it).
+    if force_disable_thinking and not thinking_enabled and "extra_body" in getattr(model_class, "model_fields", {}):
+        model_settings_from_config["extra_body"] = _deep_merge_dicts(
+            model_settings_from_config.get("extra_body"),
+            {"chat_template_kwargs": {"enable_thinking": False}, "thinking": {"type": "disabled"}},
+        )
+
     _enable_stream_usage_by_default(model_config.use, model_settings_from_config)
 
     # For Codex Responses API models: map thinking mode to reasoning_effort
diff --git a/test_open_chat_api.py b/test_open_chat_api.py
new file mode 100644
index 00000000..fc50b070
--- /dev/null
+++ b/test_open_chat_api.py
@@ -0,0 +1,72 @@
+# -*- coding: utf-8 -*-
+"""开放问答接口 /api/open/chat 测试脚本(Jupyter 版)。
+
+接口只返回「最终结果」:智能体把调研 / 工具调用全部跑完后,只把最后那段
+文本(markdown)放在 answer 字段里返回 —— 不含中间过程、工具调用、思考。
+
+用法:在 Jupyter 里把下面几段按顺序运行即可。
+  - 「配置」段:改成你要的模型 / 智能体 / 问题 / 是否思考
+  - 「调用」段:运行后 answer 变量就是最终 md 文本
+
+无第三方依赖(纯标准库 urllib)。
+"""
+
+import json
+import time
+import urllib.error
+import urllib.request
+
+# ====================== 配置(按需修改这里) ======================
+BASE_URL = "http://47.94.209.59:2026"   # 后端地址(走 nginx 用 :2026,Gateway 直连用 :8001)
+
+MESSAGE = "帮我调研一下 XXX,输出 markdown 报告。"   # 要问的问题
+MODEL_NAME = None                        # 模型名(config.yaml models[].name);None=默认模型
+AGENT_NAME = None                        # 智能体中文名称;None=默认 lead agent。不确定填啥就先跑下面的「列出智能体」
+THINKING_ENABLED = False                 # False=关闭思考(内网模型也会被强制关闭);True=开启
+TIMEOUT = 600                            # 请求超时(秒)
+# ===============================================================
+
+
+def open_chat(message, model_name=None, agent_name=None, thinking_enabled=False,
+              base_url=BASE_URL, timeout=TIMEOUT):
+    """调用 /api/open/chat,返回最终 md 文本(只要结果,不要过程)。"""
+    url = base_url.rstrip("/") + "/api/open/chat"
+    payload = {
+        "message": message,
+        "model_name": model_name,
+        "agent_name": agent_name,   # 按中文名称选智能体
+        "thinking_enabled": thinking_enabled,
+    }
+    data = json.dumps(payload, ensure_ascii=False).encode("utf-8")
+    req = urllib.request.Request(
+        url,
+        data=data,
+        headers={"Content-Type": "application/json; charset=utf-8"},
+        method="POST",
+    )
+    with urllib.request.urlopen(req, timeout=timeout) as resp:
+        result = json.loads(resp.read().decode("utf-8"))
+    return result["answer"]   # 只返回最终结果
+
+
+# ====================== (可选)列出可用智能体名称 ======================
+# 不知道 AGENT_NAME 填什么时,先跑这段看有哪些中文名:
+# import urllib.request, json
+# with urllib.request.urlopen(BASE_URL.rstrip("/") + "/api/open/agents", timeout=30) as r:
+#     for a in json.loads(r.read().decode("utf-8"))["agents"]:
+#         print(a["name"])
+
+# ====================== 调用 ======================
+_t0 = time.time()
+answer = open_chat(MESSAGE, MODEL_NAME, AGENT_NAME, THINKING_ENABLED)
+print(f"(耗时 {time.time() - _t0:.1f}s)\n")
+print(answer)        # answer 就是最终 markdown 文本
+
+
+# ====================== (可选)在 Jupyter 里渲染成 markdown ======================
+# from IPython.display import Markdown, display
+# display(Markdown(answer))
+
+# ====================== (可选)存成 .md 文件 ======================
+# with open("result.md", "w", encoding="utf-8") as f:
+#     f.write(answer)

From 16ba5deb0eaaafb2397d25dfe2e54023d448a0ae Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E5=90=89=E9=91=AB=E5=8D=87?= <xsji@kland.com.cn>
Date: Mon, 22 Jun 2026 11:30:15 +0800
Subject: [PATCH 04/17] =?UTF-8?q?feat(agents):=20=E6=99=BA=E8=83=BD?=
 =?UTF-8?q?=E4=BD=93=E5=BD=92=E5=B1=9E/=E5=8F=91=E5=B8=83=E6=93=8D?=
 =?UTF-8?q?=E4=BD=9C=E5=89=8D=E5=A2=9E=E5=8A=A0=E3=80=8C=E8=B4=A6=E5=8F=B7?=
 =?UTF-8?q?=E5=B7=B2=E5=88=87=E6=8D=A2=E3=80=8D=E6=8F=90=E9=86=92?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- 新增 useLiveAccount 钩子:跟踪「实时」登录账号(随 storage/focus 刷新),
  并与组件挂载时捕获的账号对比;token 存 localStorage 会被同浏览器其他标签页
  共享,另一标签页登录会静默替换本页 API 所用账号
- agent-card / agent-detail-panel / publish-square-picker-dialog / NewAgentPage
  接入该钩子,切换账号后对归属/发布类操作做提醒,避免误归属

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 .../workspace/agents/agent-card.tsx           | 44 +++++++++-
 .../workspace/agents/agent-detail-panel.tsx   | 44 +++++++++-
 .../publish-square-picker-dialog.tsx          |  6 +-
 .../src/core/auth/use-live-account.ts         | 82 +++++++++++++++++++
 frontend-web/src/pages/NewAgentPage.tsx       | 52 +++++++++++-
 5 files changed, 221 insertions(+), 7 deletions(-)
 create mode 100644 frontend-web/src/core/auth/use-live-account.ts

diff --git a/frontend-web/src/components/workspace/agents/agent-card.tsx b/frontend-web/src/components/workspace/agents/agent-card.tsx
index cfd7e8c2..d80487c2 100644
--- a/frontend-web/src/components/workspace/agents/agent-card.tsx
+++ b/frontend-web/src/components/workspace/agents/agent-card.tsx
@@ -77,6 +77,7 @@ import {
 import { getAgent } from "@/core/agents/api";
 import type { Agent, AgentQuestion, UpdateAgentRequest } from "@/core/agents";
 import { getStoredAuth } from "@/core/auth";
+import { useLiveAccount } from "@/core/auth/use-live-account";
 import { useI18n } from "@/core/i18n/hooks";
 import { useSkills } from "@/core/skills/hooks";
 import { formatDateTime } from "@/core/utils/datetime";
@@ -246,6 +247,8 @@ export function AgentCard({ agent, layout = "card", onPublisherClick }: AgentCar
 
   const storedAuth = useMemo(() => getStoredAuth(), []);
   const currentUserId = storedAuth?.user_id ?? null;
+  // 发布请求实际以「实时」token 执行 —— 用 live 账号来提示发布者并检测账号是否被切换。
+  const live = useLiveAccount();
   const isAdmin = storedAuth?.system_role === "admin";
   const isBuiltin = agent.builtin;
   const isOwned =
@@ -414,18 +417,38 @@ export function AgentCard({ agent, layout = "card", onPublisherClick }: AgentCar
       setPublishPickerOpen(true);
       return;
     }
+    if (!warnIfAccountSwitched()) {
+      return;
+    }
     try {
       await updateAgent.mutateAsync({
         id: agent.id,
         request: { published: !agent.published },
       });
-      toast.success(agent.published ? "已取消发布" : "已发布");
+      toast.success(
+        agent.published ? "已取消发布" : `已发布（操作账号：${live.displayName}）`,
+      );
     } catch (err) {
       toast.error(err instanceof Error ? err.message : String(err));
     }
   }
 
+  // 发布操作以实时 token 执行。若会话中途账号被切换（常见于另一标签页登录了别的
+  // 账号），先弹一次确认 —— 避免误以本人之外的账号执行发布。返回 false = 用户取消。
+  function warnIfAccountSwitched(): boolean {
+    if (!live.switched) {
+      return true;
+    }
+    return window.confirm(
+      `当前登录账号已从「${live.mountDisplayName}」切换为「${live.displayName}」。\n` +
+        `本次发布将以「${live.displayName}」执行。如需以「${live.mountDisplayName}」操作，请取消并刷新页面。\n\n仍要继续吗？`,
+    );
+  }
+
   async function handlePublishToSquare(squareId: string) {
+    if (!warnIfAccountSwitched()) {
+      return;
+    }
     try {
       await updateAgent.mutateAsync({
         id: agent.id,
@@ -433,7 +456,7 @@ export function AgentCard({ agent, layout = "card", onPublisherClick }: AgentCar
       });
       const sqName =
         publishSquares?.squares.find((s) => s.id === squareId)?.name ?? "广场";
-      toast.success(`已发布到「${sqName}」`);
+      toast.success(`已发布到「${sqName}」（操作账号：${live.displayName}）`);
       setPublishPickerOpen(false);
     } catch (err) {
       toast.error(err instanceof Error ? err.message : String(err));
@@ -1094,6 +1117,23 @@ export function AgentCard({ agent, layout = "card", onPublisherClick }: AgentCar
         loading={updateAgent.isPending}
         onConfirm={(squareId) => void handlePublishToSquare(squareId)}
         description="请选择要把该智能体发布到哪个广场。每个智能体只能归属一个广场，不会重复出现。"
+        accountNotice={
+          <div className="text-xs">
+            <p className="text-muted-foreground">
+              发布操作账号：
+              <span className="text-foreground font-medium">
+                「{live.displayName}」
+              </span>
+            </p>
+            {live.switched && (
+              <p className="text-destructive mt-1">
+                ⚠ 当前登录账号已从「{live.mountDisplayName}」切换为「
+                {live.displayName}」。如需以「{live.mountDisplayName}
+                」操作，请刷新页面确认。
+              </p>
+            )}
+          </div>
+        }
       />
     </>
   );
diff --git a/frontend-web/src/components/workspace/agents/agent-detail-panel.tsx b/frontend-web/src/components/workspace/agents/agent-detail-panel.tsx
index 9a40afc3..1273088c 100644
--- a/frontend-web/src/components/workspace/agents/agent-detail-panel.tsx
+++ b/frontend-web/src/components/workspace/agents/agent-detail-panel.tsx
@@ -59,6 +59,7 @@ import {
   type UpdateAgentRequest,
 } from "@/core/agents";
 import { getStoredAuth } from "@/core/auth";
+import { useLiveAccount } from "@/core/auth/use-live-account";
 import { isAgentSkillCandidate } from "@/core/skills";
 import { useSkills } from "@/core/skills/hooks";
 import { usePublishSquares } from "@/core/system-settings/hooks";
@@ -156,6 +157,8 @@ export function AgentDetailPanel({
   const [deleteOpen, setDeleteOpen] = useState(false);
   const [publishPickerOpen, setPublishPickerOpen] = useState(false);
   const [detailSkillName, setDetailSkillName] = useState<string | null>(null);
+  // 发布以实时 token 执行 —— 用 live 账号提示发布者并检测账号是否被切换。
+  const live = useLiveAccount();
 
   const updateAgent = useUpdateAgent();
   const deleteAgent = useDeleteAgent();
@@ -338,19 +341,39 @@ export function AgentDetailPanel({
       setPublishPickerOpen(true);
       return;
     }
+    if (!warnIfAccountSwitched()) {
+      return;
+    }
     try {
       await updateAgent.mutateAsync({
         id: agent.id,
         request: { published: !agent.published },
       });
-      toast.success(agent.published ? "已下架" : "已发布");
+      toast.success(
+        agent.published ? "已下架" : `已发布（操作账号：${live.displayName}）`,
+      );
     } catch (err) {
       toast.error(err instanceof Error ? err.message : String(err));
     }
   }
 
+  // 发布以实时 token 执行。会话中途账号被切换（常见于另一标签页登录了别的账号）时
+  // 先弹确认，避免误以本人之外的账号执行发布。返回 false = 用户取消。
+  function warnIfAccountSwitched(): boolean {
+    if (!live.switched) {
+      return true;
+    }
+    return window.confirm(
+      `当前登录账号已从「${live.mountDisplayName}」切换为「${live.displayName}」。\n` +
+        `本次发布将以「${live.displayName}」执行。如需以「${live.mountDisplayName}」操作，请取消并刷新页面。\n\n仍要继续吗？`,
+    );
+  }
+
   async function handlePublishToSquare(squareId: string) {
     if (!agent) return;
+    if (!warnIfAccountSwitched()) {
+      return;
+    }
     try {
       await updateAgent.mutateAsync({
         id: agent.id,
@@ -358,7 +381,7 @@ export function AgentDetailPanel({
       });
       const sqName =
         publishSquares?.squares.find((s) => s.id === squareId)?.name ?? "广场";
-      toast.success(`已发布到「${sqName}」`);
+      toast.success(`已发布到「${sqName}」（操作账号：${live.displayName}）`);
       setPublishPickerOpen(false);
     } catch (err) {
       toast.error(err instanceof Error ? err.message : String(err));
@@ -901,6 +924,23 @@ export function AgentDetailPanel({
         loading={updateAgent.isPending}
         onConfirm={(squareId) => void handlePublishToSquare(squareId)}
         description="请选择要把该智能体发布到哪个广场。每个智能体只能归属一个广场，不会重复出现。"
+        accountNotice={
+          <div className="text-xs">
+            <p className="text-muted-foreground">
+              发布操作账号：
+              <span className="text-foreground font-medium">
+                「{live.displayName}」
+              </span>
+            </p>
+            {live.switched && (
+              <p className="text-destructive mt-1">
+                ⚠ 当前登录账号已从「{live.mountDisplayName}」切换为「
+                {live.displayName}」。如需以「{live.mountDisplayName}
+                」操作，请刷新页面确认。
+              </p>
+            )}
+          </div>
+        }
       />
     </>
   );
diff --git a/frontend-web/src/components/workspace/publish-square-picker-dialog.tsx b/frontend-web/src/components/workspace/publish-square-picker-dialog.tsx
index 65aa7a51..fd6fecc2 100644
--- a/frontend-web/src/components/workspace/publish-square-picker-dialog.tsx
+++ b/frontend-web/src/components/workspace/publish-square-picker-dialog.tsx
@@ -1,7 +1,7 @@
 "use client";
 
 import { CheckIcon } from "lucide-react";
-import { useEffect, useState } from "react";
+import { type ReactNode, useEffect, useState } from "react";
 
 import { Button } from "@/components/ui/button";
 import {
@@ -26,6 +26,8 @@ interface PublishSquarePickerDialogProps {
   loading?: boolean;
   title?: string;
   description?: string;
+  /** 可选提示节点，渲染在描述下方 —— 用于显示「本次发布将以哪个账号进行」及账号切换告警。 */
+  accountNotice?: ReactNode;
 }
 
 /** A "publish to which square (发布广场)?" chooser. Squares are rendered as a
@@ -39,6 +41,7 @@ export function PublishSquarePickerDialog({
   loading,
   title = "选择发布广场",
   description = "请选择要把它发布到哪个广场。每个项目只能归属一个广场，不会重复出现。",
+  accountNotice,
 }: PublishSquarePickerDialogProps) {
   const [selected, setSelected] = useState<string>(
     defaultSquareId || squares[0]?.id || "",
@@ -58,6 +61,7 @@ export function PublishSquarePickerDialog({
           <DialogTitle>{title}</DialogTitle>
           <DialogDescription>{description}</DialogDescription>
         </DialogHeader>
+        {accountNotice}
         <div className="max-h-[50vh] space-y-1.5 overflow-y-auto py-1">
           {squares.map((sq) => {
             const active = selected === sq.id;
diff --git a/frontend-web/src/core/auth/use-live-account.ts b/frontend-web/src/core/auth/use-live-account.ts
new file mode 100644
index 00000000..a9417694
--- /dev/null
+++ b/frontend-web/src/core/auth/use-live-account.ts
@@ -0,0 +1,82 @@
+"use client";
+
+import { useEffect, useRef, useState } from "react";
+
+import { getAccountDisplayName, getStoredAuth } from "@/core/auth";
+import { isEmbedActive } from "@/core/embed/embed-session";
+
+const AUTH_STORAGE_KEY = "deerflow.auth";
+
+export interface LiveAccount {
+  /** 实时登录用户 id（随 storage/focus 刷新）。null = 未登录。 */
+  userId: string | null;
+  /** 实时显示名（邮箱 @ 前缀，否则 user_id）。 */
+  displayName: string;
+  /** 组件首次挂载时捕获的用户 id。 */
+  mountUserId: string | null;
+  /** 组件首次挂载时捕获的显示名。 */
+  mountDisplayName: string;
+  /**
+   * 当前登录账号与挂载时不一致 —— 即本视图加载后登录被切换过（通常是在另一个
+   * 标签页登录了别的账号）。归属/发布类操作必须据此提醒，避免误归属。
+   */
+  switched: boolean;
+}
+
+/**
+ * 跟踪「实时」登录账号，而不是挂载时缓存的值。
+ *
+ * 鉴权 token 存在 localStorage（非嵌入态下被同浏览器所有标签页共享），所以在
+ * 另一个标签页登录会**静默替换**本标签页 API 调用所用的 token。任何会决定归属的
+ * 界面（创建/发布智能体）都必须显示请求实际将以哪个账号执行，并在会话中途被切换
+ * 账号时给出提示 —— 否则会出现「用 A 账号创建却归到 B 账号名下」。
+ */
+export function useLiveAccount(): LiveAccount {
+  const mountRef = useRef<{ id: string | null; name: string } | null>(null);
+  if (mountRef.current === null) {
+    const auth = getStoredAuth();
+    mountRef.current = {
+      id: auth?.user_id ?? null,
+      name: getAccountDisplayName(auth),
+    };
+  }
+  const [live, setLive] = useState(() => mountRef.current!);
+
+  useEffect(() => {
+    function refresh() {
+      const auth = getStoredAuth();
+      const id = auth?.user_id ?? null;
+      const name = getAccountDisplayName(auth);
+      setLive((prev) =>
+        prev.id === id && prev.name === name ? prev : { id, name },
+      );
+    }
+    // 嵌入(iframe)态鉴权走按标签页的 sessionStorage、不跨标签，收不到 storage 事件
+    // —— 只挂在正常态；focus/visibilitychange 兜底刷新两种形态都有效。
+    function onStorage(e: StorageEvent) {
+      if (e.key === AUTH_STORAGE_KEY) {
+        refresh();
+      }
+    }
+    if (!isEmbedActive()) {
+      window.addEventListener("storage", onStorage);
+    }
+    window.addEventListener("focus", refresh);
+    document.addEventListener("visibilitychange", refresh);
+    refresh();
+    return () => {
+      window.removeEventListener("storage", onStorage);
+      window.removeEventListener("focus", refresh);
+      document.removeEventListener("visibilitychange", refresh);
+    };
+  }, []);
+
+  const mount = mountRef.current!;
+  return {
+    userId: live.id,
+    displayName: live.name,
+    mountUserId: mount.id,
+    mountDisplayName: mount.name,
+    switched: live.id !== null && mount.id !== null && live.id !== mount.id,
+  };
+}
diff --git a/frontend-web/src/pages/NewAgentPage.tsx b/frontend-web/src/pages/NewAgentPage.tsx
index 4ab2564a..0906cd7c 100644
--- a/frontend-web/src/pages/NewAgentPage.tsx
+++ b/frontend-web/src/pages/NewAgentPage.tsx
@@ -48,6 +48,7 @@ import {
   updateAgent,
 } from "@/core/agents/api";
 import { getStoredAuth } from "@/core/auth";
+import { useLiveAccount } from "@/core/auth/use-live-account";
 import { useI18n } from "@/core/i18n/hooks";
 import { useModels } from "@/core/models/hooks";
 import { isAgentSkillCandidate, type Skill } from "@/core/skills";
@@ -96,6 +97,10 @@ export default function NewAgentPage() {
   const { skills, isLoading: isSkillsLoading } = useSkills();
   const { models, isLoading: isModelsLoading } = useModels();
 
+  // 创建请求的归属账号取自「实时」token（apiFetch 每次都重读 localStorage），而非
+  // 本页加载时缓存的身份 —— 用 live 账号提示归属并检测账号是否在会话中途被切换。
+  const live = useLiveAccount();
+
   // 可配置的技能限定为「内置 + 自己的 + 已发布到广场的」，
   // 其他用户未发布的自定义技能不可选。
   const storedAuth = useMemo(() => getStoredAuth(), []);
@@ -287,13 +292,37 @@ export default function NewAgentPage() {
   ]);
 
   const handleSaveAgent = useCallback(async () => {
+    // 新建时归属账号 = 创建请求所用的实时 token 对应账号。若本页加载后登录被切换
+    // （常见于另一标签页登录了别的账号），现在创建会把智能体静默归到新账号名下。
+    // 先弹确认，避免「用 A 账号创建却归到 B 账号」。更新已有智能体不改归属，故跳过。
+    if (!agentId && live.switched) {
+      const ok = window.confirm(
+        `当前登录账号已从「${live.mountDisplayName}」切换为「${live.displayName}」。\n` +
+          `现在创建将把该智能体归属到「${live.displayName}」名下。\n` +
+          `如需归属「${live.mountDisplayName}」，请取消并刷新页面后重新登录。\n\n仍要以「${live.displayName}」创建吗？`,
+      );
+      if (!ok) {
+        return;
+      }
+    }
     const saved = await ensureAgent();
     if (!saved) {
       return;
     }
-    toast.success(`智能体「${saved.name}」已保存`);
+    toast.success(
+      agentId
+        ? `智能体「${saved.name}」已保存`
+        : `智能体「${saved.name}」已创建（归属账号：${live.displayName}）`,
+    );
     router.push("/page/workspace/agents");
-  }, [ensureAgent, router]);
+  }, [
+    agentId,
+    ensureAgent,
+    live.displayName,
+    live.mountDisplayName,
+    live.switched,
+    router,
+  ]);
 
   const handleChatSubmit = useCallback(
     async (text: string) => {
@@ -661,6 +690,25 @@ export default function NewAgentPage() {
                   <p className="text-destructive text-sm">{formError}</p>
                 ) : null}
 
+                <div className="rounded-md border px-3 py-2 text-xs dark:border-[#1B477E] dark:bg-[#022248]">
+                  <p className="text-muted-foreground">
+                    {agentId ? "当前登录账号：" : "创建后归属账号："}
+                    <span className="text-foreground font-medium">
+                      「{live.displayName}」
+                    </span>
+                  </p>
+                  {live.switched && (
+                    <p className="text-destructive mt-1">
+                      ⚠ 本页加载时为「{live.mountDisplayName}」，现已切换为「
+                      {live.displayName}」。
+                      {agentId
+                        ? "继续保存将以当前账号执行。"
+                        : `继续创建将归属「${live.displayName}」。`}
+                      建议刷新页面确认账号。
+                    </p>
+                  )}
+                </div>
+
                 <Button
                   className="w-full"
                   onClick={() => void handleSaveAgent()}

From 32476c7ad42327d36121aa884b502207ac4e3403 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E5=90=89=E9=91=AB=E5=8D=87?= <xsji@kland.com.cn>
Date: Mon, 22 Jun 2026 14:47:43 +0800
Subject: [PATCH 05/17] =?UTF-8?q?feat:=20=E6=99=BA=E8=83=BD=E4=BD=93?=
 =?UTF-8?q?=E8=AF=A6=E6=83=85=E6=96=B0=E5=A2=9E=E3=80=8C=E6=9C=80=E8=BF=91?=
 =?UTF-8?q?=E5=AF=B9=E8=AF=9D=E3=80=8D=E5=88=97=E8=A1=A8=EF=BC=8C=E5=AE=8C?=
 =?UTF-8?q?=E5=96=84=E6=8E=A8=E8=8D=90=E9=97=AE=E9=A2=98=E4=B8=8E=E4=BB=BB?=
 =?UTF-8?q?=E5=8A=A1=E6=B7=B1=E9=93=BE=E9=85=8D=E7=BD=AE?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- 新增 AgentRecentChats 组件，在智能体详情面板底部按 agent_id 展示最近对话
- 调整问答/智能体侧边栏与最近对话列表展示逻辑
- 补充 dashboard-input、task-deeplink 相关配置与线程工具方法
- 后端推荐问题接口标签处理及 position_sync 微调，新增相关测试

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 .../workspace/agents/agent-chat-sidebar.tsx   | 31 +++++--
 .../workspace/agents/agent-detail-panel.tsx   |  7 ++
 .../workspace/agents/agent-recent-chats.tsx   | 79 ++++++++++++++++++
 .../workspace/general-qa-sidebar.tsx          | 37 ++++++---
 .../components/workspace/recent-chat-list.tsx | 32 +++++--
 frontend-web/src/core/auth/dashboard-input.ts | 41 +++++++++
 .../src/core/auth/task-deeplink-config.ts     |  5 +-
 frontend-web/src/core/threads/utils.ts        | 26 ++++++
 frontend-web/src/pages/AgentChatPage.tsx      |  1 +
 frontend-web/src/pages/NewAgentPage.tsx       | 11 ++-
 frontend-web/src/pages/WorkspaceRoutes.tsx    | 16 ++++
 .../task-components/TaskFxWorkspace.tsx       | 83 ++++++++++++++++++-
 offline-backend-20260512/backend/CLAUDE.md    | 13 ++-
 .../gateway/routers/recommended_questions.py  | 35 ++++----
 .../deerflow/config/task_deeplink_config.py   |  7 +-
 .../deerflow/services/position_sync.py        | 17 ++--
 .../tests/test_recommended_questions_tags.py  | 62 ++++++++++++++
 offline-backend-20260512/config.yaml          | 14 +++-
 18 files changed, 457 insertions(+), 60 deletions(-)
 create mode 100644 frontend-web/src/components/workspace/agents/agent-recent-chats.tsx

diff --git a/frontend-web/src/components/workspace/agents/agent-chat-sidebar.tsx b/frontend-web/src/components/workspace/agents/agent-chat-sidebar.tsx
index a87be6ba..65b504f0 100644
--- a/frontend-web/src/components/workspace/agents/agent-chat-sidebar.tsx
+++ b/frontend-web/src/components/workspace/agents/agent-chat-sidebar.tsx
@@ -70,6 +70,8 @@ import {
 } from "@/core/threads/hooks";
 import type { AgentThread, AgentThreadState } from "@/core/threads/types";
 import {
+  bootstrapResumePath,
+  isBootstrapThread,
   isThreadFavorite,
   isThreadPinned,
   sortThreadsPinnedFirst,
@@ -338,14 +340,14 @@ function RecentChatsPane({ activeAgentId }: { activeAgentId?: string }) {
   const [renameValue, setRenameValue] = useState("");
   const [shareThreadId, setShareThreadId] = useState<string | null>(null);
 
-  // 排除调度/引导/会商类会话，与首页一致；置顶优先。
+  // 排除调度/会商类系统会话，与首页一致；智能体创建（bootstrap）会话保留展示
+  // （带机器人图标、点击回创建页）；置顶优先。
   const visibleThreads = useMemo(
     () =>
       sortThreadsPinnedFirst(
         threads.filter(
           (thread) =>
             thread.metadata?.thread_type !== "scheduler" &&
-            thread.metadata?.thread_type !== "bootstrap" &&
             thread.metadata?.thread_type !== "roundtable",
         ),
       ),
@@ -450,7 +452,8 @@ function RecentChatsPane({ activeAgentId }: { activeAgentId?: string }) {
           </div>
         ) : (
           filteredThreads.map((thread) => {
-            const href = pathOf(thread);
+            const isBoot = isBootstrapThread(thread);
+            const href = isBoot ? bootstrapResumePath(thread.thread_id) : pathOf(thread);
             const isActive = href === pathname;
             const pinned = isThreadPinned(thread);
             const favorited = isThreadFavorite(thread);
@@ -468,12 +471,22 @@ function RecentChatsPane({ activeAgentId }: { activeAgentId?: string }) {
                       : "text-muted-foreground hover:bg-accent/60 hover:text-foreground",
                   )}
                 >
-                  <MessageSquare
-                    className={cn(
-                      "size-3.5 shrink-0",
-                      isActive ? "text-primary" : "text-muted-foreground/70",
-                    )}
-                  />
+                  {isBoot ? (
+                    <Bot
+                      className={cn(
+                        "size-3.5 shrink-0",
+                        isActive ? "text-primary" : "text-muted-foreground/70",
+                      )}
+                      aria-label="智能体创建会话"
+                    />
+                  ) : (
+                    <MessageSquare
+                      className={cn(
+                        "size-3.5 shrink-0",
+                        isActive ? "text-primary" : "text-muted-foreground/70",
+                      )}
+                    />
+                  )}
                   {pinned && (
                     <Pin className="text-muted-foreground/70 size-3 shrink-0" />
                   )}
diff --git a/frontend-web/src/components/workspace/agents/agent-detail-panel.tsx b/frontend-web/src/components/workspace/agents/agent-detail-panel.tsx
index 1273088c..615d6227 100644
--- a/frontend-web/src/components/workspace/agents/agent-detail-panel.tsx
+++ b/frontend-web/src/components/workspace/agents/agent-detail-panel.tsx
@@ -43,6 +43,7 @@ import {
 import { Switch } from "@/components/ui/switch";
 import { Textarea } from "@/components/ui/textarea";
 import { AgentQuestionEditor } from "@/components/workspace/agents/agent-question-editor";
+import { AgentRecentChats } from "@/components/workspace/agents/agent-recent-chats";
 import { useArtifactsOptional } from "@/components/workspace/artifacts";
 import { EnabledSkillList } from "@/components/workspace/agents/enabled-skill-list";
 import { SkillPickerSheet } from "@/components/workspace/agents/skill-picker-sheet";
@@ -133,6 +134,7 @@ export function AgentDetailPanel({
   onAskQuestion,
   disabled,
   defaultCollapsed,
+  showRecentChats,
 }: {
   agent: Agent | null | undefined;
   agentName: string;
@@ -140,6 +142,8 @@ export function AgentDetailPanel({
   disabled?: boolean;
   /** 初始收起（任务工作区进入时默认收起智能体信息面板）。 */
   defaultCollapsed?: boolean;
+  /** 在面板底部展示「有关此智能体的最近对话」（仅智能体问答页开启，任务工作区不展示）。 */
+  showRecentChats?: boolean;
 }) {
   const router = useRouter();
   const [collapsed, setCollapsed] = useState(defaultCollapsed ?? false);
@@ -794,6 +798,9 @@ export function AgentDetailPanel({
           onViewDetail={setDetailSkillName}
         />
       </Section>
+
+      {/* ── 最近对话（右下角，仅智能体问答页展示，编辑态隐藏）──────────── */}
+      {showRecentChats && !editing && <AgentRecentChats agentId={agent.id} />}
     </div>
   );
 
diff --git a/frontend-web/src/components/workspace/agents/agent-recent-chats.tsx b/frontend-web/src/components/workspace/agents/agent-recent-chats.tsx
new file mode 100644
index 00000000..de416639
--- /dev/null
+++ b/frontend-web/src/components/workspace/agents/agent-recent-chats.tsx
@@ -0,0 +1,79 @@
+"use client";
+
+import { MessageSquareIcon } from "lucide-react";
+import Link from "next/link";
+import { usePathname } from "next/navigation";
+
+import { useThreadsPage } from "@/core/threads/hooks";
+import { pathOfThread, titleOfThread } from "@/core/threads/utils";
+import { desensitize } from "@/lib/desensitize";
+import { cn } from "@/lib/utils";
+
+const RECENT_LIMIT = 8;
+
+/**
+ * 「有关此智能体的最近对话」—— 渲染在智能体详情面板底部（右下角）。
+ * 按 `metadata.agent_id` 过滤出该智能体的会话，点击直接打开对应对话。
+ */
+export function AgentRecentChats({ agentId }: { agentId: string }) {
+  const pathname = usePathname();
+  const { data } = useThreadsPage({
+    page: 1,
+    pageSize: RECENT_LIMIT,
+    metadata: { agent_id: agentId },
+  });
+  // 该智能体的会话里同样排除调度 / 引导 / 会商类系统会话。
+  const threads = (data?.threads ?? []).filter(
+    (th) =>
+      th.metadata?.thread_type !== "bootstrap" &&
+      th.metadata?.thread_type !== "scheduler" &&
+      th.metadata?.thread_type !== "roundtable",
+  );
+
+  return (
+    <section className="bg-card/60 rounded-xl border p-4">
+      <div className="mb-3 flex items-center justify-between gap-2">
+        <h4 className="text-foreground flex items-center gap-1.5 text-sm font-semibold">
+          <MessageSquareIcon className="size-3.5" />
+          最近对话
+        </h4>
+        <span className="text-muted-foreground text-xs tabular-nums">
+          {threads.length} 个
+        </span>
+      </div>
+      {threads.length === 0 ? (
+        <p className="text-muted-foreground/60 text-sm">暂无对话</p>
+      ) : (
+        <div className="flex flex-col gap-1">
+          {threads.map((thread) => {
+            const href = pathOfThread(thread);
+            const isActive = href === pathname;
+            return (
+              <Link
+                key={thread.thread_id}
+                href={href}
+                title={desensitize(titleOfThread(thread, "新对话"))}
+                className={cn(
+                  "flex min-w-0 items-center gap-2 rounded-lg px-2 py-1.5 text-sm transition-colors",
+                  isActive
+                    ? "bg-primary/5 text-primary font-medium"
+                    : "text-muted-foreground hover:bg-accent hover:text-foreground",
+                )}
+              >
+                <MessageSquareIcon
+                  className={cn(
+                    "size-3.5 shrink-0",
+                    isActive ? "text-primary" : "text-muted-foreground/70",
+                  )}
+                />
+                <span className="truncate">
+                  {desensitize(titleOfThread(thread, "新对话"))}
+                </span>
+              </Link>
+            );
+          })}
+        </div>
+      )}
+    </section>
+  );
+}
diff --git a/frontend-web/src/components/workspace/general-qa-sidebar.tsx b/frontend-web/src/components/workspace/general-qa-sidebar.tsx
index d9062d86..b06e4123 100644
--- a/frontend-web/src/components/workspace/general-qa-sidebar.tsx
+++ b/frontend-web/src/components/workspace/general-qa-sidebar.tsx
@@ -75,8 +75,10 @@ import {
 } from "@/core/threads/hooks";
 import type { AgentThread, AgentThreadState } from "@/core/threads/types";
 import {
+  isBootstrapThread,
   isThreadFavorite,
   isThreadPinned,
+  listPathOfThread,
   pathOfThread,
   sortThreadsPinnedFirst,
   titleOfThread,
@@ -213,11 +215,12 @@ export function GeneralQaSidebar() {
 
   const visibleThreads = useMemo(
     () =>
+      // 智能体创建（bootstrap）会话也展示（带机器人图标、点击回创建页）；
+      // 仅排除调度 / 会商类系统会话。
       sortThreadsPinnedFirst(
         threads.filter(
           (thread) =>
             thread.metadata?.thread_type !== "scheduler" &&
-            thread.metadata?.thread_type !== "bootstrap" &&
             thread.metadata?.thread_type !== "roundtable",
         ),
       ),
@@ -249,9 +252,9 @@ export function GeneralQaSidebar() {
         let nextThreadPath = pathOfThread("new", { agent_id: agentIdFromPath });
         if (threadIndex > -1) {
           if (threads[threadIndex + 1]) {
-            nextThreadPath = pathOfThread(threads[threadIndex + 1]!);
+            nextThreadPath = listPathOfThread(threads[threadIndex + 1]!);
           } else if (threads[threadIndex - 1]) {
-            nextThreadPath = pathOfThread(threads[threadIndex - 1]!);
+            nextThreadPath = listPathOfThread(threads[threadIndex - 1]!);
           }
         }
         void router.push(nextThreadPath);
@@ -417,7 +420,9 @@ export function GeneralQaSidebar() {
                   </div>
                 ) : (
                   filteredThreads.map((thread) => {
-                    const isActive = pathOfThread(thread) === pathname;
+                    const href = listPathOfThread(thread);
+                    const isActive = href === pathname;
+                    const isBoot = isBootstrapThread(thread);
                     const pinned = isThreadPinned(thread);
                     const favorited = isThreadFavorite(thread);
                     return (
@@ -426,7 +431,7 @@ export function GeneralQaSidebar() {
                         className="group/item relative flex items-center"
                       >
                         <Link
-                          href={pathOfThread(thread)}
+                          href={href}
                           className={cn(
                             "relative flex h-9 min-w-0 flex-1 items-center gap-2 rounded-lg px-2.5 text-[13px] transition-colors",
                             isActive
@@ -434,12 +439,22 @@ export function GeneralQaSidebar() {
                               : "text-muted-foreground hover:bg-accent/60 hover:text-foreground",
                           )}
                         >
-                          <MessageSquare
-                            className={cn(
-                              "size-3.5 shrink-0",
-                              isActive ? "text-primary" : "text-muted-foreground/70",
-                            )}
-                          />
+                          {isBoot ? (
+                            <Bot
+                              className={cn(
+                                "size-3.5 shrink-0",
+                                isActive ? "text-primary" : "text-muted-foreground/70",
+                              )}
+                              aria-label="智能体创建会话"
+                            />
+                          ) : (
+                            <MessageSquare
+                              className={cn(
+                                "size-3.5 shrink-0",
+                                isActive ? "text-primary" : "text-muted-foreground/70",
+                              )}
+                            />
+                          )}
                           {pinned && (
                             <Pin className="size-3 shrink-0 text-muted-foreground/70" />
                           )}
diff --git a/frontend-web/src/components/workspace/recent-chat-list.tsx b/frontend-web/src/components/workspace/recent-chat-list.tsx
index 702bcaa7..b996d0a3 100644
--- a/frontend-web/src/components/workspace/recent-chat-list.tsx
+++ b/frontend-web/src/components/workspace/recent-chat-list.tsx
@@ -1,6 +1,7 @@
 "use client";
 
 import {
+  Bot,
   ChevronDown,
   ChevronRight,
   Download,
@@ -64,8 +65,10 @@ import {
 } from "@/core/threads/hooks";
 import type { AgentThread, AgentThreadState } from "@/core/threads/types";
 import {
+  isBootstrapThread,
   isThreadFavorite,
   isThreadPinned,
+  listPathOfThread,
   pathOfThread,
   sortThreadsPinnedFirst,
   titleOfThread,
@@ -130,9 +133,9 @@ export function RecentChatList({
         });
         if (threadIndex > -1) {
           if (threads[threadIndex + 1]) {
-            nextThreadPath = pathOfThread(threads[threadIndex + 1]!);
+            nextThreadPath = listPathOfThread(threads[threadIndex + 1]!);
           } else if (threads[threadIndex - 1]) {
-            nextThreadPath = pathOfThread(threads[threadIndex - 1]!);
+            nextThreadPath = listPathOfThread(threads[threadIndex - 1]!);
           }
         }
         void router.push(nextThreadPath);
@@ -189,11 +192,12 @@ export function RecentChatList({
     [t],
   );
 
+  // 智能体创建（bootstrap）会话也展示在常用对话里（带机器人图标、点击回创建页）；
+  // 仅排除调度 / 会商类系统会话。
   const visibleThreads = sortThreadsPinnedFirst(
     threads.filter(
       (thread) =>
         thread.metadata?.thread_type !== "scheduler" &&
-        thread.metadata?.thread_type !== "bootstrap" &&
         thread.metadata?.thread_type !== "roundtable",
     ),
   );
@@ -239,7 +243,9 @@ export function RecentChatList({
           <SidebarMenu>
             <div className="flex w-full flex-col gap-0.5">
               {shownThreads.map((thread) => {
-                const isActive = pathOfThread(thread) === pathname;
+                const href = listPathOfThread(thread);
+                const isActive = href === pathname;
+                const isBoot = isBootstrapThread(thread);
                 const pinned = isThreadPinned(thread);
                 const favorited = isThreadFavorite(thread);
                 return (
@@ -261,17 +267,29 @@ export function RecentChatList({
                     >
                       <Link
                         className="flex w-full items-center gap-1 whitespace-nowrap group-hover/side-menu-item:overflow-hidden"
-                        href={pathOfThread(thread)}
+                        href={href}
                       >
-                        {compact && (
-                          <MessageSquare
+                        {isBoot ? (
+                          <Bot
                             className={
                               "size-3.5 shrink-0 " +
                               (isActive
                                 ? "text-[var(--primary)]"
                                 : "text-muted-foreground/70")
                             }
+                            aria-label="智能体创建会话"
                           />
+                        ) : (
+                          compact && (
+                            <MessageSquare
+                              className={
+                                "size-3.5 shrink-0 " +
+                                (isActive
+                                  ? "text-[var(--primary)]"
+                                  : "text-muted-foreground/70")
+                              }
+                            />
+                          )
                         )}
                         {pinned && (
                           <Pin className="size-3 shrink-0 text-muted-foreground/70" />
diff --git a/frontend-web/src/core/auth/dashboard-input.ts b/frontend-web/src/core/auth/dashboard-input.ts
index 145259b4..389c4fe6 100644
--- a/frontend-web/src/core/auth/dashboard-input.ts
+++ b/frontend-web/src/core/auth/dashboard-input.ts
@@ -18,6 +18,7 @@
 
 import { requestCrudService } from "@/service/request-axios-service";
 import { readCopTaskDetail } from "@/core/auth/task-deeplink";
+import { fetchTaskDeeplinkConfig } from "@/core/auth/task-deeplink-config";
 
 /** 本页输入 JSON 暂存的 sessionStorage key（拉取一次后复用，避免重复请求）。 */
 export const DASHBOARD_INPUT_STORAGE_KEY = "dashboard.inputJson";
@@ -221,3 +222,43 @@ export async function fetchDashboardInputJson(
   if (resolved) persistDashboardInputJson(resolved);
   return resolved;
 }
+
+/* ---------------------- rwfx 进入预检：3q 是否完成 ---------------------- */
+
+/**
+ * 「3q（研判情报报告）是否完成」状态：
+ *   - ``complete``   = selectSqReport 数组里有有效内容；
+ *   - ``incomplete`` = 数组为空 / 每条 contentJson 都是 null/空（3q 未完成）；
+ *   - ``unknown``    = 缺 consumer 地址 / 请求异常（无法判定，调用方应静默放行）。
+ */
+export type SqReportStatus = "complete" | "incomplete" | "unknown";
+
+/**
+ * rwfx（任务分析）进入时的「3q 是否完成」预检：调 selectSqReport，复用 ``extractSqReport``
+ * 判定该任务是否已有有效研判情报。判定口径与大屏绘制取数完全一致（空/全 null → 无内容）。
+ *   - 有可用内容 → ``complete``（正常）；
+ *   - 空 / 全 null → ``incomplete``（rwfx 据此弹框引导去 3qfx）；
+ *   - 缺地址 / 请求失败 → ``unknown``（不打扰用户）。
+ *
+ * **数据源由后端配置 ``task_deeplink.test`` 决定**（线上 selectSqReport 外部不可用时）：
+ *   - ``test=true``  → 直接用 ``config.yaml`` 的 ``mock.sq_report`` 假数据判定，**不发请求**
+ *     （与目的树查询/行动解析的 mock 同一套机制，见 [[task-deeplink-config]]）；
+ *   - ``test=false`` → 调真实接口 ``GET {VITE_AIPI_CONSUMER_URL}/cop/selectSqReport?taskId=``。
+ */
+export async function fetchSqReportStatus(taskId: string): Promise<SqReportStatus> {
+  if (!taskId) return "unknown";
+  const cfg = await fetchTaskDeeplinkConfig();
+  // test=true：外部接口不可用，直接用后端 config 的假数据判定（不打真实接口）。
+  if (cfg.test) {
+    return extractSqReport(cfg.mock.sqReport, taskId) ? "complete" : "incomplete";
+  }
+  // 真实模式：打真实接口。service 拦截器统一注入鉴权并返回整包 body；空/全 null → null。
+  const url = sqReportUrl();
+  if (!url) return "unknown";
+  try {
+    const body = await requestCrudService.get(url, { taskId });
+    return extractSqReport(body, taskId) ? "complete" : "incomplete";
+  } catch {
+    return "unknown";
+  }
+}
diff --git a/frontend-web/src/core/auth/task-deeplink-config.ts b/frontend-web/src/core/auth/task-deeplink-config.ts
index 86d5b716..be9b18ac 100644
--- a/frontend-web/src/core/auth/task-deeplink-config.ts
+++ b/frontend-web/src/core/auth/task-deeplink-config.ts
@@ -34,11 +34,13 @@ export interface TaskDeeplinkConfig {
   /** 「3q详情」详情页 URL（rwfx 目的树面板链接，跳转追加 ?id= 任务id）；留空则前端提示未配置。 */
   detail3qUrl: string;
   /** 测试假数据：purposeDetail = 原始 {state,msg,data}；delete = {message,success}；
-   *  actionDetail = 原始 {state,msg,data:{taskId,...}}。 */
+   *  actionDetail = 原始 {state,msg,data:{taskId,...}}；sqReport = 原始 {state,msg,data:[...]}
+   *  （rwfx 3q 预检用，test 为真时生效）。 */
   mock: {
     purposeDetail?: unknown;
     delete?: { message?: string; success?: boolean };
     actionDetail?: unknown;
+    sqReport?: unknown;
   };
 }
 
@@ -81,6 +83,7 @@ function normalize(body: unknown): TaskDeeplinkConfig {
           ? (mock.delete as { message?: string; success?: boolean })
           : undefined,
       actionDetail: mock.action_detail,
+      sqReport: mock.sq_report,
     },
   };
 }
diff --git a/frontend-web/src/core/threads/utils.ts b/frontend-web/src/core/threads/utils.ts
index 7a3d8769..adc0ded4 100644
--- a/frontend-web/src/core/threads/utils.ts
+++ b/frontend-web/src/core/threads/utils.ts
@@ -81,6 +81,32 @@ export function titleOfThread(thread: AgentThread, fallback = "Untitled") {
   return fallback;
 }
 
+/**
+ * 智能体创建过程产生的「引导会话」(bootstrap)。这类会话用 metadata.thread_type
+ * === "bootstrap" 标记，点击后应回到「创建智能体」页继续，而不是普通问答页。
+ */
+export function isBootstrapThread(thread: {
+  metadata?: Record<string, unknown> | null;
+}): boolean {
+  return thread.metadata?.thread_type === "bootstrap";
+}
+
+/**
+ * bootstrap 会话点击后回到「创建智能体」页并恢复该会话。该页只挂在
+ * /page/workspace 下（见 WorkspaceRoutes 的 agents/new/:thread_id），故路径固定，
+ * 不随当前 route base 变化。
+ */
+export function bootstrapResumePath(threadId: string): string {
+  return `/page/workspace/agents/new/${threadId}`;
+}
+
+/** 列表项点击目标：bootstrap 会话回创建页恢复，其余走普通问答路由。 */
+export function listPathOfThread(thread: AgentThread): string {
+  return isBootstrapThread(thread)
+    ? bootstrapResumePath(thread.thread_id)
+    : pathOfThread(thread);
+}
+
 export function isThreadPinned(thread: AgentThread): boolean {
   return (thread.metadata as Record<string, unknown> | null)?.is_pinned === true;
 }
diff --git a/frontend-web/src/pages/AgentChatPage.tsx b/frontend-web/src/pages/AgentChatPage.tsx
index b8da2890..659507ba 100644
--- a/frontend-web/src/pages/AgentChatPage.tsx
+++ b/frontend-web/src/pages/AgentChatPage.tsx
@@ -469,6 +469,7 @@ export default function AgentChatPage({
           disabled={thread.isLoading || isDemoMode}
           onAskQuestion={(text) => handleSubmit({ text, files: [] })}
           defaultCollapsed={taskCtxOn}
+          showRecentChats={!taskCtxOn}
         />
       </div>
     </ThreadContext.Provider>
diff --git a/frontend-web/src/pages/NewAgentPage.tsx b/frontend-web/src/pages/NewAgentPage.tsx
index 0906cd7c..bbe05b4d 100644
--- a/frontend-web/src/pages/NewAgentPage.tsx
+++ b/frontend-web/src/pages/NewAgentPage.tsx
@@ -7,7 +7,7 @@ import {
   SaveIcon,
   SearchIcon,
 } from "lucide-react";
-import { useRouter } from "next/navigation";
+import { useParams, useRouter } from "next/navigation";
 import { useCallback, useEffect, useMemo, useState } from "react";
 import { toast } from "sonner";
 
@@ -76,7 +76,11 @@ function getCreateAgentErrorMessage(
 export default function NewAgentPage() {
   const { t } = useI18n();
   const router = useRouter();
-  const threadId = useMemo(() => uuid(), []);
+  // 从「常用对话」点击某条智能体创建（bootstrap）会话回到本页时，URL 带 thread_id，
+  // 用它恢复该会话（加载历史消息、继续对话）；正常进入则新开一个会话。
+  const { thread_id: resumeThreadId } = useParams<{ thread_id?: string }>();
+  const generatedThreadId = useMemo(() => uuid(), []);
+  const threadId = resumeThreadId || generatedThreadId;
 
   const [nameInput, setNameInput] = useState("");
   const [descriptionInput, setDescriptionInput] = useState("");
@@ -93,7 +97,8 @@ export default function NewAgentPage() {
   const [isSkillSelectionTouched, setIsSkillSelectionTouched] = useState(false);
   const [detailSkillName, setDetailSkillName] = useState<string | null>(null);
   const [skillSearch, setSkillSearch] = useState("");
-  const [threadPrepared, setThreadPrepared] = useState(false);
+  // 恢复已有 bootstrap 会话时该 thread 已存在，无需再 create（置为 true，跳过创建）。
+  const [threadPrepared, setThreadPrepared] = useState(!!resumeThreadId);
   const { skills, isLoading: isSkillsLoading } = useSkills();
   const { models, isLoading: isModelsLoading } = useModels();
 
diff --git a/frontend-web/src/pages/WorkspaceRoutes.tsx b/frontend-web/src/pages/WorkspaceRoutes.tsx
index 4810ed01..66d5dce4 100644
--- a/frontend-web/src/pages/WorkspaceRoutes.tsx
+++ b/frontend-web/src/pages/WorkspaceRoutes.tsx
@@ -57,6 +57,12 @@ function NotebookDetailPageWithKey() {
   return <StudioNotebookDetailPage key={notebookId} />;
 }
 
+// 恢复某条 bootstrap 会话时按 thread_id 重挂，切换会话不串台。
+function NewAgentPageWithKey() {
+  const { thread_id } = useParams<{ thread_id: string }>();
+  return <NewAgentPage key={thread_id} />;
+}
+
 function isInIframe(): boolean {
   try {
     return window.self !== window.top;
@@ -279,6 +285,16 @@ export default function WorkspaceRoutes({
             </WorkspaceLayout>
           }
         />
+        {/* 从「常用对话」点击智能体创建（bootstrap）会话回到创建页并恢复该会话。
+            key 绑定 thread_id —— 切换不同 bootstrap 会话时重挂一个干净的页面。 */}
+        <Route
+          path="agents/new/:thread_id"
+          element={
+            <WorkspaceLayout {...layoutProps}>
+              <NewAgentPageWithKey />
+            </WorkspaceLayout>
+          }
+        />
         <Route
           path="agents/:agent_id/chats"
           element={<Navigate to="/page/workspace/agents" replace />}
diff --git a/frontend-web/src/strategy-components/components/task-components/TaskFxWorkspace.tsx b/frontend-web/src/strategy-components/components/task-components/TaskFxWorkspace.tsx
index a8f699f0..2a3fb8c6 100644
--- a/frontend-web/src/strategy-components/components/task-components/TaskFxWorkspace.tsx
+++ b/frontend-web/src/strategy-components/components/task-components/TaskFxWorkspace.tsx
@@ -30,7 +30,13 @@ import {
   DialogTitle,
 } from "@/components/ui/dialog";
 import { deleteTaskPurpose, readTaskContext } from "@/core/auth/task-purpose";
-import { fetchActionDetail, readCopTaskDetail, type ActionDetail } from "@/core/auth/task-deeplink";
+import {
+  fetchActionDetail,
+  readCopTaskDetail,
+  resolveTaskDeeplinkTarget,
+  type ActionDetail,
+} from "@/core/auth/task-deeplink";
+import { fetchSqReportStatus } from "@/core/auth/dashboard-input";
 import { TASK_BUSINESSES, type TaskBusiness } from "@/core/auth/task-buttons";
 import { useTaskThreads } from "@/core/threads/use-task-threads";
 import { titleOfThread } from "@/core/threads/utils";
@@ -276,6 +282,79 @@ function TaskClearButton({ taskId }: { taskId: string }) {
   );
 }
 
+/**
+ * rwfx 专用「3q 完成预检」闸门：进入任务分析工作区时调研判情报报告接口
+ * （``GET {VITE_AIPI_CONSUMER_URL}/cop/selectSqReport?taskId=``）。若该任务的报告为空
+ * （数组为空 / 内容全为 null，即 3q 未完成）→ 弹框提示「3q 未完成，是否进行 3qfx」，
+ * 并提供「前往 3qfx」按钮切换到**同一任务**的三情分析工作区（复用深链解析重绑任务上下文）。
+ * 报告正常 / 接口不可判定（缺地址、请求失败）时不打扰用户。见 [[dashboard-input]] / [[task-deeplink]]。
+ */
+function RwfxSqReportGate({ taskId }: { taskId: string }) {
+  const navigate = useNavigate();
+  const [open, setOpen] = useState(false);
+  const [jumping, setJumping] = useState(false);
+
+  useEffect(() => {
+    if (!taskId) return;
+    let alive = true;
+    void fetchSqReportStatus(taskId).then((status) => {
+      // 仅「明确为空（3q 未完成）」才弹框；complete / unknown 静默放行。
+      if (alive && status === "incomplete") setOpen(true);
+    });
+    return () => {
+      alive = false;
+    };
+  }, [taskId]);
+
+  const handleGo3qfx = async () => {
+    const detail = readCopTaskDetail();
+    if (!detail || !taskId) {
+      toast.error("缺少任务信息，无法跳转");
+      return;
+    }
+    setJumping(true);
+    try {
+      // 复用深链解析：会清掉当前 rwfx 任务上下文并按 3qfx 业务重新绑定（同一 taskId）。
+      const target = await resolveTaskDeeplinkTarget("3qfx", detail, taskId);
+      if (target) {
+        setOpen(false);
+        navigate(target.path, { state: target.state });
+      } else {
+        toast.error("未配置三情分析（3qfx）对应的业务智能体");
+      }
+    } catch {
+      toast.error("跳转失败，请稍后再试");
+    } finally {
+      setJumping(false);
+    }
+  };
+
+  return (
+    <Dialog open={open} onOpenChange={(o) => !jumping && setOpen(o)}>
+      <DialogContent>
+        <DialogHeader>
+          <DialogTitle>3q 未完成</DialogTitle>
+          <DialogDescription>
+            未查询到该任务的研判情报报告（3q 未完成），是否前往该任务的三情分析（3qfx）？
+          </DialogDescription>
+        </DialogHeader>
+        <DialogFooter>
+          <Button variant="outline" onClick={() => setOpen(false)} disabled={jumping}>
+            暂不
+          </Button>
+          <Button
+            className="bg-[var(--color-main)] text-white hover:bg-[var(--brand-950)]"
+            onClick={handleGo3qfx}
+            disabled={jumping}
+          >
+            {jumping ? <Loader2 className="h-4 w-4 animate-spin" /> : "前往 3qfx"}
+          </Button>
+        </DialogFooter>
+      </DialogContent>
+    </Dialog>
+  );
+}
+
 /**
  * 任务工作区布局：左列表 + 右 Outlet。
  * ``lite``（3qfx/xdfx）= 精简版：左列表无跳转按钮、无底部任务操作面板。
@@ -305,6 +384,8 @@ export function TaskFxWorkspace({
         </div>
       </div>
       <TaskJumpBar business={business} taskId={taskId} />
+      {/* rwfx 专属：进入时预检 3q 是否完成，未完成则弹框引导去 3qfx。 */}
+      {business === "rwfx" && <RwfxSqReportGate taskId={taskId} />}
     </div>
   );
 }
diff --git a/offline-backend-20260512/backend/CLAUDE.md b/offline-backend-20260512/backend/CLAUDE.md
index efe0577d..c6715a5e 100644
--- a/offline-backend-20260512/backend/CLAUDE.md
+++ b/offline-backend-20260512/backend/CLAUDE.md
@@ -393,7 +393,10 @@ query-time type, `recommended_question`.
 Sync only ever adds/removes `origin='position'` rows, so a user's own
 favorites/subscriptions are never clobbered. Default sets (no binding for a
 type): agents = built-ins (`user_id IS NULL`); skills = empty (square stays
-full, nothing added to 我的); scheduled tasks = empty; questions = unrestricted.
+full, nothing added to 我的); scheduled tasks = empty; questions = **untagged
+only** (默认岗位 / 无标签绑定 只看到未打标的题目；打标的题目只在绑定了对应标签的
+岗位展示 — the router interprets `resolve_visible → None` as "show only untagged",
+not "show everything").
 `origin`/`position_id` columns were added to `agent_favorites` and
 `scheduled_task_subscriptions` — see migration `20260616_01` +
 `scripts/sql/column_additions_mysql.sql`. The running app also auto-adds these
@@ -415,8 +418,12 @@ unassigned users inherit, else `null`); powers the top-bar 岗位 label
 changes: `GET /api/recommended-questions` is now
 filtered by the caller's position (`?all=1` bypasses for admin management) and
 admin can tag questions via `/api/tags/assignments/recommended_question/{id}`.
-Each listed question now carries its assigned `tags` (bulk-attached via
-`tag_store.list_assignments`), and the list accepts a repeatable `tag_ids`
+**Default-position rule**: tags are bulk-attached to **all** questions first,
+then the visibility filter applies — a position binding recommended_question 标签
+sees only those tagged questions; a 默认岗位 / 无绑定 (`resolve_visible → None`)
+sees only the **untagged** questions (打标的题目不再泄漏到默认岗位). `?all=1`
+(admin management) skips this filter and returns every question with its tags.
+The list also accepts a repeatable `tag_ids`
 OR-filter — so the 推荐问题配置 page can display per-question 标签 and search by
 them (frontend `RecommendedQuestionsPage` + `TagSearchBar`). Tests:
 `tests/test_recommended_questions_tags.py`.
diff --git a/offline-backend-20260512/backend/app/gateway/routers/recommended_questions.py b/offline-backend-20260512/backend/app/gateway/routers/recommended_questions.py
index a267c901..cb0e5591 100644
--- a/offline-backend-20260512/backend/app/gateway/routers/recommended_questions.py
+++ b/offline-backend-20260512/backend/app/gateway/routers/recommended_questions.py
@@ -67,24 +67,11 @@ async def list_recommended_questions(
 ) -> RecommendedQuestionsListResponse:
     store = _get_store(request)
     rows = await store.list_all()
-    if not all:
-        # Filter by the caller's 岗位 (position) visibility. ``None`` =
-        # unrestricted (no skill/question tag binding → show all default
-        # questions); a set = show only the questions carrying the bound tags.
-        sync = getattr(request.app.state, "position_sync", None)
-        if sync is not None:
-            try:
-                visible = await sync.resolve_visible(_current_user_id(request), "recommended_question")
-            except Exception:
-                logger.debug("recommended-question visibility resolve failed", exc_info=True)
-                visible = None
-            if visible is not None:
-                rows = [r for r in rows if r["id"] in visible]
     responses = [RecommendedQuestionResponse(**r) for r in rows]
 
     # Bulk-attach the tags assigned to each question so the management list can
-    # display them and filter by 标签. Best-effort: a missing/failing tag store
-    # never breaks the question list.
+    # display them and the 岗位 filter below can tell tagged from untagged.
+    # Best-effort: a missing/failing tag store never breaks the question list.
     tag_store = getattr(request.app.state, "tag_store", None)
     if tag_store is not None and responses:
         try:
@@ -95,6 +82,24 @@ async def list_recommended_questions(
         for resp in responses:
             resp.tags = assignments.get(resp.id, [])
 
+    if not all:
+        # Filter by the caller's 岗位 (position) visibility:
+        #   * ``set`` — the position binds recommended_question 标签 → show only
+        #     the questions carrying those bound tags.
+        #   * ``None`` — 默认岗位 / 无标签绑定 → show only **untagged** questions.
+        #     打标的题目只在绑定了对应标签的岗位展示，不再泄漏到默认岗位。
+        sync = getattr(request.app.state, "position_sync", None)
+        if sync is not None:
+            try:
+                visible = await sync.resolve_visible(_current_user_id(request), "recommended_question")
+            except Exception:
+                logger.debug("recommended-question visibility resolve failed", exc_info=True)
+                visible = None
+            if visible is not None:
+                responses = [r for r in responses if r.id in visible]
+            else:
+                responses = [r for r in responses if not r.tags]
+
     if tag_ids:
         wanted = set(tag_ids)
         responses = [r for r in responses if any(t.get("id") in wanted for t in r.tags)]
diff --git a/offline-backend-20260512/backend/packages/harness/deerflow/config/task_deeplink_config.py b/offline-backend-20260512/backend/packages/harness/deerflow/config/task_deeplink_config.py
index e5089c2a..535e8faf 100644
--- a/offline-backend-20260512/backend/packages/harness/deerflow/config/task_deeplink_config.py
+++ b/offline-backend-20260512/backend/packages/harness/deerflow/config/task_deeplink_config.py
@@ -5,7 +5,8 @@
 随时改 ``config.yaml`` 即可（Gateway 有 mtime 热重载，无需重新构建前端）。
 
 测试开关（线上接口未通时用假数据，互相独立）：
-  - ``test``        → 目的树查询（``purpose_detail_url``）走 ``mock.purpose_detail`` 假数据。
+  - ``test``        → 目的树查询（``purpose_detail_url``）走 ``mock.purpose_detail`` 假数据；
+    同时 rwfx 进入「3q 是否完成」预检（``selectSqReport``，外部接口不可用）走 ``mock.sq_report`` 假数据。
   - ``delete_test`` → 删除接口（``delete_url``）走 ``mock.delete`` 假数据（该接口尚未开发完时打开）。
   - ``action_test`` → 行动详情接口（``action_detail_url``）走 ``mock.action_detail`` 假数据。
     xdfx 深链传入的是「行动id」，需先调该接口拿 ``data.taskId`` 解析出真正的「任务id」。
@@ -25,7 +26,7 @@ class TaskDeeplinkConfig(BaseModel):
 
     test: bool = Field(
         default=False,
-        description="为真时前端目的树查询走 mock.purpose_detail 假数据（不打真实接口）",
+        description="为真时前端目的树查询走 mock.purpose_detail 假数据；同时 rwfx 进入「3q 是否完成」预检（selectSqReport）走 mock.sq_report 假数据（都不打真实接口）",
     )
     delete_test: bool = Field(
         default=False,
@@ -61,5 +62,5 @@ class TaskDeeplinkConfig(BaseModel):
     )
     mock: dict[str, Any] = Field(
         default_factory=dict,
-        description="test/delete_test/action_test 为真时返回的假数据：{purpose_detail: {...}, delete: {message, success}, action_detail: {state, msg, data: {taskId, ...}}}",
+        description="test/delete_test/action_test 为真时返回的假数据：{purpose_detail: {...}, delete: {message, success}, action_detail: {state, msg, data: {taskId, ...}}, sq_report: {state, msg, data: [...]}}（sq_report 供 rwfx 3q 预检，test 为真时用）",
     )
diff --git a/offline-backend-20260512/backend/packages/harness/deerflow/services/position_sync.py b/offline-backend-20260512/backend/packages/harness/deerflow/services/position_sync.py
index bc3ee1e1..2a6e86e8 100644
--- a/offline-backend-20260512/backend/packages/harness/deerflow/services/position_sync.py
+++ b/offline-backend-20260512/backend/packages/harness/deerflow/services/position_sync.py
@@ -16,8 +16,11 @@ Resolution rule (per resource type):
   - ``scheduled_task``      → empty (no built-in tasks)
   - ``skill``               → empty (the 广场 still shows every visible skill;
     a position only *adds* its tagged skills to the member's "我的技能")
-  - ``recommended_question`` → *unrestricted* (handled at query time, never
-    materialized — see :meth:`resolve_visible`)
+  - ``recommended_question`` → *untagged-only* (handled at query time, never
+    materialized — see :meth:`resolve_visible`). The default set is the
+    questions carrying **no** 标签; tagged questions surface only for a position
+    that binds their tag. (The recommended-questions router interprets a
+    ``None`` from :meth:`resolve_visible` as "show only untagged".)
 
 Materialization (``sync_user``) only ever touches rows it created itself
 (``origin='position'``), so a user's own favorites / subscriptions are never
@@ -67,12 +70,14 @@ class PositionSyncService:
         return await self._positions.get_default_position_id()
 
     async def resolve_visible(self, user_id: str, resource_type: str) -> set[str] | None:
-        """Return the set of visible ``resource_type`` ids, or ``None`` = unrestricted.
+        """Return the set of visible ``resource_type`` ids, or ``None`` = no tag binding.
 
         Used by the recommended-questions filter (the only remaining query-time
-        visibility type; skills are now *materialized*, not filtered). ``None``
-        means "no position restriction → show everything"; a set means "restrict
-        to exactly these ids".
+        visibility type; skills are now *materialized*, not filtered). A set
+        means "restrict to exactly these ids". ``None`` means the effective
+        position binds no tag for this type — the recommended-questions router
+        treats that as "默认岗位 → show only untagged questions" (see
+        ``app/gateway/routers/recommended_questions.py``).
         """
         pid = await self._effective_position_id(user_id)
         if not pid:
diff --git a/offline-backend-20260512/backend/tests/test_recommended_questions_tags.py b/offline-backend-20260512/backend/tests/test_recommended_questions_tags.py
index f4d19e78..054d9225 100644
--- a/offline-backend-20260512/backend/tests/test_recommended_questions_tags.py
+++ b/offline-backend-20260512/backend/tests/test_recommended_questions_tags.py
@@ -77,3 +77,65 @@ async def test_list_without_tag_store_does_not_break() -> None:
 
     assert [q.id for q in resp.questions] == ["q1", "q2"]
     assert all(q.tags == [] for q in resp.questions)
+
+
+class _FakeSync:
+    """Stub position-sync whose ``resolve_visible`` returns a fixed value.
+
+    ``None`` mimics 默认岗位 / 无标签绑定; a set mimics a position that binds
+    recommended_question 标签.
+    """
+
+    def __init__(self, visible: set[str] | None) -> None:
+        self._visible = visible
+
+    async def resolve_visible(self, _user_id, _resource_type):  # noqa: ANN001
+        return self._visible
+
+
+@pytest.mark.asyncio
+async def test_default_position_shows_only_untagged() -> None:
+    """默认岗位（resolve_visible → None）只看到未打标的题目。"""
+    tag_store = MemoryTagStore()
+    tag = await tag_store.create_tag(
+        {"id": "t1", "name": "理赔", "tag_type": "recommended_question"}
+    )
+    await tag_store.set_target_tags("recommended_question", "q2", [tag["id"]])
+
+    req = _request(_FakeQuestionStore(_rows()), tag_store)
+    req.app.state.position_sync = _FakeSync(None)
+    resp = await rq_router.list_recommended_questions(req, all=False)
+
+    assert [q.id for q in resp.questions] == ["q1"]
+
+
+@pytest.mark.asyncio
+async def test_bound_position_shows_only_its_tagged_questions() -> None:
+    """绑定了标签的岗位（resolve_visible → set）只看到对应打标的题目。"""
+    tag_store = MemoryTagStore()
+    tag = await tag_store.create_tag(
+        {"id": "t1", "name": "理赔", "tag_type": "recommended_question"}
+    )
+    await tag_store.set_target_tags("recommended_question", "q2", [tag["id"]])
+
+    req = _request(_FakeQuestionStore(_rows()), tag_store)
+    req.app.state.position_sync = _FakeSync({"q2"})
+    resp = await rq_router.list_recommended_questions(req, all=False)
+
+    assert [q.id for q in resp.questions] == ["q2"]
+
+
+@pytest.mark.asyncio
+async def test_admin_view_ignores_position_filter() -> None:
+    """``all=True`` 管理视图照旧返回全部（含打标），不受岗位过滤影响。"""
+    tag_store = MemoryTagStore()
+    tag = await tag_store.create_tag(
+        {"id": "t1", "name": "理赔", "tag_type": "recommended_question"}
+    )
+    await tag_store.set_target_tags("recommended_question", "q2", [tag["id"]])
+
+    req = _request(_FakeQuestionStore(_rows()), tag_store)
+    req.app.state.position_sync = _FakeSync(None)
+    resp = await rq_router.list_recommended_questions(req, all=True)
+
+    assert [q.id for q in resp.questions] == ["q1", "q2"]
diff --git a/offline-backend-20260512/config.yaml b/offline-backend-20260512/config.yaml
index 5c70686c..3fd5da8e 100644
--- a/offline-backend-20260512/config.yaml
+++ b/offline-backend-20260512/config.yaml
@@ -331,7 +331,7 @@ summarization:
 # ?taskId= / ?id=），随时改这里即可（Gateway mtime 热重载，前端无需重建）。
 task_deeplink:
   # 线上接口未通时用假数据；三个开关互相独立。
-  test: true            # 目的树查询走 mock.purpose_detail 假数据
+  test: true            # 目的树查询走 mock.purpose_detail 假数据；同时 rwfx 进入「3q 是否完成」预检走 mock.sq_report 假数据
   delete_test: true     # 删除接口走 mock.delete 假数据（该接口尚未开发完）
   action_test: true     # 行动→任务解析走 mock.action_detail 假数据（xdfx 行动id→任务id）
   purpose_detail_url: "https://ch1.b.uat.4.cn/consumer/taskAnalyseSearch/getTaskPurposeDetail"
@@ -382,3 +382,15 @@ task_deeplink:
         actionName: ""
         actionStatus: 25
         narration: ""
+    # test=true 时 rwfx 进入「3q 是否完成」预检（selectSqReport，外部不可用）走这份假数据。
+    # 结构同真实 GET /consumer/cop/selectSqReport?taskId= 返回；前端判定 data 为空 / 每条
+    # contentJson 都是 null/空 = 「3q 未完成」→ 弹框引导去 3qfx；否则视为已完成（正常，不弹框）。
+    # 默认给一条有内容的数据 = 已完成（不打扰）。要测「3q 未完成」弹框：把下面 data 改成 []。
+    sq_report:
+      state: "200"
+      msg: "操作成功！"
+      data:
+        - id: "sq-mock-1"
+          taskId: 3338
+          categoryType: "敌情关键事件 JSON"
+          contentJson: "[{\"topic\":\"示例研判事件\",\"count\":1}]"

From e28277b21be620c78b6252c7eace70926ad5b5cb Mon Sep 17 00:00:00 2001
From: fengxiangdebaozi <10845145+fengxiangdebaozi@user.noreply.gitee.com>
Date: Mon, 22 Jun 2026 15:43:55 +0800
Subject: [PATCH 06/17] =?UTF-8?q?feat(layout):=20=E5=85=A8=E9=87=8F?=
 =?UTF-8?q?=E6=A8=A1=E5=BC=8F=E5=8F=B3=E4=B8=8A=E8=A7=92=E4=B9=9F=E5=8A=A0?=
 =?UTF-8?q?=E7=95=8C=E9=9D=A2=E6=A8=A1=E5=BC=8F=E5=88=87=E6=8D=A2?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

复用 appearance-controls 的 LayoutModeSwitcher，Header 操作区在换肤旁
补上界面模式下拉，与简洁模式右上角保持一致。LayoutModeSwitcher 新增
fallback 属性指明 mode 未设置时的兜底界面（Header 传 full，ChatPage 传 compact）。

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 .../workspace/appearance-controls.tsx          | 18 +++++++++++++-----
 .../src/strategy-components/Header.tsx         |  7 ++++++-
 2 files changed, 19 insertions(+), 6 deletions(-)

diff --git a/frontend-web/src/components/workspace/appearance-controls.tsx b/frontend-web/src/components/workspace/appearance-controls.tsx
index 8decbe91..02efa7c8 100644
--- a/frontend-web/src/components/workspace/appearance-controls.tsx
+++ b/frontend-web/src/components/workspace/appearance-controls.tsx
@@ -126,13 +126,21 @@ const LAYOUT_OPTIONS = [
 ];
 
 /**
- * 界面模式（全量 / 简洁）下拉切换。简洁模式下全局顶栏被隐藏，故把外观设置里的
- * 「界面模式」切换也搬到页面级顶栏右上角，图标 + 下拉，与换肤并排。
+ * 界面模式（全量 / 简洁）下拉切换。换肤旁并排，图标 + 下拉：
+ *   - 简洁模式 ChatPage 顶栏（全局顶栏被隐藏）。
+ *   - 全量模式 Header 右上角。
+ * 用户未显式设置过界面模式时（mode === undefined），当前态按调用方所在的实际界面
+ * 兜底——`fallback` 指明这一回退值（Header 传 "full"，ChatPage 传 "compact"）。
  */
-export function LayoutModeSwitcher({ tone }: { tone: AppearanceControlTone }) {
+export function LayoutModeSwitcher({
+  tone,
+  fallback = "compact",
+}: {
+  tone: AppearanceControlTone;
+  fallback?: "full" | "compact";
+}) {
   const [mode, setLayoutMode] = useLayoutMode();
-  // 未显式设置过时简洁模式才会渲染本控件（全量模式有全局顶栏），故当前态按 compact 兜底。
-  const current = mode === "full" ? "full" : "compact";
+  const current = mode ?? fallback;
   const TriggerIcon = current === "full" ? PanelLeftIcon : LayoutIcon;
 
   return (
diff --git a/frontend-web/src/strategy-components/Header.tsx b/frontend-web/src/strategy-components/Header.tsx
index 57d7480e..0bcb8334 100644
--- a/frontend-web/src/strategy-components/Header.tsx
+++ b/frontend-web/src/strategy-components/Header.tsx
@@ -4,7 +4,7 @@ import { useLocation, useNavigate } from "react-router-dom";
 
 import { Button } from "@/components/ui/button";
 import "@/components/page-sdebar.css";
-import { ThemeSwitcher } from "@/components/workspace/appearance-controls";
+import { LayoutModeSwitcher, ThemeSwitcher } from "@/components/workspace/appearance-controls";
 import { NotificationDialog } from "@/components/workspace/notification-dialog";
 import { getStoredAuth, getAccountDisplayName } from "@/core/auth";
 import { isWujieEnv } from "@/core/embed/wujie";
@@ -126,6 +126,11 @@ const Header = ({ collapsed }: HeaderProps) => {
     <>
       <ThemeSwitcher tone={isDark ? "banner-dark" : "banner-light"} />
 
+      <LayoutModeSwitcher
+        tone={isDark ? "banner-dark" : "banner-light"}
+        fallback="full"
+      />
+
       <Button
         type="button"
         variant="ghost"

From 7748661963d53f024a7ebfbe0297f7b488434073 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E8=91=A3=E5=AE=8F=E5=87=AF?= <hkdong@kland.com.cn>
Date: Mon, 22 Jun 2026 16:35:58 +0800
Subject: [PATCH 07/17] =?UTF-8?q?feat(roundtable):=20=E5=8E=86=E5=8F=B2?=
 =?UTF-8?q?=E5=8D=A1=E7=89=87=E6=98=BE=E7=A4=BA=E6=89=80=E7=94=A8=E4=B8=9A?=
 =?UTF-8?q?=E5=8A=A1=E9=93=BE(=E8=87=AA=E7=94=B1=E6=A8=A1=E5=BC=8F)=20+=20?=
 =?UTF-8?q?=E8=AF=8A=E6=96=AD=E5=8F=AA=E6=94=B6=E6=8A=A5=E9=94=99=E6=97=A5?=
 =?UTF-8?q?=E5=BF=97?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- 历史「分析记录」卡片/侧边栏列表：把「第X步」改为本次会商所用业务链名，
  无链条显示「自由模式」。后端 list 轻量 meta 从 step2 正则抽取 chain_title
  (不整体解析大块 JSON)，个人草稿 + task 草稿两个 store 都覆盖。
- 会商诊断改为只收集报错日志：record_diagnostic 写库前丢弃 level==info
  运行日志(begin/consensus/gather/job_started)，只留 warning/error。
  单点收口覆盖前台/后台/recorder 全部写路径；诊断页标题改「会商报错日志」。

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 frontend-web/.env.development                 |  2 +-
 .../workspace/recent-roundtable-list.tsx      |  3 ++-
 .../src/pages/RoundtableDiagnosticsPage.tsx   |  4 +--
 .../src/roundtable-planning/api/drafts.ts     |  7 +++++
 .../pages/RoundtablePlanningPage.tsx          | 13 ++++++----
 offline-backend-20260512/backend/CLAUDE.md    |  6 ++---
 .../app/gateway/routers/roundtable_drafts.py  |  2 ++
 .../gateway/routers/roundtable_task_drafts.py |  2 ++
 .../roundtable_diagnostics/__init__.py        |  9 ++++++-
 .../persistence/roundtable_drafts/sql.py      | 26 +++++++++++++++++++
 .../persistence/roundtable_task_drafts/sql.py | 24 +++++++++++++++++
 11 files changed, 85 insertions(+), 13 deletions(-)

diff --git a/frontend-web/.env.development b/frontend-web/.env.development
index 3837c075..b78f859a 100644
--- a/frontend-web/.env.development
+++ b/frontend-web/.env.development
@@ -8,7 +8,7 @@ VITE_USE_REAL_DATA=true
 VITE_MOCK_COP_TASK_DETAIL=true
 
 # 研判情报报告（大屏绘制智能体深链 selectSqReport）数据源：true 走本地样例假数据（真实接口未就绪时调试用）；联调通过后置 false。
-VITE_MOCK_SQ_REPORT=true
+VITE_MOCK_SQ_REPORT=false
 
 # 深链任务详情（cop-task-detail：任务名称/描述）数据源：true 走本地假数据（真实接口未就绪时调试用）；联调通过后置 false。
 VITE_MOCK_COP_TASK_DETAIL=true
diff --git a/frontend-web/src/components/workspace/recent-roundtable-list.tsx b/frontend-web/src/components/workspace/recent-roundtable-list.tsx
index c1af707f..67482aab 100644
--- a/frontend-web/src/components/workspace/recent-roundtable-list.tsx
+++ b/frontend-web/src/components/workspace/recent-roundtable-list.tsx
@@ -71,9 +71,10 @@ function StatusBadge({ draft }: { draft: DraftMeta }) {
   if (draft.status === "error") {
     return <span className="text-[10px] font-medium text-rose-500">出错</span>;
   }
+  // 非进行中/待澄清/出错时，展示本次会商所用业务链（无链条 → 自由模式）。
   return (
     <span className="text-[10px] font-medium text-muted-foreground">
-      第 {draft.furthestStep} 步
+      {draft.chainTitle || "自由模式"}
     </span>
   );
 }
diff --git a/frontend-web/src/pages/RoundtableDiagnosticsPage.tsx b/frontend-web/src/pages/RoundtableDiagnosticsPage.tsx
index 1804cd56..fd96a297 100644
--- a/frontend-web/src/pages/RoundtableDiagnosticsPage.tsx
+++ b/frontend-web/src/pages/RoundtableDiagnosticsPage.tsx
@@ -225,9 +225,9 @@ export default function RoundtableDiagnosticsPage() {
           返回
         </Button>
         <div>
-          <h1 className="text-base font-semibold text-slate-800 dark:text-foreground">会商诊断日志</h1>
+          <h1 className="text-base font-semibold text-slate-800 dark:text-foreground">会商报错日志</h1>
           <p className="text-xs text-slate-400 dark:text-muted-foreground">
-            多智能体会商全流程报错与关键事件（含后台挂起卡死、超时、产物缺失等）
+            多智能体会商全流程报错日志（席位失败 / 空交付 / 超时 / 模型容错 / 产物缺失 / 崩溃等；运行进度类 info 不再收集）
           </p>
         </div>
         <Button
diff --git a/frontend-web/src/roundtable-planning/api/drafts.ts b/frontend-web/src/roundtable-planning/api/drafts.ts
index 7641baba..40bcc3e9 100644
--- a/frontend-web/src/roundtable-planning/api/drafts.ts
+++ b/frontend-web/src/roundtable-planning/api/drafts.ts
@@ -188,6 +188,11 @@ export interface DraftMeta {
   /** Auto-derived task summary (objective → first user message → timestamp). */
   title: string;
   furthestStep: 1 | 2 | 3;
+  /**
+   * 本次会商所用业务链名（历史卡片展示）。无业务链（自由/recommend 模式）→ null/undefined，
+   * 由前端展示为「自由模式」。后端从 step2 快照里抽取（见 roundtable_drafts/sql.py）。
+   */
+  chainTitle?: string | null;
   createdAt: string;
   updatedAt: string;
   /**
@@ -250,6 +255,7 @@ export interface ServerDraftMeta {
   task_id?: string | null;
   title: string;
   furthest_step: number;
+  chain_title?: string | null;
   created_at: string | null;
   updated_at: string | null;
   status?: string | null;
@@ -269,6 +275,7 @@ export function toMeta(s: ServerDraftMeta): DraftMeta {
     taskId: s.task_id ?? null,
     title: s.title ?? "",
     furthestStep: step,
+    chainTitle: s.chain_title ?? null,
     createdAt: s.created_at ?? "",
     updatedAt: s.updated_at ?? "",
     status: (s.status as JobStatus | null | undefined) ?? null,
diff --git a/frontend-web/src/roundtable-planning/pages/RoundtablePlanningPage.tsx b/frontend-web/src/roundtable-planning/pages/RoundtablePlanningPage.tsx
index bb7b0733..3fb86805 100644
--- a/frontend-web/src/roundtable-planning/pages/RoundtablePlanningPage.tsx
+++ b/frontend-web/src/roundtable-planning/pages/RoundtablePlanningPage.tsx
@@ -3168,13 +3168,16 @@ function RoundtableEntryRail({
                         {completed ? "已完成" : "未完成"}
                       </span>
                     </div>
-                    {/* 下：时间（左下） + 步骤（右下） */}
-                    <div className="mt-2 flex items-center justify-between">
-                      <span className="font-mono text-[10px] text-neutral-400">
+                    {/* 下：时间（左下） + 业务链（右下，无链条显示「自由模式」） */}
+                    <div className="mt-2 flex items-center justify-between gap-2">
+                      <span className="font-mono text-[10px] text-neutral-400 shrink-0">
                         {formatTime(rec.updatedAt)}
                       </span>
-                      <span className="rounded-[5px] bg-neutral-100 px-1.5 py-0.5 text-[10px] font-medium text-neutral-500">
-                        第 {rec.furthestStep} 步
+                      <span
+                        className="max-w-[60%] truncate rounded-[5px] bg-neutral-100 px-1.5 py-0.5 text-[10px] font-medium text-neutral-500"
+                        title={rec.chainTitle || "自由模式"}
+                      >
+                        {rec.chainTitle || "自由模式"}
                       </span>
                     </div>
                   </button>
diff --git a/offline-backend-20260512/backend/CLAUDE.md b/offline-backend-20260512/backend/CLAUDE.md
index 22eca306..68f8f72f 100644
--- a/offline-backend-20260512/backend/CLAUDE.md
+++ b/offline-backend-20260512/backend/CLAUDE.md
@@ -245,9 +245,9 @@ FastAPI application on port 8001 with health check at `GET /health`. Set `GATEWA
 | **Runs** (`/api/runs`) | `POST /stream` - stateless run + SSE; `POST /wait` - stateless run + block; `GET /{rid}/messages` - paginated messages by run_id `{data, has_more}` (cursor: `after_seq`/`before_seq`); `GET /{rid}/feedback` - list feedback by run_id |
 | **Thread Shares** (`/api/thread-shares`) | Per-conversation sharing with two modes (`thread_shares.mode`). **`import`**: `POST /` `{thread_id, mode}` - create (or reuse) a revocable share code; `GET /{code}/preview` + `POST /{code}/import` - copy that conversation into the caller's account (fresh `thread_id` via `app/gateway/thread_copy.py` + new `threads_meta` row + `user-data` copy; dedup on `metadata.shared_origin`). **`view`**: a public, read-only static page ? see Public Shares below. Both modes: `GET /` - list my codes; `DELETE /{code}` - revoke. |
 | **Public Shares** (`/api/public/shares`) | **Unauthenticated** (the `/api/public/` prefix is whitelisted in `auth_middleware._PUBLIC_PATH_PREFIXES`). `GET /{code}` - read-only conversation snapshot (title + serialized messages + artifacts) for a `view`-mode share; messages are read straight from the latest checkpoint. `GET /{code}/files/{path}` - serves a file referenced by that conversation (images, Markdown, text, PDF, ?) so the shared page's links are openable; path resolution is confined to the owner's thread directory. Active content (`text/html`, `application/xhtml+xml`, `image/svg+xml`) is forced to download (`_public_share_disposition`); everything else is served inline. Missing/revoked/non-`view` codes all return 404. |
-| **Roundtable Drafts** (`/api/roundtable-drafts`) | **Strictly per-user** persistence for the roundtable-planning page (replaces the old browser `localStorage`). `GET /` - list drafts (lightweight meta); `POST /` - create; `GET/PUT/DELETE /{id}` - full read / partial update / delete (all owner-scoped). A draft = one full session (Step 1 intent + Step 2 roundtable), step snapshots stored as JSON in `PortableLongText`. Recommendation history bound to a draft: `GET /{id}/recommendations` - list newest-first; `POST /{id}/recommendations` - append one run (objective/status/rationale/picks/candidates). taskId 深链会商记录 are **not** here — they live in the separate `/api/roundtable-task-drafts` store (see below) so personal history never mixes with task records. Backed by `deerflow.persistence.roundtable_drafts` (tables `roundtable_drafts`, `roundtable_recommend_history`); store wired in `deps.py` as `app.state.roundtable_draft_store`. Tests: `tests/test_roundtable_task_drafts.py`. |
-| **Roundtable Task Drafts** (`/api/roundtable-task-drafts`) | **Task-scoped, unpartitioned** store for taskId-deeplink roundtable 聊天记录 (无界嵌入抽屉). Keyed **only by `task_id`** (one task → many records), **never partitioned by user** — any authenticated caller may read/write/delete a task's records (`created_by` is audit-only). Physically separate from `roundtable_drafts` so the two never mix. `GET /by-task/{task_id}` - all drafts for a task, newest-first (history dropdown), enriched with the latest background-job status per draft via `RoundtableJobRepository.status_map_by_drafts`; `GET /by-task/{task_id}/latest` - newest draft (回显最新一条); `POST /` - create (`task_id` required); `GET/PUT/DELETE /{id}` - full read / partial update / delete (all user-agnostic). Backed by `deerflow.persistence.roundtable_task_drafts` (table `roundtable_task_drafts`, auto-created by `create_all`; migration `20260618_01`); store wired in `deps.py` as `app.state.roundtable_task_draft_store`. Existing `task_id`-bearing `roundtable_drafts` rows are migrated over by `scripts/migrate_roundtable_task_drafts.py`. Tests: `tests/test_roundtable_task_drafts.py`, `tests/test_roundtable_drafts_by_task.py`. |
-| **Roundtable Diagnostics** (`/api/roundtable-diagnostics`) | 会商全流程诊断日志 (报错 + 关键事件). One row per event in `roundtable_diagnostics`: `{scope (background = 后台挂起 / foreground = 前台服务端观测 / client = 前端上报), stage (step1_intent/step2_init/step2_leader/step2_seat/step2_recommend/step2_orchestration/step3_summary/step3_report/step3_dashboard/step3_structure/upstream/job), level (info/warning/error), event, message, detail (JSON), job_id/draft_id/task_id/user_id/agent_id/agent_name/cycle, created_at}`. **Reads are admin-only**: `GET /` - filter (scope/stage/level/event/job_id/draft_id/task_id/user_id/q/since/until; `scope` accepts a comma-separated set so 前端「前台调用」= `foreground,client`) + paginate (newest-first), each item enriched with `userName` (user_id → 邮箱 `@`-prefix via `get_local_provider().get_user`, best-effort batched cache); `GET /facets` - distinct values for filter dropdowns; `POST /admin/cleanup` - retention purge. **`POST /report` is NOT admin-gated** — any authenticated user's frontend posts the errors *the user actually hit* (HTTP 502 before any SSE frame / network drop / SSE error frame) so they're collected too; stored as `scope="client"`. Backed by `deerflow.persistence.roundtable_diagnostics` (table `roundtable_diagnostics`, auto-created by `create_all`; migration `20260619_01`); store wired in `deps.py` as `app.state.roundtable_diagnostic_store` + a process-level default store (`register_default_store`) so the harness-side engine/gateway can write across the app/harness boundary. **Writers** (all best-effort — diagnostics never break a run): background path uses `DiagnosticsRecorder` (bound to job ctx) threaded through `RoundtableJobExecutor` → `JobProgressWriter.diag(...)` (engine transitions/step3 failures/empty-delivery) + `InProcessRoundtableGateway` (per-turn timeout/report·summary·dashboard truncation+missing); foreground path uses `app/gateway/roundtable_diag.py::record_foreground_diag` in `multi_agent.py`/`intent.py`/`recommend.py` error frames. **Bug-fix shipped alongside**: `InProcessRoundtableGateway._run_agent_turn` guards each agent run with an **inactivity (idle) timeout** (`_await_run_with_idle_guard`, `ROUNDTABLE_SEAT_IDLE_TIMEOUT_SECONDS` default 1200s = 20min, tuned high for unstable intranet models) — **not** a total-duration cap, so a slow-but-alive run that keeps streaming tokens runs as long as it needs; only a run that emits **zero StreamBridge events for `idle_timeout` seconds straight** (a true freeze: 内网模型连接池挂死 / sqlite 锁在挂载卷上挂死) is cancelled + raises (顶层把 job 标 error，而非永久冻结) + records a `seat_turn_timeout` diagnostic. Activity is observed via `app.state.stream_bridge` (the worker publishes every chunk there). Optional absolute cap `ROUNDTABLE_SEAT_MAX_TOTAL_SECONDS` (default 0 = off). Frontend: admin page `pages/RoundtableDiagnosticsPage.tsx` (route `/page/strategy/admin/roundtable-diagnostics`), opened from the **会商页右上角设置按钮** (admin only — `handleSettingsClick` in `roundtable-planning/pages/RoundtablePlanningPage.tsx`); admin read client `strategy-components/api/roundtable-diagnostics.ts`. **Client error reporter** `roundtable-planning/api/diagnostics-report.ts` (`reportRoundtableError` fire-and-forget + `keepalive`, `setRoundtableErrorContext` for draft/task attribution set by the page, `stageForRoundtableAgent`) is called at every client-observable failure: `multi-agent.ts` (`streamMultiAgent` — covers Step2 leader/seat AND Step3 report/summary/dashboard/structure — reports HTTP error + SSE `{status:"error"}` frame + worker `event:error` frame [模型调用没成功] + fetch-reject [网络不可达] + mid-stream drop [连接重置/超时], deduped via a `reported` flag; `initMultiAgent` HTTP/error-frame/incomplete/mid-stream), `intent.ts` (Step1 HTTP+error-frame), `recommend.ts` (Step2 recommend HTTP+error-frame), `roundtable-jobs.ts` (job api/stream). The admin page's first filter is a **后台挂起 / 前台调用** toggle (frontend = `foreground,client`); the stage dropdown options adapt to it (background → step2+step3+job; frontend → step1+step2+step3) and each row shows the reporting user's name + 👤. Tests: `tests/test_roundtable_diagnostics.py`, `tests/test_roundtable_inprocess_timeout.py`. **Seat skill-recognition (弱模型补偿)**: a seat = one custom agent run whose configured skills (`agent config.yaml → skills`) are injected **only** via the system prompt's `<skill_system>` block (model must `read_file` the SKILL.md itself) — weak models (deepseek-chat 等) ignore it and never use their skills. `app/gateway/roundtable_seat_skills.py::append_seat_skill_directive(agent_id, task, *, chain_enabled=False)` appends an explicit Chinese directive (each configured-and-enabled skill's name + description + SKILL.md container path + "先 read_file 再按其流程取信息/执行") to the **seat's per-turn task (human turn)** — far stronger pull on weak models than the system prompt. Wired into **both** seat paths: foreground `multi_agent._special_run` (only `role in {seat, seat_parallel}`; Step3 singletons report/summary/dashboard/ingest excluded) + background `InProcessRoundtableGateway.run_seat`. No-op (returns task unchanged) when the seat has no configured skills. **2026-06-20: default-off + two opt-in switches (OR semantics)** — was global-on, now injects only when **either** the seat agent's own switch (`AgentConfig.seat_skill_directive`, edited in the agent edit dialog `agent-card.tsx` + `agent-detail-panel.tsx`) **or** the **per-seat** switch on that agent **within the selected business chain** is on. The per-seat chain switch lives **in the chain's `seats` JSON** (`ChainSeatSchema.seat_skill_directive`, no DB column — rides the existing JSON like the per-seat `model` override), edited via a small ✨ icon → dropdown (Switch + description) on each agent card in `BusinessChainEditorPage`. It is threaded **per-seat** to each seat run: foreground `multi_agent` per-call field `seat_skill_directive` (orchestration looks it up per agentId from a `{agentId: bool}` session map) → `_special_run` → `chain_enabled`; background as the job field `seat_skill_directives: dict[str,bool]` (StartJobRequest → StartParams → deps factory → `InProcessRoundtableGateway._seat_skill_directives`) → `run_seat` looks up `agent_id` → `chain_enabled`. `ROUNDTABLE_SEAT_SKILL_DIRECTIVE` is only a **master kill-switch** (default allow; set `0/false` to hard-disable regardless of the switches — never force-enables). Tests: `tests/test_roundtable_seat_skills.py`, `tests/test_agents_backend.py`. **2026-06-20 Step2 提速三件套**：(1) **每席位推理深度覆盖** —— 业务链条 `seats` JSON 里逐席位可配 `seat_mode`（flash/thinking/pro/ultra，覆盖链条级 `seatMode`；编辑器席位卡右侧 ⏱Gauge 图标下拉，null=跟随链条默认）。前台经 `useStep2Orchestration` 三处 dispatch 的 `seatModeToRequestFields(seatModes[id] ?? selectedMode)` 透传；后台经 job `seat_modes: dict[str,str]` → `StartParams` → `InProcessRoundtableGateway._seat_modes`，`run_seat` 用 `_SEAT_MODE_PARAMS`（与前端 `seatModeToRunParams` 逐字对齐）解析覆盖作业级 seat_*（未命中→跟随作业级→policy 默认）。无 DB 列（ride `seats` JSON，但 `ChainSeatSchema.seat_mode` 须显式声明否则 Pydantic 剥掉）。Tests: `tests/test_roundtable_seat_modes.py`。(2) **recommend 批内并行** —— 后端 `engine._run_recommend` 由 for-await 串行改 `asyncio.gather`（对齐前端 2026-06-13 已并行；总控一轮派多席位=判定相互独立→并发；批内用派活前 deliveries 快照、批后统一并入，语义不变）。(3) **研讨前并行『取数』两段式**（业务链条 `gather_first` 列，逐条 opt-in，migration `20260620_02`，editor「研讨前并行取数」开关）—— `gather_first=True` 时 `run_orchestration` 在分发前先 `engine._run_gather_phase`（全席位 `asyncio.gather` 跑 `_GATHER_TASK`「只取数不研讨」），产出 `gathered: dict` 经 `_seat_message(gathered=...)` 作为「已收集资料」块注入 recommend/chain/dag 各研讨轮席位任务（与「前序交付」块分开），把最慢的技能延迟前置并行、给（尤其 chain 串行的）研讨段提速；研讨阶段不限制再调技能。前台对称实现：`useStep2Orchestration.runGatherPhase` 在 `runActiveOrchestration` 分发前跑一次（`gatherPhaseDoneRef` 闸门，resetInitGate 复位），产出存 `gatheredDeliveriesRef` 经 `peerDeliveries`/初始 `priorDeliveries` 注入研讨席位。链条级 `gatherFirst`(camel)↔`gather_first`(snake) 经 `api/chains.ts` 映射；job 经 `StartJobRequest.gather_first`→`StartParams`→`run_orchestration(gather_first=…)`。Tests: `tests/test_roundtable_gather_phase.py`。(4) **削减重复广播**（前端 only，多智能体 foreground 路径）—— `suppress_pre_broadcast`（仅 leader：抑制「总控本轮输入→各席位 thread」预广播 + 「总控给X派活」→兄弟席位的 post 广播，二者一个 flag）此前只在 DAG 派活轮开；现扩到 recommend（`safetyCounter > 1`，即第 2 轮起的续轮/综合）+ chain 逐席位（`i > 0`，第 2 个席位起）+ chain 收口轮（恒 true）。**保留每种模式的首轮预广播**（它为席位打底任务背景），只 suppress 重复的续轮广播（N 次读写巨大席位 thread 的纯浪费 + 弱模型模仿派活）；席位的前序交付另经「席位交付广播」`broadcastSeatDeliveryToOthers` 送达，不受影响。后台引擎路径无此广播（`_seat_message` 直接拼上下文），故 #4 仅前端。**2026-06-20 深链接业务链抽取规范**：深链接会商（taskId+goPath，businessCode 如 rwfx→6BF）下，结构化抽取(flow-json)与方案总结报告都按所选业务链**完整层级**强制校正。结构化抽取实为 `roundtable-structure` 智能体（`flow_extract.py` 的 `_system_prompt()=load_agent_soul(STRUCTURE_AGENT_ID)`，抽取→校验两段式 `model.astream`）；其 SOUL 全局一份，旧实现把某业务层级写死进 SOUL→所有业务都按「目的→行为体」抽（bug）。修法：**单份领域知识、两角色用法**的 `deerflow.agents.roundtable_orchestrator.extraction_spec`（放 harness 层，app+harness 都可 import；`build_structure_directive`/`build_summary_directive` 按 businessCode 取；rwfx=6BF 写实「任务→目的→行为体→预期行为→驱动因素→叙事，预期行为↔驱动因素 1:1、余 1:N、label 不含步骤/阶段名、优先报告回退席位」，其余业务通用规范=强约束完整性+命名+取数且明令不套固定模板；business_code 空→""零副作用）。注入：(1) 结构化抽取 `flow_extract._hierarchy_hint`+`_validate_messages` 用 `build_structure_directive` 权威覆盖默认层级（前端 `streamFlowExtract` 传 `business_code`，源 `ingestBusinessCode`）；(2) 方案总结报告 `build_summary_directive`——前台 `multi_agent._special_run`(role==summary 且 business_code) 追加、后台 `engine.build_summary_prompt(business_code=…)`（StartParams/StartJobRequest 透传）。**仅深链接生效**。种子 `roundtable-structure/SOUL.md` 层级顺序改 spec-driven、连接示例去 rwfx 写死（存量 live SOUL 不被 seed 覆盖，但注入规范权威可纠正）。Tests: `tests/test_roundtable_extraction_spec.py`。 |
+| **Roundtable Drafts** (`/api/roundtable-drafts`) | **Strictly per-user** persistence for the roundtable-planning page (replaces the old browser `localStorage`). `GET /` - list drafts (lightweight meta — incl. `chain_title`, the business-chain name used this session, cheaply regex-extracted from the in-memory `step2` blob in `_draft_to_meta`/`_extract_chain_title`; null = no chain / 自由模式; surfaced on the 第一步「分析记录」history card + sidebar list instead of the old「第 X 步」); `POST /` - create; `GET/PUT/DELETE /{id}` - full read / partial update / delete (all owner-scoped). A draft = one full session (Step 1 intent + Step 2 roundtable), step snapshots stored as JSON in `PortableLongText`. Recommendation history bound to a draft: `GET /{id}/recommendations` - list newest-first; `POST /{id}/recommendations` - append one run (objective/status/rationale/picks/candidates). taskId 深链会商记录 are **not** here — they live in the separate `/api/roundtable-task-drafts` store (see below) so personal history never mixes with task records. Backed by `deerflow.persistence.roundtable_drafts` (tables `roundtable_drafts`, `roundtable_recommend_history`); store wired in `deps.py` as `app.state.roundtable_draft_store`. Tests: `tests/test_roundtable_task_drafts.py`. |
+| **Roundtable Task Drafts** (`/api/roundtable-task-drafts`) | **Task-scoped, unpartitioned** store for taskId-deeplink roundtable 聊天记录 (无界嵌入抽屉). Keyed **only by `task_id`** (one task → many records), **never partitioned by user** — any authenticated caller may read/write/delete a task's records (`created_by` is audit-only). Physically separate from `roundtable_drafts` so the two never mix. `GET /by-task/{task_id}` - all drafts for a task, newest-first (history dropdown), enriched with the latest background-job status per draft via `RoundtableJobRepository.status_map_by_drafts` (each meta also carries `chain_title`, same `_extract_chain_title` regex as the personal store); `GET /by-task/{task_id}/latest` - newest draft (回显最新一条); `POST /` - create (`task_id` required); `GET/PUT/DELETE /{id}` - full read / partial update / delete (all user-agnostic). Backed by `deerflow.persistence.roundtable_task_drafts` (table `roundtable_task_drafts`, auto-created by `create_all`; migration `20260618_01`); store wired in `deps.py` as `app.state.roundtable_task_draft_store`. Existing `task_id`-bearing `roundtable_drafts` rows are migrated over by `scripts/migrate_roundtable_task_drafts.py`. Tests: `tests/test_roundtable_task_drafts.py`, `tests/test_roundtable_drafts_by_task.py`. |
+| **Roundtable Diagnostics** (`/api/roundtable-diagnostics`) | 会商全流程诊断日志 (报错 + 关键事件). One row per event in `roundtable_diagnostics`: `{scope (background = 后台挂起 / foreground = 前台服务端观测 / client = 前端上报), stage (step1_intent/step2_init/step2_leader/step2_seat/step2_recommend/step2_orchestration/step3_summary/step3_report/step3_dashboard/step3_structure/upstream/job), level (warning/error — **`info` 级运行日志不再收集**：`record_diagnostic` 在写库前丢弃 `level=="info"`，只留报错；begin/consensus/gather/job_started 等纯进度事件即使仍调 `.diag(level="info")` 也成 no-op，故诊断页 = 报错日志页), event, message, detail (JSON), job_id/draft_id/task_id/user_id/agent_id/agent_name/cycle, created_at}`. **Reads are admin-only**: `GET /` - filter (scope/stage/level/event/job_id/draft_id/task_id/user_id/q/since/until; `scope` accepts a comma-separated set so 前端「前台调用」= `foreground,client`) + paginate (newest-first), each item enriched with `userName` (user_id → 邮箱 `@`-prefix via `get_local_provider().get_user`, best-effort batched cache); `GET /facets` - distinct values for filter dropdowns; `POST /admin/cleanup` - retention purge. **`POST /report` is NOT admin-gated** — any authenticated user's frontend posts the errors *the user actually hit* (HTTP 502 before any SSE frame / network drop / SSE error frame) so they're collected too; stored as `scope="client"`. Backed by `deerflow.persistence.roundtable_diagnostics` (table `roundtable_diagnostics`, auto-created by `create_all`; migration `20260619_01`); store wired in `deps.py` as `app.state.roundtable_diagnostic_store` + a process-level default store (`register_default_store`) so the harness-side engine/gateway can write across the app/harness boundary. **Writers** (all best-effort — diagnostics never break a run): background path uses `DiagnosticsRecorder` (bound to job ctx) threaded through `RoundtableJobExecutor` → `JobProgressWriter.diag(...)` (engine transitions/step3 failures/empty-delivery) + `InProcessRoundtableGateway` (per-turn timeout/report·summary·dashboard truncation+missing); foreground path uses `app/gateway/roundtable_diag.py::record_foreground_diag` in `multi_agent.py`/`intent.py`/`recommend.py` error frames. **Bug-fix shipped alongside**: `InProcessRoundtableGateway._run_agent_turn` guards each agent run with an **inactivity (idle) timeout** (`_await_run_with_idle_guard`, `ROUNDTABLE_SEAT_IDLE_TIMEOUT_SECONDS` default 1200s = 20min, tuned high for unstable intranet models) — **not** a total-duration cap, so a slow-but-alive run that keeps streaming tokens runs as long as it needs; only a run that emits **zero StreamBridge events for `idle_timeout` seconds straight** (a true freeze: 内网模型连接池挂死 / sqlite 锁在挂载卷上挂死) is cancelled + raises (顶层把 job 标 error，而非永久冻结) + records a `seat_turn_timeout` diagnostic. Activity is observed via `app.state.stream_bridge` (the worker publishes every chunk there). Optional absolute cap `ROUNDTABLE_SEAT_MAX_TOTAL_SECONDS` (default 0 = off). Frontend: admin page `pages/RoundtableDiagnosticsPage.tsx` (route `/page/strategy/admin/roundtable-diagnostics`), opened from the **会商页右上角设置按钮** (admin only — `handleSettingsClick` in `roundtable-planning/pages/RoundtablePlanningPage.tsx`); admin read client `strategy-components/api/roundtable-diagnostics.ts`. **Client error reporter** `roundtable-planning/api/diagnostics-report.ts` (`reportRoundtableError` fire-and-forget + `keepalive`, `setRoundtableErrorContext` for draft/task attribution set by the page, `stageForRoundtableAgent`) is called at every client-observable failure: `multi-agent.ts` (`streamMultiAgent` — covers Step2 leader/seat AND Step3 report/summary/dashboard/structure — reports HTTP error + SSE `{status:"error"}` frame + worker `event:error` frame [模型调用没成功] + fetch-reject [网络不可达] + mid-stream drop [连接重置/超时], deduped via a `reported` flag; `initMultiAgent` HTTP/error-frame/incomplete/mid-stream), `intent.ts` (Step1 HTTP+error-frame), `recommend.ts` (Step2 recommend HTTP+error-frame), `roundtable-jobs.ts` (job api/stream). The admin page's first filter is a **后台挂起 / 前台调用** toggle (frontend = `foreground,client`); the stage dropdown options adapt to it (background → step2+step3+job; frontend → step1+step2+step3) and each row shows the reporting user's name + 👤. Tests: `tests/test_roundtable_diagnostics.py`, `tests/test_roundtable_inprocess_timeout.py`. **Seat skill-recognition (弱模型补偿)**: a seat = one custom agent run whose configured skills (`agent config.yaml → skills`) are injected **only** via the system prompt's `<skill_system>` block (model must `read_file` the SKILL.md itself) — weak models (deepseek-chat 等) ignore it and never use their skills. `app/gateway/roundtable_seat_skills.py::append_seat_skill_directive(agent_id, task, *, chain_enabled=False)` appends an explicit Chinese directive (each configured-and-enabled skill's name + description + SKILL.md container path + "先 read_file 再按其流程取信息/执行") to the **seat's per-turn task (human turn)** — far stronger pull on weak models than the system prompt. Wired into **both** seat paths: foreground `multi_agent._special_run` (only `role in {seat, seat_parallel}`; Step3 singletons report/summary/dashboard/ingest excluded) + background `InProcessRoundtableGateway.run_seat`. No-op (returns task unchanged) when the seat has no configured skills. **2026-06-20: default-off + two opt-in switches (OR semantics)** — was global-on, now injects only when **either** the seat agent's own switch (`AgentConfig.seat_skill_directive`, edited in the agent edit dialog `agent-card.tsx` + `agent-detail-panel.tsx`) **or** the **per-seat** switch on that agent **within the selected business chain** is on. The per-seat chain switch lives **in the chain's `seats` JSON** (`ChainSeatSchema.seat_skill_directive`, no DB column — rides the existing JSON like the per-seat `model` override), edited via a small ✨ icon → dropdown (Switch + description) on each agent card in `BusinessChainEditorPage`. It is threaded **per-seat** to each seat run: foreground `multi_agent` per-call field `seat_skill_directive` (orchestration looks it up per agentId from a `{agentId: bool}` session map) → `_special_run` → `chain_enabled`; background as the job field `seat_skill_directives: dict[str,bool]` (StartJobRequest → StartParams → deps factory → `InProcessRoundtableGateway._seat_skill_directives`) → `run_seat` looks up `agent_id` → `chain_enabled`. `ROUNDTABLE_SEAT_SKILL_DIRECTIVE` is only a **master kill-switch** (default allow; set `0/false` to hard-disable regardless of the switches — never force-enables). Tests: `tests/test_roundtable_seat_skills.py`, `tests/test_agents_backend.py`. **2026-06-20 Step2 提速三件套**：(1) **每席位推理深度覆盖** —— 业务链条 `seats` JSON 里逐席位可配 `seat_mode`（flash/thinking/pro/ultra，覆盖链条级 `seatMode`；编辑器席位卡右侧 ⏱Gauge 图标下拉，null=跟随链条默认）。前台经 `useStep2Orchestration` 三处 dispatch 的 `seatModeToRequestFields(seatModes[id] ?? selectedMode)` 透传；后台经 job `seat_modes: dict[str,str]` → `StartParams` → `InProcessRoundtableGateway._seat_modes`，`run_seat` 用 `_SEAT_MODE_PARAMS`（与前端 `seatModeToRunParams` 逐字对齐）解析覆盖作业级 seat_*（未命中→跟随作业级→policy 默认）。无 DB 列（ride `seats` JSON，但 `ChainSeatSchema.seat_mode` 须显式声明否则 Pydantic 剥掉）。Tests: `tests/test_roundtable_seat_modes.py`。(2) **recommend 批内并行** —— 后端 `engine._run_recommend` 由 for-await 串行改 `asyncio.gather`（对齐前端 2026-06-13 已并行；总控一轮派多席位=判定相互独立→并发；批内用派活前 deliveries 快照、批后统一并入，语义不变）。(3) **研讨前并行『取数』两段式**（业务链条 `gather_first` 列，逐条 opt-in，migration `20260620_02`，editor「研讨前并行取数」开关）—— `gather_first=True` 时 `run_orchestration` 在分发前先 `engine._run_gather_phase`（全席位 `asyncio.gather` 跑 `_GATHER_TASK`「只取数不研讨」），产出 `gathered: dict` 经 `_seat_message(gathered=...)` 作为「已收集资料」块注入 recommend/chain/dag 各研讨轮席位任务（与「前序交付」块分开），把最慢的技能延迟前置并行、给（尤其 chain 串行的）研讨段提速；研讨阶段不限制再调技能。前台对称实现：`useStep2Orchestration.runGatherPhase` 在 `runActiveOrchestration` 分发前跑一次（`gatherPhaseDoneRef` 闸门，resetInitGate 复位），产出存 `gatheredDeliveriesRef` 经 `peerDeliveries`/初始 `priorDeliveries` 注入研讨席位。链条级 `gatherFirst`(camel)↔`gather_first`(snake) 经 `api/chains.ts` 映射；job 经 `StartJobRequest.gather_first`→`StartParams`→`run_orchestration(gather_first=…)`。Tests: `tests/test_roundtable_gather_phase.py`。(4) **削减重复广播**（前端 only，多智能体 foreground 路径）—— `suppress_pre_broadcast`（仅 leader：抑制「总控本轮输入→各席位 thread」预广播 + 「总控给X派活」→兄弟席位的 post 广播，二者一个 flag）此前只在 DAG 派活轮开；现扩到 recommend（`safetyCounter > 1`，即第 2 轮起的续轮/综合）+ chain 逐席位（`i > 0`，第 2 个席位起）+ chain 收口轮（恒 true）。**保留每种模式的首轮预广播**（它为席位打底任务背景），只 suppress 重复的续轮广播（N 次读写巨大席位 thread 的纯浪费 + 弱模型模仿派活）；席位的前序交付另经「席位交付广播」`broadcastSeatDeliveryToOthers` 送达，不受影响。后台引擎路径无此广播（`_seat_message` 直接拼上下文），故 #4 仅前端。**2026-06-20 深链接业务链抽取规范**：深链接会商（taskId+goPath，businessCode 如 rwfx→6BF）下，结构化抽取(flow-json)与方案总结报告都按所选业务链**完整层级**强制校正。结构化抽取实为 `roundtable-structure` 智能体（`flow_extract.py` 的 `_system_prompt()=load_agent_soul(STRUCTURE_AGENT_ID)`，抽取→校验两段式 `model.astream`）；其 SOUL 全局一份，旧实现把某业务层级写死进 SOUL→所有业务都按「目的→行为体」抽（bug）。修法：**单份领域知识、两角色用法**的 `deerflow.agents.roundtable_orchestrator.extraction_spec`（放 harness 层，app+harness 都可 import；`build_structure_directive`/`build_summary_directive` 按 businessCode 取；rwfx=6BF 写实「任务→目的→行为体→预期行为→驱动因素→叙事，预期行为↔驱动因素 1:1、余 1:N、label 不含步骤/阶段名、优先报告回退席位」，其余业务通用规范=强约束完整性+命名+取数且明令不套固定模板；business_code 空→""零副作用）。注入：(1) 结构化抽取 `flow_extract._hierarchy_hint`+`_validate_messages` 用 `build_structure_directive` 权威覆盖默认层级（前端 `streamFlowExtract` 传 `business_code`，源 `ingestBusinessCode`）；(2) 方案总结报告 `build_summary_directive`——前台 `multi_agent._special_run`(role==summary 且 business_code) 追加、后台 `engine.build_summary_prompt(business_code=…)`（StartParams/StartJobRequest 透传）。**仅深链接生效**。种子 `roundtable-structure/SOUL.md` 层级顺序改 spec-driven、连接示例去 rwfx 写死（存量 live SOUL 不被 seed 覆盖，但注入规范权威可纠正）。Tests: `tests/test_roundtable_extraction_spec.py`。 |
 | **Roundtable Draft Shares** (`/api/roundtable-draft-shares`) | Per-draft sharing for the roundtable-planning page, mirroring Thread Shares but the shared resource is a `roundtable_drafts` row. Two modes (`roundtable_draft_shares.mode`). **`import`**: `POST /` `{draft_id, mode}` - create (or reuse) a revocable code; `GET /{code}/preview` + `POST /{code}/import` - copy that draft (step1/2/3) into the caller's account under a deterministic per-(code,recipient) id (`imp_<sha1>` ? re-import is a no-op). **`view`**: a public, read-only result page ? see Public Roundtable Shares. Both: `GET /` - list my codes; `DELETE /{code}` - revoke. Backed by `deerflow.persistence.roundtable_draft_shares` (table `roundtable_draft_shares`, migration `20260606_01`); store wired in `deps.py` as `app.state.roundtable_draft_share_store`. Tests: `tests/test_roundtable_draft_shares.py`. |
 | **Public Roundtable Shares** (`/api/public/roundtable-shares`) | **Unauthenticated** (`/api/public/` whitelisted). `GET /{code}` - read-only snapshot for a `view`-mode draft share: returns the Step 3??????report HTML (`step3.html`, a self-contained document) + title/summary/`has_report`. The frontend renders it in a sandboxed `iframe` (no `allow-scripts`) at `#/roundtable-share/<code>`. Missing/revoked/non-`view` codes all return 404. |
 | **Scheduled Tasks** (`/api/scheduled-tasks`) | Per-user cron-like agent tasks. `GET/POST /` - list/create owned tasks; `PATCH/DELETE /{id}` - update/delete (owner); `POST /{id}/{pause,resume,trigger}`; `GET /{id}/runs`, `GET /runs/{rid}`, run-file preview/edit/download; `GET /runs/{rid}/messages` - agent messages for the run's underlying agent run (used to inspect a stuck/running run; authorized via `_load_run`); `GET /runs/{rid}/html` - generated HTML document for an `html_page` run; `DELETE /runs/{rid}` - owner-scoped, with an admin fallback so admins can delete any user's run; publish/subscribe under `/{id}/{publish,unpublish,subscribe}` and `GET /published`. **Admin (cross-user):** `GET /admin/all` - every task; `PATCH/DELETE /admin/{id}` - edit/delete any task; `POST /admin/{id}/trigger`; `GET /admin/{id}/runs`. Admin routes require `system_role == "admin"`. |
diff --git a/offline-backend-20260512/backend/app/gateway/routers/roundtable_drafts.py b/offline-backend-20260512/backend/app/gateway/routers/roundtable_drafts.py
index 9ef5ccd5..75574acf 100644
--- a/offline-backend-20260512/backend/app/gateway/routers/roundtable_drafts.py
+++ b/offline-backend-20260512/backend/app/gateway/routers/roundtable_drafts.py
@@ -47,6 +47,8 @@ class DraftMetaResponse(BaseModel):
     task_id: str | None = None
     title: str = ""
     furthest_step: int = 1
+    # 本次会商所用业务链名（历史卡片展示）；无链条（自由/recommend 模式）时为 None。
+    chain_title: str | None = None
     created_at: datetime | str | None = None
     updated_at: datetime | str | None = None
     # 最近一次后台作业的状态（Phase 6 历史下拉「进行中」loading 用）；无作业时为 None。
diff --git a/offline-backend-20260512/backend/app/gateway/routers/roundtable_task_drafts.py b/offline-backend-20260512/backend/app/gateway/routers/roundtable_task_drafts.py
index b0696cdd..1bea7a18 100644
--- a/offline-backend-20260512/backend/app/gateway/routers/roundtable_task_drafts.py
+++ b/offline-backend-20260512/backend/app/gateway/routers/roundtable_task_drafts.py
@@ -42,6 +42,8 @@ class TaskDraftMetaResponse(BaseModel):
     task_id: str
     title: str = ""
     furthest_step: int = 1
+    # 本次会商所用业务链名（历史卡片展示）；无链条（自由/recommend 模式）时为 None。
+    chain_title: str | None = None
     created_at: datetime | str | None = None
     updated_at: datetime | str | None = None
     # Latest background-job status for this draft (history-dropdown loading badge).
diff --git a/offline-backend-20260512/backend/packages/harness/deerflow/persistence/roundtable_diagnostics/__init__.py b/offline-backend-20260512/backend/packages/harness/deerflow/persistence/roundtable_diagnostics/__init__.py
index 2afba98c..99204114 100644
--- a/offline-backend-20260512/backend/packages/harness/deerflow/persistence/roundtable_diagnostics/__init__.py
+++ b/offline-backend-20260512/backend/packages/harness/deerflow/persistence/roundtable_diagnostics/__init__.py
@@ -41,10 +41,17 @@ def get_default_store() -> RoundtableDiagnosticRepository | None:
 
 
 async def record_diagnostic(store: RoundtableDiagnosticRepository | None = None, **fields: Any) -> None:
-    """Best-effort 写一条诊断事件。``store`` 缺省用进程级默认仓储。任何异常静默吞掉。"""
+    """Best-effort 写一条诊断事件。``store`` 缺省用进程级默认仓储。任何异常静默吞掉。
+
+    会商诊断**只收集报错日志**：``info`` 级是纯运行进度（begin/consensus/gather/job_started
+    等「运行日志」），对排障没价值、还会淹没真正的报错，直接不落库；只保留 ``warning`` /
+    ``error``（席位失败、空交付、超时、模型容错、产物缺失、崩溃等）。
+    """
     repo = store or _default_store
     if repo is None:
         return
+    if str(fields.get("level") or "").strip().lower() == "info":
+        return
     try:
         await repo.record(**fields)
     except Exception:  # noqa: BLE001 — 诊断日志写失败绝不打断会商
diff --git a/offline-backend-20260512/backend/packages/harness/deerflow/persistence/roundtable_drafts/sql.py b/offline-backend-20260512/backend/packages/harness/deerflow/persistence/roundtable_drafts/sql.py
index aeb74b88..2ca3ddcb 100644
--- a/offline-backend-20260512/backend/packages/harness/deerflow/persistence/roundtable_drafts/sql.py
+++ b/offline-backend-20260512/backend/packages/harness/deerflow/persistence/roundtable_drafts/sql.py
@@ -11,6 +11,7 @@ All write/read paths are scoped by ``user_id`` for ownership isolation.
 from __future__ import annotations
 
 import json
+import re
 from datetime import UTC, datetime
 from typing import Any
 
@@ -24,6 +25,29 @@ from deerflow.persistence.roundtable_drafts.model import (
 
 _UNSET = object()
 
+# 业务链名抽取：在已加载进内存的 step2 JSON 串里直接找 `"chain":{...,"title":"X"}`，
+# 不整体 json.loads（step2 是 LONGTEXT 大块），只做一次廉价正则扫描。`"chain":null`
+# （recommend/自由模式）不匹配 → 返回 None。`[^{}]*?` 限定在 chain 对象内（{id,title}
+# 无嵌套花括号）。
+_CHAIN_TITLE_RE = re.compile(r'"chain"\s*:\s*\{[^{}]*?"title"\s*:\s*"((?:\\.|[^"\\])*)"')
+
+
+def _extract_chain_title(step2_raw: Any) -> str | None:
+    """从 step2 原始 JSON 串里抽取所用业务链名（无链条 → None，前端显示「自由模式」）。"""
+    if not isinstance(step2_raw, str) or not step2_raw:
+        return None
+    m = _CHAIN_TITLE_RE.search(step2_raw)
+    if not m:
+        return None
+    title = m.group(1)
+    try:
+        # 还原 JSON 字符串转义（\" \\ \n \uXXXX 等）。
+        title = json.loads(f'"{title}"')
+    except Exception:
+        pass
+    title = title.strip()
+    return title or None
+
 
 def _loads(value: Any) -> Any:
     if not isinstance(value, str):
@@ -56,6 +80,8 @@ class RoundtableDraftRepository:
             "task_id": row.task_id,
             "title": row.title,
             "furthest_step": row.furthest_step,
+            # 所用业务链名（历史卡片展示用）；无链条 → None → 前端显示「自由模式」。
+            "chain_title": _extract_chain_title(row.step2),
             "created_at": row.created_at.isoformat() if isinstance(row.created_at, datetime) else row.created_at,
             "updated_at": row.updated_at.isoformat() if isinstance(row.updated_at, datetime) else row.updated_at,
         }
diff --git a/offline-backend-20260512/backend/packages/harness/deerflow/persistence/roundtable_task_drafts/sql.py b/offline-backend-20260512/backend/packages/harness/deerflow/persistence/roundtable_task_drafts/sql.py
index c415f62d..fad0ef0a 100644
--- a/offline-backend-20260512/backend/packages/harness/deerflow/persistence/roundtable_task_drafts/sql.py
+++ b/offline-backend-20260512/backend/packages/harness/deerflow/persistence/roundtable_task_drafts/sql.py
@@ -15,6 +15,7 @@ in ``PortableLongText`` and round-tripped with ``json.dumps`` / ``json.loads``.
 from __future__ import annotations
 
 import json
+import re
 from datetime import UTC, datetime
 from typing import Any
 
@@ -25,6 +26,27 @@ from deerflow.persistence.roundtable_task_drafts.model import RoundtableTaskDraf
 
 _UNSET = object()
 
+# 业务链名抽取：在已加载进内存的 step2 JSON 串里直接找 `"chain":{...,"title":"X"}`，
+# 不整体 json.loads（step2 是 LONGTEXT 大块）。`"chain":null`（自由模式）不匹配 → None。
+# 与 roundtable_drafts/sql.py 保持一致。
+_CHAIN_TITLE_RE = re.compile(r'"chain"\s*:\s*\{[^{}]*?"title"\s*:\s*"((?:\\.|[^"\\])*)"')
+
+
+def _extract_chain_title(step2_raw: Any) -> str | None:
+    """从 step2 原始 JSON 串里抽取所用业务链名（无链条 → None，前端显示「自由模式」）。"""
+    if not isinstance(step2_raw, str) or not step2_raw:
+        return None
+    m = _CHAIN_TITLE_RE.search(step2_raw)
+    if not m:
+        return None
+    title = m.group(1)
+    try:
+        title = json.loads(f'"{title}"')
+    except Exception:
+        pass
+    title = title.strip()
+    return title or None
+
 
 def _loads(value: Any) -> Any:
     if not isinstance(value, str):
@@ -55,6 +77,8 @@ class RoundtableTaskDraftRepository:
             "task_id": row.task_id,
             "title": row.title,
             "furthest_step": row.furthest_step,
+            # 所用业务链名（历史卡片展示用）；无链条 → None → 前端显示「自由模式」。
+            "chain_title": _extract_chain_title(row.step2),
             "created_at": row.created_at.isoformat() if isinstance(row.created_at, datetime) else row.created_at,
             "updated_at": row.updated_at.isoformat() if isinstance(row.updated_at, datetime) else row.updated_at,
         }

From 23743ee1930a329a5e46d3338fe57650ca32f113 Mon Sep 17 00:00:00 2001
From: fengxiangdebaozi <10845145+fengxiangdebaozi@user.noreply.gitee.com>
Date: Mon, 22 Jun 2026 15:52:35 +0800
Subject: [PATCH 08/17] =?UTF-8?q?chore:=20=E7=B2=BE=E7=AE=80=20AI=20?=
 =?UTF-8?q?=E5=86=99=E4=BD=9C=E7=A4=BA=E4=BE=8B=E6=96=87=E6=A1=88?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

ai-writing 表单/检索/欢迎页的占位与示例改为通用文案。

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 .../src/open-canvas/components/ai-writing/AIWritingForm.tsx     | 2 +-
 .../src/open-canvas/components/ai-writing/MaterialsToolbar.tsx  | 2 +-
 .../src/open-canvas/components/ai-writing/WritingWelcome.tsx    | 2 +-
 3 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/frontend-web/src/open-canvas/components/ai-writing/AIWritingForm.tsx b/frontend-web/src/open-canvas/components/ai-writing/AIWritingForm.tsx
index aa6a7d60..c89efb9f 100644
--- a/frontend-web/src/open-canvas/components/ai-writing/AIWritingForm.tsx
+++ b/frontend-web/src/open-canvas/components/ai-writing/AIWritingForm.tsx
@@ -192,7 +192,7 @@ export function AIWritingForm({
             </span>
           </div>
           <Textarea
-            placeholder="描述你想写的文章，例如：写一篇分析今年 AI 大模型发展趋势的文章，重点关注国内企业的布局..."
+            placeholder="描述你想写的文章"
             value={intent}
             onChange={(e) => setIntent(e.target.value)}
             disabled={disabled}
diff --git a/frontend-web/src/open-canvas/components/ai-writing/MaterialsToolbar.tsx b/frontend-web/src/open-canvas/components/ai-writing/MaterialsToolbar.tsx
index 9cdb1418..f6ad5b8e 100644
--- a/frontend-web/src/open-canvas/components/ai-writing/MaterialsToolbar.tsx
+++ b/frontend-web/src/open-canvas/components/ai-writing/MaterialsToolbar.tsx
@@ -73,7 +73,7 @@ export function MaterialsToolbar({
           value={userQuery}
           onChange={handleInput}
           onKeyDown={handleKeyDown}
-          placeholder={placeholder ?? '例如：检索新能源汽车销量数据。按 Enter 检索并追加…'}
+          placeholder={placeholder ?? '按 Enter 检索并追加…'}
           disabled={disabled}
           className="w-full resize-none rounded-md border border-input bg-background px-2.5 py-1.5 text-xs leading-5 placeholder:text-muted-foreground focus-visible:outline-none focus-visible:ring-1 focus-visible:ring-ring disabled:opacity-50 min-h-[28px] max-h-24"
         />
diff --git a/frontend-web/src/open-canvas/components/ai-writing/WritingWelcome.tsx b/frontend-web/src/open-canvas/components/ai-writing/WritingWelcome.tsx
index 6c0cd0e5..5ab0ca03 100644
--- a/frontend-web/src/open-canvas/components/ai-writing/WritingWelcome.tsx
+++ b/frontend-web/src/open-canvas/components/ai-writing/WritingWelcome.tsx
@@ -44,7 +44,7 @@ export function WritingWelcome() {
             <p className="mt-2 max-w-xl text-sm leading-relaxed text-slate-500 dark:text-white/60">
               这是一个专业的自动化写作系统。你只需在下方对话框中
               <strong className="font-semibold text-slate-700 dark:text-white/80">
-                输入你的本意（例如：「帮我写一篇 1500 字的新能源汽车报告」）
+                输入你的本意（例如：「帮我写一篇 1500 字的周报」）
               </strong>
               ，系统即会自动为你在右侧填充所需的字数、体裁等格式进行对齐；你也可在右侧进行纯人工的参数深度定制，并一键开启 AI 深度协作创作。
             </p>

From b59cbd51d9087fcdf11f8c593e54cfe3f5dceb89 Mon Sep 17 00:00:00 2001
From: fengxiangdebaozi <10845145+fengxiangdebaozi@user.noreply.gitee.com>
Date: Mon, 22 Jun 2026 16:36:33 +0800
Subject: [PATCH 09/17] =?UTF-8?q?feat(ai-writing):=20=E5=90=84=E9=98=B6?=
 =?UTF-8?q?=E6=AE=B5=E6=A8=A1=E5=9E=8B=E8=B0=83=E7=94=A8=E8=87=AA=E5=8A=A8?=
 =?UTF-8?q?=E5=AE=B9=E9=94=99=EF=BC=8C=E5=A4=B1=E8=B4=A5=E6=97=B6=E6=8D=A2?=
 =?UTF-8?q?=E4=B8=8B=E4=B8=80=E4=B8=AA=E6=A8=A1=E5=9E=8B=E7=BB=A7=E7=BB=AD?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

参考多智能体会商的容错思路：意图解析/大纲/正文/审核四个 LLM
环节失败时，按 config.yaml models 顺序换下一个模型重跑该环节，
并通过 model_switch 事件在前端时间线提示「『X』模型<原因>，
已自动切换到『Y』继续」。正文逐章节、审核逐维度独立切换，
已生成的章节/维度保留；意图解析成功的模型还会写回 state
供下游阶段沿用，避免每阶段重复经历同一次失败。

- nodes/_model_fallback.py: 候选模型链 + 异常分类 + model_switch
  事件构造（沿用 ROUNDTABLE_MODEL_FALLBACK[_MAX] env 开关）
- intent_parser/writer_outline/writer_draft/editor: 失败循环换模型
- 前端 types/reducer/timeline: 新增 model_switch 事件类型，按
  stage 复位对应流式缓冲并以 amber 系统提示渲染

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 .../ai-writing/AIWritingTimeline.tsx          |   1 +
 .../components/ai-writing/TimelineStep.tsx    |   8 +-
 .../open-canvas/hooks/useAIWritingStream.ts   |  68 ++++++
 .../src/open-canvas/types/ai-writing.ts       |  11 +
 .../ai_writing/nodes/_model_fallback.py       | 197 ++++++++++++++++++
 .../agents/ai_writing/nodes/editor.py         |  72 +++++--
 .../agents/ai_writing/nodes/intent_parser.py  | 112 ++++++----
 .../agents/ai_writing/nodes/writer_draft.py   | 149 ++++++++-----
 .../agents/ai_writing/nodes/writer_outline.py | 124 +++++++----
 9 files changed, 598 insertions(+), 144 deletions(-)
 create mode 100644 offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/_model_fallback.py

diff --git a/frontend-web/src/open-canvas/components/ai-writing/AIWritingTimeline.tsx b/frontend-web/src/open-canvas/components/ai-writing/AIWritingTimeline.tsx
index eedf8a69..c86a652d 100644
--- a/frontend-web/src/open-canvas/components/ai-writing/AIWritingTimeline.tsx
+++ b/frontend-web/src/open-canvas/components/ai-writing/AIWritingTimeline.tsx
@@ -19,6 +19,7 @@ const DISPLAYED_TYPES = new Set([
   'await_user',
   'resumed',
   'notice',
+  'model_switch',
   'done',
   'error',
 ]);
diff --git a/frontend-web/src/open-canvas/components/ai-writing/TimelineStep.tsx b/frontend-web/src/open-canvas/components/ai-writing/TimelineStep.tsx
index b1289ca6..f5ed8327 100644
--- a/frontend-web/src/open-canvas/components/ai-writing/TimelineStep.tsx
+++ b/frontend-web/src/open-canvas/components/ai-writing/TimelineStep.tsx
@@ -13,6 +13,7 @@ interface TimelineStepProps {
 function statusOf(event: ProgressEvent, isLast: boolean): StepStatus {
   if (event.type === 'error') return 'error';
   if (event.type === 'notice') return 'warning';
+  if (event.type === 'model_switch') return 'warning';
   if (event.type === 'await_user') return 'paused';
   // step_start 只有在列表末尾（后续事件还未到达）时才显示转圈，否则已有后续事件说明该步骤已完成
   if (event.type === 'step_start') return isLast ? 'running' : 'done';
@@ -30,8 +31,11 @@ const AGENT_COLOR: Record<string, string> = {
 
 export function TimelineStep({ event, isLast, children }: TimelineStepProps) {
   const status = statusOf(event, isLast ?? true);
-  // notice=非致命告警（如 ES 波动、模型兜底），error=致命错误，两者都展示「标题 + 真实原因」
-  const isAlert = event.type === 'error' || event.type === 'notice';
+  // notice=非致命告警（如 ES 波动、模型兜底），error=致命错误，model_switch=模型自动容错切换（amber 同 notice）
+  const isAlert =
+    event.type === 'error' ||
+    event.type === 'notice' ||
+    event.type === 'model_switch';
   const isError = event.type === 'error';
 
   return (
diff --git a/frontend-web/src/open-canvas/hooks/useAIWritingStream.ts b/frontend-web/src/open-canvas/hooks/useAIWritingStream.ts
index 991a6941..04607a21 100644
--- a/frontend-web/src/open-canvas/hooks/useAIWritingStream.ts
+++ b/frontend-web/src/open-canvas/hooks/useAIWritingStream.ts
@@ -388,6 +388,37 @@ function handleEvent(prev: AIWritingState, event: ProgressEvent): AIWritingState
     case 'notice':
       return base;
 
+    case 'model_switch': {
+      // 模型自动容错：丢弃当前阶段已累积的流式缓冲，避免上一个失败模型的半截输出
+      // 与下一个模型的输出粘在一起；时间线展示侧由 TimelineStep 按 amber 渲染（同 notice）。
+      const stage = event.modelSwitch?.stage;
+      const sectionIndex = event.modelSwitch?.sectionIndex;
+      const resetByStage: Partial<AIWritingState> = (() => {
+        switch (stage) {
+          case 'intent':
+            return { intentStreamingText: '', intentThinkingText: '' };
+          case 'outline':
+            return { outlineStreamingText: '', outlineThinkingText: '' };
+          case 'draft': {
+            // 只清掉失败那一节的流式片段，已写好的章节保持不动
+            if (typeof sectionIndex === 'number') {
+              return {
+                streamingDraftSections: prev.streamingDraftSections.filter(
+                  (s) => s.sectionIndex !== sectionIndex,
+                ),
+              };
+            }
+            return { streamingDraftSections: [], streamingDraftTitle: '' };
+          }
+          case 'review':
+            return { streamingReviewItems: [] };
+          default:
+            return {};
+        }
+      })();
+      return { ...base, ...resetByStage };
+    }
+
     default:
       return base;
   }
@@ -506,6 +537,22 @@ export function normalizeProgressDict(parsed: Record<string, unknown>, sessionId
     out.reviewItemIssues = (out.issues as Record<string, unknown>[]).map(normalizeReviewIssue);
   }
 
+  // model_switch（updates 通道里的 progress_events 也可能带）：折成 modelSwitch 子对象
+  if (out.type === 'model_switch' && !out.modelSwitch) {
+    out.modelSwitch = {
+      stage: (out.stage as 'intent' | 'outline' | 'draft' | 'review') ?? 'intent',
+      failedModel: String(out.failed_model ?? out.failedModel ?? ''),
+      nextModel: String(out.next_model ?? out.nextModel ?? ''),
+      reason: String(out.reason ?? ''),
+      sectionIndex: typeof out.section_index === 'number'
+        ? (out.section_index as number)
+        : (typeof out.sectionIndex === 'number' ? (out.sectionIndex as number) : undefined),
+      sectionTitle: typeof out.section_title === 'string'
+        ? (out.section_title as string)
+        : (typeof out.sectionTitle === 'string' ? (out.sectionTitle as string) : undefined),
+    };
+  }
+
   if (!out.message) out.message = '';
   return out as unknown as ProgressEvent;
 }
@@ -522,6 +569,7 @@ export function customToProgressEvent(raw: unknown, sessionId: string): Progress
     'research_progress', 'outline_chunk', 'draft_chunk',
     'review_chunk', 'review_item_done',
     'skill_search_step', 'skill_select_chunk',
+    'model_switch',
   ];
   if (!allowed.includes(type as ProgressEventType)) return null;
 
@@ -536,10 +584,30 @@ export function customToProgressEvent(raw: unknown, sessionId: string): Progress
   if (Array.isArray(merged.issues)) {
     merged.reviewItemIssues = (merged.issues as Record<string, unknown>[]).map(normalizeReviewIssue);
   }
+  if ('agent_name' in merged && !merged.agentName) merged.agentName = merged.agent_name;
   // skill_search_step：把字段折成 skillStep 子对象，方便 reducer / UI 取用
   if (type === 'skill_search_step') {
     merged.skillStep = normalizeSkillStep(o);
   }
+  // model_switch：把 snake_case 字段折成 modelSwitch 子对象（reducer + TimelineStep 取用）
+  if (type === 'model_switch') {
+    merged.modelSwitch = {
+      stage: (o.stage as 'intent' | 'outline' | 'draft' | 'review') ?? 'intent',
+      failedModel: String(o.failed_model ?? (o as Record<string, unknown>).failedModel ?? ''),
+      nextModel: String(o.next_model ?? (o as Record<string, unknown>).nextModel ?? ''),
+      reason: String(o.reason ?? ''),
+      sectionIndex: typeof o.section_index === 'number'
+        ? (o.section_index as number)
+        : (typeof (o as Record<string, unknown>).sectionIndex === 'number'
+          ? ((o as Record<string, unknown>).sectionIndex as number)
+          : undefined),
+      sectionTitle: typeof o.section_title === 'string'
+        ? (o.section_title as string)
+        : (typeof (o as Record<string, unknown>).sectionTitle === 'string'
+          ? ((o as Record<string, unknown>).sectionTitle as string)
+          : undefined),
+    };
+  }
   return merged as unknown as ProgressEvent;
 }
 
diff --git a/frontend-web/src/open-canvas/types/ai-writing.ts b/frontend-web/src/open-canvas/types/ai-writing.ts
index 361cb090..e3f2f58d 100644
--- a/frontend-web/src/open-canvas/types/ai-writing.ts
+++ b/frontend-web/src/open-canvas/types/ai-writing.ts
@@ -172,6 +172,7 @@ export type ProgressEventType =
   | "await_user"
   | "resumed"
   | "notice"
+  | "model_switch"
   | "skill_search_step"
   | "skill_select_chunk"
   | "done"
@@ -215,6 +216,16 @@ export interface ProgressEvent {
   blockedSections?: BlockedSection[];
   // skill_search_step：技能检索步骤事件（list_skills / select_skills / invoke_skill）
   skillStep?: SkillSearchStep;
+  // model_switch：模型自动容错切换（某模型挂掉，换下一个模型继续）
+  // stage ∈ intent / outline / draft / review，前端 reducer 据此清掉对应阶段的流式缓冲。
+  modelSwitch?: {
+    stage: "intent" | "outline" | "draft" | "review";
+    failedModel: string;
+    nextModel: string;
+    reason: string;
+    sectionIndex?: number;
+    sectionTitle?: string;
+  };
 }
 
 // ── 用户干预提交体 ────────────────────────────────────────────────────────────
diff --git a/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/_model_fallback.py b/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/_model_fallback.py
new file mode 100644
index 00000000..17fb5a8e
--- /dev/null
+++ b/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/_model_fallback.py
@@ -0,0 +1,197 @@
+"""AI 写作模型自动容错（按 ``config.yaml models[]`` 顺序换下一个模型继续）。
+
+与多智能体会商 ``app/gateway/roundtable_model_fallback.py`` 同思路，但**只放在 harness 层**
+避免 ``app→harness`` 单向依赖被反向打破：
+
+  - ``fallback_chain(current)``：候选模型链 ``[当前, …其余]``（按 config.yaml 顺序，去重保序）；
+  - ``classify_exception(exc)``：把异常归到一个用户可读的失败原因码；
+  - ``reason_text(reason)``：原因码 → 简短中文（用于「『X』模型 <原因>，已切换『Y』继续」）；
+  - ``emit_model_switch_event(writer, ...)``：把 ``model_switch`` 推到 LangGraph custom 通道
+    供前端时间线展示；
+  - ``build_switch_progress_event(...)``：节点 return 的 ``progress_events`` 里同样塞一份
+    （走 updates 通道，前端 reducer 据此刷新缓冲并落到时间线）。
+
+总开关 / 上限沿用会商的 env：
+  - ``ROUNDTABLE_MODEL_FALLBACK``：默认开；``0/false/off/no`` 关闭（链只含当前模型）；
+  - ``ROUNDTABLE_MODEL_FALLBACK_MAX``：候选数上限（含当前），0/未设 = 不限。
+
+放在 ai_writing 节点目录下而非 harness/agents/middlewares：本模块不是中间件，
+且仅服务 ai_writing 的几个流式节点；如果未来圆桌也想下沉合并，再统一拎出去。
+"""
+from __future__ import annotations
+
+import logging
+import os
+from typing import Any, Callable
+
+logger = logging.getLogger(__name__)
+
+
+_REASON_TEXT: dict[str, str] = {
+    "quota": "额度不足/账单异常",
+    "auth": "鉴权失败/无访问权限",
+    "busy": "服务繁忙",
+    "transient": "网络/连接异常",
+    "upstream": "上游响应异常",
+    "config": "模型未配置或配置异常",
+    "exception": "调用异常",
+    "generic": "调用失败",
+}
+
+
+def reason_text(reason: str | None) -> str:
+    """原因码 → 简短中文（未知码 → 「调用失败」）。"""
+    if not reason:
+        return "调用失败"
+    return _REASON_TEXT.get(reason, "调用失败")
+
+
+_QUOTA_PATTERNS = ("insufficient_quota", "quota", "billing", "credit", "payment",
+                   "余额不足", "超出限额", "额度不足", "欠费")
+_AUTH_PATTERNS = ("authentication", "unauthorized", "invalid api key", "invalid_api_key",
+                  "permission", "forbidden", "access denied", "无权", "未授权")
+_BUSY_PATTERNS = ("server busy", "temporarily unavailable", "try again later", "please retry",
+                  "please try again", "overloaded", "high demand", "rate limit",
+                  "负载较高", "服务繁忙", "稍后重试", "请稍后重试")
+_UPSTREAM_PATTERNS = ("bad_response_status_code", "bad response status",
+                      "expecting ',' delimiter", "expecting value", "expecting property name",
+                      "unexpected character", "unterminated string", "jsondecodeerror")
+_TRANSIENT_EXC_NAMES = {"APITimeoutError", "APIConnectionError", "InternalServerError",
+                        "ReadError", "RemoteProtocolError", "TimeoutError"}
+
+
+def classify_exception(exc: BaseException) -> str:
+    """异常 → 失败原因码（quota/auth/busy/transient/upstream/config/generic/exception）。
+
+    与 ``LLMErrorHandlingMiddleware._classify_error`` 同口径，简化版：不区分是否可重试
+    （ai_writing 的容错只换模型，不在同模型上重试）。``create_chat_model`` 抛
+    ``ValueError("Model X not found in config")`` 这类配置缺失另归 ``config``。
+    """
+    name = exc.__class__.__name__
+    detail = str(exc).lower()
+
+    if isinstance(exc, ValueError) and "not found in config" in detail:
+        return "config"
+    if any(p in detail for p in _QUOTA_PATTERNS):
+        return "quota"
+    if any(p in detail for p in _AUTH_PATTERNS):
+        return "auth"
+    if name in _TRANSIENT_EXC_NAMES:
+        return "transient"
+    if any(p in detail for p in _BUSY_PATTERNS):
+        return "busy"
+    if any(p in detail for p in _UPSTREAM_PATTERNS):
+        return "upstream"
+    return "exception"
+
+
+def _fallback_enabled() -> bool:
+    raw = (os.getenv("ROUNDTABLE_MODEL_FALLBACK", "1") or "1").strip().lower()
+    return raw not in {"0", "false", "no", "off"}
+
+
+def _max_attempts() -> int:
+    try:
+        return max(0, int(os.getenv("ROUNDTABLE_MODEL_FALLBACK_MAX", "0") or "0"))
+    except (TypeError, ValueError):
+        return 0
+
+
+def _config_model_names() -> list[str]:
+    try:
+        from deerflow.config import get_app_config
+        cfg = get_app_config()
+        return [m.name for m in (cfg.models or []) if getattr(m, "name", None)]
+    except Exception:  # noqa: BLE001 — 拿不到配置（测试桩）时退化为空
+        logger.debug("ai_writing fallback_chain: get_app_config().models unavailable", exc_info=True)
+        return []
+
+
+def fallback_chain(current: str | None) -> list[str]:
+    """``[当前模型, …config.yaml 其余模型]``（去重保序）。
+
+    - ``current`` 为空 → 链头落到 ``models[0]``；
+    - ``current`` 在配置内 → 排第一，余者按配置顺序补齐；
+    - ``current`` 不在配置内 → 仍排第一（让 ``create_chat_model`` 在第一轮按原行为抛，
+      之后再换模型），再补齐全部配置模型；
+    - 总开关关闭 → 只返回头一个；
+    - 拿不到配置 → ``[current]``（current 为空则空列表）。
+    """
+    all_names = _config_model_names()
+    cur = (current or "").strip()
+    ordered: list[str] = []
+    if cur:
+        ordered.append(cur)
+    for n in all_names:
+        if n and n not in ordered:
+            ordered.append(n)
+    if not ordered:
+        return [cur] if cur else []
+    if not _fallback_enabled():
+        return ordered[:1]
+    cap = _max_attempts()
+    if cap > 0:
+        ordered = ordered[:cap]
+    return ordered
+
+
+def build_switch_progress_event(
+    *,
+    agent_name: str,
+    stage: str,
+    failed_model: str,
+    next_model: str,
+    reason: str,
+    section_index: int | None = None,
+    section_title: str | None = None,
+) -> dict:
+    """构造一条 ``model_switch`` ``progress_event`` 字典（供节点 return 时塞进 ``progress_events`` 数组）。
+
+    ``stage`` ∈ {intent, outline, draft, review}：前端 reducer 据此清掉对应阶段的流式缓冲，
+    避免半段输出被下一个模型的输出粘在一起。
+    """
+    fm = (failed_model or "").strip() or "当前模型"
+    nm = (next_model or "").strip() or "下一个模型"
+    rt = reason_text(reason)
+    detail = f"「{agent_name}」所用模型「{fm}」{rt}，已自动切换到「{nm}」继续。"
+    payload: dict[str, Any] = {
+        "type": "model_switch",
+        "agent_name": agent_name,
+        "title": "模型容错",
+        "detail": detail,
+        "message": detail,
+        "stage": stage,
+        "failed_model": fm,
+        "next_model": nm,
+        "reason": reason,
+    }
+    if section_index is not None:
+        payload["section_index"] = section_index
+    if section_title:
+        payload["section_title"] = section_title
+    return payload
+
+
+def emit_model_switch_event(
+    writer: Callable[[dict], None] | None,
+    *,
+    agent_name: str,
+    stage: str,
+    failed_model: str,
+    next_model: str,
+    reason: str,
+    section_index: int | None = None,
+    section_title: str | None = None,
+) -> None:
+    """把 ``model_switch`` 事件推到 LangGraph custom 通道（前端 useStream onCustomEvent 收）。"""
+    if writer is None:
+        return
+    payload = build_switch_progress_event(
+        agent_name=agent_name, stage=stage,
+        failed_model=failed_model, next_model=next_model, reason=reason,
+        section_index=section_index, section_title=section_title,
+    )
+    try:
+        writer(payload)
+    except Exception:  # noqa: BLE001 — writer 在无 runtime 时是 no-op，理论不抛
+        logger.debug("emit_model_switch_event: writer raised", exc_info=True)
diff --git a/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/editor.py b/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/editor.py
index abafe383..191cc602 100644
--- a/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/editor.py
+++ b/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/editor.py
@@ -21,7 +21,13 @@ from ._agent_runner import (
     run_writing_agent,
     use_builtin_agent,
 )
-from ._utils import get_ai_writing_config, get_writing_model, parse_json, _content_to_text, reasoning_delta, ThinkTagParser, safe_get_stream_writer
+from ._model_fallback import (
+    build_switch_progress_event,
+    classify_exception,
+    emit_model_switch_event,
+    fallback_chain,
+)
+from ._utils import get_ai_writing_config, parse_json, resolve_writing_model_name, _content_to_text, reasoning_delta, ThinkTagParser, safe_get_stream_writer
 
 logger = logging.getLogger(__name__)
 
@@ -154,22 +160,17 @@ async def editor_node(state: AIWritingState, config: RunnableConfig) -> dict:
                     "title": "模型错误", "detail": str(e), "message": "审核失败",
                 }],
             }
+    # 旧路径下模型由 fallback_chain 在每维度审查前按候选顺序解析；模型挂掉自动换下一个继续。
+    editor_chain: list[str] = []
+    current_model_name: str | None = None
     if not use_agent:
-        # 模型创建守卫：模型名配错时 create_chat_model 抛 ValueError，
-        # 不拦截会硬崩溃整个图；此处转成可恢复的 status:error。
-        try:
-            model = get_writing_model(config, state)
-        except Exception as e:
-            logger.error("editor model creation failed: %s", e, exc_info=True)
-            return {
-                "status": "error",
-                "progress_events": [{
-                    "type": "error", "agent_name": "编辑",
-                    "title": "模型错误", "detail": str(e), "message": "审核失败",
-                }],
-            }
+        initial_name, _ud = resolve_writing_model_name(config, state)
+        editor_chain = fallback_chain(initial_name)
+        current_model_name = editor_chain[0] if editor_chain else None
+        from deerflow.models import create_chat_model
     results: dict[str, dict] = {}
     review_errors: list[str] = []  # 分项审查的模型异常，供完成后非致命提示
+    switch_notices: list[dict] = []  # 模型容错切换记录，最终塞进 progress_events 走时间线
 
     _CATEGORY_LABEL = {"fact": "事实性", "logic": "逻辑性", "language": "语言文字"}
 
@@ -192,7 +193,47 @@ async def editor_node(state: AIWritingState, config: RunnableConfig) -> dict:
                 if res.get("_llm_error"):
                     review_errors.append(f"「{_CATEGORY_LABEL.get(category, category)}审查」{res['_llm_error']}")
             else:
-                full_text = await _stream_one_review(model, user_prompt, category, writer)
+                # 旧路径：按 fallback_chain 试候选模型，失败就换下一个继续本维度。
+                # 与 writer_draft 同思路：以「上一维度成功使用的模型」为头部。
+                dim_chain: list[str] = []
+                if current_model_name:
+                    dim_chain.append(current_model_name)
+                for n in editor_chain:
+                    if n and n not in dim_chain:
+                        dim_chain.append(n)
+                if not dim_chain:
+                    dim_chain = [None]  # type: ignore[list-item]
+                full_text = ""
+                dim_succeeded = False
+                last_dim_exc: BaseException | None = None
+                for d_idx, candidate in enumerate(dim_chain):
+                    try:
+                        m = create_chat_model(name=candidate) if candidate else create_chat_model()
+                        full_text = await _stream_one_review(m, user_prompt, category, writer)
+                        dim_succeeded = True
+                        current_model_name = candidate or current_model_name
+                        break
+                    except Exception as e:
+                        last_dim_exc = e
+                        reason = classify_exception(e)
+                        logger.error(
+                            "editor review [%s] failed on model=%s (reason=%s): %s",
+                            category, candidate, reason, e, exc_info=True,
+                        )
+                        next_candidate = dim_chain[d_idx + 1] if d_idx + 1 < len(dim_chain) else None
+                        if next_candidate:
+                            emit_model_switch_event(
+                                writer, agent_name=agent_label, stage="review",
+                                failed_model=candidate or "", next_model=next_candidate, reason=reason,
+                            )
+                            switch_notices.append(build_switch_progress_event(
+                                agent_name=agent_label, stage="review",
+                                failed_model=candidate or "", next_model=next_candidate, reason=reason,
+                            ))
+                            continue
+                if not dim_succeeded:
+                    assert last_dim_exc is not None
+                    raise last_dim_exc
                 res = parse_json(full_text, fallback)
         except Exception as e:
             logger.warning("editor review [%s] failed: %s", category, e)
@@ -252,6 +293,7 @@ async def editor_node(state: AIWritingState, config: RunnableConfig) -> dict:
         "review_result": review,
         "progress_events": [
             {"type": "step_start", "agent_name": agent_label, "message": "开始审核"},
+            *switch_notices,
             *notices,
             {
                 "type": "review_ready",
diff --git a/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/intent_parser.py b/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/intent_parser.py
index ac94bc91..3e66f7b0 100644
--- a/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/intent_parser.py
+++ b/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/intent_parser.py
@@ -9,10 +9,15 @@ from langchain_core.runnables import RunnableConfig
 from langchain_core.messages import HumanMessage, SystemMessage
 
 from ..state import AIWritingState
+from ._model_fallback import (
+    build_switch_progress_event,
+    classify_exception,
+    emit_model_switch_event,
+    fallback_chain,
+)
 from ._utils import (
     ThinkTagParser,
     _content_to_text,
-    get_model,
     parse_json,
     reasoning_delta,
     safe_get_stream_writer,
@@ -207,47 +212,74 @@ async def intent_parser_node(state: AIWritingState, config: RunnableConfig) -> d
     # 流式调用：与 writer_outline / writer_draft 同套机制 —— LLM 逐字吐出 JSON，
     # 通过 ``intent_chunk`` 推给前端打字机；完成后 ``parse_json`` 解析结构化结果，
     # 再通过 ``intent_ready`` 推一份结构化字段给前端结果卡片。
+    #
+    # 模型自动容错：失败时按 fallback_chain 顺序换下一个模型重试本环节。每次切换
+    # 都通过 ``model_switch`` custom 事件 + ``progress_events`` 同步告诉前端，前端
+    # 清掉本环节已累积的 intent_chunk 缓冲，避免半段输出和下个模型的输出拼接错乱。
     writer = safe_get_stream_writer()
-    llm_error: str | None = None
+    initial_model = (configurable.get("model_name") or "").strip() or None
+    chain = fallback_chain(initial_model)
+    progress_events: list[dict] = []
+    from deerflow.models import create_chat_model
+
     full_text = ""
-    parser = ThinkTagParser()
-    try:
-        model = get_model(config)
-        async for chunk in model.astream(
-            [SystemMessage(content=system), HumanMessage(content=user)]
-        ):
-            # 推理模型的思考走 reasoning_content（content 为空）：作为思考实时推送
-            reasoning = reasoning_delta(chunk)
-            if reasoning:
-                writer({"type": "intent_chunk", "text": reasoning, "is_thinking": True})
-            delta = _content_to_text(chunk.content)
-            if not delta:
+    success_model: str | None = None
+    last_error: BaseException | None = None
+    for idx, candidate in enumerate(chain or [None]):
+        full_text = ""  # 每次重试都从头累积，避免不同模型的输出粘在一起
+        parser = ThinkTagParser()
+        try:
+            model = create_chat_model(name=candidate) if candidate else create_chat_model()
+            async for chunk in model.astream(
+                [SystemMessage(content=system), HumanMessage(content=user)]
+            ):
+                # 推理模型的思考走 reasoning_content（content 为空）：作为思考实时推送
+                reasoning = reasoning_delta(chunk)
+                if reasoning:
+                    writer({"type": "intent_chunk", "text": reasoning, "is_thinking": True})
+                delta = _content_to_text(chunk.content)
+                if not delta:
+                    continue
+                for is_thinking, text in parser.feed(delta):
+                    if not is_thinking:
+                        full_text += text
+                    writer({
+                        "type": "intent_chunk",
+                        "text": text,
+                        "is_thinking": is_thinking,
+                    })
+            success_model = candidate or ""
+            break
+        except Exception as exc:
+            last_error = exc
+            reason = classify_exception(exc)
+            logger.error(
+                "intent_parser streaming failed on model=%s (reason=%s): %s",
+                candidate, reason, exc, exc_info=True,
+            )
+            next_candidate = chain[idx + 1] if idx + 1 < len(chain) else None
+            if next_candidate:
+                emit_model_switch_event(
+                    writer, agent_name="系统", stage="intent",
+                    failed_model=candidate or "", next_model=next_candidate, reason=reason,
+                )
+                progress_events.append(build_switch_progress_event(
+                    agent_name="系统", stage="intent",
+                    failed_model=candidate or "", next_model=next_candidate, reason=reason,
+                ))
                 continue
-            for is_thinking, text in parser.feed(delta):
-                if not is_thinking:
-                    full_text += text
-                writer({
-                    "type": "intent_chunk",
-                    "text": text,
-                    "is_thinking": is_thinking,
-                })
-    except Exception as exc:
-        # 模型 / 网络挂掉 —— 不阻断流程，用 fallback 继续。
-        logger.error("intent_parser streaming failed: %s", exc, exc_info=True)
-        llm_error = f"{type(exc).__name__}: {exc}"
+            # 链尾仍失败：保留 fallback 默认值，向用户报一条非致命 notice，流程继续。
+            progress_events.append({
+                "type": "notice",
+                "agent_name": "系统",
+                "title": "模型错误",
+                "detail": (
+                    f"意图解析失败，已按默认配置兜底。原因：{type(exc).__name__}: {exc}"
+                ),
+            })
 
     result = parse_json(full_text, fallback_result) if full_text else fallback_result
 
-    # 意图解析模型调用失败时以非致命 notice 提示用户（fallback 已兜底，流程仍可继续）
-    progress_events: list[dict] = []
-    if llm_error:
-        progress_events.append({
-            "type": "notice",
-            "agent_name": "系统",
-            "title": "模型错误",
-            "detail": f"意图解析失败，已按默认配置兜底。原因：{llm_error}",
-        })
-
     final_article_type, final_target_audience, final_word_count, final_refined_intent = _resolve_intent_fields(
         state, result, fallback_result, valid_keys, default_key
     )
@@ -267,9 +299,11 @@ async def intent_parser_node(state: AIWritingState, config: RunnableConfig) -> d
     # 捕获用户表单选定的主模型进 state（仅本节点能可靠拿到 configurable.model_name —
     # 它运行在首次 invocation；后续大纲/写作/审核在 resume 后跑，config 已不带它）。
     # 大纲/写作/审核经 get_writing_model 严格读 state['model_name']，不回退默认。
-    selected_model = (configurable.get("model_name") or "").strip()
-    if selected_model:
-        updates["model_name"] = selected_model
+    # 容错切换后用实际成功的模型作为后续阶段起点，避免每个阶段都重复经历同一次失败。
+    persisted_model = (success_model or initial_model or "").strip()
+    if persisted_model:
+        updates["model_name"] = persisted_model
+    _ = last_error  # 链尾失败已生成 notice，这里保留变量供调试时断点查看
     updates["article_type"] = final_article_type
     updates["target_audience"] = final_target_audience
     updates["word_count_target"] = final_word_count
diff --git a/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/writer_draft.py b/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/writer_draft.py
index 7b7a082d..5f96f458 100644
--- a/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/writer_draft.py
+++ b/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/writer_draft.py
@@ -21,6 +21,12 @@ from ._agent_runner import (
     stream_writing_agent,
     use_builtin_agent,
 )
+from ._model_fallback import (
+    build_switch_progress_event,
+    classify_exception,
+    emit_model_switch_event,
+    fallback_chain,
+)
 from ._utils import resolve_writing_model_name, _content_to_text, count_words, reasoning_delta, ThinkTagParser, safe_get_stream_writer
 
 logger = logging.getLogger(__name__)
@@ -234,31 +240,25 @@ async def writer_draft_node(state: AIWritingState, config: RunnableConfig) -> di
                     "title": "模型错误", "detail": str(e), "message": "草稿生成失败",
                 }],
             }
+    # 旧路径的模型由 fallback_chain 在每章流式调用前按候选顺序解析（per-section），
+    # 不再在循环外预创建：模型名配错 / 模型挂掉时不再硬崩溃，按链顺序换下一个模型重写本章节。
+    draft_chain: list[str] = []
     if not use_agent:
-        # 模型创建守卫：模型名配错时 create_chat_model 抛 ValueError，
-        # 不拦截会硬崩溃整个图；此处转成可恢复的 status:error。
-        try:
-            model_name, used_default = resolve_writing_model_name(config, state)
-            if used_default:
-                model_default_notice = {
-                    "type": "notice", "agent_name": "作家",
-                    "title": "模型提示", "detail": "未检测到所选写作模型，已使用默认模型继续写作。",
-                }
-            from deerflow.models import create_chat_model
-            model = create_chat_model(name=model_name)
-        except Exception as e:
-            logger.error("draft model creation failed: %s", e, exc_info=True)
-            return {
-                "status": "error",
-                "progress_events": [{
-                    "type": "error", "agent_name": "作家",
-                    "title": "模型错误", "detail": str(e), "message": "草稿生成失败",
-                }],
+        initial_name, used_default = resolve_writing_model_name(config, state)
+        if used_default:
+            model_default_notice = {
+                "type": "notice", "agent_name": "作家",
+                "title": "模型提示", "detail": "未检测到所选写作模型，已使用默认模型继续写作。",
             }
+        draft_chain = fallback_chain(initial_name)
+        from deerflow.models import create_chat_model
     section_titles: list[str] = []
     section_bufs: list[str] = []   # 各章节正文（不含 ## 标题），供扩写兜底逐节替换
     section_errors: list[str] = []  # 分节写作的模型异常，供完成后非致命提示
     blocked_sections: list[dict] = []  # 严格模式下素材不足、无法写作的章节
+    switch_notices: list[dict] = []  # 模型容错切换记录，最终塞进 progress_events 走时间线
+    # 上一节成功使用的模型 —— 优先尝试，命中则零开销；与会商「当前模型先试」一致。
+    current_model_name: str | None = (draft_chain[0] if draft_chain else None)
 
     for i, section in enumerate(sections):
         sec_title  = _section_title(i, section)
@@ -292,9 +292,12 @@ async def writer_draft_node(state: AIWritingState, config: RunnableConfig) -> di
         )
 
         section_buf = ""   # 只累积正文（非思考）用于拼接 Markdown
-        parser = ThinkTagParser()
+        section_succeeded = False
         try:
             if use_agent:
+                parser = ThinkTagParser()  # 内置 agent 走自有路径，本变量保留以避免类型告警
+                _ = parser
+
                 def _on_text(text: str, is_thinking: bool, _i: int = i, _t: str = sec_title) -> None:
                     writer({
                         "type": "draft_chunk",
@@ -312,36 +315,78 @@ async def writer_draft_node(state: AIWritingState, config: RunnableConfig) -> di
                     config=config,
                     on_text=_on_text,
                 )
+                section_succeeded = True
             else:
-                async for chunk in model.astream([
-                    SystemMessage(content=DRAFT_SECTION_SYSTEM),
-                    HumanMessage(content=user_msg),
-                ]):
-                    # 推理模型的思考走 reasoning_content（content 为空）：作为思考实时推送
-                    reasoning = reasoning_delta(chunk)
-                    if reasoning:
-                        writer({
-                            "type": "draft_chunk",
-                            "section_index": i,
-                            "section_title": sec_title,
-                            "text": reasoning,
-                            "is_thinking": True,
-                            "is_section_done": False,
-                        })
-                    delta = _content_to_text(chunk.content)
-                    if not delta:
-                        continue
-                    for is_thinking, text in parser.feed(delta):
-                        if not is_thinking:
-                            section_buf += text
-                        writer({
-                            "type": "draft_chunk",
-                            "section_index": i,
-                            "section_title": sec_title,
-                            "text": text,
-                            "is_thinking": is_thinking,
-                            "is_section_done": False,
-                        })
+                # 旧路径：按 fallback_chain 试候选模型，失败就换下一个继续本章节。
+                # 以「上一节成功使用的模型」为头部 —— 同 config.yaml 顺序的链尾再补全。
+                section_chain: list[str] = []
+                if current_model_name:
+                    section_chain.append(current_model_name)
+                for n in draft_chain:
+                    if n and n not in section_chain:
+                        section_chain.append(n)
+                if not section_chain:
+                    section_chain = [None]  # type: ignore[list-item]
+                last_section_exc: BaseException | None = None
+                for s_idx, candidate in enumerate(section_chain):
+                    section_buf = ""
+                    parser = ThinkTagParser()
+                    try:
+                        m = create_chat_model(name=candidate) if candidate else create_chat_model()
+                        async for chunk in m.astream([
+                            SystemMessage(content=DRAFT_SECTION_SYSTEM),
+                            HumanMessage(content=user_msg),
+                        ]):
+                            reasoning = reasoning_delta(chunk)
+                            if reasoning:
+                                writer({
+                                    "type": "draft_chunk",
+                                    "section_index": i,
+                                    "section_title": sec_title,
+                                    "text": reasoning,
+                                    "is_thinking": True,
+                                    "is_section_done": False,
+                                })
+                            delta = _content_to_text(chunk.content)
+                            if not delta:
+                                continue
+                            for is_thinking, text in parser.feed(delta):
+                                if not is_thinking:
+                                    section_buf += text
+                                writer({
+                                    "type": "draft_chunk",
+                                    "section_index": i,
+                                    "section_title": sec_title,
+                                    "text": text,
+                                    "is_thinking": is_thinking,
+                                    "is_section_done": False,
+                                })
+                        section_succeeded = True
+                        current_model_name = candidate or current_model_name
+                        break
+                    except Exception as e:
+                        last_section_exc = e
+                        reason = classify_exception(e)
+                        logger.error(
+                            "draft section [%s] failed on model=%s (reason=%s): %s",
+                            sec_title, candidate, reason, e, exc_info=True,
+                        )
+                        next_candidate = section_chain[s_idx + 1] if s_idx + 1 < len(section_chain) else None
+                        if next_candidate:
+                            emit_model_switch_event(
+                                writer, agent_name=agent_label, stage="draft",
+                                failed_model=candidate or "", next_model=next_candidate,
+                                reason=reason, section_index=i, section_title=sec_title,
+                            )
+                            switch_notices.append(build_switch_progress_event(
+                                agent_name=agent_label, stage="draft",
+                                failed_model=candidate or "", next_model=next_candidate,
+                                reason=reason, section_index=i, section_title=sec_title,
+                            ))
+                            continue
+                if not section_succeeded:
+                    assert last_section_exc is not None
+                    raise last_section_exc
 
             # 章节完成
             writer({
@@ -387,6 +432,7 @@ async def writer_draft_node(state: AIWritingState, config: RunnableConfig) -> di
             "status": "awaiting_section_help",
             "progress_events": [
                 {"type": "step_start", "agent_name": agent_label, "message": f"开始写作（第{revision_count+1}稿）"},
+                *switch_notices,
                 {
                     "type": "notice", "agent_name": agent_label, "title": "素材不足",
                     "detail": f"有 {len(blocked_sections)} 个章节缺少足够素材支撑写作，需要你决定如何处理。",
@@ -432,7 +478,9 @@ async def writer_draft_node(state: AIWritingState, config: RunnableConfig) -> di
                     on_text=None,
                 )).strip()
             else:
-                resp = await model.ainvoke([
+                # 扩写是兜底优化，不再做模型链兜底：拿当前已成功的模型试一次即可，失败就保留原文。
+                expand_model = create_chat_model(name=current_model_name) if current_model_name else create_chat_model()
+                resp = await expand_model.ainvoke([
                     SystemMessage(content=DRAFT_SECTION_SYSTEM),
                     HumanMessage(content=expand_msg),
                 ])
@@ -479,6 +527,7 @@ async def writer_draft_node(state: AIWritingState, config: RunnableConfig) -> di
         "status": "awaiting_draft_confirm",
         "progress_events": [
             {"type": "step_start", "agent_name": agent_label, "message": f"开始写作（第{revision_count+1}稿）"},
+            *switch_notices,
             *notices,
             {
                 "type": "draft_ready",
diff --git a/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/writer_outline.py b/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/writer_outline.py
index 8e50a4f0..1a938d25 100644
--- a/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/writer_outline.py
+++ b/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/writer_outline.py
@@ -29,7 +29,13 @@ from ._agent_runner import (
     stream_writing_agent,
     use_builtin_agent,
 )
-from ._utils import ThinkTagParser, _content_to_text, get_writing_model, llm_json, parse_json, reasoning_delta, safe_get_stream_writer
+from ._model_fallback import (
+    build_switch_progress_event,
+    classify_exception,
+    emit_model_switch_event,
+    fallback_chain,
+)
+from ._utils import ThinkTagParser, _content_to_text, llm_json, parse_json, reasoning_delta, resolve_writing_model_name, safe_get_stream_writer
 
 logger = logging.getLogger(__name__)
 
@@ -292,10 +298,11 @@ async def writer_outline_node(state: AIWritingState, config: RunnableConfig) ->
     writer = safe_get_stream_writer()
 
     full_text = ""   # 只累积正文（非思考）用于 JSON 解析
-    parser = ThinkTagParser()
+    extra_progress_events: list[dict] = []
 
-    try:
-        if use_agent:
+    if use_agent:
+        # 内置 agent 路径走 stream_writing_agent，不在本模块做模型链兜底（agent 自带模型解析）。
+        try:
             def _on_text(text: str, is_thinking: bool) -> None:
                 writer({"type": "outline_chunk", "text": text, "is_thinking": is_thinking})
 
@@ -306,41 +313,81 @@ async def writer_outline_node(state: AIWritingState, config: RunnableConfig) ->
                 config=config,
                 on_text=_on_text,
             )
-        else:
-            # get_writing_model 放进 try：用户模型缺失 / 模型名配错时抛 ValueError，
-            # 留在 try 外会硬崩溃；放进来则统一走下方的 status:error 兜底。
-            model = get_writing_model(config, state)
-            async for chunk in model.astream([
-                SystemMessage(content=system_prompt),
-                HumanMessage(content=user_prompt),
-            ]):
-                # 推理模型的思考走 reasoning_content（content 为空）：作为思考实时推送
-                reasoning = reasoning_delta(chunk)
-                if reasoning:
-                    writer({"type": "outline_chunk", "text": reasoning, "is_thinking": True})
-                delta = _content_to_text(chunk.content)
-                if not delta:
+        except Exception as e:
+            logger.error("outline streaming failed: %s", e, exc_info=True)
+            return {
+                "status": "error",
+                "progress_events": [{
+                    "type": "error",
+                    "agent_name": agent_label,
+                    "title": "模型错误",
+                    "detail": str(e),
+                    "message": "大纲生成失败",
+                }],
+            }
+    else:
+        # 旧路径：按 fallback_chain 顺序尝试，失败即换下一个模型重跑该环节。
+        initial_name, _used_default = resolve_writing_model_name(config, state)
+        chain = fallback_chain(initial_name)
+        from deerflow.models import create_chat_model
+
+        success = False
+        last_exc: BaseException | None = None
+        for idx, candidate in enumerate(chain or [None]):
+            full_text = ""
+            parser = ThinkTagParser()
+            try:
+                model = create_chat_model(name=candidate) if candidate else create_chat_model()
+                async for chunk in model.astream([
+                    SystemMessage(content=system_prompt),
+                    HumanMessage(content=user_prompt),
+                ]):
+                    reasoning = reasoning_delta(chunk)
+                    if reasoning:
+                        writer({"type": "outline_chunk", "text": reasoning, "is_thinking": True})
+                    delta = _content_to_text(chunk.content)
+                    if not delta:
+                        continue
+                    for is_thinking, text in parser.feed(delta):
+                        if not is_thinking:
+                            full_text += text
+                        writer({
+                            "type": "outline_chunk",
+                            "text": text,
+                            "is_thinking": is_thinking,
+                        })
+                success = True
+                break
+            except Exception as e:
+                last_exc = e
+                reason = classify_exception(e)
+                logger.error(
+                    "outline streaming failed on model=%s (reason=%s): %s",
+                    candidate, reason, e, exc_info=True,
+                )
+                next_candidate = chain[idx + 1] if idx + 1 < len(chain) else None
+                if next_candidate:
+                    emit_model_switch_event(
+                        writer, agent_name=agent_label, stage="outline",
+                        failed_model=candidate or "", next_model=next_candidate, reason=reason,
+                    )
+                    extra_progress_events.append(build_switch_progress_event(
+                        agent_name=agent_label, stage="outline",
+                        failed_model=candidate or "", next_model=next_candidate, reason=reason,
+                    ))
                     continue
-                for is_thinking, text in parser.feed(delta):
-                    if not is_thinking:
-                        full_text += text
-                    writer({
-                        "type": "outline_chunk",
-                        "text": text,
-                        "is_thinking": is_thinking,
-                    })
-    except Exception as e:
-        logger.error("outline streaming failed: %s", e, exc_info=True)
-        return {
-            "status": "error",
-            "progress_events": [{
-                "type": "error",
-                "agent_name": agent_label,
-                "title": "模型错误",
-                "detail": str(e),
-                "message": "大纲生成失败",
-            }],
-        }
+        if not success:
+            assert last_exc is not None
+            return {
+                "status": "error",
+                "progress_events": extra_progress_events + [{
+                    "type": "error",
+                    "agent_name": agent_label,
+                    "title": "模型错误",
+                    "detail": str(last_exc),
+                    "message": "大纲生成失败（所有候选模型均失败）",
+                }],
+            }
 
     # 编辑模式解析失败时回退为「当前大纲」，避免把用户已有大纲弄丢
     if is_revise:
@@ -387,6 +434,7 @@ async def writer_outline_node(state: AIWritingState, config: RunnableConfig) ->
         "status": "awaiting_outline_confirm",
         "progress_events": [
             {"type": "step_start", "agent_name": agent_label, "message": "开始规划大纲"},
+            *extra_progress_events,
             {"type": "outline_ready", "agent_name": agent_label, "message": f"大纲已生成，共 {section_count} 个章节"},
         ],
     }

From 0a456e91465c3f97a9a901f1e7b7bb2bc6925e17 Mon Sep 17 00:00:00 2001
From: fengxiangdebaozi <10845145+fengxiangdebaozi@user.noreply.gitee.com>
Date: Mon, 22 Jun 2026 17:07:28 +0800
Subject: [PATCH 10/17] =?UTF-8?q?feat(ai-writing):=20=E5=90=8E=E5=8F=B0?=
 =?UTF-8?q?=E6=8C=82=E8=B5=B7=E6=94=B9=E4=B8=BA=E5=8D=B3=E6=97=B6=E8=BF=94?=
 =?UTF-8?q?=E5=9B=9E=EF=BC=8C=E4=B9=90=E8=A7=82=E6=A0=87=E8=AE=B0=E5=8E=86?=
 =?UTF-8?q?=E5=8F=B2=E3=80=8C=E6=8C=82=E8=B5=B7=E4=B8=AD=E3=80=8D?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

点击「转后台」后立刻 reset 回初始页，挂起请求 fire-and-forget；
历史下拉靠 markBackground/clearBackground 乐观标记马上显示「挂起中」，
不等后端确认，与「多智能体会商」交互一致。

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 .../components/ai-writing/AIWritingPanel.tsx  | 33 +++++++++-------
 .../open-canvas/hooks/useAIWritingSessions.ts | 38 ++++++++++++++++++-
 2 files changed, 56 insertions(+), 15 deletions(-)

diff --git a/frontend-web/src/open-canvas/components/ai-writing/AIWritingPanel.tsx b/frontend-web/src/open-canvas/components/ai-writing/AIWritingPanel.tsx
index 9439c133..e38984ed 100644
--- a/frontend-web/src/open-canvas/components/ai-writing/AIWritingPanel.tsx
+++ b/frontend-web/src/open-canvas/components/ai-writing/AIWritingPanel.tsx
@@ -91,7 +91,7 @@ export function AIWritingPanel({
   const { state, startWriting, submitIntervention, pauseWriting, resumeWriting, reset } = useAIWriting();
   const { historySession, isHistoryMode, openHistory, closeHistory } = useAIWritingHistory();
   const writingForm = useWritingFormOptional();
-  const { sessions, loading: sessionsLoading, refresh, rename, remove } = useAIWritingSessions();
+  const { sessions, loading: sessionsLoading, refresh, rename, remove, markBackground, clearBackground } = useAIWritingSessions();
   const { articleTypes, refresh: refreshArticleTypes } = useArticleTypes();
 
   useEffect(() => { refreshArticleTypes(); }, [refreshArticleTypes]);
@@ -207,24 +207,31 @@ export function AIWritingPanel({
   }, [closeHistory, reset]);
 
   // ── 后台挂起：把当前写作转后台，后端自动驱动到定稿（离页/刷新/切号不打断） ──
-  const handleBackgroundSuspend = useCallback(async () => {
+  // 与「多智能体会商」一致：点击后**立刻**回到初始页面，挂起请求 fire-and-forget；
+  // 历史下拉靠乐观标记马上显示「挂起中」，不等后端确认。
+  const handleBackgroundSuspend = useCallback(() => {
     const sid = state.sessionId;
     if (!sid) {
       toast.error('当前没有可挂起的写作会话');
       return;
     }
-    try {
-      await suspendAIWritingToBackground(sid);
-    } catch (e) {
-      toast.error(e instanceof Error ? e.message : '后台挂起失败');
-      return;
-    }
-    toast.success('已转入后台自动完成，稍后可在「历史记录」查看进度与结果');
-    // reset() 会 stop 客户端 SSE 订阅并回到 idle —— 后端驱动器接管续跑，不冲突。
+    // 1) 立刻：乐观标记 + 重置 UI 回到 idle（reset() 会 stop 客户端 SSE 订阅，
+    //    后端驱动器接管续跑，不冲突）。
+    markBackground(sid);
     reset();
-    // 刷新历史列表，让「挂起中」徽标立刻出现。
-    refresh();
-  }, [state.sessionId, reset, refresh]);
+    toast.success('已转入后台自动完成，稍后可在「历史记录」查看进度与结果');
+    // 2) 后台：fire-and-forget 通知后端拉起驱动；完成后刷新真实状态并清除乐观标记。
+    void (async () => {
+      try {
+        await suspendAIWritingToBackground(sid);
+      } catch (e) {
+        toast.error(e instanceof Error ? e.message : '后台挂起失败');
+      } finally {
+        await refresh();
+        clearBackground(sid);
+      }
+    })();
+  }, [state.sessionId, reset, refresh, markBackground, clearBackground]);
 
   // 离页 / 刷新 / 关标签页时，若写作仍在进行中则自动转后台（与圆桌「自动挂起」一致）。
   // 用 ref 持有最新的「是否在跑 + 会话 id」，让卸载/卸载前回调读到即时值。
diff --git a/frontend-web/src/open-canvas/hooks/useAIWritingSessions.ts b/frontend-web/src/open-canvas/hooks/useAIWritingSessions.ts
index ffa7ae6d..ac98220f 100644
--- a/frontend-web/src/open-canvas/hooks/useAIWritingSessions.ts
+++ b/frontend-web/src/open-canvas/hooks/useAIWritingSessions.ts
@@ -1,4 +1,4 @@
-import { useCallback, useState } from 'react';
+import { useCallback, useMemo, useState } from 'react';
 
 import {
   listAIWritingSessions,
@@ -10,6 +10,27 @@ import type { AIWritingSessionSummary } from '../api/ai-writing-sessions';
 export function useAIWritingSessions() {
   const [sessions, setSessions] = useState<AIWritingSessionSummary[]>([]);
   const [loading, setLoading] = useState(false);
+  // 「后台挂起」乐观标记：点击挂起后立刻让历史下拉显示「挂起中」，不必等后端确认。
+  // 后端状态（status='background'）刷新到位后由调用方 clearBackground 清除。
+  const [pendingBackground, setPendingBackground] = useState<Set<string>>(() => new Set());
+
+  const markBackground = useCallback((id: string) => {
+    setPendingBackground((prev) => {
+      if (prev.has(id)) return prev;
+      const next = new Set(prev);
+      next.add(id);
+      return next;
+    });
+  }, []);
+
+  const clearBackground = useCallback((id: string) => {
+    setPendingBackground((prev) => {
+      if (!prev.has(id)) return prev;
+      const next = new Set(prev);
+      next.delete(id);
+      return next;
+    });
+  }, []);
 
   const refresh = useCallback(async () => {
     setLoading(true);
@@ -41,5 +62,18 @@ export function useAIWritingSessions() {
     }
   }, []);
 
-  return { sessions, loading, refresh, rename, remove };
+  // 把乐观「挂起中」标记叠加到真实列表上：已在列表里的会话立刻显示 background。
+  const displaySessions = useMemo(
+    () =>
+      pendingBackground.size === 0
+        ? sessions
+        : sessions.map((s) =>
+            pendingBackground.has(s.id) && s.status !== 'done'
+              ? { ...s, status: 'background' }
+              : s,
+          ),
+    [sessions, pendingBackground],
+  );
+
+  return { sessions: displaySessions, loading, refresh, rename, remove, markBackground, clearBackground };
 }

From 6b480fdbe97d608fe42b9fa5eb2c73e0f9dd016f Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E5=90=89=E9=91=AB=E5=8D=87?= <xsji@kland.com.cn>
Date: Mon, 22 Jun 2026 17:17:54 +0800
Subject: [PATCH 11/17] =?UTF-8?q?feat:=20=E6=96=B0=E5=A2=9E=E6=8A=80?=
 =?UTF-8?q?=E8=83=BD=E5=9C=B0=E5=9D=80=E7=AE=A1=E7=90=86=E4=B8=8E=E5=BC=80?=
 =?UTF-8?q?=E6=94=BE=E9=97=AE=E7=AD=94=E6=8E=A5=E5=8F=A3=EF=BC=8C=E5=AE=8C?=
 =?UTF-8?q?=E5=96=84=E4=BB=BB=E5=8A=A1=E5=B7=A5=E4=BD=9C=E5=8C=BA?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- 新增「技能地址管理」页面与 /api/skill-addresses 路由（扫描/预览/批量替换技能内地址）
- 完善 /api/open/chat 开放问答接口与模型工厂、网关配置
- 调整任务工作区、侧边栏与路由
- 将本地 0622/ 临时目录与 __pycache__/ 加入忽略，不提交

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 .gitignore                                    |   6 +
 CLAUDE.md                                     |   1 +
 frontend-web/src/components/page-sidebar.tsx  |  15 +
 frontend-web/src/pages/SkillAddressesPage.tsx | 522 +++++++++++++++
 frontend-web/src/pages/StrategyRoutes.tsx     |  11 +
 .../api/skill-addresses.ts                    | 272 ++++++++
 .../task-components/TaskFxWorkspace.tsx       |   5 +-
 offline-backend-20260512/backend/CLAUDE.md    |  41 +-
 .../backend/app/gateway/app.py                |   4 +
 .../backend/app/gateway/routers/open_chat.py  | 133 +++-
 .../app/gateway/routers/skill_addresses.py    | 631 ++++++++++++++++++
 offline-backend-20260512/backend/config.yaml  |   6 +
 .../harness/deerflow/models/factory.py        |   8 +-
 offline-backend-20260512/config.yaml          |   6 +
 test_open_chat_api.py                         | 128 +++-
 15 files changed, 1749 insertions(+), 40 deletions(-)
 create mode 100644 frontend-web/src/pages/SkillAddressesPage.tsx
 create mode 100644 frontend-web/src/strategy-components/api/skill-addresses.ts
 create mode 100644 offline-backend-20260512/backend/app/gateway/routers/skill_addresses.py

diff --git a/.gitignore b/.gitignore
index de6dbc5f..0b5f0dfe 100644
--- a/.gitignore
+++ b/.gitignore
@@ -104,3 +104,9 @@ docs/backups
 
 # uv lockfile (not tracked per project preference)
 offline-backend-20260512/backend/uv.lock
+
+# Local-only scratch directory (not committed)
+0622/
+
+# Python bytecode cache
+__pycache__/
diff --git a/CLAUDE.md b/CLAUDE.md
index f2dd1701..16f0ef3b 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -82,6 +82,7 @@ The frontend (`frontend-web/`) is a React 19 + Vite SPA:
 - **轻应用中心 (Light App Center)**: a sibling of 任务管理 in the sidebar with two pages — 应用管理 (admin registers `window_open` or `iframe` mini-apps with a route-param builder: login params resolved from `localStorage.userInfo`, theme params literal) and 应用列表 (card grid of `window_open` apps, `window.open` launch). `iframe` apps are mounted under a chosen parent menu (XK/PG/TY/综合管理) and **dynamically injected into the sidebar** by `core/page-layout/use-sidebar-menu.ts` (`useSidebarMenuItems`), opening in the `LightAppIframeView` host. Client + React Query hooks in `strategy-components/api/light-apps.ts`; URL builder/constants in `strategy-components/lib/light-app.ts`. Backend: `/api/light-apps` (global/shared, admin-gated writes).
 - **菜单管理 (Menu Management, admin-only)**: an **overlay** over the static sidebar menu — `MenuManagementPage` (`pages/MenuManagementPage.tsx`, route `/page/strategy/admin/menu-management`, opened from the admin-only 「菜单管理」 item in the **bottom-left settings dropdown** of `components/page-sidebar.tsx`). Admins rename / 停用·启用 / reorder (↑↓ or @dnd-kit drag) / re-parent (horizontal drag **or** the edit dialog's 「移动到」parent selector) any node across 一级/二级/三级, **including injected iframe light-app nodes** (`light-app-<appId>`). The static tree (`core/page-layout/sidebar-menu.ts`) stays the source of icons/routes; only per-node deltas (`label`/`disabled`/`parentId`/`sortOrder`) persist to the backend. Merge logic in `core/page-layout/menu-overrides.ts` (`applyMenuOverrides`, applied in `use-sidebar-menu.ts` after light-app injection, before role filter; disabling a node hides its whole subtree) — strict **3-level cap** (`wouldExceedDepth`). Client `strategy-components/api/menu-overrides.ts`. Backend: `/api/menu-overrides` (global/shared; open read, admin atomic replace-all).
 - **敏感词管理 (Sensitive Word Management)**: configures the **展示层脱敏映射** (原始词 → 替换词). **Gated on the username `lqq`** (not the admin role) — `getAccountDisplayName().toLowerCase() === "lqq"` (username = email `@` prefix, else `user_id`, same as the profile page); the dropdown entry, the route guard, and the page-level guard all use this check. `src/lib/desensitize.ts` ships a frozen seed map `SEED_DESENSITIZE_MAP` (the deployment seed) plus a live mutable `DESENSITIZE_MAP` (initialized from the seed so `desensitize()` works before any fetch). `DesensitizeWordsLoader` (mounted in `App.tsx` `AppProviders`) fetches `/api/sensitive-words` on startup and, if non-empty, calls `replaceDesensitizeMap(...)` to make the backend config the source of truth (empty/failed → keep seed). Page `SensitiveWordsPage.tsx` (route `/page/strategy/admin/sensitive-words`, opened from the 「敏感词管理」 item in the **bottom-left settings dropdown** of `components/page-sidebar.tsx`, present in **both** full + collapsed sidebar modes) lists/adds/edits/启停/deletes words and saves the **whole set atomically** (replace-all); a 「导入种子词」 button merges any missing `SEED_DESENSITIZE_MAP` entries into the editor. Client + hooks in `strategy-components/api/sensitive-words.ts` (`wordsToMap` folds enabled rows into the runtime map). Backend: `/api/sensitive-words` (global/shared; open read, write allowed for admin **or** the `lqq` account), mirroring 菜单管理's store shape. **侧边栏 + 导航栏 menu labels are desensitized** at the data source: `useSidebarMenuItems`/`useActiveThirdLevel` (`core/page-layout/use-sidebar-menu.ts`) map every node's `label` through a desensitizer built **directly from the `/api/sensitive-words` query** (`makeDesensitizer(wordsToMap(words))`, falling back to the global seed `desensitize` while empty) — this sidesteps the `replaceDesensitizeMap` effect-timing race. Only `label` is rewritten; `path`/`id`/`icon` are untouched so routing/highlight are unaffected. All nav surfaces (`page-sidebar.tsx`, `page-sidebar-dark.tsx`, `page-sidebar-v1.tsx`, `strategy-components/Header.tsx`, `PageRoutes.tsx`) consume this hook and are covered; `MenuManagementPage` uses the raw tree so admins still edit real labels.
+- **技能地址管理 (Skill Address Management, admin-only)**: 扫描运行中的技能（`skills/{public,custom}` 下所有 `.md`/`.py`）里的 **http(s) URL 与 IP/IP:端口**，按唯一地址聚合全部出现位置（跨 .md/.py），支持**批量替换**（如把 UAT 地址一次性改成生产）。页面 `pages/SkillAddressesPage.tsx`（route `/page/strategy/admin/skill-addresses`，入口「技能地址管理」在 `components/page-sidebar.tsx` 左下角设置下拉，full+collapsed 两种模式）：打开自动扫描并带**进度条**（扫描逐技能 / 应用逐文件 X/N，应对 150+ 技能），URL/IP 类型过滤 + 搜索 + 「片段批量套用」（如所有 `uat.4.cn` → `prod.4.cn`）、每地址「替换为」输入框（可展开看 `file:line` 片段）、内联**预览**、**应用**、**撤销上次**（回滚）。client + NDJSON 流式解析在 `strategy-components/api/skill-addresses.ts`。后端 `/api/skill-addresses`（**仅管理员**）：所有文件 IO 走 `asyncio.to_thread` 不阻塞事件循环；`GET /scan`(+`/scan/stream`)、`POST /preview`、`POST /apply`(+`/apply/stream`)、`GET /backups`、`POST /rollback/{id}`；**子串安全替换**（基于 token span，不会误伤子串）、`expected_total` 乐观锁（409）、写前自动磁盘备份到 `{skills_root}/.address-edits/`、应用后刷新技能系统提示缓存。纯引擎 `deerflow/skills/address_scan.py`，router `app/gateway/routers/skill_addresses.py`，tests `tests/test_skill_addresses.py`。范围仅运行中技能，不动 `deploy/minimal/skills/`。
 - **按钮管理 (Task Button Management, admin-only)**: configures the **可配置跳转按钮** rendered in the left task panel of the task deep-link workspaces (rwfx / 3qfx / xdfx, see `TaskFxWorkspace`). Each button = `{business, label, linkType, target, appendTaskId, idParam, idKind, enabled, sortOrder}`; `linkType` is `url` (外链 `window.open`, optional append of `?{idParam}={id}` — **`idParam` is the editable query-param name**, e.g. `id`/`task_id`), `business` (内部切换到另一业务, reuses `resolveTaskDeeplinkTarget`, always task id), or `purpose` (弹出当前任务的「任务目的」表格 via the shared `TaskPurposeDialog`, no target — configurable for any page incl. xdfx). **`idKind` (`task`/`action`) only matters for xdfx**: its deep-link `?taskId=` is an 行动id and the 任务id is resolved at login (`fetchTaskIdFromAction`) — `TaskContext` now also stashes the raw `actionId`, so a url button can append either (default `task`). Page `TaskButtonsPage.tsx` (route `/page/strategy/admin/task-buttons`, **admin-gated**, entry「按钮管理」in the bottom-left settings dropdown of `components/page-sidebar.tsx`, both full + collapsed modes) has per-business tabs, edits name/type/target/启停/排序/增删, saves the **whole set atomically** (replace-all), plus 「导入默认」 to merge code-seed defaults. **Seed/fallback** (like 敏感词管理): only when the whole table is empty (fresh deploy, never saved) do the workspace + page fall back to `defaultTaskButtons(cfg)` — rwfx's url buttons (编辑/下一步/3q详情) built from the **live** `/api/public/config/task-deeplink` plus cross-business switch buttons for **all three** businesses (each switches to the other two — rwfx→三情/行动, 3qfx→任务/行动, xdfx→任务/三情). Once saved, the store is the sole source of truth. The old hardcoded jump buttons were migrated out: `TaskFxWorkspace` no longer hardcodes 切换到三情分析, `TaskActionsPanel` keeps only the non-jump 清除记录, and `TaskPurposePanel` dropped its 3q详情 button (now configurable) **and defaults the 目的树 table to collapsed** (chevron toggle). Helpers `core/auth/task-buttons.ts` (`TaskButton`, `defaultTaskButtons`, `resolveTaskButtons`, `buttonsForBusiness`, `runTaskButton`), client+hooks `strategy-components/api/task-buttons.ts`. Backend: `/api/task-buttons` (global/shared; open read, admin replace-all), store `deerflow.persistence.task_buttons` (table `task_buttons`, auto-created by `create_all`), wired `app.state.task_button_store`. Tests: `tests/test_task_buttons.py`.
 - **State**: TanStack Query for server state, context providers (`SubtasksProvider`, `ArtifactsProvider`, `PromptInputProvider`) scoped to `WorkspaceRoutes`
 - **LLM streaming**: `@langchain/langgraph-sdk` streams messages from the backend via SSE
diff --git a/frontend-web/src/components/page-sidebar.tsx b/frontend-web/src/components/page-sidebar.tsx
index d01d6d6e..92b4eaa8 100644
--- a/frontend-web/src/components/page-sidebar.tsx
+++ b/frontend-web/src/components/page-sidebar.tsx
@@ -9,6 +9,7 @@ import {
   FolderOpen,
   ListTree,
   MousePointerClick,
+  Replace,
   Settings,
   ShieldAlert,
   User,
@@ -222,6 +223,14 @@ function SecondaryNavSidebar() {
                       按钮管理
                     </DropdownMenuItem>
                   )}
+                  {isAdmin && (
+                    <DropdownMenuItem
+                      onClick={() => navigate("/page/strategy/admin/skill-addresses")}
+                    >
+                      <Replace className="size-4" />
+                      技能地址管理
+                    </DropdownMenuItem>
+                  )}
                   {canManageSensitiveWords() && (
                     <DropdownMenuItem
                       onClick={() => navigate("/page/strategy/admin/sensitive-words")}
@@ -508,6 +517,12 @@ function LegacyAccordionSidebar({
                   按钮管理
                 </DropdownMenuItem>
               )}
+              {isAdmin && (
+                <DropdownMenuItem onClick={() => navigate("/page/strategy/admin/skill-addresses")}>
+                  <Replace className="size-4" />
+                  技能地址管理
+                </DropdownMenuItem>
+              )}
               {canManageSensitiveWords() && (
                 <DropdownMenuItem onClick={() => navigate("/page/strategy/admin/sensitive-words")}>
                   <ShieldAlert className="size-4" />
diff --git a/frontend-web/src/pages/SkillAddressesPage.tsx b/frontend-web/src/pages/SkillAddressesPage.tsx
new file mode 100644
index 00000000..b1cc1642
--- /dev/null
+++ b/frontend-web/src/pages/SkillAddressesPage.tsx
@@ -0,0 +1,522 @@
+"use client";
+
+import { useCallback, useEffect, useMemo, useState } from "react";
+import { Navigate } from "react-router-dom";
+import {
+  AlertTriangle,
+  ChevronDown,
+  ChevronRight,
+  Eye,
+  FileCode,
+  Globe,
+  Info,
+  Loader2,
+  Network,
+  Replace,
+  ScanLine,
+  Search,
+  Undo2,
+  Wand2,
+} from "lucide-react";
+import { toast } from "sonner";
+
+import { getStoredAuth } from "@/core/auth";
+import {
+  type AddressGroup,
+  type AddressKind,
+  type ApplyProgress,
+  type PreviewResult,
+  type Replacement,
+  type ScanProgress,
+  type ScanResult,
+  applyReplacementsStream,
+  previewReplacements,
+  rollbackBackup,
+  scanAddressesStream,
+  useSkillAddressBackups,
+} from "@/strategy-components/api/skill-addresses";
+
+const ALL_KINDS: AddressKind[] = ["url", "ip"];
+
+function ProgressBar({ done, total, label }: { done: number; total: number; label: string }) {
+  const pct = total > 0 ? Math.round((done / total) * 100) : 0;
+  return (
+    <div className="rounded-xl border border-blue-200/70 bg-blue-50/60 p-3.5">
+      <div className="mb-1.5 flex items-center justify-between text-xs font-medium text-blue-700">
+        <span className="flex items-center gap-1.5">
+          <Loader2 className="size-3.5 animate-spin" />
+          {label}
+        </span>
+        <span className="font-mono">
+          {done} / {total}（{pct}%）
+        </span>
+      </div>
+      <div className="h-2 w-full overflow-hidden rounded-full bg-blue-100">
+        <div className="h-full rounded-full bg-blue-500 transition-all duration-200" style={{ width: `${pct}%` }} />
+      </div>
+    </div>
+  );
+}
+
+function KindBadge({ kind }: { kind: AddressKind }) {
+  const isUrl = kind === "url";
+  return (
+    <span
+      className={`inline-flex items-center gap-1 rounded-md px-1.5 py-0.5 text-[11px] font-semibold ${
+        isUrl ? "bg-indigo-50 text-indigo-600" : "bg-emerald-50 text-emerald-600"
+      }`}
+    >
+      {isUrl ? <Globe className="size-3" /> : <Network className="size-3" />}
+      {isUrl ? "URL" : "IP"}
+    </span>
+  );
+}
+
+export default function SkillAddressesPage() {
+  const isAdmin = getStoredAuth()?.system_role === "admin";
+
+  const [kinds, setKinds] = useState<AddressKind[]>(ALL_KINDS);
+  const [scan, setScan] = useState<ScanResult | null>(null);
+  const [scanning, setScanning] = useState(false);
+  const [scanProg, setScanProg] = useState<ScanProgress | null>(null);
+
+  const [edits, setEdits] = useState<Record<string, string>>({});
+  const [query, setQuery] = useState("");
+  const [expanded, setExpanded] = useState<Set<string>>(new Set());
+
+  const [preview, setPreview] = useState<PreviewResult | null>(null);
+  const [applying, setApplying] = useState(false);
+  const [applyProg, setApplyProg] = useState<ApplyProgress | null>(null);
+  const [lastBackupId, setLastBackupId] = useState<string | null>(null);
+
+  const [segFind, setSegFind] = useState("");
+  const [segRepl, setSegRepl] = useState("");
+
+  const { data: backups = [], refetch: refetchBackups } = useSkillAddressBackups();
+
+  const runScan = useCallback(
+    async (useKinds: AddressKind[]) => {
+      setScanning(true);
+      setScanProg({ done: 0, total: 0 });
+      setPreview(null);
+      try {
+        const result = await scanAddressesStream({ kinds: useKinds, onProgress: setScanProg });
+        setScan(result);
+      } catch (err) {
+        toast.error(err instanceof Error ? err.message : "扫描失败");
+      } finally {
+        setScanning(false);
+        setScanProg(null);
+      }
+    },
+    [],
+  );
+
+  // 首次进入自动扫描。
+  useEffect(() => {
+    if (isAdmin) void runScan(ALL_KINDS);
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, [isAdmin]);
+
+  const groups = scan?.groups ?? [];
+
+  const toggleKind = (k: AddressKind) => {
+    const next = kinds.includes(k) ? kinds.filter((x) => x !== k) : [...kinds, k];
+    if (next.length === 0) return; // 至少保留一种
+    setKinds(next);
+    void runScan(next);
+  };
+
+  const setEdit = (addr: string, val: string) => setEdits((p) => ({ ...p, [addr]: val }));
+
+  const toggleExpand = (addr: string) =>
+    setExpanded((p) => {
+      const n = new Set(p);
+      if (n.has(addr)) n.delete(addr);
+      else n.add(addr);
+      return n;
+    });
+
+  /** 片段批量套用：把所有地址里出现的「查找片段」整体替换为「替换片段」，填入各行的「替换为」。 */
+  const applySegment = () => {
+    const find = segFind.trim();
+    if (!find) {
+      toast.error("请先填写要查找的片段（如 uat.4.cn）");
+      return;
+    }
+    let n = 0;
+    setEdits((prev) => {
+      const next = { ...prev };
+      for (const g of groups) {
+        if (g.address.includes(find)) {
+          const replaced = g.address.split(find).join(segRepl);
+          if (replaced !== g.address) {
+            next[g.address] = replaced;
+            n++;
+          }
+        }
+      }
+      return next;
+    });
+    if (n === 0) toast.info(`没有地址包含片段「${find}」`);
+    else toast.success(`已为 ${n} 个地址套用替换（请预览后应用）`);
+  };
+
+  const buildReplacements = useCallback((): Replacement[] => {
+    const out: Replacement[] = [];
+    for (const g of groups) {
+      const to = (edits[g.address] ?? "").trim();
+      if (to && to !== g.address) out.push({ from: g.address, to, kind: g.kind });
+    }
+    return out;
+  }, [groups, edits]);
+
+  const pendingCount = useMemo(() => buildReplacements().length, [buildReplacements]);
+
+  const runPreview = async () => {
+    const repls = buildReplacements();
+    if (repls.length === 0) {
+      toast.error("还没有填写任何「替换为」");
+      return;
+    }
+    try {
+      const result = await previewReplacements(repls);
+      setPreview(result);
+      if (result.total === 0) toast.info("没有任何文件会被改动（请检查填写的替换值）");
+    } catch (err) {
+      toast.error(err instanceof Error ? err.message : "预览失败");
+    }
+  };
+
+  const runApply = async () => {
+    if (!preview) return;
+    const repls = buildReplacements();
+    if (repls.length === 0) return;
+    setApplying(true);
+    setApplyProg({ done: 0, totalFiles: 0 });
+    try {
+      const result = await applyReplacementsStream(repls, {
+        expectedTotal: preview.expectedTotal,
+        backup: true,
+        onProgress: setApplyProg,
+      });
+      toast.success(`已修改 ${result.changedFiles.length} 个文件，共替换 ${result.replacedTotal} 处`);
+      setLastBackupId(result.backupId);
+      setEdits({});
+      setPreview(null);
+      await refetchBackups();
+      await runScan(kinds);
+    } catch (err) {
+      toast.error(err instanceof Error ? err.message : "应用失败");
+    } finally {
+      setApplying(false);
+      setApplyProg(null);
+    }
+  };
+
+  const runRollback = async () => {
+    const id = lastBackupId ?? backups[0]?.backupId;
+    if (!id) {
+      toast.error("没有可撤销的备份");
+      return;
+    }
+    try {
+      const { restored } = await rollbackBackup(id);
+      toast.success(`已从备份恢复 ${restored} 个文件`);
+      setLastBackupId(null);
+      await refetchBackups();
+      await runScan(kinds);
+    } catch (err) {
+      toast.error(err instanceof Error ? err.message : "撤销失败");
+    }
+  };
+
+  const q = query.trim().toLowerCase();
+  const visible = useMemo(
+    () => (!q ? groups : groups.filter((g) => g.address.toLowerCase().includes(q) || g.occurrences.some((o) => o.skill.toLowerCase().includes(q)))),
+    [groups, q],
+  );
+
+  const totalOccurrences = useMemo(() => groups.reduce((s, g) => s + g.count, 0), [groups]);
+  const canRollback = !!(lastBackupId || backups.length > 0);
+
+  if (!isAdmin) return <Navigate to="/page/workspace/chats/new" replace />;
+
+  return (
+    <div className="h-full overflow-auto bg-slate-50/60 px-4 py-6 sm:px-6 lg:px-8">
+      <div className="mx-auto max-w-5xl">
+        {/* 顶部信息卡 */}
+        <div className="relative mb-6 overflow-hidden rounded-3xl border border-slate-200/60 bg-white p-6 shadow-sm md:p-8">
+          <div className="pointer-events-none absolute right-0 top-0 select-none p-8 opacity-[0.03]">
+            <Replace className="size-44 text-slate-900" />
+          </div>
+          <div className="relative">
+            <div className="mb-4 flex flex-col justify-between gap-4 md:flex-row md:items-center">
+              <div>
+                <span className="mb-3 inline-flex select-none items-center gap-1.5 rounded-full border border-blue-100 bg-blue-50 px-3 py-1 text-xs font-semibold text-blue-600">
+                  <Wand2 className="size-3" />
+                  技能维护工具
+                </span>
+                <h1 className="flex flex-wrap items-center gap-3 text-2xl font-bold tracking-tight text-slate-800 md:text-3xl">
+                  技能地址管理
+                  <span className="rounded-lg border border-slate-200/60 bg-slate-100 px-2.5 py-0.5 font-mono text-xs font-medium text-slate-500">
+                    {groups.length} 地址 / {totalOccurrences} 处
+                    {scan ? ` / ${scan.scannedSkills} 技能` : ""}
+                  </span>
+                </h1>
+              </div>
+
+              <div className="flex items-center gap-2.5">
+                <button
+                  type="button"
+                  onClick={runRollback}
+                  disabled={!canRollback || scanning || applying}
+                  className="inline-flex cursor-pointer items-center gap-2 rounded-xl border border-slate-200 bg-white px-3.5 py-2 text-sm font-semibold text-slate-600 shadow-sm transition-all hover:border-amber-300 hover:bg-amber-50 hover:text-amber-700 disabled:cursor-not-allowed disabled:opacity-50"
+                  title="从最近一次的磁盘备份恢复"
+                >
+                  <Undo2 className="size-[15px]" />
+                  撤销上次
+                </button>
+                <button
+                  type="button"
+                  onClick={() => runScan(kinds)}
+                  disabled={scanning || applying}
+                  className="inline-flex cursor-pointer items-center gap-2 rounded-xl border border-blue-600 bg-blue-600 px-5 py-2 text-sm font-semibold text-white shadow-sm transition-all hover:bg-blue-700 active:scale-95 disabled:cursor-not-allowed disabled:opacity-50"
+                >
+                  {scanning ? <Loader2 className="size-[15px] animate-spin" /> : <ScanLine className="size-[15px]" />}
+                  {scanning ? "扫描中…" : "重新扫描"}
+                </button>
+              </div>
+            </div>
+
+            <p className="flex max-w-3xl items-start gap-2.5 rounded-2xl border border-slate-200/50 bg-slate-50/70 p-3.5 text-sm leading-relaxed text-slate-600">
+              <Info className="mt-0.5 size-4 shrink-0 text-slate-400" />
+              <span>
+                扫描运行中的技能（<code className="rounded bg-slate-100 px-1">skills/public</code> 与{" "}
+                <code className="rounded bg-slate-100 px-1">skills/custom</code> 下所有 .md / .py），按<strong>唯一地址</strong>聚合其全部出现位置（跨 .md 与 .py）。填写「替换为」后<strong>预览</strong>命中再<strong>应用</strong>；应用前会自动磁盘备份，可一键<strong>撤销</strong>。替换为精确整串匹配，不会误伤子串。
+              </span>
+            </p>
+          </div>
+        </div>
+
+        {/* 进度条 */}
+        {scanning && scanProg && (
+          <div className="mb-6">
+            <ProgressBar done={scanProg.done} total={scanProg.total} label={`正在扫描技能${scanProg.skill ? `：${scanProg.skill}` : "…"}`} />
+          </div>
+        )}
+        {applying && applyProg && (
+          <div className="mb-6">
+            <ProgressBar done={applyProg.done} total={applyProg.totalFiles} label={`正在修改文件${applyProg.file ? `：${applyProg.file}` : "…"}`} />
+          </div>
+        )}
+
+        {/* 工具栏：类型过滤 + 搜索 + 片段批量套用 */}
+        <div className="mb-6 space-y-4 rounded-2xl border border-slate-200/60 bg-white p-4 shadow-sm md:p-5">
+          <div className="flex flex-col items-start justify-between gap-4 md:flex-row md:items-center">
+            <div className="flex items-center gap-2">
+              {ALL_KINDS.map((k) => (
+                <button
+                  key={k}
+                  type="button"
+                  onClick={() => toggleKind(k)}
+                  disabled={scanning || applying}
+                  className={`inline-flex items-center gap-1.5 rounded-lg border px-3 py-1.5 text-sm font-medium transition-all disabled:opacity-50 ${
+                    kinds.includes(k)
+                      ? "border-blue-300 bg-blue-50 text-blue-700"
+                      : "border-slate-200 bg-white text-slate-500 hover:border-slate-300"
+                  }`}
+                >
+                  {k === "url" ? <Globe className="size-3.5" /> : <Network className="size-3.5" />}
+                  {k === "url" ? "URL" : "IP"}
+                </button>
+              ))}
+            </div>
+            <div className="relative w-full md:w-80">
+              <Search className="absolute left-3.5 top-1/2 size-4 -translate-y-1/2 text-slate-400" />
+              <input
+                type="text"
+                placeholder="搜索地址 / 技能名…"
+                value={query}
+                onChange={(e) => setQuery(e.target.value)}
+                className="w-full rounded-xl border border-slate-200 bg-slate-50/50 py-2 pl-10 pr-4 text-sm transition-all hover:border-slate-300 focus:border-blue-500 focus:outline-none focus:ring-4 focus:ring-blue-500/5"
+              />
+            </div>
+          </div>
+
+          {/* 片段批量套用 */}
+          <div className="flex flex-col items-stretch gap-2.5 rounded-xl border border-dashed border-slate-200 bg-slate-50/50 p-3 md:flex-row md:items-center">
+            <span className="flex shrink-0 items-center gap-1.5 text-xs font-semibold text-slate-500">
+              <Wand2 className="size-3.5" />
+              片段批量套用
+            </span>
+            <input
+              type="text"
+              placeholder="查找片段，如 uat.4.cn"
+              value={segFind}
+              onChange={(e) => setSegFind(e.target.value)}
+              className="min-w-0 flex-1 rounded-lg border border-slate-200 bg-white px-3 py-1.5 font-mono text-sm focus:border-blue-500 focus:outline-none focus:ring-2 focus:ring-blue-500/10"
+            />
+            <span className="shrink-0 text-slate-400">→</span>
+            <input
+              type="text"
+              placeholder="替换为，如 prod.4.cn"
+              value={segRepl}
+              onChange={(e) => setSegRepl(e.target.value)}
+              className="min-w-0 flex-1 rounded-lg border border-slate-200 bg-white px-3 py-1.5 font-mono text-sm focus:border-blue-500 focus:outline-none focus:ring-2 focus:ring-blue-500/10"
+            />
+            <button
+              type="button"
+              onClick={applySegment}
+              className="inline-flex shrink-0 cursor-pointer items-center justify-center gap-1.5 rounded-lg border border-slate-200 bg-white px-3.5 py-1.5 text-sm font-semibold text-slate-700 transition-all hover:border-blue-300 hover:bg-blue-50 hover:text-blue-600"
+            >
+              套用到匹配项
+            </button>
+          </div>
+        </div>
+
+        {/* 预览面板 */}
+        {preview && (
+          <div className="mb-6 overflow-hidden rounded-2xl border border-blue-200/70 bg-white shadow-sm">
+            <div className="flex flex-col items-start justify-between gap-3 border-b border-blue-100 bg-blue-50/50 px-4 py-3 md:flex-row md:items-center">
+              <div className="text-sm font-semibold text-slate-700">
+                预览：将改动 <span className="font-mono text-blue-700">{preview.total}</span> 处
+                （{preview.items.filter((i) => i.count > 0).length} 个地址命中）
+              </div>
+              <button
+                type="button"
+                onClick={runApply}
+                disabled={applying || preview.total === 0}
+                className="inline-flex cursor-pointer items-center gap-2 rounded-xl border border-blue-600 bg-blue-600 px-5 py-2 text-sm font-semibold text-white shadow-sm transition-all hover:bg-blue-700 active:scale-95 disabled:cursor-not-allowed disabled:opacity-50"
+              >
+                {applying ? <Loader2 className="size-[15px] animate-spin" /> : <Replace className="size-[15px]" />}
+                {applying ? "应用中…" : "确认应用"}
+              </button>
+            </div>
+            {preview.warnings.length > 0 && (
+              <div className="border-b border-amber-100 bg-amber-50/60 px-4 py-2.5">
+                {preview.warnings.map((w, i) => (
+                  <div key={i} className="flex items-center gap-1.5 text-xs text-amber-700">
+                    <AlertTriangle className="size-3.5 shrink-0" />
+                    {w}
+                  </div>
+                ))}
+              </div>
+            )}
+            <ul className="max-h-72 divide-y divide-slate-100 overflow-auto">
+              {preview.items.map((it) => (
+                <li key={it.from} className="px-4 py-2.5 text-sm">
+                  <div className="flex items-center gap-2">
+                    <KindBadge kind={it.kind} />
+                    <span className="font-mono text-slate-500 line-through">{it.from}</span>
+                    <span className="text-slate-400">→</span>
+                    <span className="font-mono font-medium text-blue-700">{it.to}</span>
+                    <span className="ml-auto rounded bg-slate-100 px-1.5 py-0.5 font-mono text-xs text-slate-500">{it.count} 处</span>
+                  </div>
+                </li>
+              ))}
+            </ul>
+          </div>
+        )}
+
+        {/* 地址列表 */}
+        <div className="overflow-hidden rounded-2xl border border-slate-200/60 bg-white shadow-sm">
+          <div className="flex items-center justify-between border-b border-slate-200/60 bg-slate-50/70 px-4 py-2.5 text-xs font-semibold text-slate-500">
+            <span>地址（{visible.length}）</span>
+            {pendingCount > 0 && (
+              <button
+                type="button"
+                onClick={runPreview}
+                disabled={scanning || applying}
+                className="inline-flex cursor-pointer items-center gap-1.5 rounded-lg border border-blue-300 bg-blue-50 px-3 py-1 text-xs font-semibold text-blue-700 transition-all hover:bg-blue-100 disabled:opacity-50"
+              >
+                <Eye className="size-3.5" />
+                预览改动（{pendingCount}）
+              </button>
+            )}
+          </div>
+
+          {scanning && !scan ? (
+            <div className="px-4 py-12 text-center text-sm text-slate-400">扫描中…</div>
+          ) : visible.length === 0 ? (
+            <div className="px-4 py-12 text-center text-sm text-slate-400">
+              {groups.length === 0 ? "未发现任何 URL / IP 地址" : "没有匹配的地址"}
+            </div>
+          ) : (
+            <ul className="divide-y divide-slate-100">
+              {visible.map((g) => (
+                <GroupRow
+                  key={g.address}
+                  group={g}
+                  value={edits[g.address] ?? ""}
+                  expanded={expanded.has(g.address)}
+                  onToggle={() => toggleExpand(g.address)}
+                  onChange={(v) => setEdit(g.address, v)}
+                />
+              ))}
+            </ul>
+          )}
+        </div>
+      </div>
+    </div>
+  );
+}
+
+function GroupRow({
+  group,
+  value,
+  expanded,
+  onToggle,
+  onChange,
+}: {
+  group: AddressGroup;
+  value: string;
+  expanded: boolean;
+  onToggle: () => void;
+  onChange: (v: string) => void;
+}) {
+  const dirty = value.trim() !== "" && value.trim() !== group.address;
+  return (
+    <li className={`px-4 py-3 transition-colors hover:bg-slate-50/60 ${dirty ? "bg-blue-50/30" : ""}`}>
+      <div className="flex flex-col gap-2.5 lg:flex-row lg:items-center">
+        <div className="flex min-w-0 flex-1 items-center gap-2">
+          <button type="button" onClick={onToggle} className="flex size-6 shrink-0 items-center justify-center rounded text-slate-400 hover:bg-slate-100 hover:text-slate-600">
+            {expanded ? <ChevronDown className="size-4" /> : <ChevronRight className="size-4" />}
+          </button>
+          <KindBadge kind={group.kind} />
+          <span className="truncate font-mono text-sm text-slate-700" title={group.address}>
+            {group.address}
+          </span>
+          <span className="shrink-0 rounded bg-slate-100 px-1.5 py-0.5 font-mono text-xs text-slate-500">{group.count} 处</span>
+        </div>
+        <input
+          type="text"
+          value={value}
+          placeholder="替换为（留空 = 不改）"
+          onChange={(e) => onChange(e.target.value)}
+          className="w-full rounded-lg border border-slate-200 bg-white px-3 py-1.5 font-mono text-sm transition-all focus:border-blue-500 focus:outline-none focus:ring-2 focus:ring-blue-500/10 lg:w-96"
+        />
+      </div>
+
+      {expanded && (
+        <ul className="mt-2.5 ml-8 space-y-1 border-l border-slate-100 pl-3">
+          {group.occurrences.map((o, i) => (
+            <li key={i} className="flex items-start gap-2 text-xs text-slate-500">
+              <FileCode className="mt-0.5 size-3.5 shrink-0 text-slate-400" />
+              <div className="min-w-0">
+                <span className="font-medium text-slate-600">{o.skill}</span>
+                <span className="mx-1 text-slate-300">·</span>
+                <span className="font-mono">
+                  {o.file}:{o.line}
+                </span>
+                <div className="truncate font-mono text-slate-400" title={o.snippet}>
+                  {o.snippet}
+                </div>
+              </div>
+            </li>
+          ))}
+        </ul>
+      )}
+    </li>
+  );
+}
diff --git a/frontend-web/src/pages/StrategyRoutes.tsx b/frontend-web/src/pages/StrategyRoutes.tsx
index 0bab56b0..44b0883d 100644
--- a/frontend-web/src/pages/StrategyRoutes.tsx
+++ b/frontend-web/src/pages/StrategyRoutes.tsx
@@ -40,6 +40,7 @@ import LightAppListPage from "./LightAppListPage";
 import LightAppManagementPage from "./LightAppManagementPage";
 import MenuManagementPage from "./MenuManagementPage";
 import SensitiveWordsPage from "./SensitiveWordsPage";
+import SkillAddressesPage from "./SkillAddressesPage";
 import TaskButtonsPage from "./TaskButtonsPage";
 import LightAppNewNotePage from "./LightAppNewNotePage";
 import LightAppNoteListPage from "./LightAppNoteListPage";
@@ -393,6 +394,16 @@ export default function StrategyRoutes() {
                 )
               }
             />
+            <Route
+              path="admin/skill-addresses"
+              element={
+                getStoredAuth()?.system_role === "admin" ? (
+                  <SkillAddressesPage />
+                ) : (
+                  <Navigate to="/page/workspace/chats/new" replace />
+                )
+              }
+            />
             <Route
               path="admin/scheduled-tasks"
               element={
diff --git a/frontend-web/src/strategy-components/api/skill-addresses.ts b/frontend-web/src/strategy-components/api/skill-addresses.ts
new file mode 100644
index 00000000..d52582a0
--- /dev/null
+++ b/frontend-web/src/strategy-components/api/skill-addresses.ts
@@ -0,0 +1,272 @@
+/**
+ * 技能地址(URL/IP)批量替换 API 客户端 + React Query hooks。
+ *
+ * 后端：`/api/skill-addresses`（见 app/gateway/routers/skill_addresses.py）。
+ * 全部仅管理员可用。扫描/应用提供 **NDJSON 流式进度**端点（150+ 技能时显示
+ * 「已处理 X / 共 N」进度条），并保留一次性端点作为兜底。
+ * snake_case ⇄ camelCase 在此边界收敛。
+ */
+
+import { useQuery } from "@tanstack/react-query";
+
+import { apiFetch } from "@/core/api/fetch-client";
+import { getBackendBaseURL } from "@/core/config";
+
+const BASE = () => `${getBackendBaseURL()}/api/skill-addresses`;
+
+export type AddressKind = "url" | "ip";
+
+export interface AddressOccurrence {
+  skill: string;
+  category: string; // "public" | "custom"
+  file: string; // 相对技能根的 POSIX 路径
+  line: number;
+  col: number;
+  snippet: string;
+}
+
+export interface AddressGroup {
+  address: string;
+  kind: AddressKind;
+  count: number;
+  occurrences: AddressOccurrence[];
+}
+
+export interface ScanResult {
+  groups: AddressGroup[];
+  scannedFiles: number;
+  scannedSkills: number;
+  kinds: string[];
+  scannedAt: string;
+}
+
+export interface Replacement {
+  from: string;
+  to: string;
+  kind?: AddressKind;
+}
+
+export interface PreviewFileHit {
+  file: string;
+  line: number;
+  before: string;
+  after: string;
+}
+
+export interface PreviewItem {
+  from: string;
+  to: string;
+  kind: AddressKind;
+  count: number;
+  files: PreviewFileHit[];
+}
+
+export interface PreviewResult {
+  items: PreviewItem[];
+  total: number;
+  warnings: string[];
+  expectedTotal: number;
+}
+
+export interface ChangedFile {
+  file: string;
+  count: number;
+}
+
+export interface ApplyResult {
+  changedFiles: ChangedFile[];
+  replacedTotal: number;
+  backupId: string | null;
+  warnings: string[];
+}
+
+export interface BackupEntry {
+  backupId: string;
+  files: number;
+  createdAt: string;
+}
+
+/** 扫描进度（逐技能）。 */
+export interface ScanProgress {
+  done: number;
+  total: number;
+  skill?: string;
+}
+
+/** 应用进度（逐文件）。 */
+export interface ApplyProgress {
+  done: number;
+  totalFiles: number;
+  file?: string;
+}
+
+// ── 错误处理 ───────────────────────────────────────────────────────────────
+
+async function ensureOk(res: Response, ctx: string): Promise<void> {
+  if (res.ok) return;
+  const data = await res.json().catch(() => null as unknown);
+  const detail =
+    data && typeof data === "object" && "detail" in data
+      ? String((data as { detail: unknown }).detail)
+      : `${ctx} 失败 (${res.status})`;
+  throw new Error(detail);
+}
+
+async function readJson<T>(res: Response, ctx: string): Promise<T> {
+  await ensureOk(res, ctx);
+  return (await res.json()) as T;
+}
+
+// ── 序列化辅助 ─────────────────────────────────────────────────────────────
+
+function toScanResult(d: any): ScanResult {
+  return {
+    groups: (d.groups ?? []) as AddressGroup[],
+    scannedFiles: d.scanned_files ?? 0,
+    scannedSkills: d.scanned_skills ?? 0,
+    kinds: d.kinds ?? [],
+    scannedAt: d.scanned_at ?? "",
+  };
+}
+
+function toPreviewResult(d: any): PreviewResult {
+  return {
+    items: (d.items ?? []).map((i: any) => ({
+      from: i.from,
+      to: i.to,
+      kind: i.kind,
+      count: i.count ?? 0,
+      files: i.files ?? [],
+    })),
+    total: d.total ?? 0,
+    warnings: d.warnings ?? [],
+    expectedTotal: d.expected_total ?? 0,
+  };
+}
+
+function toApplyResult(d: any): ApplyResult {
+  return {
+    changedFiles: (d.changed_files ?? []).map((c: any) => ({ file: c.file, count: c.count })),
+    replacedTotal: d.replaced_total ?? 0,
+    backupId: d.backup_id ?? null,
+    warnings: d.warnings ?? [],
+  };
+}
+
+// ── NDJSON 流式读取 ────────────────────────────────────────────────────────
+
+/** 逐行解析 NDJSON 流，对每个事件对象调用 onEvent。 */
+async function readNdjson(res: Response, ctx: string, onEvent: (ev: any) => void): Promise<void> {
+  await ensureOk(res, ctx);
+  if (!res.body) throw new Error(`${ctx}：响应无数据流`);
+  const reader = res.body.getReader();
+  const decoder = new TextDecoder();
+  let buf = "";
+  const flush = (chunk: string) => {
+    buf += chunk;
+    let idx: number;
+    while ((idx = buf.indexOf("\n")) >= 0) {
+      const line = buf.slice(0, idx).trim();
+      buf = buf.slice(idx + 1);
+      if (line) onEvent(JSON.parse(line));
+    }
+  };
+  for (;;) {
+    const { done, value } = await reader.read();
+    if (done) break;
+    flush(decoder.decode(value, { stream: true }));
+  }
+  const tail = (buf + decoder.decode()).trim();
+  if (tail) onEvent(JSON.parse(tail));
+}
+
+// ── 端点 ───────────────────────────────────────────────────────────────────
+
+/** 一次性扫描（兜底，无进度）。 */
+export async function scanAddresses(kinds?: AddressKind[], skill?: string): Promise<ScanResult> {
+  const params = new URLSearchParams();
+  if (kinds && kinds.length) params.set("kinds", kinds.join(","));
+  if (skill) params.set("skill", skill);
+  const qs = params.toString();
+  const res = await apiFetch(`${BASE()}/scan${qs ? `?${qs}` : ""}`, { method: "GET" });
+  return toScanResult(await readJson<any>(res, "扫描"));
+}
+
+/** 流式扫描：逐技能上报进度，返回最终聚合结果。 */
+export async function scanAddressesStream(
+  opts: { kinds?: AddressKind[]; skill?: string; onProgress?: (p: ScanProgress) => void } = {},
+): Promise<ScanResult> {
+  const params = new URLSearchParams();
+  if (opts.kinds && opts.kinds.length) params.set("kinds", opts.kinds.join(","));
+  if (opts.skill) params.set("skill", opts.skill);
+  const qs = params.toString();
+  const res = await apiFetch(`${BASE()}/scan/stream${qs ? `?${qs}` : ""}`, { method: "GET" });
+  let result: ScanResult | null = null;
+  await readNdjson(res, "扫描", (ev) => {
+    if (ev.type === "start") opts.onProgress?.({ done: 0, total: ev.total ?? 0 });
+    else if (ev.type === "progress") opts.onProgress?.({ done: ev.done, total: ev.total, skill: ev.skill });
+    else if (ev.type === "result") result = toScanResult(ev);
+  });
+  if (!result) throw new Error("扫描未返回结果");
+  return result;
+}
+
+/** 干跑预览。 */
+export async function previewReplacements(replacements: Replacement[]): Promise<PreviewResult> {
+  const res = await apiFetch(`${BASE()}/preview`, {
+    method: "POST",
+    headers: { "Content-Type": "application/json" },
+    body: JSON.stringify({ replacements }),
+  });
+  return toPreviewResult(await readJson<any>(res, "预览"));
+}
+
+/** 流式应用：逐文件上报进度，返回最终结果。`expectedTotal` 做乐观锁。 */
+export async function applyReplacementsStream(
+  replacements: Replacement[],
+  opts: { expectedTotal?: number; backup?: boolean; onProgress?: (p: ApplyProgress) => void } = {},
+): Promise<ApplyResult> {
+  const body = {
+    replacements,
+    expected_total: opts.expectedTotal ?? null,
+    backup: opts.backup ?? true,
+  };
+  const res = await apiFetch(`${BASE()}/apply/stream`, {
+    method: "POST",
+    headers: { "Content-Type": "application/json" },
+    body: JSON.stringify(body),
+  });
+  let result: ApplyResult | null = null;
+  await readNdjson(res, "应用", (ev) => {
+    if (ev.type === "start") opts.onProgress?.({ done: 0, totalFiles: ev.total_files ?? 0 });
+    else if (ev.type === "progress") opts.onProgress?.({ done: ev.done, totalFiles: ev.total_files, file: ev.file });
+    else if (ev.type === "result") result = toApplyResult(ev);
+  });
+  if (!result) throw new Error("应用未返回结果");
+  return result;
+}
+
+export async function listBackups(): Promise<BackupEntry[]> {
+  const res = await apiFetch(`${BASE()}/backups`, { method: "GET" });
+  const data = await readJson<{ backups: any[] }>(res, "读取备份");
+  return (data.backups ?? []).map((b) => ({ backupId: b.backup_id, files: b.files ?? 0, createdAt: b.created_at ?? "" }));
+}
+
+export async function rollbackBackup(backupId: string): Promise<{ restored: number }> {
+  const res = await apiFetch(`${BASE()}/rollback/${encodeURIComponent(backupId)}`, { method: "POST" });
+  const data = await readJson<{ restored_files: string[] }>(res, "撤销");
+  return { restored: (data.restored_files ?? []).length };
+}
+
+// ── React Query ────────────────────────────────────────────────────────────
+
+export const SKILL_ADDRESS_BACKUPS_KEY = "skill-address-backups";
+
+export function useSkillAddressBackups() {
+  return useQuery({
+    queryKey: [SKILL_ADDRESS_BACKUPS_KEY],
+    queryFn: listBackups,
+    refetchOnWindowFocus: false,
+    staleTime: 30_000,
+  });
+}
diff --git a/frontend-web/src/strategy-components/components/task-components/TaskFxWorkspace.tsx b/frontend-web/src/strategy-components/components/task-components/TaskFxWorkspace.tsx
index 2a3fb8c6..f21ba0b1 100644
--- a/frontend-web/src/strategy-components/components/task-components/TaskFxWorkspace.tsx
+++ b/frontend-web/src/strategy-components/components/task-components/TaskFxWorkspace.tsx
@@ -343,7 +343,10 @@ function RwfxSqReportGate({ taskId }: { taskId: string }) {
             暂不
           </Button>
           <Button
-            className="bg-[var(--color-main)] text-white hover:bg-[var(--brand-950)]"
+            // 弹框经 Radix Portal 渲染到 body，在 `.strategy-data` 作用域之外，故不能用
+            // 局部的 `--color-main`（红主题侥幸有值、蓝主题失效→按钮消失）。改用全局按主题
+            // 定义的品牌色 token（:root/.dark/.lightblue 都有），三套主题都能正确取色。
+            className="bg-[var(--workspace-brand)] text-[var(--workspace-brand-foreground)] hover:opacity-90"
             onClick={handleGo3qfx}
             disabled={jumping}
           >
diff --git a/offline-backend-20260512/backend/CLAUDE.md b/offline-backend-20260512/backend/CLAUDE.md
index c6715a5e..df32030d 100644
--- a/offline-backend-20260512/backend/CLAUDE.md
+++ b/offline-backend-20260512/backend/CLAUDE.md
@@ -243,7 +243,7 @@ FastAPI application on port 8001 with health check at `GET /health`. Set `GATEWA
 | **Thread Runs** (`/api/threads/{id}/runs`) | `POST /` - create background run; `POST /stream` - create + SSE stream; `POST /wait` - create + block; `GET /` - list runs; `GET /{rid}` - run details; `POST /{rid}/cancel` - cancel; `GET /{rid}/join` - join SSE; `GET /{rid}/messages` - paginated messages `{data, has_more}`; `GET /{rid}/events` - full event stream; `GET /../messages` - thread messages with feedback; `GET /../token-usage` - aggregate tokens |
 | **Feedback** (`/api/threads/{id}/runs/{rid}/feedback`) | `PUT /` - upsert feedback; `DELETE /` - delete user feedback; `POST /` - create feedback; `GET /` - list feedback; `GET /stats` - aggregate stats; `DELETE /{fid}` - delete specific |
 | **Runs** (`/api/runs`) | `POST /stream` - stateless run + SSE; `POST /wait` - stateless run + block; `GET /{rid}/messages` - paginated messages by run_id `{data, has_more}` (cursor: `after_seq`/`before_seq`); `GET /{rid}/feedback` - list feedback by run_id |
-| **Open Chat** (`/api/open/chat`) | **Unauthenticated** open Q&A entry for external systems (the `/api/open/` prefix is whitelisted in `auth_middleware._PUBLIC_PATH_PREFIXES` + CSRF-exempt in `csrf_middleware.should_check_csrf`). `POST /chat` `{message \| messages[], model_name?, agent_name?, agent_id?, thinking_enabled=false, strip_think=true, thread_id?, recursion_limit=100}` → resolves the agent by **Chinese display name** (`agent_name`, matched against each `.deer-flow/agents/*/config.yaml` `name` via `_resolve_agent_id`; explicit `agent_id` wins if given), builds the lead agent via `make_lead_agent` (so the chosen **agent's** SOUL/tools/skills all apply), `ainvoke`s it once (stateless — no checkpointer, no thread persistence; runs under the `"default"` user bucket), and returns **only the final** `{answer, model_name, agent_name, agent_id, thinking_enabled, thread_id}` (no intermediate tool-call/process). `GET /agents` lists `{name, id}` so callers know what `agent_name` to send. **Thinking off is reliable**: when `thinking_enabled=false` the route also sets `thinking_force_disabled=true` so the lead agent calls `create_chat_model(force_disable_thinking=True)`, which injects `extra_body.chat_template_kwargs.enable_thinking=false` (+ GLM `thinking.type="disabled"`) even for internal vLLM/Qwen models that only declare `supports_thinking: true` — the same trick `TitleMiddleware._force_disable_thinking` uses for title generation. Sets `is_scheduled_run=true` internally to skip the auto-title LLM call. Router `app/gateway/routers/open_chat.py`; standalone caller `test_open_chat_api.py` (repo root); tests `tests/test_open_chat.py`. |
+| **Open Chat** (`/api/open/chat`) | **Unauthenticated** open Q&A entry for external systems (the `/api/open/` prefix is whitelisted in `auth_middleware._PUBLIC_PATH_PREFIXES` + CSRF-exempt in `csrf_middleware.should_check_csrf`). `POST /chat` `{message \| messages[], model_name?, agent_name?, agent_id?, thinking_enabled=false, strip_think=true, thread_id?, recursion_limit=100}` → resolves the agent by **Chinese display name** (`agent_name`, matched against each `.deer-flow/agents/*/config.yaml` `name` via `_resolve_agent_id`; explicit `agent_id` wins if given), builds the lead agent via `make_lead_agent` (so the chosen **agent's** SOUL/tools/skills all apply), `ainvoke`s it once (stateless — no checkpointer, no thread persistence; runs under the `"default"` user bucket), and returns **only the final** `{answer, model_name, agent_name, agent_id, thinking_enabled, thread_id}` (no intermediate tool-call/process). **`POST /chat/stream`** is the **SSE streaming variant** (same request body) for **long tasks** — the blocking `/chat` produces no downstream bytes for the whole run, so a front nginx `proxy_read_timeout` (default 60s) cuts it off with a **504**; the stream keeps the connection warm (a `: keepalive` SSE comment every `_HEARTBEAT_SECONDS=15`s via a background producer-task + queue, plus `event: delta` AI-text increments and `event: progress`), and ends with **`event: result`** carrying the same final `{answer, …}` JSON (or `event: error`). Clients only need to read `event: result`; intermediate events are ignorable. Both endpoints share `_build_run_context` (config/agent/thread). `GET /agents` lists `{name, id}` so callers know what `agent_name` to send. **Thinking off is reliable**: when `thinking_enabled=false` the route also sets `thinking_force_disabled=true` so the lead agent calls `create_chat_model(force_disable_thinking=True)`, which injects **three** OpenAI-compatible "thinking off" shapes into `extra_body` — top-level `enable_thinking=false` (**SiliconFlow / DashScope hybrid models** like `deepseek-ai/DeepSeek-V4-*` on `api.siliconflow.cn` — the knob the nested `chat_template_kwargs` shape does NOT reach), `chat_template_kwargs.enable_thinking=false` (vLLM/Qwen native), and `thinking.type="disabled"` (Zhipu GLM) — even for models that only declare `supports_thinking: true`. For the **web-UI path** (which sends `thinking_enabled=false` but NOT `force_disable`), SiliconFlow models need `when_thinking_disabled: {extra_body: {enable_thinking: false}}` in their `config.yaml` entry (added to `deepseek-ai/DeepSeek-V4-Flash`) for thinking-off to take effect. Sets `is_scheduled_run=true` internally to skip the auto-title LLM call. Router `app/gateway/routers/open_chat.py`; standalone caller `test_open_chat_api.py` (repo root, streaming by default); tests `tests/test_open_chat.py`. |
 | **Thread Shares** (`/api/thread-shares`) | Per-conversation sharing with two modes (`thread_shares.mode`). **`import`**: `POST /` `{thread_id, mode}` - create (or reuse) a revocable share code; `GET /{code}/preview` + `POST /{code}/import` - copy that conversation into the caller's account (fresh `thread_id` via `app/gateway/thread_copy.py` + new `threads_meta` row + `user-data` copy; dedup on `metadata.shared_origin`). **`view`**: a public, read-only static page ? see Public Shares below. Both modes: `GET /` - list my codes; `DELETE /{code}` - revoke. |
 | **Public Shares** (`/api/public/shares`) | **Unauthenticated** (the `/api/public/` prefix is whitelisted in `auth_middleware._PUBLIC_PATH_PREFIXES`). `GET /{code}` - read-only conversation snapshot (title + serialized messages + artifacts) for a `view`-mode share; messages are read straight from the latest checkpoint. `GET /{code}/files/{path}` - serves a file referenced by that conversation (images, Markdown, text, PDF, ?) so the shared page's links are openable; path resolution is confined to the owner's thread directory. Active content (`text/html`, `application/xhtml+xml`, `image/svg+xml`) is forced to download (`_public_share_disposition`); everything else is served inline. Missing/revoked/non-`view` codes all return 404. |
 | **Roundtable Drafts** (`/api/roundtable-drafts`) | **Strictly per-user** persistence for the roundtable-planning page (replaces the old browser `localStorage`). `GET /` - list drafts (lightweight meta); `POST /` - create; `GET/PUT/DELETE /{id}` - full read / partial update / delete (all owner-scoped). A draft = one full session (Step 1 intent + Step 2 roundtable), step snapshots stored as JSON in `PortableLongText`. Recommendation history bound to a draft: `GET /{id}/recommendations` - list newest-first; `POST /{id}/recommendations` - append one run (objective/status/rationale/picks/candidates). taskId 深链会商记录 are **not** here — they live in the separate `/api/roundtable-task-drafts` store (see below) so personal history never mixes with task records. Backed by `deerflow.persistence.roundtable_drafts` (tables `roundtable_drafts`, `roundtable_recommend_history`); store wired in `deps.py` as `app.state.roundtable_draft_store`. Tests: `tests/test_roundtable_task_drafts.py`. |
@@ -527,6 +527,45 @@ Operators upgrading an existing DB run the whole file (`USE <db>;` then execute)
 - **Installation**: `POST /api/skills/install` extracts .skill ZIP archive to custom/ directory
 - **Chinese name (`name_zh`)**: a Chinese display name stored in the **`skills` DB table** (`SkillRow.name_zh`, `String(128)`, migration `20260609_01`) ? never written to the on-disk SKILL.md, so it works for built-in skills too. Threaded through `SkillStore` create/ensure_legacy/update/update_admin exactly like `detail`; surfaced on `SkillResponse.name_zh` from the ownership record in `_skill_to_response`. Authorization mirrors `detail`: custom skill ? owner or admin; built-in/public ? admin only (row materialized on demand via `ensure_legacy`). Endpoints (in `app/gateway/routers/skills.py`): `PUT /api/skills/{name}/name-zh` (set/clear), `POST /api/skills/{name}/name-zh/generate` (LLM generates one), `POST /api/skills/name-zh/generate-missing` (????: **admin-only** bulk-generate for every skill that lacks one, bounded-concurrency `asyncio.gather`). **Optional model selection**: both generate endpoints accept an optional JSON body `{model_name}` ? an explicit name wins, else the curator model setting, else the main model (`_resolve_skill_name_zh_model_name`). **Model-error surfacing**: `_generate_skill_name_zh` raises `SkillNameZhModelError` when the model itself fails (build/timeout/API error) ? distinct from the model returning an unusable name (? `None`). The single endpoint maps that to `502 ?????,?????????:?`; the bulk endpoint fails fast on a model-build error (502) and otherwise carries a representative `model_error` string in `SkillNameZhBulkResponse` so the UI can say "N ?????????". The frontend (`skill-settings-page.tsx` list + `skill-detail-dialog.tsx` editor, plus the agent skill pickers/badges in `NewAgentPage.tsx` + `agent-card.tsx`) shows `name_zh || name`, offers per-skill "AI ??" plus an **admin-only** header "???????" button, and a compact model dropdown (`components/workspace/model-name-select.tsx`, default = configured model) beside each action.
 
+#### Skill Address Management (技能地址管理 — URL/IP 批量替换)
+
+Admin-only maintenance tool to find every `http(s)://` URL and IPv4 (`ip[:port]`)
+address in the **running** skills (`skills/{public,custom}`, all `.md` + `.py`,
+recursing into `scripts/`/`references/`/`templates/`) and rewrite them in bulk —
+e.g. swapping a UAT endpoint for production across all skills at once. The same
+address commonly lives in both a skill's `SKILL.md` and its `scripts/*.py`, so
+results are **aggregated by unique address** with every occurrence (`file:line` +
+snippet) and replaced together.
+
+- **Pure engine** — `deerflow/skills/address_scan.py` (harness, no `app.*`):
+  `find_spans` / `scan_occurrences` (URL regex with trailing-punctuation strip;
+  IPv4 with 0–255 octet check + optional port; an IP inside a URL is *not*
+  separately surfaced), `apply_replacements` (**substring-safe**: re-detects exact
+  token spans and only rewrites a span whose text equals a requested `from`, so
+  `1.2.3.4` is never rewritten inside `11.2.3.40`), `infer_kind`,
+  `is_valid_replacement`. Tested in `tests/test_skill_addresses.py`.
+- **Router** — `app/gateway/routers/skill_addresses.py` (`/api/skill-addresses`,
+  **admin only**). **Availability at scale (150+ skills):** every filesystem
+  walk/read/write runs via `asyncio.to_thread` so the event loop is never blocked.
+  Routes: `GET /scan` (aggregated, one-shot) + `GET /scan/stream` (per-skill NDJSON
+  progress `{type:start|progress|result}`); `POST /preview` (dry-run diff +
+  validation warnings); `POST /apply` + `POST /apply/stream` (per-file NDJSON
+  progress); `GET /backups`; `POST /rollback/{backup_id}`. **Apply safety:**
+  optimistic lock via `expected_total` (409 if files changed since the scan),
+  path-guard (`is_relative_to` the skills root), auto disk backup to
+  `{skills_root}/.address-edits/{timestamp}/` (the `.`-prefixed dir is excluded
+  from scans), then `refresh_skills_system_prompt_cache_async()` so edits take
+  effect immediately. Registered in `app.py` after `sensitive_words.router`.
+- **Frontend** — admin page `pages/SkillAddressesPage.tsx` (route
+  `/page/strategy/admin/skill-addresses`, entry「技能地址管理」in the bottom-left
+  settings dropdown of `components/page-sidebar.tsx`, both full + collapsed modes):
+  auto-scan on open with a **progress bar** (扫描逐技能 / 应用逐文件 X/N), URL/IP
+  type filter, search, a「片段批量套用」helper (e.g. all `uat.4.cn` → `prod.4.cn`),
+  per-address「替换为」input with expandable occurrences, inline 预览 panel, 应用,
+  and 撤销上次 (rollback). Client + NDJSON stream parsing in
+  `strategy-components/api/skill-addresses.ts`. Scope is the running skills only;
+  the `deploy/minimal/skills/` copy is not touched (sync separately).
+
 #### Skill Compression (技能压缩)
 
 When the skill catalogue grows large, injecting every enabled skill full-text
diff --git a/offline-backend-20260512/backend/app/gateway/app.py b/offline-backend-20260512/backend/app/gateway/app.py
index 995f76b3..65b035b6 100644
--- a/offline-backend-20260512/backend/app/gateway/app.py
+++ b/offline-backend-20260512/backend/app/gateway/app.py
@@ -69,6 +69,7 @@ from app.gateway.routers import (
     runs,
     scheduled_tasks,
     sensitive_words,
+    skill_addresses,
     skills,
     suggestions,
     system_settings,
@@ -1359,6 +1360,9 @@ This gateway provides custom endpoints for models, MCP configuration, skills, an
     # Sensitive-word map (敏感词管理 / 脱敏词) at /api/sensitive-words
     app.include_router(sensitive_words.router)
 
+    # Skill URL/IP batch replace (技能地址管理) at /api/skill-addresses
+    app.include_router(skill_addresses.router)
+
     # 大屏绘制智能体独立页面会话（按 taskId 存取，不分权）at /api/dashboard-sessions
     app.include_router(dashboard_sessions.router)
 
diff --git a/offline-backend-20260512/backend/app/gateway/routers/open_chat.py b/offline-backend-20260512/backend/app/gateway/routers/open_chat.py
index 1b8f0b78..849e0085 100644
--- a/offline-backend-20260512/backend/app/gateway/routers/open_chat.py
+++ b/offline-backend-20260512/backend/app/gateway/routers/open_chat.py
@@ -33,12 +33,16 @@
 
 from __future__ import annotations
 
+import asyncio
+import json
 import logging
 import re
 import uuid
+from collections.abc import AsyncIterator
 from typing import Any, Literal
 
 from fastapi import APIRouter, HTTPException
+from fastapi.responses import StreamingResponse
 from langchain_core.messages import AIMessage, BaseMessage, HumanMessage, SystemMessage
 from pydantic import BaseModel, Field
 
@@ -167,15 +171,15 @@ def _extract_answer(result: Any, *, strip_think: bool) -> str:
     return answer
 
 
-@router.post("/chat", response_model=OpenChatResponse)
-async def open_chat(body: OpenChatRequest) -> OpenChatResponse:
-    """指定模型 + 智能体做一次问答(无状态、可关闭思考)。"""
-    lc_messages = _build_lc_messages(body)
+# 流式响应每隔这么久没有真实事件就发一个 SSE 注释心跳,持续制造下行字节,
+# 让前置反向代理(nginx proxy_read_timeout 默认 60s 等)不会把长任务判定成空闲超时。
+_HEARTBEAT_SECONDS = 15.0
 
+
+def _build_run_context(body: OpenChatRequest) -> tuple[dict[str, Any], str | None, str]:
+    """把请求体翻译成 (config, agent_id, thread_id)。两个端点共用,确保关思考等行为一致。"""
     agent_id = _resolve_agent_id(body.agent_id, body.agent_name)
-
     thread_id = body.thread_id or f"open-{uuid.uuid4()}"
-
     configurable: dict[str, Any] = {
         "thread_id": thread_id,
         "model_name": body.model_name,
@@ -187,8 +191,26 @@ async def open_chat(body: OpenChatRequest) -> OpenChatResponse:
     }
     if agent_id:
         configurable["agent_id"] = agent_id
-
     config: dict[str, Any] = {"configurable": configurable, "recursion_limit": body.recursion_limit}
+    return config, agent_id, thread_id
+
+
+def _sse(event: str | None, data: Any) -> str:
+    """序列化一条 SSE 消息。event=None 时只发 data 行。"""
+    payload = json.dumps(data, ensure_ascii=False)
+    head = f"event: {event}\n" if event else ""
+    return f"{head}data: {payload}\n\n"
+
+
+@router.post("/chat", response_model=OpenChatResponse)
+async def open_chat(body: OpenChatRequest) -> OpenChatResponse:
+    """指定模型 + 智能体做一次问答(无状态、可关闭思考)。
+
+    同步阻塞版:整段调研跑完才返回。**长任务(联网调研)请改用 /chat/stream**——
+    本端点期间不产生任何下行字节,容易被前置反代的 proxy_read_timeout 掐成 504。
+    """
+    lc_messages = _build_lc_messages(body)
+    config, agent_id, thread_id = _build_run_context(body)
 
     try:
         agent = make_lead_agent(config)
@@ -211,6 +233,103 @@ async def open_chat(body: OpenChatRequest) -> OpenChatResponse:
     )
 
 
+async def _open_chat_event_stream(body: OpenChatRequest) -> AsyncIterator[str]:
+    """驱动智能体并产出 SSE 事件流。
+
+    - 边跑边转发 AI 文本增量(``event: delta``,仅供实时观看,客户端可忽略);
+    - 没有真实事件时每 ``_HEARTBEAT_SECONDS`` 发一个心跳注释行保活;
+    - 全部跑完后用 ``event: result`` 给出最终(已按需剥离 <think> 的)答案;
+    - 出错则 ``event: error``。
+
+    用「后台生产任务 + 队列 + 超时取队列」的方式把「智能体进度」与「定时心跳」解耦,
+    所以即便某次工具调用静默很久,心跳仍照常下行,连接不会被反代判空闲。
+    """
+    lc_messages = _build_lc_messages(body)
+    config, agent_id, thread_id = _build_run_context(body)
+    strip_think = body.strip_think and not body.thinking_enabled
+
+    queue: asyncio.Queue[tuple[str, Any]] = asyncio.Queue()
+    final_state: dict[str, Any] = {}
+
+    async def produce() -> None:
+        nonlocal final_state
+        try:
+            agent = make_lead_agent(config)
+            async for mode, chunk in agent.astream({"messages": lc_messages}, config=config, stream_mode=["values", "messages"]):
+                if mode == "values" and isinstance(chunk, dict):
+                    final_state = chunk
+                    await queue.put(("progress", {"messages": len(chunk.get("messages") or [])}))
+                elif mode == "messages":
+                    msg = chunk[0] if isinstance(chunk, tuple) and chunk else chunk
+                    # 仅转发 AI 正文增量;DeepSeek 等把思考放在 additional_kwargs.reasoning_content,
+                    # 不在 content 里,所以这里天然不会泄漏 <think>。
+                    if isinstance(msg, AIMessage):
+                        text = _extract_text(getattr(msg, "content", "")).strip()
+                        if text:
+                            await queue.put(("delta", {"text": text}))
+        except Exception as exc:  # noqa: BLE001 — 开放接口需把内部错误归一成 error 事件
+            logger.exception("open_chat_stream 调用智能体失败 (model=%s, agent=%s)", body.model_name, agent_id)
+            await queue.put(("error", {"detail": f"智能体调用失败:{exc}"}))
+        finally:
+            await queue.put(("__end__", None))
+
+    task = asyncio.create_task(produce())
+    try:
+        while True:
+            try:
+                event, data = await asyncio.wait_for(queue.get(), timeout=_HEARTBEAT_SECONDS)
+            except asyncio.TimeoutError:
+                yield ": keepalive\n\n"  # SSE 注释行,客户端忽略,仅用于保活
+                continue
+            if event == "__end__":
+                break
+            if event == "error":
+                yield _sse("error", data)
+                return
+            yield _sse(event, data)
+
+        answer = _extract_answer(final_state, strip_think=strip_think)
+        if not answer:
+            yield _sse("error", {"detail": "智能体未返回任何文本答案。"})
+            return
+        yield _sse(
+            "result",
+            {
+                "answer": answer,
+                "model_name": body.model_name,
+                "agent_name": body.agent_name,
+                "agent_id": agent_id,
+                "thinking_enabled": body.thinking_enabled,
+                "thread_id": thread_id,
+            },
+        )
+    finally:
+        if not task.done():
+            task.cancel()
+
+
+@router.post("/chat/stream")
+async def open_chat_stream(body: OpenChatRequest) -> StreamingResponse:
+    """流式版 /chat:边跑边发心跳保活,跑完用 ``event: result`` 给最终答案。
+
+    解决两件事:(1) 长任务(联网调研、报告)同步阻塞会被前置反代 proxy_read_timeout
+    掐成 504;流式持续有字节下行,代理不再判定空闲超时。(2) 客户端只需读到
+    ``event: result`` 解析 ``answer`` 即可,中间的 ``delta``/``progress``/心跳都可忽略。
+    """
+    # 入参校验放在流外:非法请求直接抛 4xx,而不是开了流再报错。
+    _build_lc_messages(body)
+    return StreamingResponse(
+        _open_chat_event_stream(body),
+        media_type="text/event-stream",
+        headers={
+            "Cache-Control": "no-cache",
+            "Connection": "keep-alive",
+            # 关掉 nginx 对该响应的缓冲,确保 SSE/心跳实时下行(即便前置 nginx 默认开缓冲)。
+            "X-Accel-Buffering": "no",
+        },
+    )
+
+
 @router.get("/agents")
 async def open_list_agents() -> dict[str, Any]:
     """列出可用智能体的名称(方便知道 agent_name 该填什么)。"""
diff --git a/offline-backend-20260512/backend/app/gateway/routers/skill_addresses.py b/offline-backend-20260512/backend/app/gateway/routers/skill_addresses.py
new file mode 100644
index 00000000..82d3f679
--- /dev/null
+++ b/offline-backend-20260512/backend/app/gateway/routers/skill_addresses.py
@@ -0,0 +1,631 @@
+"""技能地址(URL/IP)批量替换 API (技能地址管理).
+
+Scans the **running** skills directory (``skills/{public,custom}``) — the files
+the agent actually reads — for ``http(s)://`` URLs and IPv4 (``ip[:port]``)
+addresses, aggregates each unique address with every place it appears (across
+``.md`` and ``.py``, across files), and lets an admin rewrite them in bulk.
+
+Why this shape: the same address routinely lives in both a skill's ``SKILL.md``
+(documented endpoint) and its ``scripts/*.py`` (the actual call), so the natural
+unit is *"one address → replace all of its occurrences at once"*. The pure
+scan/replace logic lives in ``deerflow.skills.address_scan`` (harness layer,
+substring-safe span replacement); this router only walks the tree, enforces
+admin auth + path safety, backs up before writing, and refreshes the skills
+system-prompt cache so edits take effect immediately.
+
+**Scale / availability (150+ skills):** every filesystem walk / read / write is
+offloaded to a worker thread via ``asyncio.to_thread`` so the event loop is
+never blocked while scanning or rewriting — the Gateway stays responsive. The
+streaming endpoints additionally emit NDJSON progress (``已处理 X / 共 N``) so the
+UI can render a progress bar for both the scan and the apply phases.
+
+Routes (prefix ``/api/skill-addresses``, **admin only**):
+    GET  /scan?kinds=url,ip&skill=<name>   read-only aggregated scan (one shot)
+    GET  /scan/stream?kinds=...&skill=...   scan with per-skill NDJSON progress
+    POST /preview                          dry-run diff of a replacement set
+    POST /apply                            write changes (one shot, with backup)
+    POST /apply/stream                     write changes with per-file progress
+    GET  /backups                          list restorable backups (newest first)
+    POST /rollback/{backup_id}             restore a backup
+"""
+
+from __future__ import annotations
+
+import asyncio
+import json
+import logging
+import shutil
+from collections.abc import AsyncIterator
+from datetime import UTC, datetime
+from pathlib import Path
+
+from fastapi import APIRouter, HTTPException, Query, Request
+from fastapi.responses import StreamingResponse
+from pydantic import BaseModel, ConfigDict, Field
+
+from app.gateway.deps import get_optional_user_from_request
+from deerflow.agents.lead_agent.prompt import refresh_skills_system_prompt_cache_async
+from deerflow.skills.address_scan import (
+    VALID_KINDS,
+    apply_replacements,
+    infer_kind,
+    is_valid_replacement,
+    normalize_kinds,
+    scan_occurrences,
+)
+from deerflow.skills.storage import get_or_new_skill_storage
+
+logger = logging.getLogger(__name__)
+
+router = APIRouter(prefix="/api/skill-addresses", tags=["skill-addresses"])
+
+_BACKUP_DIRNAME = ".address-edits"
+_SCAN_SUBDIRS = ("public", "custom")
+_SCAN_EXTS = (".md", ".py")
+
+
+# ── schemas ────────────────────────────────────────────────────────────────
+
+
+class OccurrenceModel(BaseModel):
+    skill: str
+    category: str  # "public" | "custom"
+    file: str  # POSIX path relative to the skills root
+    line: int
+    col: int
+    snippet: str
+
+
+class AddressGroup(BaseModel):
+    address: str
+    kind: str  # "url" | "ip"
+    count: int
+    occurrences: list[OccurrenceModel]
+
+
+class ScanResponse(BaseModel):
+    groups: list[AddressGroup]
+    scanned_files: int
+    scanned_skills: int
+    kinds: list[str]
+    scanned_at: str
+
+
+class Replacement(BaseModel):
+    model_config = ConfigDict(populate_by_name=True)
+
+    from_: str = Field(alias="from", min_length=1)
+    to: str = Field(min_length=1)
+    kind: str | None = None
+
+
+class PreviewRequest(BaseModel):
+    replacements: list[Replacement] = Field(default_factory=list)
+
+
+class PreviewFileHit(BaseModel):
+    file: str
+    line: int
+    before: str
+    after: str
+
+
+class PreviewItem(BaseModel):
+    model_config = ConfigDict(populate_by_name=True)
+
+    from_: str = Field(serialization_alias="from")
+    to: str
+    kind: str
+    count: int
+    files: list[PreviewFileHit]
+
+
+class PreviewResponse(BaseModel):
+    items: list[PreviewItem]
+    total: int
+    warnings: list[str]
+    expected_total: int
+
+
+class ApplyRequest(BaseModel):
+    replacements: list[Replacement] = Field(default_factory=list)
+    expected_total: int | None = None
+    backup: bool = True
+
+
+class ChangedFile(BaseModel):
+    file: str
+    count: int
+
+
+class ApplyResponse(BaseModel):
+    changed_files: list[ChangedFile]
+    replaced_total: int
+    backup_id: str | None
+    warnings: list[str]
+
+
+class BackupEntry(BaseModel):
+    backup_id: str
+    files: int
+    created_at: str
+
+
+class BackupListResponse(BaseModel):
+    backups: list[BackupEntry]
+
+
+class RollbackResponse(BaseModel):
+    backup_id: str
+    restored_files: list[str]
+
+
+# ── auth / path helpers ──────────────────────────────────────────────────────
+
+
+async def _require_admin(request: Request) -> None:
+    """Admin-only. Lets the call through when auth is disabled (no user)."""
+
+    user = await get_optional_user_from_request(request)
+    if user is None:
+        return
+    if getattr(user, "system_role", None) != "admin":
+        raise HTTPException(status_code=403, detail="技能地址管理仅限管理员")
+
+
+def _skills_root() -> Path:
+    return Path(get_or_new_skill_storage().get_skills_root_path()).resolve()
+
+
+def _is_within(root: Path, p: Path) -> bool:
+    try:
+        p.resolve().relative_to(root)
+        return True
+    except (ValueError, OSError):
+        return False
+
+
+def _skill_of(root: Path, path: Path) -> tuple[str, str, str]:
+    """Return ``(category, skill_name, posix_rel_path)`` for a scanned file."""
+
+    rel = path.relative_to(root)
+    parts = rel.parts
+    category = parts[0] if parts else ""
+    skill_name = parts[1] if len(parts) > 1 else ""
+    return category, skill_name, rel.as_posix()
+
+
+def _read_text(path: Path) -> str | None:
+    """Read UTF-8 text preserving original line endings; skip binary/bad files."""
+
+    try:
+        with open(path, "r", encoding="utf-8", newline="") as fh:
+            return fh.read()
+    except (UnicodeDecodeError, OSError):
+        return None
+
+
+def _write_text(path: Path, text: str) -> None:
+    with open(path, "w", encoding="utf-8", newline="") as fh:
+        fh.write(text)
+
+
+# ── synchronous workers (run via asyncio.to_thread; never block the loop) ─────
+
+
+def _iter_skill_files(root: Path, skill: str | None = None) -> list[Path]:
+    """All ``*.md`` / ``*.py`` under ``{root}/{public,custom}``.
+
+    Hidden directories (``.address-edits`` backups, ``.history``) are skipped.
+    When *skill* is given, only that skill's directory is walked.
+    """
+
+    files: list[Path] = []
+    for sub in _SCAN_SUBDIRS:
+        base = root / sub
+        if skill:
+            base = base / skill
+        if not base.is_dir():
+            continue
+        for p in base.rglob("*"):
+            if not p.is_file() or p.suffix.lower() not in _SCAN_EXTS:
+                continue
+            try:
+                rel_parts = p.relative_to(root).parts
+            except ValueError:
+                continue
+            if any(part.startswith(".") for part in rel_parts):
+                continue
+            files.append(p)
+    return files
+
+
+def _group_files_by_skill(root: Path, files: list[Path]) -> dict[tuple[str, str], list[Path]]:
+    """Bucket scanned files under their owning ``(category, skill_name)``."""
+
+    buckets: dict[tuple[str, str], list[Path]] = {}
+    for p in files:
+        category, skill_name, _ = _skill_of(root, p)
+        buckets.setdefault((category, skill_name), []).append(p)
+    return buckets
+
+
+def _occurrence_kind(value: str) -> str:
+    return infer_kind(value) or "url"
+
+
+def _merge_occurrences(groups: dict[str, AddressGroup], pairs: list[tuple[str, OccurrenceModel]]) -> None:
+    """Merge ``(address, occurrence)`` pairs into the aggregate group map."""
+
+    for address, occ in pairs:
+        g = groups.get(address)
+        if g is None:
+            g = AddressGroup(address=address, kind=_occurrence_kind(address), count=0, occurrences=[])
+            groups[address] = g
+        g.count += 1
+        g.occurrences.append(occ)
+
+
+def _plan_apply(root: Path, mapping: dict[str, str]) -> tuple[list[tuple[Path, str, int, str]], int]:
+    """Compute the rewrite plan without touching disk → (plan, current_total)."""
+
+    plan: list[tuple[Path, str, int, str]] = []
+    current_total = 0
+    for path in _iter_skill_files(root):
+        text = _read_text(path)
+        if text is None:
+            continue
+        new_text, cnt = apply_replacements(text, mapping, VALID_KINDS)
+        if cnt <= 0:
+            continue
+        if not _is_within(root, path):
+            raise HTTPException(status_code=400, detail=f"拒绝写入技能目录之外的文件：{path}")
+        _, _, rel = _skill_of(root, path)
+        plan.append((path, new_text, cnt, rel))
+        current_total += cnt
+    return plan, current_total
+
+
+def _backup_file(root: Path, backup_id: str, path: Path, rel: str) -> None:
+    dest = root / _BACKUP_DIRNAME / backup_id / rel
+    dest.parent.mkdir(parents=True, exist_ok=True)
+    shutil.copy2(path, dest)
+
+
+# ── scan extraction shared by one-shot + stream ───────────────────────────────
+
+
+def _scan_files_with_addresses(root: Path, files: list[Path], wanted: tuple[str, ...]) -> tuple[list[tuple[str, OccurrenceModel]], int]:
+    """Scan files → ([(address, occurrence)], files_read)."""
+
+    pairs: list[tuple[str, OccurrenceModel]] = []
+    read = 0
+    for path in files:
+        text = _read_text(path)
+        if text is None:
+            continue
+        read += 1
+        category, skill_name, rel = _skill_of(root, path)
+        for o in scan_occurrences(text, wanted):
+            pairs.append(
+                (
+                    o.value,
+                    OccurrenceModel(skill=skill_name, category=category, file=rel, line=o.line, col=o.col, snippet=o.snippet),
+                )
+            )
+    return pairs, read
+
+
+def _ordered_groups(groups: dict[str, AddressGroup]) -> list[AddressGroup]:
+    return sorted(groups.values(), key=lambda g: (-g.count, g.address))
+
+
+def _build_mapping(replacements: list[Replacement]) -> tuple[dict[str, str], list[tuple[str, str, str]], list[str]]:
+    """Validate replacements → ``(mapping, [(from,to,kind)], warnings)``.
+
+    Drops (with a warning) any entry that is empty, duplicated, not a recognized
+    URL/IP, a no-op (``from == to``), or whose ``to`` is malformed for its kind.
+    """
+
+    mapping: dict[str, str] = {}
+    norm: list[tuple[str, str, str]] = []
+    warnings: list[str] = []
+    seen: set[str] = set()
+    for r in replacements:
+        frm = r.from_.strip()
+        to = r.to.strip()
+        if not frm:
+            continue
+        if frm in seen:
+            warnings.append(f"重复的原始地址已忽略：{frm}")
+            continue
+        kind = (r.kind or "").strip().lower() or infer_kind(frm)
+        if kind not in VALID_KINDS:
+            warnings.append(f"无法识别为 URL/IP，已忽略：{frm}")
+            continue
+        if to == frm:
+            warnings.append(f"替换值与原值相同，已忽略：{frm}")
+            continue
+        if not is_valid_replacement(kind, to):
+            warnings.append(f"替换值不是合法的 {kind}，已忽略：{to}")
+            continue
+        seen.add(frm)
+        mapping[frm] = to
+        norm.append((frm, to, kind))
+    return mapping, norm, warnings
+
+
+def _ndjson(obj: dict) -> bytes:
+    return (json.dumps(obj, ensure_ascii=False) + "\n").encode("utf-8")
+
+
+# ── routes: scan ─────────────────────────────────────────────────────────────
+
+
+@router.get("/scan", response_model=ScanResponse)
+async def scan_addresses(
+    request: Request,
+    kinds: str | None = Query(default=None, description="逗号分隔：url,ip（默认全部）"),
+    skill: str | None = Query(default=None, description="仅扫描指定技能目录"),
+) -> ScanResponse:
+    await _require_admin(request)
+    root = _skills_root()
+    wanted = normalize_kinds(kinds)
+    files = await asyncio.to_thread(_iter_skill_files, root, skill)
+    buckets = _group_files_by_skill(root, files)
+    groups: dict[str, AddressGroup] = {}
+    scanned = 0
+    for paths in buckets.values():
+        pairs, read = await asyncio.to_thread(_scan_files_with_addresses, root, paths, wanted)
+        _merge_occurrences(groups, pairs)
+        scanned += read
+    return ScanResponse(
+        groups=_ordered_groups(groups),
+        scanned_files=scanned,
+        scanned_skills=len(buckets),
+        kinds=list(wanted),
+        scanned_at=datetime.now(UTC).isoformat(),
+    )
+
+
+@router.get("/scan/stream")
+async def scan_addresses_stream(
+    request: Request,
+    kinds: str | None = Query(default=None),
+    skill: str | None = Query(default=None),
+) -> StreamingResponse:
+    """Scan with per-skill NDJSON progress.
+
+    Emits ``{"type":"start","total":N}`` then one
+    ``{"type":"progress","done":i,"total":N,"skill":name}`` per skill, finally
+    ``{"type":"result", ...ScanResponse...}``.
+    """
+
+    await _require_admin(request)
+    root = _skills_root()
+    wanted = normalize_kinds(kinds)
+
+    async def gen() -> AsyncIterator[bytes]:
+        files = await asyncio.to_thread(_iter_skill_files, root, skill)
+        buckets = _group_files_by_skill(root, files)
+        total = len(buckets)
+        yield _ndjson({"type": "start", "total": total})
+        groups: dict[str, AddressGroup] = {}
+        scanned = 0
+        done = 0
+        for (category, skill_name), paths in buckets.items():
+            pairs, read = await asyncio.to_thread(_scan_files_with_addresses, root, paths, wanted)
+            _merge_occurrences(groups, pairs)
+            scanned += read
+            done += 1
+            yield _ndjson({"type": "progress", "done": done, "total": total, "skill": skill_name or category})
+        result = ScanResponse(
+            groups=_ordered_groups(groups),
+            scanned_files=scanned,
+            scanned_skills=total,
+            kinds=list(wanted),
+            scanned_at=datetime.now(UTC).isoformat(),
+        )
+        yield _ndjson({"type": "result", **result.model_dump()})
+
+    return StreamingResponse(gen(), media_type="application/x-ndjson")
+
+
+# ── routes: preview ──────────────────────────────────────────────────────────
+
+
+def _compute_preview(root: Path, mapping: dict[str, str], norm: list[tuple[str, str, str]]) -> tuple[list[PreviewItem], int]:
+    counts: dict[str, int] = {frm: 0 for frm, _, _ in norm}
+    files: dict[str, list[PreviewFileHit]] = {frm: [] for frm, _, _ in norm}
+    total = 0
+    if mapping:
+        for path in _iter_skill_files(root):
+            text = _read_text(path)
+            if text is None:
+                continue
+            _, _, rel = _skill_of(root, path)
+            for o in scan_occurrences(text, VALID_KINDS):
+                if o.value not in mapping:
+                    continue
+                to = mapping[o.value]
+                counts[o.value] += 1
+                total += 1
+                files[o.value].append(PreviewFileHit(file=rel, line=o.line, before=o.snippet, after=o.snippet.replace(o.value, to)))
+    items = [PreviewItem(from_=frm, to=to, kind=kind, count=counts[frm], files=files[frm]) for frm, to, kind in norm]
+    return items, total
+
+
+@router.post("/preview", response_model=PreviewResponse)
+async def preview_replacements(request: Request, body: PreviewRequest) -> PreviewResponse:
+    await _require_admin(request)
+    root = _skills_root()
+    mapping, norm, warnings = _build_mapping(body.replacements)
+    items, total = await asyncio.to_thread(_compute_preview, root, mapping, norm)
+    return PreviewResponse(items=items, total=total, warnings=warnings, expected_total=total)
+
+
+# ── routes: apply ────────────────────────────────────────────────────────────
+
+
+def _new_backup_id() -> str:
+    return datetime.now(UTC).strftime("%Y%m%d-%H%M%S-%f")
+
+
+@router.post("/apply", response_model=ApplyResponse)
+async def apply_replacements_route(request: Request, body: ApplyRequest) -> ApplyResponse:
+    await _require_admin(request)
+    root = _skills_root()
+    mapping, _norm, warnings = _build_mapping(body.replacements)
+    if not mapping:
+        return ApplyResponse(changed_files=[], replaced_total=0, backup_id=None, warnings=warnings)
+
+    plan, current_total = await asyncio.to_thread(_plan_apply, root, mapping)
+    if body.expected_total is not None and body.expected_total != current_total:
+        raise HTTPException(
+            status_code=409,
+            detail=f"文件自上次扫描后已变化（预期 {body.expected_total} 处，现为 {current_total} 处），请重新扫描后再应用。",
+        )
+    if not plan:
+        return ApplyResponse(changed_files=[], replaced_total=0, backup_id=None, warnings=warnings)
+
+    backup_id: str | None = None
+    if body.backup:
+        backup_id = _new_backup_id()
+        for path, _new, _cnt, rel in plan:
+            await asyncio.to_thread(_backup_file, root, backup_id, path, rel)
+
+    changed: list[ChangedFile] = []
+    replaced_total = 0
+    for path, new_text, cnt, rel in plan:
+        await asyncio.to_thread(_write_text, path, new_text)
+        changed.append(ChangedFile(file=rel, count=cnt))
+        replaced_total += cnt
+
+    await _refresh_cache_safe()
+    logger.info("Skill address apply: %d files, %d replacements, backup=%s", len(changed), replaced_total, backup_id)
+    return ApplyResponse(changed_files=changed, replaced_total=replaced_total, backup_id=backup_id, warnings=warnings)
+
+
+@router.post("/apply/stream")
+async def apply_replacements_stream(request: Request, body: ApplyRequest) -> StreamingResponse:
+    """Apply with per-file NDJSON progress.
+
+    Validation + the optimistic-lock check happen *before* streaming (so a 409
+    is a normal HTTP error). Then: ``{"type":"start","total_files":M,...}`` →
+    one ``{"type":"progress","done":k,...}`` per written file →
+    ``{"type":"result", ...ApplyResponse...}``.
+    """
+
+    await _require_admin(request)
+    root = _skills_root()
+    mapping, _norm, warnings = _build_mapping(body.replacements)
+
+    plan: list[tuple[Path, str, int, str]] = []
+    current_total = 0
+    if mapping:
+        plan, current_total = await asyncio.to_thread(_plan_apply, root, mapping)
+        if body.expected_total is not None and body.expected_total != current_total:
+            raise HTTPException(
+                status_code=409,
+                detail=f"文件自上次扫描后已变化（预期 {body.expected_total} 处，现为 {current_total} 处），请重新扫描后再应用。",
+            )
+
+    async def gen() -> AsyncIterator[bytes]:
+        if not plan:
+            result = ApplyResponse(changed_files=[], replaced_total=0, backup_id=None, warnings=warnings)
+            yield _ndjson({"type": "result", **result.model_dump()})
+            return
+
+        backup_id: str | None = None
+        if body.backup:
+            backup_id = _new_backup_id()
+
+        yield _ndjson({"type": "start", "total_files": len(plan), "total_replacements": current_total})
+
+        changed: list[ChangedFile] = []
+        replaced_total = 0
+        done = 0
+        for path, new_text, cnt, rel in plan:
+            if backup_id:
+                await asyncio.to_thread(_backup_file, root, backup_id, path, rel)
+            await asyncio.to_thread(_write_text, path, new_text)
+            changed.append(ChangedFile(file=rel, count=cnt))
+            replaced_total += cnt
+            done += 1
+            yield _ndjson({"type": "progress", "done": done, "total_files": len(plan), "file": rel, "count": cnt})
+
+        await _refresh_cache_safe()
+        logger.info("Skill address apply(stream): %d files, %d replacements, backup=%s", len(changed), replaced_total, backup_id)
+        result = ApplyResponse(changed_files=changed, replaced_total=replaced_total, backup_id=backup_id, warnings=warnings)
+        yield _ndjson({"type": "result", **result.model_dump()})
+
+    return StreamingResponse(gen(), media_type="application/x-ndjson")
+
+
+# ── routes: backups / rollback ───────────────────────────────────────────────
+
+
+def _list_backups(root: Path) -> list[BackupEntry]:
+    backup_root = root / _BACKUP_DIRNAME
+    entries: list[BackupEntry] = []
+    if backup_root.is_dir():
+        for d in backup_root.iterdir():
+            if not d.is_dir():
+                continue
+            file_count = sum(1 for p in d.rglob("*") if p.is_file())
+            try:
+                created = datetime.fromtimestamp(d.stat().st_mtime, UTC).isoformat()
+            except OSError:
+                created = ""
+            entries.append(BackupEntry(backup_id=d.name, files=file_count, created_at=created))
+    entries.sort(key=lambda e: e.backup_id, reverse=True)
+    return entries
+
+
+@router.get("/backups", response_model=BackupListResponse)
+async def list_backups(request: Request) -> BackupListResponse:
+    await _require_admin(request)
+    root = _skills_root()
+    entries = await asyncio.to_thread(_list_backups, root)
+    return BackupListResponse(backups=entries)
+
+
+def _restore_backup(root: Path, backup_root: Path) -> list[str]:
+    restored: list[str] = []
+    for src in backup_root.rglob("*"):
+        if not src.is_file():
+            continue
+        rel = src.relative_to(backup_root)
+        dest = root / rel
+        if not _is_within(root, dest):
+            continue
+        dest.parent.mkdir(parents=True, exist_ok=True)
+        shutil.copy2(src, dest)
+        restored.append(rel.as_posix())
+    return restored
+
+
+@router.post("/rollback/{backup_id}", response_model=RollbackResponse)
+async def rollback(request: Request, backup_id: str) -> RollbackResponse:
+    await _require_admin(request)
+    root = _skills_root()
+    if "/" in backup_id or "\\" in backup_id or backup_id in (".", ".."):
+        raise HTTPException(status_code=400, detail="非法的备份编号")
+    backup_root = (root / _BACKUP_DIRNAME / backup_id).resolve()
+    if not _is_within(root, backup_root) or not backup_root.is_dir():
+        raise HTTPException(status_code=404, detail="备份不存在")
+
+    restored = await asyncio.to_thread(_restore_backup, root, backup_root)
+    await _refresh_cache_safe()
+    logger.info("Skill address rollback: backup=%s, restored %d files", backup_id, len(restored))
+    return RollbackResponse(backup_id=backup_id, restored_files=restored)
+
+
+# ── misc ─────────────────────────────────────────────────────────────────────
+
+
+async def _refresh_cache_safe() -> None:
+    try:
+        await refresh_skills_system_prompt_cache_async()
+    except Exception:  # noqa: BLE001 — cache refresh is best-effort
+        logger.warning("Failed to refresh skills prompt cache", exc_info=True)
diff --git a/offline-backend-20260512/backend/config.yaml b/offline-backend-20260512/backend/config.yaml
index b12eec02..6b8593f0 100644
--- a/offline-backend-20260512/backend/config.yaml
+++ b/offline-backend-20260512/backend/config.yaml
@@ -123,6 +123,12 @@ models:
     temperature: 1.0
     supports_vision: true
     supports_thinking: true
+    # SiliconFlow 混合推理模型关思考用「请求体顶层 enable_thinking: false」,
+    # 不认 chat_template_kwargs / thinking.type 那两种嵌套形状。声明这个后,
+    # 网页端「关闭思考」对本模型才生效(开放接口另由 force_disable_thinking 兜底)。
+    when_thinking_disabled:
+      extra_body:
+        enable_thinking: false
   - name: Pro/MiniMaxAI/MiniMax-M2.5
     display_name: Pro/MiniMaxAI/MiniMax-M2.5
     use: langchain_openai:ChatOpenAI
diff --git a/offline-backend-20260512/backend/packages/harness/deerflow/models/factory.py b/offline-backend-20260512/backend/packages/harness/deerflow/models/factory.py
index 50096990..67f52a26 100644
--- a/offline-backend-20260512/backend/packages/harness/deerflow/models/factory.py
+++ b/offline-backend-20260512/backend/packages/harness/deerflow/models/factory.py
@@ -134,10 +134,16 @@ def create_chat_model(name: str | None = None, thinking_enabled: bool = False, *
     # Hard override: force-disable thinking regardless of the per-model config shape.
     # Injected only for OpenAI-compatible classes (those declaring `extra_body`);
     # other providers would reject the kwarg, so they are skipped (they ignore it).
+    # Three disable shapes are injected so the common OpenAI-compatible gateways are
+    # all covered: SiliconFlow / DashScope-style **top-level** ``enable_thinking``
+    # (DeepSeek-V3.1+/Qwen3/GLM hybrid models on api.siliconflow.cn — the knob the
+    # nested ``chat_template_kwargs`` shape does NOT reach), vLLM/Qwen native
+    # ``chat_template_kwargs.enable_thinking``, and Zhipu GLM ``thinking.type``.
+    # Unknown fields are ignored by these gateways, so sending all three is safe.
     if force_disable_thinking and not thinking_enabled and "extra_body" in getattr(model_class, "model_fields", {}):
         model_settings_from_config["extra_body"] = _deep_merge_dicts(
             model_settings_from_config.get("extra_body"),
-            {"chat_template_kwargs": {"enable_thinking": False}, "thinking": {"type": "disabled"}},
+            {"enable_thinking": False, "chat_template_kwargs": {"enable_thinking": False}, "thinking": {"type": "disabled"}},
         )
 
     _enable_stream_usage_by_default(model_config.use, model_settings_from_config)
diff --git a/offline-backend-20260512/config.yaml b/offline-backend-20260512/config.yaml
index 3fd5da8e..06e067bb 100644
--- a/offline-backend-20260512/config.yaml
+++ b/offline-backend-20260512/config.yaml
@@ -77,6 +77,12 @@ models:
   temperature: 1.0
   supports_vision: true
   supports_thinking: true
+  # SiliconFlow 混合推理模型关思考用「请求体顶层 enable_thinking: false」,
+  # 不认 chat_template_kwargs / thinking.type 那两种嵌套形状。声明这个后,
+  # 网页端「关闭思考」对本模型才生效(开放接口另由 force_disable_thinking 兜底)。
+  when_thinking_disabled:
+    extra_body:
+      enable_thinking: false
 - name: Pro/MiniMaxAI/MiniMax-M2.5
   display_name: Pro/MiniMaxAI/MiniMax-M2.5
   use: langchain_openai:ChatOpenAI
diff --git a/test_open_chat_api.py b/test_open_chat_api.py
index fc50b070..cd9e587c 100644
--- a/test_open_chat_api.py
+++ b/test_open_chat_api.py
@@ -1,13 +1,16 @@
 # -*- coding: utf-8 -*-
-"""开放问答接口 /api/open/chat 测试脚本(Jupyter 版)。
+"""开放问答接口测试脚本(流式版,Jupyter / 命令行均可)。
 
 接口只返回「最终结果」:智能体把调研 / 工具调用全部跑完后,只把最后那段
-文本(markdown)放在 answer 字段里返回 —— 不含中间过程、工具调用、思考。
+文本(markdown)放在 answer 里返回 —— 不含中间过程、工具调用、思考。
 
-用法:在 Jupyter 里把下面几段按顺序运行即可。
-  - 「配置」段:改成你要的模型 / 智能体 / 问题 / 是否思考
-  - 「调用」段:运行后 answer 变量就是最终 md 文本
+为什么默认用流式(/api/open/chat/stream):
+  - 长任务(联网调研、出报告)同步阻塞那版整段没有下行字节,容易被前置 nginx 反代
+    的 proxy_read_timeout(默认 60s)掐成 504。流式版每隔几秒就有心跳/增量下行,
+    代理不会判定空闲超时,**多久都不会超时**。
+  - 客户端只需读到 `event: result` 解析 answer 即可,中途 delta/progress/心跳都忽略。
 
+用法:改下面「配置」段,然后整段运行;answer 变量就是最终 md 文本。
 无第三方依赖(纯标准库 urllib)。
 """
 
@@ -17,19 +20,22 @@ import urllib.error
 import urllib.request
 
 # ====================== 配置(按需修改这里) ======================
-BASE_URL = "http://47.94.209.59:2026"   # 后端地址(走 nginx 用 :2026,Gateway 直连用 :8001)
+BASE_URL = "http://47.94.209.59:2026"   # 走 nginx 用 :2026;Gateway 直连用 :8001
+STREAM = True                            # True=流式(推荐,不超时);False=同步阻塞(短问题用)
 
-MESSAGE = "帮我调研一下 XXX,输出 markdown 报告。"   # 要问的问题
-MODEL_NAME = None                        # 模型名(config.yaml models[].name);None=默认模型
-AGENT_NAME = None                        # 智能体中文名称;None=默认 lead agent。不确定填啥就先跑下面的「列出智能体」
-THINKING_ENABLED = False                 # False=关闭思考(内网模型也会被强制关闭);True=开启
-TIMEOUT = 600                            # 请求超时(秒)
+MESSAGE = "帮我调研一下 赖清德,输出 markdown 报告。"   # 要问的问题
+MODEL_NAME = "deepseek-ai/DeepSeek-V4-Flash"          # 模型名(config.yaml models[].name);None=默认模型
+AGENT_NAME = "你好君"                     # 智能体中文名称;None=默认 lead agent。不确定填啥先跑 ping_agents()
+THINKING_ENABLED = False                 # False=关闭思考(SiliconFlow/内网模型也会被强制关掉);True=开启
+SHOW_LIVE = True                         # 流式时是否实时打印增量正文(看着它一点点出),只看结果可设 False
+READ_TIMEOUT = 120                       # 单次 socket 读超时(秒)。服务端每 ~15s 发心跳,故 120s 绰绰有余
+TIMEOUT = 6000                           # 同步(非流式)模式的整体超时(秒)
 # ===============================================================
 
 
 def open_chat(message, model_name=None, agent_name=None, thinking_enabled=False,
               base_url=BASE_URL, timeout=TIMEOUT):
-    """调用 /api/open/chat,返回最终 md 文本(只要结果,不要过程)。"""
+    """同步阻塞版:整段跑完才返回。短问题用;长任务请用 open_chat_stream。"""
     url = base_url.rstrip("/") + "/api/open/chat"
     payload = {
         "message": message,
@@ -39,8 +45,7 @@ def open_chat(message, model_name=None, agent_name=None, thinking_enabled=False,
     }
     data = json.dumps(payload, ensure_ascii=False).encode("utf-8")
     req = urllib.request.Request(
-        url,
-        data=data,
+        url, data=data,
         headers={"Content-Type": "application/json; charset=utf-8"},
         method="POST",
     )
@@ -49,24 +54,87 @@ def open_chat(message, model_name=None, agent_name=None, thinking_enabled=False,
     return result["answer"]   # 只返回最终结果
 
 
-# ====================== (可选)列出可用智能体名称 ======================
-# 不知道 AGENT_NAME 填什么时,先跑这段看有哪些中文名:
-# import urllib.request, json
-# with urllib.request.urlopen(BASE_URL.rstrip("/") + "/api/open/agents", timeout=30) as r:
-#     for a in json.loads(r.read().decode("utf-8"))["agents"]:
-#         print(a["name"])
+def open_chat_stream(message, model_name=None, agent_name=None, thinking_enabled=False,
+                     base_url=BASE_URL, read_timeout=READ_TIMEOUT, show_live=SHOW_LIVE):
+    """流式版:读 SSE,边跑边收心跳/增量(不超时),最终从 event: result 取 answer。"""
+    url = base_url.rstrip("/") + "/api/open/chat/stream"
+    payload = {
+        "message": message,
+        "model_name": model_name,
+        "agent_name": agent_name,
+        "thinking_enabled": thinking_enabled,
+    }
+    data = json.dumps(payload, ensure_ascii=False).encode("utf-8")
+    req = urllib.request.Request(
+        url, data=data,
+        headers={"Content-Type": "application/json; charset=utf-8", "Accept": "text/event-stream"},
+        method="POST",
+    )
+
+    answer = None
+    error = None
+    event_name = None
+    data_lines = []
+
+    with urllib.request.urlopen(req, timeout=read_timeout) as resp:
+        for raw in resp:                       # 按行读取,服务端心跳让读取永不空闲超时
+            line = raw.decode("utf-8").rstrip("\r\n")
+            if line.startswith(":"):           # SSE 注释行(心跳),忽略
+                continue
+            if line == "":                     # 空行 = 一个事件结束,分发
+                if data_lines:
+                    payload_str = "\n".join(data_lines)
+                    try:
+                        obj = json.loads(payload_str)
+                    except json.JSONDecodeError:
+                        obj = {}
+                    if event_name == "result":
+                        answer = obj.get("answer")
+                    elif event_name == "error":
+                        error = obj.get("detail") or payload_str
+                    elif event_name == "delta" and show_live:
+                        print(obj.get("text", ""), end="", flush=True)
+                event_name, data_lines = None, []
+                continue
+            if line.startswith("event:"):
+                event_name = line[len("event:"):].strip()
+            elif line.startswith("data:"):
+                data_lines.append(line[len("data:"):].lstrip())
+
+    if error:
+        raise RuntimeError(f"接口返回错误:{error}")
+    if answer is None:
+        raise RuntimeError("未收到 event: result(流可能被中途断开)。")
+    return answer
+
+
+def ping_agents(base_url=BASE_URL, timeout=30):
+    """探活:GET /api/open/agents。秒回说明链路 OK,顺带看有哪些 agent_name 可填。"""
+    with urllib.request.urlopen(base_url.rstrip("/") + "/api/open/agents", timeout=timeout) as r:
+        return json.loads(r.read().decode("utf-8"))
+
 
 # ====================== 调用 ======================
-_t0 = time.time()
-answer = open_chat(MESSAGE, MODEL_NAME, AGENT_NAME, THINKING_ENABLED)
-print(f"(耗时 {time.time() - _t0:.1f}s)\n")
-print(answer)        # answer 就是最终 markdown 文本
+if __name__ == "__main__":
+    # 先探活(失败说明是链路/部署问题,而非模型问题)
+    try:
+        info = ping_agents()
+        print(f"[探活OK] 可用智能体 {info.get('count', 0)} 个\n")
+    except Exception as exc:                    # noqa: BLE001
+        print(f"[探活失败] {exc!r} —— 先排查 BASE_URL / 端口 / 服务是否在跑\n")
 
+    _t0 = time.time()
+    if STREAM:
+        answer = open_chat_stream(MESSAGE, MODEL_NAME, AGENT_NAME, THINKING_ENABLED)
+    else:
+        answer = open_chat(MESSAGE, MODEL_NAME, AGENT_NAME, THINKING_ENABLED)
+    print(f"\n\n(耗时 {time.time() - _t0:.1f}s)\n")
+    print(answer)        # answer 就是最终 markdown 文本
 
-# ====================== (可选)在 Jupyter 里渲染成 markdown ======================
-# from IPython.display import Markdown, display
-# display(Markdown(answer))
+    # ============ (可选)Jupyter 里渲染 markdown ============
+    # from IPython.display import Markdown, display
+    # display(Markdown(answer))
 
-# ====================== (可选)存成 .md 文件 ======================
-# with open("result.md", "w", encoding="utf-8") as f:
-#     f.write(answer)
+    # ============ (可选)存成 .md ============
+    # with open("result.md", "w", encoding="utf-8") as f:
+    #     f.write(answer)

From f8228847ab6afedddc0584bdcefee7b8a4edb85b Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E5=90=89=E9=91=AB=E5=8D=87?= <xsji@kland.com.cn>
Date: Mon, 22 Jun 2026 17:29:10 +0800
Subject: [PATCH 12/17] =?UTF-8?q?fix(skill-addresses):=20=E8=A1=A5?=
 =?UTF-8?q?=E6=8F=90=E4=BA=A4=E8=A2=AB=20gitignore=20=E6=BC=8F=E6=8E=89?=
 =?UTF-8?q?=E7=9A=84=20address=5Fscan=20=E5=BC=95=E6=93=8E=E6=A8=A1?=
 =?UTF-8?q?=E5=9D=97?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

skills/ 忽略规则误伤了 packages/harness/deerflow/skills/address_scan.py，
导致部署环境缺模块、启动报 ModuleNotFoundError: deerflow.skills.address_scan。
强制纳入跟踪。

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 .../harness/deerflow/skills/address_scan.py   | 255 ++++++++++++++++++
 1 file changed, 255 insertions(+)
 create mode 100644 offline-backend-20260512/backend/packages/harness/deerflow/skills/address_scan.py

diff --git a/offline-backend-20260512/backend/packages/harness/deerflow/skills/address_scan.py b/offline-backend-20260512/backend/packages/harness/deerflow/skills/address_scan.py
new file mode 100644
index 00000000..473b3de5
--- /dev/null
+++ b/offline-backend-20260512/backend/packages/harness/deerflow/skills/address_scan.py
@@ -0,0 +1,255 @@
+"""Scan skill text for URL / IP addresses and apply safe batch replacements.
+
+Pure-logic layer for the 技能地址(URL/IP)批量替换 feature. This module knows
+**only about text** — it never walks the filesystem. The HTTP router
+(``app/gateway/routers/skill_addresses.py``) walks ``skills/{public,custom}``,
+reads each ``*.md`` / ``*.py`` file, and feeds the text here. Keeping the
+extraction / replacement logic here (no ``app.*`` import) makes it unit-testable
+in isolation and respects the harness/app boundary.
+
+Two address kinds are supported (the feature scope):
+
+* ``"url"`` — full ``http(s)://…`` URLs.
+* ``"ip"``  — dotted-quad IPv4, optionally ``:port``. Each octet is range
+  checked (0–255) so version-number noise like ``1.2.3.400`` is not matched.
+
+The single most important invariant is **substring-safe replacement**: we never
+``str.replace`` raw file text (that would rewrite ``1.2.3.4`` *inside*
+``11.2.3.40``). Instead we re-detect the exact token spans and only rewrite a
+span whose text is *exactly* one of the requested ``from`` values.
+"""
+
+from __future__ import annotations
+
+import re
+from dataclasses import dataclass
+
+__all__ = [
+    "VALID_KINDS",
+    "AddressSpan",
+    "Occurrence",
+    "normalize_kinds",
+    "find_spans",
+    "scan_occurrences",
+    "count_matches",
+    "apply_replacements",
+    "is_valid_replacement",
+    "infer_kind",
+]
+
+# Supported address kinds (the configured feature scope: URL + IP).
+VALID_KINDS: tuple[str, ...] = ("url", "ip")
+
+# A URL body stops at whitespace, quotes, brackets, angle brackets, backticks
+# and backslashes — everything that commonly *delimits* a URL in md / py. The
+# trailing punctuation that often clings to the end (a sentence period, a
+# closing paren already excluded, a Markdown comma) is stripped afterwards.
+_URL_RE = re.compile(r"https?://[^\s'\"`<>)\]}\\]+", re.IGNORECASE)
+
+# Dotted-quad IPv4 (optionally :port). Guarded so it is not glued to a longer
+# number run (``(?<![\d.])`` / ``(?![\d.])``); octets are range-checked below.
+_IP_RE = re.compile(r"(?<![\d.])(\d{1,3}(?:\.\d{1,3}){3})(?::(\d{1,5}))?(?![\d.])")
+
+# Punctuation (ASCII + CJK) that should not be considered part of a URL token
+# when it appears at the very end of the match.
+_URL_TRAILING = set(".,;:!?'\")]}>`" + "）。，；：！？】》、…")
+
+
+@dataclass(frozen=True)
+class AddressSpan:
+    """An address token located at an exact ``[start, end)`` offset in text."""
+
+    kind: str
+    value: str
+    start: int
+    end: int
+
+
+@dataclass(frozen=True)
+class Occurrence:
+    """A located address enriched with 1-based line/column + the source line."""
+
+    kind: str
+    value: str
+    line: int
+    col: int
+    snippet: str
+
+
+def normalize_kinds(kinds: object) -> tuple[str, ...]:
+    """Coerce arbitrary input into an ordered, de-duped subset of ``VALID_KINDS``.
+
+    ``None`` / empty means "all kinds". Unknown kinds are dropped.
+    """
+
+    if kinds is None:
+        return VALID_KINDS
+    if isinstance(kinds, str):
+        raw = [p.strip() for p in kinds.split(",")]
+    else:
+        try:
+            raw = [str(p).strip() for p in kinds]  # type: ignore[assignment]
+        except TypeError:
+            return VALID_KINDS
+    seen: list[str] = []
+    for k in raw:
+        kl = k.lower()
+        if kl in VALID_KINDS and kl not in seen:
+            seen.append(kl)
+    return tuple(seen) if seen else VALID_KINDS
+
+
+def _valid_ipv4(quad: str) -> bool:
+    parts = quad.split(".")
+    if len(parts) != 4:
+        return False
+    for p in parts:
+        if not p.isdigit() or len(p) > 3:
+            return False
+        if int(p) > 255:
+            return False
+    return True
+
+
+def _valid_port(port: str | None) -> bool:
+    if port is None:
+        return True
+    return port.isdigit() and 0 <= int(port) <= 65535
+
+
+def _url_spans(text: str) -> list[AddressSpan]:
+    spans: list[AddressSpan] = []
+    for m in _URL_RE.finditer(text):
+        start, end = m.start(), m.end()
+        value = m.group(0)
+        # Strip clinging trailing punctuation.
+        while value and value[-1] in _URL_TRAILING:
+            value = value[:-1]
+            end -= 1
+        if value:
+            spans.append(AddressSpan(kind="url", value=value, start=start, end=end))
+    return spans
+
+
+def _ip_spans(text: str, exclude: list[AddressSpan]) -> list[AddressSpan]:
+    spans: list[AddressSpan] = []
+    for m in _IP_RE.finditer(text):
+        quad = m.group(1)
+        port = m.group(2)
+        if not _valid_ipv4(quad) or not _valid_port(port):
+            continue
+        start, end = m.start(), m.end()
+        # Skip an IP that lives *inside* a URL (e.g. http://1.2.3.4:8080/x) —
+        # replacing just the IP there would corrupt the URL.
+        if any(s.start <= start < s.end for s in exclude):
+            continue
+        spans.append(AddressSpan(kind="ip", value=m.group(0), start=start, end=end))
+    return spans
+
+
+def find_spans(text: str, kinds: object = None) -> list[AddressSpan]:
+    """Return every address token span in *text*, ordered by position.
+
+    URL regions are always computed first so that IPs embedded inside a URL are
+    not separately surfaced, even when only ``"ip"`` is requested.
+    """
+
+    wanted = normalize_kinds(kinds)
+    url_spans = _url_spans(text)
+    result: list[AddressSpan] = []
+    if "url" in wanted:
+        result.extend(url_spans)
+    if "ip" in wanted:
+        result.extend(_ip_spans(text, exclude=url_spans))
+    result.sort(key=lambda s: s.start)
+    return result
+
+
+def scan_occurrences(text: str, kinds: object = None, *, snippet_max: int = 240) -> list[Occurrence]:
+    """Locate addresses with 1-based line/column and a trimmed source line."""
+
+    spans = find_spans(text, kinds)
+    if not spans:
+        return []
+    # Precompute line-start offsets for O(log n) line lookup.
+    line_starts = [0]
+    for i, ch in enumerate(text):
+        if ch == "\n":
+            line_starts.append(i + 1)
+
+    occ: list[Occurrence] = []
+    import bisect
+
+    for s in spans:
+        line_idx = bisect.bisect_right(line_starts, s.start) - 1
+        line_no = line_idx + 1
+        col = s.start - line_starts[line_idx] + 1
+        line_end = line_starts[line_idx + 1] - 1 if line_idx + 1 < len(line_starts) else len(text)
+        raw_line = text[line_starts[line_idx]:line_end].rstrip("\r")
+        snippet = raw_line.strip()
+        if len(snippet) > snippet_max:
+            snippet = snippet[: snippet_max - 1] + "…"
+        occ.append(Occurrence(kind=s.kind, value=s.value, line=line_no, col=col, snippet=snippet))
+    return occ
+
+
+def count_matches(text: str, mapping: dict[str, str], kinds: object = None) -> int:
+    """How many address spans in *text* would be rewritten by *mapping*."""
+
+    if not mapping:
+        return 0
+    return sum(1 for s in find_spans(text, kinds) if s.value in mapping)
+
+
+def apply_replacements(text: str, mapping: dict[str, str], kinds: object = None) -> tuple[str, int]:
+    """Substring-safe batch replace.
+
+    Only spans whose text is *exactly* a key of *mapping* are rewritten, applied
+    right-to-left so earlier offsets stay valid. Returns ``(new_text, count)``.
+    A no-op mapping (or no hits) returns the original text and ``0``.
+    """
+
+    if not mapping:
+        return text, 0
+    spans = [s for s in find_spans(text, kinds) if s.value in mapping]
+    if not spans:
+        return text, 0
+    out = text
+    for s in sorted(spans, key=lambda s: s.start, reverse=True):
+        out = out[: s.start] + mapping[s.value] + out[s.end:]
+    return out, len(spans)
+
+
+def infer_kind(value: str) -> str | None:
+    """Return the address kind of *value* if it is a single, whole address.
+
+    Used to classify a user-supplied ``from`` value (so the matching ``to`` can
+    be validated against the right rules). Returns ``None`` when *value* is not
+    exactly one URL or IP token.
+    """
+
+    if not value:
+        return None
+    spans = find_spans(value, VALID_KINDS)
+    if len(spans) == 1 and spans[0].value == value:
+        return spans[0].kind
+    return None
+
+
+def is_valid_replacement(kind: str, value: str) -> bool:
+    """Validate that a user-supplied replacement is well-formed for its kind.
+
+    Keeps an admin from turning a URL into garbage (e.g. dropping the scheme) or
+    an IP into a non-IP. Empty replacements are rejected.
+    """
+
+    if not value or value.strip() != value or any(c.isspace() for c in value):
+        return False
+    if kind == "url":
+        return bool(re.fullmatch(r"https?://[^\s'\"`<>]+", value, re.IGNORECASE))
+    if kind == "ip":
+        m = _IP_RE.fullmatch(value)
+        if not m:
+            return False
+        return _valid_ipv4(m.group(1)) and _valid_port(m.group(2))
+    return False

From 26557b30a0510b12e2cee34f4485ad657cc5c827 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E5=90=89=E9=91=AB=E5=8D=87?= <xsji@kland.com.cn>
Date: Mon, 22 Jun 2026 17:35:42 +0800
Subject: [PATCH 13/17] =?UTF-8?q?feat(skill-addresses):=20=E5=B0=86?=
 =?UTF-8?q?=E3=80=8C=E6=8A=80=E8=83=BD=E5=9C=B0=E5=9D=80=E7=AE=A1=E7=90=86?=
 =?UTF-8?q?=E3=80=8D=E5=B9=B6=E5=85=A5=E7=AE=A1=E7=90=86=E5=91=98=E3=80=8C?=
 =?UTF-8?q?=E6=8A=80=E8=83=BD=E7=AE=A1=E7=90=86=E3=80=8D=E9=A1=B5=E7=9A=84?=
 =?UTF-8?q?=E6=A0=87=E7=AD=BE=E9=A1=B5?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- SkillAddressesPage 主体抽成可复用的 SkillAddressesPanel（默认导出保留旧路由作直链回退）
- AdminSkillsPage 顶部加 Tabs：技能列表 + 技能地址管理（嵌入 panel）
- 移除 page-sidebar 左下角设置下拉里两处独立的「技能地址管理」入口
- 同步更新前后端 CLAUDE.md 文档

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 CLAUDE.md                                     |  2 +-
 frontend-web/src/components/page-sidebar.tsx  | 15 ----
 frontend-web/src/pages/AdminSkillsPage.tsx    | 69 ++++++++++++++-----
 frontend-web/src/pages/SkillAddressesPage.tsx |  6 +-
 offline-backend-20260512/backend/CLAUDE.md    | 11 ++-
 5 files changed, 64 insertions(+), 39 deletions(-)

diff --git a/CLAUDE.md b/CLAUDE.md
index 16f0ef3b..b1f29437 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -82,7 +82,7 @@ The frontend (`frontend-web/`) is a React 19 + Vite SPA:
 - **轻应用中心 (Light App Center)**: a sibling of 任务管理 in the sidebar with two pages — 应用管理 (admin registers `window_open` or `iframe` mini-apps with a route-param builder: login params resolved from `localStorage.userInfo`, theme params literal) and 应用列表 (card grid of `window_open` apps, `window.open` launch). `iframe` apps are mounted under a chosen parent menu (XK/PG/TY/综合管理) and **dynamically injected into the sidebar** by `core/page-layout/use-sidebar-menu.ts` (`useSidebarMenuItems`), opening in the `LightAppIframeView` host. Client + React Query hooks in `strategy-components/api/light-apps.ts`; URL builder/constants in `strategy-components/lib/light-app.ts`. Backend: `/api/light-apps` (global/shared, admin-gated writes).
 - **菜单管理 (Menu Management, admin-only)**: an **overlay** over the static sidebar menu — `MenuManagementPage` (`pages/MenuManagementPage.tsx`, route `/page/strategy/admin/menu-management`, opened from the admin-only 「菜单管理」 item in the **bottom-left settings dropdown** of `components/page-sidebar.tsx`). Admins rename / 停用·启用 / reorder (↑↓ or @dnd-kit drag) / re-parent (horizontal drag **or** the edit dialog's 「移动到」parent selector) any node across 一级/二级/三级, **including injected iframe light-app nodes** (`light-app-<appId>`). The static tree (`core/page-layout/sidebar-menu.ts`) stays the source of icons/routes; only per-node deltas (`label`/`disabled`/`parentId`/`sortOrder`) persist to the backend. Merge logic in `core/page-layout/menu-overrides.ts` (`applyMenuOverrides`, applied in `use-sidebar-menu.ts` after light-app injection, before role filter; disabling a node hides its whole subtree) — strict **3-level cap** (`wouldExceedDepth`). Client `strategy-components/api/menu-overrides.ts`. Backend: `/api/menu-overrides` (global/shared; open read, admin atomic replace-all).
 - **敏感词管理 (Sensitive Word Management)**: configures the **展示层脱敏映射** (原始词 → 替换词). **Gated on the username `lqq`** (not the admin role) — `getAccountDisplayName().toLowerCase() === "lqq"` (username = email `@` prefix, else `user_id`, same as the profile page); the dropdown entry, the route guard, and the page-level guard all use this check. `src/lib/desensitize.ts` ships a frozen seed map `SEED_DESENSITIZE_MAP` (the deployment seed) plus a live mutable `DESENSITIZE_MAP` (initialized from the seed so `desensitize()` works before any fetch). `DesensitizeWordsLoader` (mounted in `App.tsx` `AppProviders`) fetches `/api/sensitive-words` on startup and, if non-empty, calls `replaceDesensitizeMap(...)` to make the backend config the source of truth (empty/failed → keep seed). Page `SensitiveWordsPage.tsx` (route `/page/strategy/admin/sensitive-words`, opened from the 「敏感词管理」 item in the **bottom-left settings dropdown** of `components/page-sidebar.tsx`, present in **both** full + collapsed sidebar modes) lists/adds/edits/启停/deletes words and saves the **whole set atomically** (replace-all); a 「导入种子词」 button merges any missing `SEED_DESENSITIZE_MAP` entries into the editor. Client + hooks in `strategy-components/api/sensitive-words.ts` (`wordsToMap` folds enabled rows into the runtime map). Backend: `/api/sensitive-words` (global/shared; open read, write allowed for admin **or** the `lqq` account), mirroring 菜单管理's store shape. **侧边栏 + 导航栏 menu labels are desensitized** at the data source: `useSidebarMenuItems`/`useActiveThirdLevel` (`core/page-layout/use-sidebar-menu.ts`) map every node's `label` through a desensitizer built **directly from the `/api/sensitive-words` query** (`makeDesensitizer(wordsToMap(words))`, falling back to the global seed `desensitize` while empty) — this sidesteps the `replaceDesensitizeMap` effect-timing race. Only `label` is rewritten; `path`/`id`/`icon` are untouched so routing/highlight are unaffected. All nav surfaces (`page-sidebar.tsx`, `page-sidebar-dark.tsx`, `page-sidebar-v1.tsx`, `strategy-components/Header.tsx`, `PageRoutes.tsx`) consume this hook and are covered; `MenuManagementPage` uses the raw tree so admins still edit real labels.
-- **技能地址管理 (Skill Address Management, admin-only)**: 扫描运行中的技能（`skills/{public,custom}` 下所有 `.md`/`.py`）里的 **http(s) URL 与 IP/IP:端口**，按唯一地址聚合全部出现位置（跨 .md/.py），支持**批量替换**（如把 UAT 地址一次性改成生产）。页面 `pages/SkillAddressesPage.tsx`（route `/page/strategy/admin/skill-addresses`，入口「技能地址管理」在 `components/page-sidebar.tsx` 左下角设置下拉，full+collapsed 两种模式）：打开自动扫描并带**进度条**（扫描逐技能 / 应用逐文件 X/N，应对 150+ 技能），URL/IP 类型过滤 + 搜索 + 「片段批量套用」（如所有 `uat.4.cn` → `prod.4.cn`）、每地址「替换为」输入框（可展开看 `file:line` 片段）、内联**预览**、**应用**、**撤销上次**（回滚）。client + NDJSON 流式解析在 `strategy-components/api/skill-addresses.ts`。后端 `/api/skill-addresses`（**仅管理员**）：所有文件 IO 走 `asyncio.to_thread` 不阻塞事件循环；`GET /scan`(+`/scan/stream`)、`POST /preview`、`POST /apply`(+`/apply/stream`)、`GET /backups`、`POST /rollback/{id}`；**子串安全替换**（基于 token span，不会误伤子串）、`expected_total` 乐观锁（409）、写前自动磁盘备份到 `{skills_root}/.address-edits/`、应用后刷新技能系统提示缓存。纯引擎 `deerflow/skills/address_scan.py`，router `app/gateway/routers/skill_addresses.py`，tests `tests/test_skill_addresses.py`。范围仅运行中技能，不动 `deploy/minimal/skills/`。
+- **技能地址管理 (Skill Address Management, admin-only)**: 扫描运行中的技能（`skills/{public,custom}` 下所有 `.md`/`.py`）里的 **http(s) URL 与 IP/IP:端口**，按唯一地址聚合全部出现位置（跨 .md/.py），支持**批量替换**（如把 UAT 地址一次性改成生产）。主体抽成可复用组件 `SkillAddressesPanel`（`pages/SkillAddressesPage.tsx` 导出，默认导出 `SkillAddressesPage` 仍保留旧路由 `/page/strategy/admin/skill-addresses` 作直链回退），**作为「技能地址管理」标签页嵌入管理员「技能管理」页 `pages/AdminSkillsPage.tsx`**（该页加 Tabs：「技能列表」+「技能地址管理」；入口在「设置和更多」下拉的「技能管理」`workspace-nav-menu.tsx` → `/page/workspace/admin/skills`，不再单列于 `components/page-sidebar.tsx` 左下角设置下拉）：打开自动扫描并带**进度条**（扫描逐技能 / 应用逐文件 X/N，应对 150+ 技能），URL/IP 类型过滤 + 搜索 + 「片段批量套用」（如所有 `uat.4.cn` → `prod.4.cn`）、每地址「替换为」输入框（可展开看 `file:line` 片段）、内联**预览**、**应用**、**撤销上次**（回滚）。client + NDJSON 流式解析在 `strategy-components/api/skill-addresses.ts`。后端 `/api/skill-addresses`（**仅管理员**）：所有文件 IO 走 `asyncio.to_thread` 不阻塞事件循环；`GET /scan`(+`/scan/stream`)、`POST /preview`、`POST /apply`(+`/apply/stream`)、`GET /backups`、`POST /rollback/{id}`；**子串安全替换**（基于 token span，不会误伤子串）、`expected_total` 乐观锁（409）、写前自动磁盘备份到 `{skills_root}/.address-edits/`、应用后刷新技能系统提示缓存。纯引擎 `deerflow/skills/address_scan.py`，router `app/gateway/routers/skill_addresses.py`，tests `tests/test_skill_addresses.py`。范围仅运行中技能，不动 `deploy/minimal/skills/`。
 - **按钮管理 (Task Button Management, admin-only)**: configures the **可配置跳转按钮** rendered in the left task panel of the task deep-link workspaces (rwfx / 3qfx / xdfx, see `TaskFxWorkspace`). Each button = `{business, label, linkType, target, appendTaskId, idParam, idKind, enabled, sortOrder}`; `linkType` is `url` (外链 `window.open`, optional append of `?{idParam}={id}` — **`idParam` is the editable query-param name**, e.g. `id`/`task_id`), `business` (内部切换到另一业务, reuses `resolveTaskDeeplinkTarget`, always task id), or `purpose` (弹出当前任务的「任务目的」表格 via the shared `TaskPurposeDialog`, no target — configurable for any page incl. xdfx). **`idKind` (`task`/`action`) only matters for xdfx**: its deep-link `?taskId=` is an 行动id and the 任务id is resolved at login (`fetchTaskIdFromAction`) — `TaskContext` now also stashes the raw `actionId`, so a url button can append either (default `task`). Page `TaskButtonsPage.tsx` (route `/page/strategy/admin/task-buttons`, **admin-gated**, entry「按钮管理」in the bottom-left settings dropdown of `components/page-sidebar.tsx`, both full + collapsed modes) has per-business tabs, edits name/type/target/启停/排序/增删, saves the **whole set atomically** (replace-all), plus 「导入默认」 to merge code-seed defaults. **Seed/fallback** (like 敏感词管理): only when the whole table is empty (fresh deploy, never saved) do the workspace + page fall back to `defaultTaskButtons(cfg)` — rwfx's url buttons (编辑/下一步/3q详情) built from the **live** `/api/public/config/task-deeplink` plus cross-business switch buttons for **all three** businesses (each switches to the other two — rwfx→三情/行动, 3qfx→任务/行动, xdfx→任务/三情). Once saved, the store is the sole source of truth. The old hardcoded jump buttons were migrated out: `TaskFxWorkspace` no longer hardcodes 切换到三情分析, `TaskActionsPanel` keeps only the non-jump 清除记录, and `TaskPurposePanel` dropped its 3q详情 button (now configurable) **and defaults the 目的树 table to collapsed** (chevron toggle). Helpers `core/auth/task-buttons.ts` (`TaskButton`, `defaultTaskButtons`, `resolveTaskButtons`, `buttonsForBusiness`, `runTaskButton`), client+hooks `strategy-components/api/task-buttons.ts`. Backend: `/api/task-buttons` (global/shared; open read, admin replace-all), store `deerflow.persistence.task_buttons` (table `task_buttons`, auto-created by `create_all`), wired `app.state.task_button_store`. Tests: `tests/test_task_buttons.py`.
 - **State**: TanStack Query for server state, context providers (`SubtasksProvider`, `ArtifactsProvider`, `PromptInputProvider`) scoped to `WorkspaceRoutes`
 - **LLM streaming**: `@langchain/langgraph-sdk` streams messages from the backend via SSE
diff --git a/frontend-web/src/components/page-sidebar.tsx b/frontend-web/src/components/page-sidebar.tsx
index 92b4eaa8..d01d6d6e 100644
--- a/frontend-web/src/components/page-sidebar.tsx
+++ b/frontend-web/src/components/page-sidebar.tsx
@@ -9,7 +9,6 @@ import {
   FolderOpen,
   ListTree,
   MousePointerClick,
-  Replace,
   Settings,
   ShieldAlert,
   User,
@@ -223,14 +222,6 @@ function SecondaryNavSidebar() {
                       按钮管理
                     </DropdownMenuItem>
                   )}
-                  {isAdmin && (
-                    <DropdownMenuItem
-                      onClick={() => navigate("/page/strategy/admin/skill-addresses")}
-                    >
-                      <Replace className="size-4" />
-                      技能地址管理
-                    </DropdownMenuItem>
-                  )}
                   {canManageSensitiveWords() && (
                     <DropdownMenuItem
                       onClick={() => navigate("/page/strategy/admin/sensitive-words")}
@@ -517,12 +508,6 @@ function LegacyAccordionSidebar({
                   按钮管理
                 </DropdownMenuItem>
               )}
-              {isAdmin && (
-                <DropdownMenuItem onClick={() => navigate("/page/strategy/admin/skill-addresses")}>
-                  <Replace className="size-4" />
-                  技能地址管理
-                </DropdownMenuItem>
-              )}
               {canManageSensitiveWords() && (
                 <DropdownMenuItem onClick={() => navigate("/page/strategy/admin/sensitive-words")}>
                   <ShieldAlert className="size-4" />
diff --git a/frontend-web/src/pages/AdminSkillsPage.tsx b/frontend-web/src/pages/AdminSkillsPage.tsx
index c942bc48..2dba69ed 100644
--- a/frontend-web/src/pages/AdminSkillsPage.tsx
+++ b/frontend-web/src/pages/AdminSkillsPage.tsx
@@ -3,6 +3,7 @@
 import {
   ChevronLeftIcon,
   ChevronRightIcon,
+  Replace,
   SearchIcon,
   SparklesIcon,
 } from "lucide-react";
@@ -10,6 +11,7 @@ import { useEffect, useMemo, useState } from "react";
 
 import { Button } from "@/components/ui/button";
 import { Input } from "@/components/ui/input";
+import { Tabs, TabsContent, TabsList, TabsTrigger } from "@/components/ui/tabs";
 import {
   WorkspaceBody,
   WorkspaceContainer,
@@ -19,6 +21,7 @@ import { SkillRow } from "@/components/workspace/settings/skill-settings-page";
 import { getStoredAuth } from "@/core/auth";
 import { useAllSkillsAdmin } from "@/core/skills/hooks";
 import { desensitize } from "@/lib/desensitize";
+import { SkillAddressesPanel } from "./SkillAddressesPage";
 
 const PAGE_SIZE = 10;
 
@@ -52,27 +55,47 @@ export default function AdminSkillsPage() {
     <WorkspaceContainer>
       <WorkspaceHeader />
       <WorkspaceBody>
-        <div className="flex size-full min-h-0 flex-col">
-          <div className="flex items-start justify-between gap-4 border-b px-6 py-4">
-            <div>
+        <Tabs
+          defaultValue="skills"
+          className="flex size-full min-h-0 flex-col gap-0"
+        >
+          <div className="flex items-center justify-between gap-4 border-b px-6 pt-4">
+            <div className="flex items-center gap-4">
               <h1 className="text-xl font-semibold">技能管理</h1>
-              <p className="text-muted-foreground mt-0.5 text-sm">
-                管理员视图：查看、编辑、删除所有用户的技能（包括内置技能与未发布的自定义技能）。
-              </p>
-            </div>
-            <div className="relative w-64 shrink-0">
-              <SearchIcon className="text-muted-foreground absolute left-2.5 top-1/2 size-4 -translate-y-1/2" />
-              <Input
-                value={search}
-                onChange={(e) => setSearch(e.target.value)}
-                placeholder="按名称搜索"
-                className="pl-8"
-              />
+              <TabsList>
+                <TabsTrigger value="skills">
+                  <SparklesIcon />
+                  技能列表
+                </TabsTrigger>
+                <TabsTrigger value="addresses">
+                  <Replace />
+                  技能地址管理
+                </TabsTrigger>
+              </TabsList>
             </div>
           </div>
 
-          <div className="flex flex-1 flex-col gap-4 overflow-y-auto p-6">
-            {isLoading && (
+          <TabsContent
+            value="skills"
+            className="mt-0 flex min-h-0 flex-1 flex-col overflow-hidden"
+          >
+            <div className="flex items-start justify-between gap-4 border-b px-6 py-4">
+              <p className="text-muted-foreground text-sm">
+                管理员视图：查看、编辑、删除所有用户的技能（包括内置技能与未发布的自定义技能）。
+              </p>
+              <div className="relative w-64 shrink-0">
+                <SearchIcon className="text-muted-foreground absolute left-2.5 top-1/2 size-4 -translate-y-1/2" />
+                <Input
+                  value={search}
+                  onChange={(e) => setSearch(e.target.value)}
+                  placeholder="按名称搜索"
+                  className="pl-8"
+                />
+              </div>
+            </div>
+
+            <div className="flex flex-1 flex-col gap-4 overflow-y-auto p-6">
+              {isLoading && (
               <div className="text-muted-foreground flex h-40 items-center justify-center text-sm">
                 加载中…
               </div>
@@ -132,8 +155,16 @@ export default function AdminSkillsPage() {
                 </Button>
               </div>
             )}
-          </div>
-        </div>
+            </div>
+          </TabsContent>
+
+          <TabsContent
+            value="addresses"
+            className="mt-0 min-h-0 flex-1 overflow-hidden"
+          >
+            <SkillAddressesPanel />
+          </TabsContent>
+        </Tabs>
       </WorkspaceBody>
     </WorkspaceContainer>
   );
diff --git a/frontend-web/src/pages/SkillAddressesPage.tsx b/frontend-web/src/pages/SkillAddressesPage.tsx
index b1cc1642..4fefa80e 100644
--- a/frontend-web/src/pages/SkillAddressesPage.tsx
+++ b/frontend-web/src/pages/SkillAddressesPage.tsx
@@ -72,7 +72,7 @@ function KindBadge({ kind }: { kind: AddressKind }) {
   );
 }
 
-export default function SkillAddressesPage() {
+export function SkillAddressesPanel() {
   const isAdmin = getStoredAuth()?.system_role === "admin";
 
   const [kinds, setKinds] = useState<AddressKind[]>(ALL_KINDS);
@@ -462,6 +462,10 @@ export default function SkillAddressesPage() {
   );
 }
 
+export default function SkillAddressesPage() {
+  return <SkillAddressesPanel />;
+}
+
 function GroupRow({
   group,
   value,
diff --git a/offline-backend-20260512/backend/CLAUDE.md b/offline-backend-20260512/backend/CLAUDE.md
index df32030d..d233650f 100644
--- a/offline-backend-20260512/backend/CLAUDE.md
+++ b/offline-backend-20260512/backend/CLAUDE.md
@@ -556,9 +556,14 @@ snippet) and replaced together.
   `{skills_root}/.address-edits/{timestamp}/` (the `.`-prefixed dir is excluded
   from scans), then `refresh_skills_system_prompt_cache_async()` so edits take
   effect immediately. Registered in `app.py` after `sensitive_words.router`.
-- **Frontend** — admin page `pages/SkillAddressesPage.tsx` (route
-  `/page/strategy/admin/skill-addresses`, entry「技能地址管理」in the bottom-left
-  settings dropdown of `components/page-sidebar.tsx`, both full + collapsed modes):
+- **Frontend** — the panel is the reusable `SkillAddressesPanel` (exported from
+  `pages/SkillAddressesPage.tsx`; its default export keeps the standalone route
+  `/page/strategy/admin/skill-addresses` as a direct-link fallback). It is
+  **embedded as a「技能地址管理」tab inside the admin「技能管理」page**
+  `pages/AdminSkillsPage.tsx` (Tabs: 技能列表 + 技能地址管理; reached from the
+  「技能管理」item in the「设置和更多」dropdown `workspace-nav-menu.tsx` →
+  `/page/workspace/admin/skills`) — it is no longer a separate entry in the
+  `components/page-sidebar.tsx` settings dropdown. The panel:
   auto-scan on open with a **progress bar** (扫描逐技能 / 应用逐文件 X/N), URL/IP
   type filter, search, a「片段批量套用」helper (e.g. all `uat.4.cn` → `prod.4.cn`),
   per-address「替换为」input with expandable occurrences, inline 预览 panel, 应用,

From 7487395e448594db239bd1e81d1735ab3c599998 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E5=90=89=E9=91=AB=E5=8D=87?= <xsji@kland.com.cn>
Date: Mon, 22 Jun 2026 17:55:44 +0800
Subject: [PATCH 14/17] =?UTF-8?q?feat(skill-addresses):=20=E5=9C=B0?=
 =?UTF-8?q?=E5=9D=80=E6=A3=80=E7=B4=A2=E5=8D=87=E7=BA=A7=E4=B8=BA=E5=A4=9A?=
 =?UTF-8?q?=E5=85=B3=E9=94=AE=E8=AF=8D=E5=8C=B9=E9=85=8D?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- 空格分隔多个关键词，全部命中（地址或技能名任一包含）才显示，便于在大量地址中收窄
- 搜索框加清除按钮 + 更清晰的占位提示
- 过滤时列表头显示「命中 / 总数」

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 frontend-web/src/pages/SkillAddressesPage.tsx | 35 ++++++++++++++-----
 1 file changed, 27 insertions(+), 8 deletions(-)

diff --git a/frontend-web/src/pages/SkillAddressesPage.tsx b/frontend-web/src/pages/SkillAddressesPage.tsx
index 4fefa80e..3fa96662 100644
--- a/frontend-web/src/pages/SkillAddressesPage.tsx
+++ b/frontend-web/src/pages/SkillAddressesPage.tsx
@@ -17,6 +17,7 @@ import {
   Search,
   Undo2,
   Wand2,
+  X,
 } from "lucide-react";
 import { toast } from "sonner";
 
@@ -231,11 +232,19 @@ export function SkillAddressesPanel() {
     }
   };
 
-  const q = query.trim().toLowerCase();
-  const visible = useMemo(
-    () => (!q ? groups : groups.filter((g) => g.address.toLowerCase().includes(q) || g.occurrences.some((o) => o.skill.toLowerCase().includes(q)))),
-    [groups, q],
+  // 多关键词检索：空格分隔，每个关键词都要命中（地址或技能名任一包含）才显示。
+  // 这样「uat 4.cn」能同时收窄到既含 uat 又含 4.cn 的地址，应对地址过多的场景。
+  const tokens = useMemo(
+    () => query.trim().toLowerCase().split(/\s+/).filter(Boolean),
+    [query],
   );
+  const visible = useMemo(() => {
+    if (tokens.length === 0) return groups;
+    return groups.filter((g) => {
+      const haystack = `${g.address} ${g.occurrences.map((o) => o.skill).join(" ")}`.toLowerCase();
+      return tokens.every((t) => haystack.includes(t));
+    });
+  }, [groups, tokens]);
 
   const totalOccurrences = useMemo(() => groups.reduce((s, g) => s + g.count, 0), [groups]);
   const canRollback = !!(lastBackupId || backups.length > 0);
@@ -332,15 +341,25 @@ export function SkillAddressesPanel() {
                 </button>
               ))}
             </div>
-            <div className="relative w-full md:w-80">
+            <div className="relative w-full md:w-96">
               <Search className="absolute left-3.5 top-1/2 size-4 -translate-y-1/2 text-slate-400" />
               <input
                 type="text"
-                placeholder="搜索地址 / 技能名…"
+                placeholder="按关键词检索地址 / 技能名（空格分隔多个，全部命中）"
                 value={query}
                 onChange={(e) => setQuery(e.target.value)}
-                className="w-full rounded-xl border border-slate-200 bg-slate-50/50 py-2 pl-10 pr-4 text-sm transition-all hover:border-slate-300 focus:border-blue-500 focus:outline-none focus:ring-4 focus:ring-blue-500/5"
+                className="w-full rounded-xl border border-slate-200 bg-slate-50/50 py-2 pl-10 pr-9 text-sm transition-all hover:border-slate-300 focus:border-blue-500 focus:outline-none focus:ring-4 focus:ring-blue-500/5"
               />
+              {query && (
+                <button
+                  type="button"
+                  onClick={() => setQuery("")}
+                  title="清除"
+                  className="absolute right-2.5 top-1/2 -translate-y-1/2 rounded p-0.5 text-slate-400 transition-colors hover:bg-slate-100 hover:text-slate-600"
+                >
+                  <X className="size-4" />
+                </button>
+              )}
             </div>
           </div>
 
@@ -422,7 +441,7 @@ export function SkillAddressesPanel() {
         {/* 地址列表 */}
         <div className="overflow-hidden rounded-2xl border border-slate-200/60 bg-white shadow-sm">
           <div className="flex items-center justify-between border-b border-slate-200/60 bg-slate-50/70 px-4 py-2.5 text-xs font-semibold text-slate-500">
-            <span>地址（{visible.length}）</span>
+            <span>地址（{visible.length}{tokens.length > 0 ? ` / ${groups.length}` : ""}）</span>
             {pendingCount > 0 && (
               <button
                 type="button"

From 06b169d7a1584aaea5f5fb7dc7702c0aeaf37f4d Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E5=90=89=E9=91=AB=E5=8D=87?= <xsji@kland.com.cn>
Date: Mon, 22 Jun 2026 18:05:05 +0800
Subject: [PATCH 15/17] =?UTF-8?q?feat(skill-addresses):=20=E6=96=B0?=
 =?UTF-8?q?=E5=A2=9E=E6=8A=80=E8=83=BD=E5=9C=B0=E5=9D=80=E4=BF=AE=E6=94=B9?=
 =?UTF-8?q?=E8=AE=B0=E5=BD=95=EF=BC=8C=E6=94=AF=E6=8C=81=E5=8D=95=E6=AC=A1?=
 =?UTF-8?q?=E8=AE=B0=E5=BD=95=E5=9B=9E=E9=80=80?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

后端：
- 每次「应用」写一份 {id}.manifest.json（记录本次 from→to 替换明细/文件数/处数），
  放在备份目录的同级，避免被回退误恢复或计入文件数
- GET /backups 返回每条记录的 replaced_total + replacements；POST /rollback/{id} 回退单条

前端：
- 把「撤销上次」升级为「修改记录」面板：每条显示时间/改了哪些地址/文件数，
  带「回退此次」（二次确认）；应用后自动展开
- BackupEntry 类型扩展 replacedTotal + replacements

同步更新前后端 CLAUDE.md。

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 CLAUDE.md                                     |   2 +-
 frontend-web/src/pages/SkillAddressesPage.tsx | 146 +++++++++++++++---
 .../api/skill-addresses.ts                    |  16 +-
 offline-backend-20260512/backend/CLAUDE.md    |  13 +-
 .../app/gateway/routers/skill_addresses.py    |  96 +++++++++++-
 5 files changed, 245 insertions(+), 28 deletions(-)

diff --git a/CLAUDE.md b/CLAUDE.md
index b1f29437..2961555f 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -82,7 +82,7 @@ The frontend (`frontend-web/`) is a React 19 + Vite SPA:
 - **轻应用中心 (Light App Center)**: a sibling of 任务管理 in the sidebar with two pages — 应用管理 (admin registers `window_open` or `iframe` mini-apps with a route-param builder: login params resolved from `localStorage.userInfo`, theme params literal) and 应用列表 (card grid of `window_open` apps, `window.open` launch). `iframe` apps are mounted under a chosen parent menu (XK/PG/TY/综合管理) and **dynamically injected into the sidebar** by `core/page-layout/use-sidebar-menu.ts` (`useSidebarMenuItems`), opening in the `LightAppIframeView` host. Client + React Query hooks in `strategy-components/api/light-apps.ts`; URL builder/constants in `strategy-components/lib/light-app.ts`. Backend: `/api/light-apps` (global/shared, admin-gated writes).
 - **菜单管理 (Menu Management, admin-only)**: an **overlay** over the static sidebar menu — `MenuManagementPage` (`pages/MenuManagementPage.tsx`, route `/page/strategy/admin/menu-management`, opened from the admin-only 「菜单管理」 item in the **bottom-left settings dropdown** of `components/page-sidebar.tsx`). Admins rename / 停用·启用 / reorder (↑↓ or @dnd-kit drag) / re-parent (horizontal drag **or** the edit dialog's 「移动到」parent selector) any node across 一级/二级/三级, **including injected iframe light-app nodes** (`light-app-<appId>`). The static tree (`core/page-layout/sidebar-menu.ts`) stays the source of icons/routes; only per-node deltas (`label`/`disabled`/`parentId`/`sortOrder`) persist to the backend. Merge logic in `core/page-layout/menu-overrides.ts` (`applyMenuOverrides`, applied in `use-sidebar-menu.ts` after light-app injection, before role filter; disabling a node hides its whole subtree) — strict **3-level cap** (`wouldExceedDepth`). Client `strategy-components/api/menu-overrides.ts`. Backend: `/api/menu-overrides` (global/shared; open read, admin atomic replace-all).
 - **敏感词管理 (Sensitive Word Management)**: configures the **展示层脱敏映射** (原始词 → 替换词). **Gated on the username `lqq`** (not the admin role) — `getAccountDisplayName().toLowerCase() === "lqq"` (username = email `@` prefix, else `user_id`, same as the profile page); the dropdown entry, the route guard, and the page-level guard all use this check. `src/lib/desensitize.ts` ships a frozen seed map `SEED_DESENSITIZE_MAP` (the deployment seed) plus a live mutable `DESENSITIZE_MAP` (initialized from the seed so `desensitize()` works before any fetch). `DesensitizeWordsLoader` (mounted in `App.tsx` `AppProviders`) fetches `/api/sensitive-words` on startup and, if non-empty, calls `replaceDesensitizeMap(...)` to make the backend config the source of truth (empty/failed → keep seed). Page `SensitiveWordsPage.tsx` (route `/page/strategy/admin/sensitive-words`, opened from the 「敏感词管理」 item in the **bottom-left settings dropdown** of `components/page-sidebar.tsx`, present in **both** full + collapsed sidebar modes) lists/adds/edits/启停/deletes words and saves the **whole set atomically** (replace-all); a 「导入种子词」 button merges any missing `SEED_DESENSITIZE_MAP` entries into the editor. Client + hooks in `strategy-components/api/sensitive-words.ts` (`wordsToMap` folds enabled rows into the runtime map). Backend: `/api/sensitive-words` (global/shared; open read, write allowed for admin **or** the `lqq` account), mirroring 菜单管理's store shape. **侧边栏 + 导航栏 menu labels are desensitized** at the data source: `useSidebarMenuItems`/`useActiveThirdLevel` (`core/page-layout/use-sidebar-menu.ts`) map every node's `label` through a desensitizer built **directly from the `/api/sensitive-words` query** (`makeDesensitizer(wordsToMap(words))`, falling back to the global seed `desensitize` while empty) — this sidesteps the `replaceDesensitizeMap` effect-timing race. Only `label` is rewritten; `path`/`id`/`icon` are untouched so routing/highlight are unaffected. All nav surfaces (`page-sidebar.tsx`, `page-sidebar-dark.tsx`, `page-sidebar-v1.tsx`, `strategy-components/Header.tsx`, `PageRoutes.tsx`) consume this hook and are covered; `MenuManagementPage` uses the raw tree so admins still edit real labels.
-- **技能地址管理 (Skill Address Management, admin-only)**: 扫描运行中的技能（`skills/{public,custom}` 下所有 `.md`/`.py`）里的 **http(s) URL 与 IP/IP:端口**，按唯一地址聚合全部出现位置（跨 .md/.py），支持**批量替换**（如把 UAT 地址一次性改成生产）。主体抽成可复用组件 `SkillAddressesPanel`（`pages/SkillAddressesPage.tsx` 导出，默认导出 `SkillAddressesPage` 仍保留旧路由 `/page/strategy/admin/skill-addresses` 作直链回退），**作为「技能地址管理」标签页嵌入管理员「技能管理」页 `pages/AdminSkillsPage.tsx`**（该页加 Tabs：「技能列表」+「技能地址管理」；入口在「设置和更多」下拉的「技能管理」`workspace-nav-menu.tsx` → `/page/workspace/admin/skills`，不再单列于 `components/page-sidebar.tsx` 左下角设置下拉）：打开自动扫描并带**进度条**（扫描逐技能 / 应用逐文件 X/N，应对 150+ 技能），URL/IP 类型过滤 + 搜索 + 「片段批量套用」（如所有 `uat.4.cn` → `prod.4.cn`）、每地址「替换为」输入框（可展开看 `file:line` 片段）、内联**预览**、**应用**、**撤销上次**（回滚）。client + NDJSON 流式解析在 `strategy-components/api/skill-addresses.ts`。后端 `/api/skill-addresses`（**仅管理员**）：所有文件 IO 走 `asyncio.to_thread` 不阻塞事件循环；`GET /scan`(+`/scan/stream`)、`POST /preview`、`POST /apply`(+`/apply/stream`)、`GET /backups`、`POST /rollback/{id}`；**子串安全替换**（基于 token span，不会误伤子串）、`expected_total` 乐观锁（409）、写前自动磁盘备份到 `{skills_root}/.address-edits/`、应用后刷新技能系统提示缓存。纯引擎 `deerflow/skills/address_scan.py`，router `app/gateway/routers/skill_addresses.py`，tests `tests/test_skill_addresses.py`。范围仅运行中技能，不动 `deploy/minimal/skills/`。
+- **技能地址管理 (Skill Address Management, admin-only)**: 扫描运行中的技能（`skills/{public,custom}` 下所有 `.md`/`.py`）里的 **http(s) URL 与 IP/IP:端口**，按唯一地址聚合全部出现位置（跨 .md/.py），支持**批量替换**（如把 UAT 地址一次性改成生产）。主体抽成可复用组件 `SkillAddressesPanel`（`pages/SkillAddressesPage.tsx` 导出，默认导出 `SkillAddressesPage` 仍保留旧路由 `/page/strategy/admin/skill-addresses` 作直链回退），**作为「技能地址管理」标签页嵌入管理员「技能管理」页 `pages/AdminSkillsPage.tsx`**（该页加 Tabs：「技能列表」+「技能地址管理」；入口在「设置和更多」下拉的「技能管理」`workspace-nav-menu.tsx` → `/page/workspace/admin/skills`，不再单列于 `components/page-sidebar.tsx` 左下角设置下拉）：打开自动扫描并带**进度条**（扫描逐技能 / 应用逐文件 X/N，应对 150+ 技能），URL/IP 类型过滤 + 搜索 + 「片段批量套用」（如所有 `uat.4.cn` → `prod.4.cn`）、每地址「替换为」输入框（可展开看 `file:line` 片段）、内联**预览**、**应用**、**多关键词检索**（空格分隔全部命中），以及**修改记录**面板——每次应用生成一条记录（显示时间/改了哪些地址 from→to/文件数/处数），可**针对单次记录回退**（二次确认）。client + NDJSON 流式解析在 `strategy-components/api/skill-addresses.ts`。后端 `/api/skill-addresses`（**仅管理员**）：所有文件 IO 走 `asyncio.to_thread` 不阻塞事件循环；`GET /scan`(+`/scan/stream`)、`POST /preview`、`POST /apply`(+`/apply/stream`)、`GET /backups`、`POST /rollback/{id}`；**子串安全替换**（基于 token span，不会误伤子串）、`expected_total` 乐观锁（409）、写前自动磁盘备份到 `{skills_root}/.address-edits/`（同时写一份 `{id}.manifest.json` 记录本次替换明细，供修改记录展示）、应用后刷新技能系统提示缓存。纯引擎 `deerflow/skills/address_scan.py`，router `app/gateway/routers/skill_addresses.py`，tests `tests/test_skill_addresses.py`。范围仅运行中技能，不动 `deploy/minimal/skills/`。
 - **按钮管理 (Task Button Management, admin-only)**: configures the **可配置跳转按钮** rendered in the left task panel of the task deep-link workspaces (rwfx / 3qfx / xdfx, see `TaskFxWorkspace`). Each button = `{business, label, linkType, target, appendTaskId, idParam, idKind, enabled, sortOrder}`; `linkType` is `url` (外链 `window.open`, optional append of `?{idParam}={id}` — **`idParam` is the editable query-param name**, e.g. `id`/`task_id`), `business` (内部切换到另一业务, reuses `resolveTaskDeeplinkTarget`, always task id), or `purpose` (弹出当前任务的「任务目的」表格 via the shared `TaskPurposeDialog`, no target — configurable for any page incl. xdfx). **`idKind` (`task`/`action`) only matters for xdfx**: its deep-link `?taskId=` is an 行动id and the 任务id is resolved at login (`fetchTaskIdFromAction`) — `TaskContext` now also stashes the raw `actionId`, so a url button can append either (default `task`). Page `TaskButtonsPage.tsx` (route `/page/strategy/admin/task-buttons`, **admin-gated**, entry「按钮管理」in the bottom-left settings dropdown of `components/page-sidebar.tsx`, both full + collapsed modes) has per-business tabs, edits name/type/target/启停/排序/增删, saves the **whole set atomically** (replace-all), plus 「导入默认」 to merge code-seed defaults. **Seed/fallback** (like 敏感词管理): only when the whole table is empty (fresh deploy, never saved) do the workspace + page fall back to `defaultTaskButtons(cfg)` — rwfx's url buttons (编辑/下一步/3q详情) built from the **live** `/api/public/config/task-deeplink` plus cross-business switch buttons for **all three** businesses (each switches to the other two — rwfx→三情/行动, 3qfx→任务/行动, xdfx→任务/三情). Once saved, the store is the sole source of truth. The old hardcoded jump buttons were migrated out: `TaskFxWorkspace` no longer hardcodes 切换到三情分析, `TaskActionsPanel` keeps only the non-jump 清除记录, and `TaskPurposePanel` dropped its 3q详情 button (now configurable) **and defaults the 目的树 table to collapsed** (chevron toggle). Helpers `core/auth/task-buttons.ts` (`TaskButton`, `defaultTaskButtons`, `resolveTaskButtons`, `buttonsForBusiness`, `runTaskButton`), client+hooks `strategy-components/api/task-buttons.ts`. Backend: `/api/task-buttons` (global/shared; open read, admin replace-all), store `deerflow.persistence.task_buttons` (table `task_buttons`, auto-created by `create_all`), wired `app.state.task_button_store`. Tests: `tests/test_task_buttons.py`.
 - **State**: TanStack Query for server state, context providers (`SubtasksProvider`, `ArtifactsProvider`, `PromptInputProvider`) scoped to `WorkspaceRoutes`
 - **LLM streaming**: `@langchain/langgraph-sdk` streams messages from the backend via SSE
diff --git a/frontend-web/src/pages/SkillAddressesPage.tsx b/frontend-web/src/pages/SkillAddressesPage.tsx
index 3fa96662..46f499a6 100644
--- a/frontend-web/src/pages/SkillAddressesPage.tsx
+++ b/frontend-web/src/pages/SkillAddressesPage.tsx
@@ -9,6 +9,7 @@ import {
   Eye,
   FileCode,
   Globe,
+  History,
   Info,
   Loader2,
   Network,
@@ -90,6 +91,11 @@ export function SkillAddressesPanel() {
   const [applyProg, setApplyProg] = useState<ApplyProgress | null>(null);
   const [lastBackupId, setLastBackupId] = useState<string | null>(null);
 
+  // 修改记录（每次「应用」生成一条可单独回退的记录）。
+  const [showHistory, setShowHistory] = useState(false);
+  const [confirmRollbackId, setConfirmRollbackId] = useState<string | null>(null);
+  const [rollingBack, setRollingBack] = useState(false);
+
   const [segFind, setSegFind] = useState("");
   const [segRepl, setSegRepl] = useState("");
 
@@ -205,6 +211,7 @@ export function SkillAddressesPanel() {
       setLastBackupId(result.backupId);
       setEdits({});
       setPreview(null);
+      setShowHistory(true); // 应用后自动展开修改记录，便于查看/回退本次
       await refetchBackups();
       await runScan(kinds);
     } catch (err) {
@@ -215,23 +222,29 @@ export function SkillAddressesPanel() {
     }
   };
 
-  const runRollback = async () => {
-    const id = lastBackupId ?? backups[0]?.backupId;
-    if (!id) {
-      toast.error("没有可撤销的备份");
-      return;
-    }
+  /** 回退单次修改记录：从该次备份恢复文件（不影响其他记录）。 */
+  const rollbackOne = async (backupId: string) => {
+    setRollingBack(true);
     try {
-      const { restored } = await rollbackBackup(id);
-      toast.success(`已从备份恢复 ${restored} 个文件`);
-      setLastBackupId(null);
+      const { restored } = await rollbackBackup(backupId);
+      toast.success(`已从该记录回退 ${restored} 个文件`);
+      setConfirmRollbackId(null);
+      if (backupId === lastBackupId) setLastBackupId(null);
       await refetchBackups();
       await runScan(kinds);
     } catch (err) {
-      toast.error(err instanceof Error ? err.message : "撤销失败");
+      toast.error(err instanceof Error ? err.message : "回退失败");
+    } finally {
+      setRollingBack(false);
     }
   };
 
+  const fmtTime = (iso: string) => {
+    if (!iso) return "";
+    const d = new Date(iso);
+    return Number.isNaN(d.getTime()) ? iso : d.toLocaleString();
+  };
+
   // 多关键词检索：空格分隔，每个关键词都要命中（地址或技能名任一包含）才显示。
   // 这样「uat 4.cn」能同时收窄到既含 uat 又含 4.cn 的地址，应对地址过多的场景。
   const tokens = useMemo(
@@ -247,7 +260,6 @@ export function SkillAddressesPanel() {
   }, [groups, tokens]);
 
   const totalOccurrences = useMemo(() => groups.reduce((s, g) => s + g.count, 0), [groups]);
-  const canRollback = !!(lastBackupId || backups.length > 0);
 
   if (!isAdmin) return <Navigate to="/page/workspace/chats/new" replace />;
 
@@ -278,13 +290,16 @@ export function SkillAddressesPanel() {
               <div className="flex items-center gap-2.5">
                 <button
                   type="button"
-                  onClick={runRollback}
-                  disabled={!canRollback || scanning || applying}
-                  className="inline-flex cursor-pointer items-center gap-2 rounded-xl border border-slate-200 bg-white px-3.5 py-2 text-sm font-semibold text-slate-600 shadow-sm transition-all hover:border-amber-300 hover:bg-amber-50 hover:text-amber-700 disabled:cursor-not-allowed disabled:opacity-50"
-                  title="从最近一次的磁盘备份恢复"
+                  onClick={() => setShowHistory((v) => !v)}
+                  className={`inline-flex cursor-pointer items-center gap-2 rounded-xl border px-3.5 py-2 text-sm font-semibold shadow-sm transition-all ${
+                    showHistory
+                      ? "border-amber-300 bg-amber-50 text-amber-700"
+                      : "border-slate-200 bg-white text-slate-600 hover:border-amber-300 hover:bg-amber-50 hover:text-amber-700"
+                  }`}
+                  title="查看每次修改记录，可针对单次回退"
                 >
-                  <Undo2 className="size-[15px]" />
-                  撤销上次
+                  <History className="size-[15px]" />
+                  修改记录{backups.length > 0 ? `（${backups.length}）` : ""}
                 </button>
                 <button
                   type="button"
@@ -302,7 +317,7 @@ export function SkillAddressesPanel() {
               <Info className="mt-0.5 size-4 shrink-0 text-slate-400" />
               <span>
                 扫描运行中的技能（<code className="rounded bg-slate-100 px-1">skills/public</code> 与{" "}
-                <code className="rounded bg-slate-100 px-1">skills/custom</code> 下所有 .md / .py），按<strong>唯一地址</strong>聚合其全部出现位置（跨 .md 与 .py）。填写「替换为」后<strong>预览</strong>命中再<strong>应用</strong>；应用前会自动磁盘备份，可一键<strong>撤销</strong>。替换为精确整串匹配，不会误伤子串。
+                <code className="rounded bg-slate-100 px-1">skills/custom</code> 下所有 .md / .py），按<strong>唯一地址</strong>聚合其全部出现位置（跨 .md 与 .py）。填写「替换为」后<strong>预览</strong>命中再<strong>应用</strong>；每次应用都会自动磁盘备份并生成一条<strong>修改记录</strong>，可在「修改记录」里针对<strong>单次记录回退</strong>。替换为精确整串匹配，不会误伤子串。
               </span>
             </p>
           </div>
@@ -320,6 +335,101 @@ export function SkillAddressesPanel() {
           </div>
         )}
 
+        {/* 修改记录：每次应用一条，可针对单次回退 */}
+        {showHistory && (
+          <div className="mb-6 overflow-hidden rounded-2xl border border-amber-200/70 bg-white shadow-sm">
+            <div className="flex items-center justify-between border-b border-amber-100 bg-amber-50/50 px-4 py-3 text-sm font-semibold text-slate-700">
+              <span className="flex items-center gap-2">
+                <History className="size-4 text-amber-600" />
+                修改记录（{backups.length}）
+              </span>
+              <button
+                type="button"
+                onClick={() => refetchBackups()}
+                className="rounded-lg border border-slate-200 bg-white px-2.5 py-1 text-xs font-medium text-slate-500 transition-colors hover:border-slate-300 hover:text-slate-700"
+              >
+                刷新
+              </button>
+            </div>
+            {backups.length === 0 ? (
+              <div className="px-4 py-10 text-center text-sm text-slate-400">
+                暂无修改记录。应用一次替换后，这里会生成一条可回退的记录。
+              </div>
+            ) : (
+              <ul className="max-h-96 divide-y divide-slate-100 overflow-auto">
+                {backups.map((b) => (
+                  <li key={b.backupId} className="px-4 py-3">
+                    <div className="flex flex-col gap-2.5 lg:flex-row lg:items-start lg:justify-between">
+                      <div className="min-w-0 flex-1">
+                        <div className="flex items-center gap-2 text-sm font-medium text-slate-700">
+                          {fmtTime(b.createdAt)}
+                          {b.backupId === lastBackupId && (
+                            <span className="rounded-md bg-amber-100 px-1.5 py-0.5 text-[11px] font-semibold text-amber-700">最近</span>
+                          )}
+                          <span className="rounded bg-slate-100 px-1.5 py-0.5 font-mono text-xs text-slate-500">
+                            {b.files} 文件 / {b.replacedTotal} 处
+                          </span>
+                        </div>
+                        {b.replacements.length > 0 ? (
+                          <div className="mt-1.5 flex flex-wrap gap-1.5">
+                            {b.replacements.slice(0, 4).map((r, i) => (
+                              <span key={i} className="inline-flex items-center gap-1 rounded-md bg-slate-50 px-1.5 py-0.5 font-mono text-[11px] text-slate-500">
+                                <span className="line-through">{r.from}</span>
+                                <span className="text-slate-400">→</span>
+                                <span className="font-medium text-blue-700">{r.to}</span>
+                              </span>
+                            ))}
+                            {b.replacements.length > 4 && (
+                              <span className="text-[11px] text-slate-400">等 {b.replacements.length} 项</span>
+                            )}
+                          </div>
+                        ) : (
+                          <div className="mt-1 font-mono text-xs text-slate-400">{b.backupId}</div>
+                        )}
+                      </div>
+
+                      <div className="flex shrink-0 items-center gap-2">
+                        {confirmRollbackId === b.backupId ? (
+                          <>
+                            <button
+                              type="button"
+                              onClick={() => rollbackOne(b.backupId)}
+                              disabled={rollingBack || scanning || applying}
+                              className="inline-flex cursor-pointer items-center gap-1.5 rounded-lg border border-amber-500 bg-amber-500 px-3 py-1.5 text-xs font-semibold text-white transition-all hover:bg-amber-600 disabled:cursor-not-allowed disabled:opacity-50"
+                            >
+                              {rollingBack ? <Loader2 className="size-3.5 animate-spin" /> : <Undo2 className="size-3.5" />}
+                              确认回退
+                            </button>
+                            <button
+                              type="button"
+                              onClick={() => setConfirmRollbackId(null)}
+                              disabled={rollingBack}
+                              className="rounded-lg border border-slate-200 bg-white px-3 py-1.5 text-xs font-medium text-slate-500 transition-colors hover:border-slate-300 hover:text-slate-700"
+                            >
+                              取消
+                            </button>
+                          </>
+                        ) : (
+                          <button
+                            type="button"
+                            onClick={() => setConfirmRollbackId(b.backupId)}
+                            disabled={scanning || applying}
+                            className="inline-flex cursor-pointer items-center gap-1.5 rounded-lg border border-slate-200 bg-white px-3 py-1.5 text-xs font-semibold text-slate-600 transition-all hover:border-amber-300 hover:bg-amber-50 hover:text-amber-700 disabled:cursor-not-allowed disabled:opacity-50"
+                            title="把文件恢复到这次修改之前"
+                          >
+                            <Undo2 className="size-3.5" />
+                            回退此次
+                          </button>
+                        )}
+                      </div>
+                    </div>
+                  </li>
+                ))}
+              </ul>
+            )}
+          </div>
+        )}
+
         {/* 工具栏：类型过滤 + 搜索 + 片段批量套用 */}
         <div className="mb-6 space-y-4 rounded-2xl border border-slate-200/60 bg-white p-4 shadow-sm md:p-5">
           <div className="flex flex-col items-start justify-between gap-4 md:flex-row md:items-center">
diff --git a/frontend-web/src/strategy-components/api/skill-addresses.ts b/frontend-web/src/strategy-components/api/skill-addresses.ts
index d52582a0..fafc387d 100644
--- a/frontend-web/src/strategy-components/api/skill-addresses.ts
+++ b/frontend-web/src/strategy-components/api/skill-addresses.ts
@@ -80,10 +80,18 @@ export interface ApplyResult {
   warnings: string[];
 }
 
+export interface BackupReplacement {
+  from: string;
+  to: string;
+  kind: AddressKind;
+}
+
 export interface BackupEntry {
   backupId: string;
   files: number;
   createdAt: string;
+  replacedTotal: number;
+  replacements: BackupReplacement[];
 }
 
 /** 扫描进度（逐技能）。 */
@@ -249,7 +257,13 @@ export async function applyReplacementsStream(
 export async function listBackups(): Promise<BackupEntry[]> {
   const res = await apiFetch(`${BASE()}/backups`, { method: "GET" });
   const data = await readJson<{ backups: any[] }>(res, "读取备份");
-  return (data.backups ?? []).map((b) => ({ backupId: b.backup_id, files: b.files ?? 0, createdAt: b.created_at ?? "" }));
+  return (data.backups ?? []).map((b) => ({
+    backupId: b.backup_id,
+    files: b.files ?? 0,
+    createdAt: b.created_at ?? "",
+    replacedTotal: b.replaced_total ?? 0,
+    replacements: (b.replacements ?? []).map((r: any) => ({ from: r.from, to: r.to, kind: r.kind })),
+  }));
 }
 
 export async function rollbackBackup(backupId: string): Promise<{ restored: number }> {
diff --git a/offline-backend-20260512/backend/CLAUDE.md b/offline-backend-20260512/backend/CLAUDE.md
index d233650f..e47e3ca8 100644
--- a/offline-backend-20260512/backend/CLAUDE.md
+++ b/offline-backend-20260512/backend/CLAUDE.md
@@ -556,6 +556,14 @@ snippet) and replaced together.
   `{skills_root}/.address-edits/{timestamp}/` (the `.`-prefixed dir is excluded
   from scans), then `refresh_skills_system_prompt_cache_async()` so edits take
   effect immediately. Registered in `app.py` after `sensitive_words.router`.
+  **Per-edit history:** each apply also writes a sibling manifest
+  `{skills_root}/.address-edits/{timestamp}.manifest.json` (best-effort) capturing
+  `{created_at, replaced_total, replacements:[{from,to,kind}], changed_files}`; the
+  sibling-not-child placement keeps it out of both the restore set and the file
+  count. `GET /backups` returns each record enriched with `replaced_total` +
+  `replacements` (from the manifest; falls back to dir mtime when absent), and
+  `POST /rollback/{backup_id}` restores a single record — so the UI shows a full
+  modification history and rolls back any one entry independently.
 - **Frontend** — the panel is the reusable `SkillAddressesPanel` (exported from
   `pages/SkillAddressesPage.tsx`; its default export keeps the standalone route
   `/page/strategy/admin/skill-addresses` as a direct-link fallback). It is
@@ -567,7 +575,10 @@ snippet) and replaced together.
   auto-scan on open with a **progress bar** (扫描逐技能 / 应用逐文件 X/N), URL/IP
   type filter, search, a「片段批量套用」helper (e.g. all `uat.4.cn` → `prod.4.cn`),
   per-address「替换为」input with expandable occurrences, inline 预览 panel, 应用,
-  and 撤销上次 (rollback). Client + NDJSON stream parsing in
+  **multi-keyword search** (space-separated, all-must-hit), and a **修改记录**
+  (modification history) panel — each apply is one record (time / which addresses
+  changed from→to / file & replacement counts) with a per-record 回退此次
+  (single-record rollback, with inline confirm). Client + NDJSON stream parsing in
   `strategy-components/api/skill-addresses.ts`. Scope is the running skills only;
   the `deploy/minimal/skills/` copy is not touched (sync separately).
 
diff --git a/offline-backend-20260512/backend/app/gateway/routers/skill_addresses.py b/offline-backend-20260512/backend/app/gateway/routers/skill_addresses.py
index 82d3f679..ed302c29 100644
--- a/offline-backend-20260512/backend/app/gateway/routers/skill_addresses.py
+++ b/offline-backend-20260512/backend/app/gateway/routers/skill_addresses.py
@@ -145,10 +145,20 @@ class ApplyResponse(BaseModel):
     warnings: list[str]
 
 
+class ReplacementSummary(BaseModel):
+    model_config = ConfigDict(populate_by_name=True)
+
+    from_: str = Field(serialization_alias="from")
+    to: str
+    kind: str
+
+
 class BackupEntry(BaseModel):
     backup_id: str
     files: int
     created_at: str
+    replaced_total: int = 0
+    replacements: list[ReplacementSummary] = Field(default_factory=list)
 
 
 class BackupListResponse(BaseModel):
@@ -292,6 +302,51 @@ def _backup_file(root: Path, backup_id: str, path: Path, rel: str) -> None:
     shutil.copy2(path, dest)
 
 
+_MANIFEST_SUFFIX = ".manifest.json"
+
+
+def _manifest_path(root: Path, backup_id: str) -> Path:
+    """Manifest lives as a *sibling* of the backup dir (``{id}.manifest.json``),
+    not inside it — so it is neither restored back into the skills tree on
+    rollback nor counted as a backed-up file."""
+
+    return root / _BACKUP_DIRNAME / f"{backup_id}{_MANIFEST_SUFFIX}"
+
+
+def _write_manifest(
+    root: Path,
+    backup_id: str,
+    norm: list[tuple[str, str, str]],
+    changed: list["ChangedFile"],
+    replaced_total: int,
+    created_at: str,
+) -> None:
+    """Record what this edit changed so the UI can show a meaningful, per-edit
+    history (原地址 → 新地址, 文件数, 处数) and roll a single record back."""
+
+    data = {
+        "created_at": created_at,
+        "replaced_total": replaced_total,
+        "replacements": [{"from": frm, "to": to, "kind": kind} for frm, to, kind in norm],
+        "changed_files": [{"file": c.file, "count": c.count} for c in changed],
+    }
+    try:  # best-effort: a metadata write failure must not fail the apply itself
+        path = _manifest_path(root, backup_id)
+        path.parent.mkdir(parents=True, exist_ok=True)
+        with open(path, "w", encoding="utf-8") as fh:
+            json.dump(data, fh, ensure_ascii=False)
+    except OSError:
+        logger.warning("Failed to write skill-address backup manifest for %s", backup_id, exc_info=True)
+
+
+def _read_manifest(root: Path, backup_id: str) -> dict | None:
+    try:
+        with open(_manifest_path(root, backup_id), "r", encoding="utf-8") as fh:
+            return json.load(fh)
+    except (OSError, json.JSONDecodeError):
+        return None
+
+
 # ── scan extraction shared by one-shot + stream ───────────────────────────────
 
 
@@ -475,7 +530,7 @@ def _new_backup_id() -> str:
 async def apply_replacements_route(request: Request, body: ApplyRequest) -> ApplyResponse:
     await _require_admin(request)
     root = _skills_root()
-    mapping, _norm, warnings = _build_mapping(body.replacements)
+    mapping, norm, warnings = _build_mapping(body.replacements)
     if not mapping:
         return ApplyResponse(changed_files=[], replaced_total=0, backup_id=None, warnings=warnings)
 
@@ -501,6 +556,9 @@ async def apply_replacements_route(request: Request, body: ApplyRequest) -> Appl
         changed.append(ChangedFile(file=rel, count=cnt))
         replaced_total += cnt
 
+    if backup_id:
+        await asyncio.to_thread(_write_manifest, root, backup_id, norm, changed, replaced_total, datetime.now(UTC).isoformat())
+
     await _refresh_cache_safe()
     logger.info("Skill address apply: %d files, %d replacements, backup=%s", len(changed), replaced_total, backup_id)
     return ApplyResponse(changed_files=changed, replaced_total=replaced_total, backup_id=backup_id, warnings=warnings)
@@ -518,7 +576,7 @@ async def apply_replacements_stream(request: Request, body: ApplyRequest) -> Str
 
     await _require_admin(request)
     root = _skills_root()
-    mapping, _norm, warnings = _build_mapping(body.replacements)
+    mapping, norm, warnings = _build_mapping(body.replacements)
 
     plan: list[tuple[Path, str, int, str]] = []
     current_total = 0
@@ -554,6 +612,9 @@ async def apply_replacements_stream(request: Request, body: ApplyRequest) -> Str
             done += 1
             yield _ndjson({"type": "progress", "done": done, "total_files": len(plan), "file": rel, "count": cnt})
 
+        if backup_id:
+            await asyncio.to_thread(_write_manifest, root, backup_id, norm, changed, replaced_total, datetime.now(UTC).isoformat())
+
         await _refresh_cache_safe()
         logger.info("Skill address apply(stream): %d files, %d replacements, backup=%s", len(changed), replaced_total, backup_id)
         result = ApplyResponse(changed_files=changed, replaced_total=replaced_total, backup_id=backup_id, warnings=warnings)
@@ -573,11 +634,32 @@ def _list_backups(root: Path) -> list[BackupEntry]:
             if not d.is_dir():
                 continue
             file_count = sum(1 for p in d.rglob("*") if p.is_file())
-            try:
-                created = datetime.fromtimestamp(d.stat().st_mtime, UTC).isoformat()
-            except OSError:
-                created = ""
-            entries.append(BackupEntry(backup_id=d.name, files=file_count, created_at=created))
+            manifest = _read_manifest(root, d.name)
+            created = ""
+            replaced_total = 0
+            replacements: list[ReplacementSummary] = []
+            if manifest:
+                created = str(manifest.get("created_at") or "")
+                replaced_total = int(manifest.get("replaced_total") or 0)
+                replacements = [
+                    ReplacementSummary(from_=str(r.get("from", "")), to=str(r.get("to", "")), kind=str(r.get("kind", "")))
+                    for r in manifest.get("replacements", [])
+                    if isinstance(r, dict)
+                ]
+            if not created:
+                try:
+                    created = datetime.fromtimestamp(d.stat().st_mtime, UTC).isoformat()
+                except OSError:
+                    created = ""
+            entries.append(
+                BackupEntry(
+                    backup_id=d.name,
+                    files=file_count,
+                    created_at=created,
+                    replaced_total=replaced_total,
+                    replacements=replacements,
+                )
+            )
     entries.sort(key=lambda e: e.backup_id, reverse=True)
     return entries
 

From 7c55a577adab218fdbd8eb7eafc28d7b3fd857b5 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E5=90=89=E9=91=AB=E5=8D=87?= <xsji@kland.com.cn>
Date: Mon, 22 Jun 2026 19:36:30 +0800
Subject: [PATCH 16/17] =?UTF-8?q?feat(open-chat):=20=E5=BC=80=E6=94=BE?=
 =?UTF-8?q?=E9=97=AE=E7=AD=94=E6=8E=A5=E5=8F=A3=E5=9B=9E=E4=BC=A0=E7=94=9F?=
 =?UTF-8?q?=E6=88=90=E6=96=87=E4=BB=B6=20+=20=E4=B8=B4=E6=97=B6=E4=B8=8D?=
 =?UTF-8?q?=E7=95=99=E5=AD=98?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

- /api/open/chat[/stream] 新增 include_files(默认开)，把智能体在 outputs 目录
  生成的文件(name/virtual_path/encoding(text|base64)/content)随结果回传，
  解决"报告写进文件、answer 只回一段总结"导致拿不到完整内容的问题；
  既无 answer 也无文件时返回 502。
- 无状态运行标记 ephemeral：lead agent 跳过记忆与知识库中间件(不召回/不留存/
  不自动沉淀)，取完文件后清理临时线程目录，做到"通过本接口的聊天记录不存储"。

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 offline-backend-20260512/backend/CLAUDE.md    |   2 +-
 .../backend/app/gateway/routers/open_chat.py  | 102 +++++++++++++++++-
 .../deerflow/agents/lead_agent/agent.py       |  16 ++-
 3 files changed, 111 insertions(+), 9 deletions(-)

diff --git a/offline-backend-20260512/backend/CLAUDE.md b/offline-backend-20260512/backend/CLAUDE.md
index e47e3ca8..6fa38cdd 100644
--- a/offline-backend-20260512/backend/CLAUDE.md
+++ b/offline-backend-20260512/backend/CLAUDE.md
@@ -243,7 +243,7 @@ FastAPI application on port 8001 with health check at `GET /health`. Set `GATEWA
 | **Thread Runs** (`/api/threads/{id}/runs`) | `POST /` - create background run; `POST /stream` - create + SSE stream; `POST /wait` - create + block; `GET /` - list runs; `GET /{rid}` - run details; `POST /{rid}/cancel` - cancel; `GET /{rid}/join` - join SSE; `GET /{rid}/messages` - paginated messages `{data, has_more}`; `GET /{rid}/events` - full event stream; `GET /../messages` - thread messages with feedback; `GET /../token-usage` - aggregate tokens |
 | **Feedback** (`/api/threads/{id}/runs/{rid}/feedback`) | `PUT /` - upsert feedback; `DELETE /` - delete user feedback; `POST /` - create feedback; `GET /` - list feedback; `GET /stats` - aggregate stats; `DELETE /{fid}` - delete specific |
 | **Runs** (`/api/runs`) | `POST /stream` - stateless run + SSE; `POST /wait` - stateless run + block; `GET /{rid}/messages` - paginated messages by run_id `{data, has_more}` (cursor: `after_seq`/`before_seq`); `GET /{rid}/feedback` - list feedback by run_id |
-| **Open Chat** (`/api/open/chat`) | **Unauthenticated** open Q&A entry for external systems (the `/api/open/` prefix is whitelisted in `auth_middleware._PUBLIC_PATH_PREFIXES` + CSRF-exempt in `csrf_middleware.should_check_csrf`). `POST /chat` `{message \| messages[], model_name?, agent_name?, agent_id?, thinking_enabled=false, strip_think=true, thread_id?, recursion_limit=100}` → resolves the agent by **Chinese display name** (`agent_name`, matched against each `.deer-flow/agents/*/config.yaml` `name` via `_resolve_agent_id`; explicit `agent_id` wins if given), builds the lead agent via `make_lead_agent` (so the chosen **agent's** SOUL/tools/skills all apply), `ainvoke`s it once (stateless — no checkpointer, no thread persistence; runs under the `"default"` user bucket), and returns **only the final** `{answer, model_name, agent_name, agent_id, thinking_enabled, thread_id}` (no intermediate tool-call/process). **`POST /chat/stream`** is the **SSE streaming variant** (same request body) for **long tasks** — the blocking `/chat` produces no downstream bytes for the whole run, so a front nginx `proxy_read_timeout` (default 60s) cuts it off with a **504**; the stream keeps the connection warm (a `: keepalive` SSE comment every `_HEARTBEAT_SECONDS=15`s via a background producer-task + queue, plus `event: delta` AI-text increments and `event: progress`), and ends with **`event: result`** carrying the same final `{answer, …}` JSON (or `event: error`). Clients only need to read `event: result`; intermediate events are ignorable. Both endpoints share `_build_run_context` (config/agent/thread). `GET /agents` lists `{name, id}` so callers know what `agent_name` to send. **Thinking off is reliable**: when `thinking_enabled=false` the route also sets `thinking_force_disabled=true` so the lead agent calls `create_chat_model(force_disable_thinking=True)`, which injects **three** OpenAI-compatible "thinking off" shapes into `extra_body` — top-level `enable_thinking=false` (**SiliconFlow / DashScope hybrid models** like `deepseek-ai/DeepSeek-V4-*` on `api.siliconflow.cn` — the knob the nested `chat_template_kwargs` shape does NOT reach), `chat_template_kwargs.enable_thinking=false` (vLLM/Qwen native), and `thinking.type="disabled"` (Zhipu GLM) — even for models that only declare `supports_thinking: true`. For the **web-UI path** (which sends `thinking_enabled=false` but NOT `force_disable`), SiliconFlow models need `when_thinking_disabled: {extra_body: {enable_thinking: false}}` in their `config.yaml` entry (added to `deepseek-ai/DeepSeek-V4-Flash`) for thinking-off to take effect. Sets `is_scheduled_run=true` internally to skip the auto-title LLM call. Router `app/gateway/routers/open_chat.py`; standalone caller `test_open_chat_api.py` (repo root, streaming by default); tests `tests/test_open_chat.py`. |
+| **Open Chat** (`/api/open/chat`) | **Unauthenticated** open Q&A entry for external systems (the `/api/open/` prefix is whitelisted in `auth_middleware._PUBLIC_PATH_PREFIXES` + CSRF-exempt in `csrf_middleware.should_check_csrf`). `POST /chat` `{message \| messages[], model_name?, agent_name?, agent_id?, thinking_enabled=false, strip_think=true, thread_id?, recursion_limit=100}` → resolves the agent by **Chinese display name** (`agent_name`, matched against each `.deer-flow/agents/*/config.yaml` `name` via `_resolve_agent_id`; explicit `agent_id` wins if given), builds the lead agent via `make_lead_agent` (so the chosen **agent's** SOUL/tools/skills all apply), `ainvoke`s it once (stateless — no checkpointer, no thread persistence; runs under the `"default"` user bucket), and returns **only the final** `{answer, model_name, agent_name, agent_id, thinking_enabled, thread_id}` (no intermediate tool-call/process). **`POST /chat/stream`** is the **SSE streaming variant** (same request body) for **long tasks** — the blocking `/chat` produces no downstream bytes for the whole run, so a front nginx `proxy_read_timeout` (default 60s) cuts it off with a **504**; the stream keeps the connection warm (a `: keepalive` SSE comment every `_HEARTBEAT_SECONDS=15`s via a background producer-task + queue, plus `event: delta` AI-text increments and `event: progress`), and ends with **`event: result`** carrying the same final `{answer, …}` JSON (or `event: error`). Clients only need to read `event: result`; intermediate events are ignorable. Both endpoints share `_build_run_context` (config/agent/thread). `GET /agents` lists `{name, id}` so callers know what `agent_name` to send. **Thinking off is reliable**: when `thinking_enabled=false` the route also sets `thinking_force_disabled=true` so the lead agent calls `create_chat_model(force_disable_thinking=True)`, which injects **three** OpenAI-compatible "thinking off" shapes into `extra_body` — top-level `enable_thinking=false` (**SiliconFlow / DashScope hybrid models** like `deepseek-ai/DeepSeek-V4-*` on `api.siliconflow.cn` — the knob the nested `chat_template_kwargs` shape does NOT reach), `chat_template_kwargs.enable_thinking=false` (vLLM/Qwen native), and `thinking.type="disabled"` (Zhipu GLM) — even for models that only declare `supports_thinking: true`. For the **web-UI path** (which sends `thinking_enabled=false` but NOT `force_disable`), SiliconFlow models need `when_thinking_disabled: {extra_body: {enable_thinking: false}}` in their `config.yaml` entry (added to `deepseek-ai/DeepSeek-V4-Flash`) for thinking-off to take effect. Sets `is_scheduled_run=true` internally to skip the auto-title LLM call. **Generated-file return (`include_files=true`, default on)**: an agent that produces its report as a file (writes to `/mnt/user-data/outputs/` + `present_files`) returns only a short summary in `answer`; to recover the real deliverable both endpoints now also return `files: [{name, virtual_path, encoding(text|base64), content}]` — `_collect_output_files` scans the run's `state['thread_data']['outputs_path']` (the open thread is stateless so artifacts can't be fetched by thread, but the endpoint runs server-side and the sandbox already wrote the files to disk), text decoded as utf-8 else base64, capped by `_MAX_OUTPUT_FILES=20` / `_MAX_OUTPUT_FILE_BYTES=5MB`. `/chat` puts `files` on the response; `/chat/stream` puts it in the `event: result` payload. A run with no answer **and** no files now 502s ("…文本答案或文件"). Router `app/gateway/routers/open_chat.py`; standalone caller `test_open_chat_api.py` (repo root, streaming by default); tests `tests/test_open_chat.py`. |
 | **Thread Shares** (`/api/thread-shares`) | Per-conversation sharing with two modes (`thread_shares.mode`). **`import`**: `POST /` `{thread_id, mode}` - create (or reuse) a revocable share code; `GET /{code}/preview` + `POST /{code}/import` - copy that conversation into the caller's account (fresh `thread_id` via `app/gateway/thread_copy.py` + new `threads_meta` row + `user-data` copy; dedup on `metadata.shared_origin`). **`view`**: a public, read-only static page ? see Public Shares below. Both modes: `GET /` - list my codes; `DELETE /{code}` - revoke. |
 | **Public Shares** (`/api/public/shares`) | **Unauthenticated** (the `/api/public/` prefix is whitelisted in `auth_middleware._PUBLIC_PATH_PREFIXES`). `GET /{code}` - read-only conversation snapshot (title + serialized messages + artifacts) for a `view`-mode share; messages are read straight from the latest checkpoint. `GET /{code}/files/{path}` - serves a file referenced by that conversation (images, Markdown, text, PDF, ?) so the shared page's links are openable; path resolution is confined to the owner's thread directory. Active content (`text/html`, `application/xhtml+xml`, `image/svg+xml`) is forced to download (`_public_share_disposition`); everything else is served inline. Missing/revoked/non-`view` codes all return 404. |
 | **Roundtable Drafts** (`/api/roundtable-drafts`) | **Strictly per-user** persistence for the roundtable-planning page (replaces the old browser `localStorage`). `GET /` - list drafts (lightweight meta); `POST /` - create; `GET/PUT/DELETE /{id}` - full read / partial update / delete (all owner-scoped). A draft = one full session (Step 1 intent + Step 2 roundtable), step snapshots stored as JSON in `PortableLongText`. Recommendation history bound to a draft: `GET /{id}/recommendations` - list newest-first; `POST /{id}/recommendations` - append one run (objective/status/rationale/picks/candidates). taskId 深链会商记录 are **not** here — they live in the separate `/api/roundtable-task-drafts` store (see below) so personal history never mixes with task records. Backed by `deerflow.persistence.roundtable_drafts` (tables `roundtable_drafts`, `roundtable_recommend_history`); store wired in `deps.py` as `app.state.roundtable_draft_store`. Tests: `tests/test_roundtable_task_drafts.py`. |
diff --git a/offline-backend-20260512/backend/app/gateway/routers/open_chat.py b/offline-backend-20260512/backend/app/gateway/routers/open_chat.py
index 849e0085..3dad51d2 100644
--- a/offline-backend-20260512/backend/app/gateway/routers/open_chat.py
+++ b/offline-backend-20260512/backend/app/gateway/routers/open_chat.py
@@ -34,11 +34,14 @@
 from __future__ import annotations
 
 import asyncio
+import base64
 import json
 import logging
 import re
+import shutil
 import uuid
 from collections.abc import AsyncIterator
+from pathlib import Path
 from typing import Any, Literal
 
 from fastapi import APIRouter, HTTPException
@@ -79,6 +82,16 @@ class OpenChatRequest(BaseModel):
     strip_think: bool = Field(default=True, description="关闭思考时,从答案里剥离残留的 <think>…</think> 块")
     thread_id: str | None = Field(default=None, description="可选;仅用于文件隔离,留空自动生成")
     recursion_limit: int = Field(default=100, ge=1, le=300, description="智能体单次问答的最大步数")
+    include_files: bool = Field(default=True, description="是否随结果回传智能体在 outputs 目录生成的文件（名称+内容）")
+
+
+class OpenChatFile(BaseModel):
+    """智能体本次运行在 /mnt/user-data/outputs 下生成的一个文件。"""
+
+    name: str  # outputs/ 下的相对路径，如 "报告.md"
+    virtual_path: str  # 沙箱虚拟路径，如 "/mnt/user-data/outputs/报告.md"
+    encoding: Literal["text", "base64"] = "text"  # text=可直接用的文本；base64=二进制需解码
+    content: str
 
 
 class OpenChatResponse(BaseModel):
@@ -88,6 +101,7 @@ class OpenChatResponse(BaseModel):
     agent_id: str | None
     thinking_enabled: bool
     thread_id: str
+    files: list[OpenChatFile] = Field(default_factory=list, description="本次生成的输出文件（含内容）")
 
 
 def _resolve_agent_id(agent_id: str | None, agent_name: str | None) -> str | None:
@@ -171,6 +185,76 @@ def _extract_answer(result: Any, *, strip_think: bool) -> str:
     return answer
 
 
+_OUTPUTS_VIRTUAL_PREFIX = "/mnt/user-data/outputs"
+_MAX_OUTPUT_FILES = 20
+_MAX_OUTPUT_FILE_BYTES = 5 * 1024 * 1024  # 单文件 5MB 上限,避免把大二进制塞进 JSON 响应
+
+
+def _collect_output_files(state: Any, *, include: bool) -> list[OpenChatFile]:
+    """读取本次运行在沙箱 outputs 目录生成的文件,随结果一并回传。
+
+    开放问答是无状态的(不持久化线程),客户端没法用 artifacts 路由按线程取件;但本端点
+    就跑在服务端进程里,沙箱把生成的文件落在 ``state['thread_data']['outputs_path']``,
+    所以这里直接扫描该目录,把文件内容带回(文本按 utf-8,二进制按 base64)。
+    受 ``_MAX_OUTPUT_FILES`` / ``_MAX_OUTPUT_FILE_BYTES`` 限制。
+    """
+    if not include or not isinstance(state, dict):
+        return []
+    thread_data = state.get("thread_data") or {}
+    outputs_path = thread_data.get("outputs_path")
+    if not outputs_path:
+        return []
+    base = Path(outputs_path)
+    if not base.is_dir():
+        return []
+    try:
+        entries = sorted((p for p in base.rglob("*") if p.is_file()), key=lambda p: p.as_posix())
+    except OSError:
+        return []
+
+    files: list[OpenChatFile] = []
+    for entry in entries:
+        if len(files) >= _MAX_OUTPUT_FILES:
+            break
+        try:
+            if entry.stat().st_size > _MAX_OUTPUT_FILE_BYTES:
+                continue
+            data = entry.read_bytes()
+        except OSError:
+            continue
+        rel = entry.relative_to(base).as_posix()
+        try:
+            content = data.decode("utf-8")
+            encoding: Literal["text", "base64"] = "text"
+        except UnicodeDecodeError:
+            content = base64.b64encode(data).decode("ascii")
+            encoding = "base64"
+        files.append(OpenChatFile(name=rel, virtual_path=f"{_OUTPUTS_VIRTUAL_PREFIX}/{rel}", encoding=encoding, content=content))
+    return files
+
+
+def _cleanup_ephemeral_thread(state: Any, thread_id: str | None) -> None:
+    """删除本次开放问答的临时线程目录(open-*),落实"聊天记录不存储"。
+
+    生成文件已在 _collect_output_files 读进内存(随响应回传),这里把磁盘上整个临时线程
+    目录清掉,避免无状态调用在服务端堆积残留。仅对 ``open-*`` 线程生效,best-effort。
+    """
+    if not thread_id or not str(thread_id).startswith("open-"):
+        return
+    thread_data = state.get("thread_data") if isinstance(state, dict) else None
+    outputs_path = (thread_data or {}).get("outputs_path")
+    if not outputs_path:
+        return
+    try:
+        p = Path(outputs_path).resolve()
+        # outputs_path 形如 .../threads/<thread_id>/user-data/outputs;只删到 <thread_id> 这层
+        thread_root = next((parent for parent in p.parents if parent.name == thread_id), None)
+        if thread_root is not None and thread_root.exists():
+            shutil.rmtree(thread_root, ignore_errors=True)
+    except Exception as exc:  # noqa: BLE001 — 清理失败不影响返回结果
+        logger.warning("清理临时线程目录失败 (%s): %s", thread_id, exc)
+
+
 # 流式响应每隔这么久没有真实事件就发一个 SSE 注释心跳,持续制造下行字节,
 # 让前置反向代理(nginx proxy_read_timeout 默认 60s 等)不会把长任务判定成空闲超时。
 _HEARTBEAT_SECONDS = 15.0
@@ -188,6 +272,10 @@ def _build_run_context(body: OpenChatRequest) -> tuple[dict[str, Any], str | Non
         "thinking_force_disabled": not body.thinking_enabled,
         # 一次性问答:跳过自动起标题(省一次 LLM 调用)、不挂定时任务工具。
         "is_scheduled_run": True,
+        # 临时/无状态运行:跳过记忆与知识库的读写(不召回、不留存、不自动沉淀),
+        # 做到"通过本接口的聊天记录不存储"(见 _build_middlewares 的 is_ephemeral)。
+        "ephemeral": True,
+        "memory_injection_enabled": False,
     }
     if agent_id:
         configurable["agent_id"] = agent_id
@@ -220,8 +308,10 @@ async def open_chat(body: OpenChatRequest) -> OpenChatResponse:
         raise HTTPException(status_code=502, detail=f"智能体调用失败:{exc}") from exc
 
     answer = _extract_answer(result, strip_think=body.strip_think and not body.thinking_enabled)
-    if not answer:
-        raise HTTPException(status_code=502, detail="智能体未返回任何文本答案。")
+    files = _collect_output_files(result, include=body.include_files)
+    _cleanup_ephemeral_thread(result, thread_id)  # 取完文件即清理临时目录,不在服务端留存
+    if not answer and not files:
+        raise HTTPException(status_code=502, detail="智能体未返回任何文本答案或文件。")
 
     return OpenChatResponse(
         answer=answer,
@@ -230,6 +320,7 @@ async def open_chat(body: OpenChatRequest) -> OpenChatResponse:
         agent_id=agent_id,
         thinking_enabled=body.thinking_enabled,
         thread_id=thread_id,
+        files=files,
     )
 
 
@@ -289,8 +380,10 @@ async def _open_chat_event_stream(body: OpenChatRequest) -> AsyncIterator[str]:
             yield _sse(event, data)
 
         answer = _extract_answer(final_state, strip_think=strip_think)
-        if not answer:
-            yield _sse("error", {"detail": "智能体未返回任何文本答案。"})
+        files = _collect_output_files(final_state, include=body.include_files)
+        _cleanup_ephemeral_thread(final_state, thread_id)  # 取完文件即清理临时目录,不在服务端留存
+        if not answer and not files:
+            yield _sse("error", {"detail": "智能体未返回任何文本答案或文件。"})
             return
         yield _sse(
             "result",
@@ -301,6 +394,7 @@ async def _open_chat_event_stream(body: OpenChatRequest) -> AsyncIterator[str]:
                 "agent_id": agent_id,
                 "thinking_enabled": body.thinking_enabled,
                 "thread_id": thread_id,
+                "files": [f.model_dump() for f in files],
             },
         )
     finally:
diff --git a/offline-backend-20260512/backend/packages/harness/deerflow/agents/lead_agent/agent.py b/offline-backend-20260512/backend/packages/harness/deerflow/agents/lead_agent/agent.py
index e2d27c2f..66cf4d88 100644
--- a/offline-backend-20260512/backend/packages/harness/deerflow/agents/lead_agent/agent.py
+++ b/offline-backend-20260512/backend/packages/harness/deerflow/agents/lead_agent/agent.py
@@ -306,17 +306,25 @@ def _build_middlewares(
     if not cfg.get("is_scheduled_run", False):
         middlewares.append(TitleMiddleware(app_config=resolved_app_config))
 
-    # Add MemoryMiddleware (after TitleMiddleware) — skip in bootstrap mode (agent creation flow)
+    # Ephemeral runs (e.g. the open Q&A API /api/open/chat) must NOT touch any
+    # server-side per-user state: no memory recall/retain, no knowledge-base
+    # recall/auto-ingest. Those runs use the shared "default" user bucket, so
+    # persisting their conversations there would both store records the caller
+    # explicitly asked NOT to keep and leak content across unrelated callers.
+    # Skip both write-capable middlewares entirely.
+    is_ephemeral = cfg.get("ephemeral", False)
+
+    # Add MemoryMiddleware (after TitleMiddleware) — skip in bootstrap mode (agent creation flow) and ephemeral runs
     memory_injection_enabled = cfg.get("memory_injection_enabled", True)
-    if not is_bootstrap and not is_writing_setup:
+    if not is_bootstrap and not is_writing_setup and not is_ephemeral:
         middlewares.append(MemoryMiddleware(agent_name=agent_name, memory_config=resolved_app_config.memory, injection_enabled=memory_injection_enabled))
 
     # KnowledgeRagMiddleware — inject knowledge-base recall (phase 2). Gated
     # per-conversation by the ``knowledge_rag_enabled`` runtime flag, defaulting
     # to the global ``knowledge.rag_enabled`` config. No-op when knowledge is
-    # disabled. Skipped in bootstrap/writing-setup like MemoryMiddleware.
+    # disabled. Skipped in bootstrap/writing-setup/ephemeral like MemoryMiddleware.
     knowledge_cfg = getattr(resolved_app_config, "knowledge", None)
-    if not is_bootstrap and not is_writing_setup and knowledge_cfg is not None and knowledge_cfg.enabled:
+    if not is_bootstrap and not is_writing_setup and not is_ephemeral and knowledge_cfg is not None and knowledge_cfg.enabled:
         from deerflow.agents.middlewares.knowledge_rag_middleware import KnowledgeRagMiddleware
 
         middlewares.append(KnowledgeRagMiddleware(limit=knowledge_cfg.rag_limit, default_enabled=knowledge_cfg.rag_enabled))

From dd942cdd6061ee1b6134be676d417eb6adaa0617 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E5=90=89=E9=91=AB=E5=8D=87?= <xsji@kland.com.cn>
Date: Mon, 22 Jun 2026 21:00:22 +0800
Subject: [PATCH 17/17] =?UTF-8?q?feat(xdfx):=20=E8=A1=8C=E5=8A=A8=E5=88=86?=
 =?UTF-8?q?=E6=9E=90=E8=81=8A=E5=A4=A9=E8=AE=B0=E5=BD=95=E6=8C=89=E8=A1=8C?=
 =?UTF-8?q?=E5=8A=A8id=E5=88=92=E5=88=86=20+=20=E9=A6=96=E9=97=AE=E8=BF=BD?=
 =?UTF-8?q?=E5=8A=A0=20action=5Fid?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

xdfx 单智能体问答不再按 task_id 划分聊天记录，改按「行动id（action_id）」划分：
- TaskContext 增加 partitionByAction 标记 + taskThreadMetadata 统一产出线程标签
- xdfx 新建线程额外打 metadata.actionId，对话列表按 actionId(+agent_id) 过滤
- 仍保留 metadata.taskId 以维持后端「任务深链对话全局共享」语义
- 首次提问开场信息追加「行动id（action_id）为…」
- useTaskThreads 增加 actionId 过滤参数；useChatsBase 从上下文派生
rwfx / 3qfx 行为不变。

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
---
 CLAUDE.md                                     |  2 +-
 frontend-web/src/core/auth/task-deeplink.ts   | 16 +++++++++---
 frontend-web/src/core/auth/task-purpose.ts    | 23 ++++++++++++++++
 .../src/core/threads/use-task-threads.ts      | 13 +++++++---
 frontend-web/src/pages/AgentChatPage.tsx      |  7 ++---
 .../task-components/TaskFxWorkspace.tsx       | 26 ++++++++++++++-----
 6 files changed, 69 insertions(+), 18 deletions(-)

diff --git a/CLAUDE.md b/CLAUDE.md
index 2961555f..3a401e82 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -88,7 +88,7 @@ The frontend (`frontend-web/`) is a React 19 + Vite SPA:
 - **LLM streaming**: `@langchain/langgraph-sdk` streams messages from the backend via SSE
 - **UI**: Radix UI primitives + TailwindCSS v4 + shadcn-style components in `src/components/ui/`
 - **Shims**: `src/shims/` provides `next/link` and `next/navigation` polyfills (original code was Next.js)
-- **Login URL params** (`LoginPage.tsx`, frontend-only — the backend never sees these): besides the auth params (`ticket`/`userId`/`env`/`theme`/`yUserId`/`token`/`access_token`/`authToken`/`password`), the page supports a **task deep-link** via `?taskId=&goPath=`. After login it calls the consumer API `GET {VITE_AIPI_CONSUMER_URL}/taskAnalyseSearch/cop-task-detail?taskId=` (header `Authorization: Bearer admin`, auto-injected by `utils/request.ts`), stashes the full `data` (`id/overview/taskDirection/taskName/taskContent`) into `sessionStorage["login.copTaskDetail"]`, then routes by `goPath`: base codes `3qfx`/`rwfx`/`xdfx` → that business's configured **single agent** Q&A inside a **任务工作区** (resolved via `/api/business-mapping`; `resolveTaskDeeplinkTarget` calls `persistTaskContext({taskId, agentId, openingText})` and navigates to the `…/chats` **index**, which opens the latest task conversation or 新建 — a 1:N task↔对话 workspace, NOT `pendingAgentMessage`); the `-A` suffix (`3qfx-A`/…) → **multi-agent roundtable** with `roundtablePrefill.businessCode` so the page auto-selects the configured 业务链条. **xdfx's deep-link `?taskId=` is actually an 行动id, not a 任务id** — `resolveDeeplinkTaskId(goPath, rawId)` (called in `LoginPage` before everything) detects `xdfx`/`xdfx-A` and resolves it via `GET {action_detail_url}/{actionId}` → `data.taskId` (`fetchTaskIdFromAction`; `action_test`+`mock.action_detail` for offline testing); the resolved 任务id then drives `cop-task-detail`, the purpose tree, and the 1:N binding (other businesses pass the id through unchanged). goPath→businessCode: `3qfx`→`3Q`, `rwfx`→`6BF`, `xdfx`→`7BF`. The roundtable target also threads the URL `taskId` into `roundtablePrefill.taskId` (`resolveTaskDeeplinkTarget(goPath, detail, taskId)`), which turns on two taskId-scoped behaviors: (1) **separate, unpartitioned per-task storage** — taskId 会商聊天记录 live in their **own** store/table (`roundtable_task_drafts`), keyed **only by `task_id`** (one taskId = many records, shared across users, **no user 分权** at all; `created_by` is audit-only), kept entirely out of the per-user personal store (`roundtable_drafts`, now strictly per-user) so the two never mix. Backend router `/api/roundtable-task-drafts` (`GET /by-task/{id}` + `/by-task/{id}/latest`, `POST`, `GET/PUT/DELETE /{id}` — all user-agnostic); frontend client `roundtable-planning/api/task-drafts.ts`, and `useDraftPersistence`/`useRoundtableDrafts` route **all** CRUD to it when a `taskId` is present (else the personal `drafts.ts`). The matching **background jobs** are shared by task too: `roundtable_jobs.task_id` marks a task job; reads/dedup/stream/resume/cancel pass `?task_id=` so any user opening the task sees the same running 研讨 (progress/flowchart), and the executor writes the final report back to the task-draft store. Existing rows are moved over by `scripts/migrate_roundtable_task_drafts.py` (migration `20260618_01`). The history dropdown filters to the task and 新建 keeps the binding; (2) **Step-3 auto structured output** — after the `roundtable-summary` agent writes the Markdown report it automatically runs a second round emitting a `flow-json` flowchart (`useStep3Summary` `autoFlow`+`chainLevels`), whose hierarchy follows the business chain's stage goals (任务 → 目的 → 行为体 → …; coordinatorPrompt-specified hierarchy wins) and whose nodes never carry 阶段编号/阶段名 (the prior bug). **Both targets render embedded under the 任务管理 layout** — nested routes under `task-management/<business-path>/{roundtable,agent/:agent_id/chats}` in `StrategyRoutes.tsx` (`<business-path>` = `plan-task-sentiment` / `plan-task/plan-task-fx` / `plan-task/plan-task-action`, matching the COH中任务 menu paths in `sidebar-menu.ts`) so the global sidebar keeps 任务管理 → COH中任务 → the matching business highlighted. **All three single-agent businesses share `TaskFxWorkspace`** (`task-components/TaskFxWorkspace.tsx`): left = a「当前任务」card (`readCopTaskDetail()` 任务名称 + 简介, desensitized, 留存 above 新建对话) + per-task 1:N conversation list (`useTaskThreads(taskId, agentId)` — filtered by **both** `metadata.taskId` AND `metadata.agent_id` so the three businesses' conversations stay isolated even though they share one taskId) + 新建对话; the `…/chats` index (`TaskFxIndexResolver`) opens the latest task thread or 新建. The right `:thread_id` host differs by business: **rwfx** → `TaskFxChat` (`AgentChatPage taskGate`) = full workspace (non-`lite`): 目的树闸门 panel + 清除记录 (placed under 新建对话); **xdfx** → also `TaskFxChat` (same 目的树闸门 query+display, same `purpose_detail_url`) but workspace passed `lite`; **3qfx** → `TaskBindChat` (`AgentChatPage taskBind`, `lite`) = 1:N list only, no gate, and 新建对话 **auto-sends** the task opening text (`taskAutoSendText`; gate businesses send it via the gate's 开始问答 instead). **rwfx auto-starts the first Q&A when the 任务目的 query returns empty** (`TaskPurposePanel autoStartWhenEmpty`, threaded from the rwfx `TaskFxChat` only). The 目的树闸门 (`TaskPurposePanel`) defaults **collapsed** and now keeps only 弹框查看 + 开始问答; the former jump buttons (编辑/下一步/3q详情/切换三情/弹框查看) all moved to the **configurable bottom 快捷跳转 bar** (`TaskJumpBar`, red 红底白字 buttons, right-aligned) driven by 按钮管理 — see the 按钮管理 bullet. `TaskPurposeDialog` (exported from `TaskPurposePanel`) backs both the gate zoom and the bottom bar's `purpose`-type button. `chatBasePath` keeps the post-send URL rewrite + new-chat button in the task-management context. Backend config `task_deeplink` (`config.yaml`, surfaced by `GET /api/public/config/task-deeplink`, mirrored in `core/auth/task-deeplink-config.ts`) adds `action_test`/`action_detail_url`/`detail_3q_url` + `mock.action_detail`. Any failure (no consumer URL / fetch error / unmapped goPath / unresolvable action id / no configured agent) silently falls back to the default home. Helpers: `core/auth/task-deeplink.ts`, `core/auth/task-purpose.ts`, `core/auth/task-deeplink-config.ts`, `core/threads/use-task-threads.ts`.
+- **Login URL params** (`LoginPage.tsx`, frontend-only — the backend never sees these): besides the auth params (`ticket`/`userId`/`env`/`theme`/`yUserId`/`token`/`access_token`/`authToken`/`password`), the page supports a **task deep-link** via `?taskId=&goPath=`. After login it calls the consumer API `GET {VITE_AIPI_CONSUMER_URL}/taskAnalyseSearch/cop-task-detail?taskId=` (header `Authorization: Bearer admin`, auto-injected by `utils/request.ts`), stashes the full `data` (`id/overview/taskDirection/taskName/taskContent`) into `sessionStorage["login.copTaskDetail"]`, then routes by `goPath`: base codes `3qfx`/`rwfx`/`xdfx` → that business's configured **single agent** Q&A inside a **任务工作区** (resolved via `/api/business-mapping`; `resolveTaskDeeplinkTarget` calls `persistTaskContext({taskId, agentId, openingText})` and navigates to the `…/chats` **index**, which opens the latest task conversation or 新建 — a 1:N task↔对话 workspace, NOT `pendingAgentMessage`); the `-A` suffix (`3qfx-A`/…) → **multi-agent roundtable** with `roundtablePrefill.businessCode` so the page auto-selects the configured 业务链条. **xdfx's deep-link `?taskId=` is actually an 行动id, not a 任务id** — `resolveDeeplinkTaskId(goPath, rawId)` (called in `LoginPage` before everything) detects `xdfx`/`xdfx-A` and resolves it via `GET {action_detail_url}/{actionId}` → `data.taskId` (`fetchTaskIdFromAction`; `action_test`+`mock.action_detail` for offline testing); the resolved 任务id then drives `cop-task-detail`, the purpose tree, and the 1:N binding (other businesses pass the id through unchanged). **xdfx additionally partitions its chat records by 行动id (action_id), NOT by task_id**: `resolveTaskDeeplinkTarget` sets `partitionByAction:true` + keeps the raw action_id as `TaskContext.actionId`, so new threads are tagged `metadata.{taskId,actionId}` (taskId kept only to preserve the backend「任务深链对话全局共享」semantics, which keys off `metadata.taskId`) and the conversation list filters by `actionId` (+`agent_id`) — different actions of the same task get separate lists. xdfx's first-question opening text also appends `行动id（action_id）为…` (see `buildOpeningText` / `taskThreadMetadata`). goPath→businessCode: `3qfx`→`3Q`, `rwfx`→`6BF`, `xdfx`→`7BF`. The roundtable target also threads the URL `taskId` into `roundtablePrefill.taskId` (`resolveTaskDeeplinkTarget(goPath, detail, taskId)`), which turns on two taskId-scoped behaviors: (1) **separate, unpartitioned per-task storage** — taskId 会商聊天记录 live in their **own** store/table (`roundtable_task_drafts`), keyed **only by `task_id`** (one taskId = many records, shared across users, **no user 分权** at all; `created_by` is audit-only), kept entirely out of the per-user personal store (`roundtable_drafts`, now strictly per-user) so the two never mix. Backend router `/api/roundtable-task-drafts` (`GET /by-task/{id}` + `/by-task/{id}/latest`, `POST`, `GET/PUT/DELETE /{id}` — all user-agnostic); frontend client `roundtable-planning/api/task-drafts.ts`, and `useDraftPersistence`/`useRoundtableDrafts` route **all** CRUD to it when a `taskId` is present (else the personal `drafts.ts`). The matching **background jobs** are shared by task too: `roundtable_jobs.task_id` marks a task job; reads/dedup/stream/resume/cancel pass `?task_id=` so any user opening the task sees the same running 研讨 (progress/flowchart), and the executor writes the final report back to the task-draft store. Existing rows are moved over by `scripts/migrate_roundtable_task_drafts.py` (migration `20260618_01`). The history dropdown filters to the task and 新建 keeps the binding; (2) **Step-3 auto structured output** — after the `roundtable-summary` agent writes the Markdown report it automatically runs a second round emitting a `flow-json` flowchart (`useStep3Summary` `autoFlow`+`chainLevels`), whose hierarchy follows the business chain's stage goals (任务 → 目的 → 行为体 → …; coordinatorPrompt-specified hierarchy wins) and whose nodes never carry 阶段编号/阶段名 (the prior bug). **Both targets render embedded under the 任务管理 layout** — nested routes under `task-management/<business-path>/{roundtable,agent/:agent_id/chats}` in `StrategyRoutes.tsx` (`<business-path>` = `plan-task-sentiment` / `plan-task/plan-task-fx` / `plan-task/plan-task-action`, matching the COH中任务 menu paths in `sidebar-menu.ts`) so the global sidebar keeps 任务管理 → COH中任务 → the matching business highlighted. **All three single-agent businesses share `TaskFxWorkspace`** (`task-components/TaskFxWorkspace.tsx`): left = a「当前任务」card (`readCopTaskDetail()` 任务名称 + 简介, desensitized, 留存 above 新建对话) + per-task 1:N conversation list (`useTaskThreads(taskId, agentId, actionId?)` — filtered by **both** `metadata.taskId` AND `metadata.agent_id` so the three businesses' conversations stay isolated even though they share one taskId; **xdfx** passes a third `actionId` so it filters by `metadata.actionId` too, partitioning chat records per 行动 — `useChatsBase` derives it from `TaskContext.partitionByAction`) + 新建对话; the `…/chats` index (`TaskFxIndexResolver`) opens the latest task thread or 新建. The right `:thread_id` host differs by business: **rwfx** → `TaskFxChat` (`AgentChatPage taskGate`) = full workspace (non-`lite`): 目的树闸门 panel + 清除记录 (placed under 新建对话); **xdfx** → also `TaskFxChat` (same 目的树闸门 query+display, same `purpose_detail_url`) but workspace passed `lite`; **3qfx** → `TaskBindChat` (`AgentChatPage taskBind`, `lite`) = 1:N list only, no gate, and 新建对话 **auto-sends** the task opening text (`taskAutoSendText`; gate businesses send it via the gate's 开始问答 instead). **rwfx auto-starts the first Q&A when the 任务目的 query returns empty** (`TaskPurposePanel autoStartWhenEmpty`, threaded from the rwfx `TaskFxChat` only). The 目的树闸门 (`TaskPurposePanel`) defaults **collapsed** and now keeps only 弹框查看 + 开始问答; the former jump buttons (编辑/下一步/3q详情/切换三情/弹框查看) all moved to the **configurable bottom 快捷跳转 bar** (`TaskJumpBar`, red 红底白字 buttons, right-aligned) driven by 按钮管理 — see the 按钮管理 bullet. `TaskPurposeDialog` (exported from `TaskPurposePanel`) backs both the gate zoom and the bottom bar's `purpose`-type button. `chatBasePath` keeps the post-send URL rewrite + new-chat button in the task-management context. Backend config `task_deeplink` (`config.yaml`, surfaced by `GET /api/public/config/task-deeplink`, mirrored in `core/auth/task-deeplink-config.ts`) adds `action_test`/`action_detail_url`/`detail_3q_url` + `mock.action_detail`. Any failure (no consumer URL / fetch error / unmapped goPath / unresolvable action id / no configured agent) silently falls back to the default home. Helpers: `core/auth/task-deeplink.ts`, `core/auth/task-purpose.ts`, `core/auth/task-deeplink-config.ts`, `core/threads/use-task-threads.ts`.
 - **iframe 嵌入模式 (外部系统用 iframe 套用深链问答/会商)** — 三个**按标签页**开关，全部由深链 URL 参数开启、存进同一份 `sessionStorage` 会话（`embed.session`，键名常量在 `core/embed/embed-session.ts`），跨登录跳转/刷新/页内导航存活，**绝不跨标签污染**同浏览器另开的正常系统页。核心隔离手法：用 `sessionStorage`（按标签页/浏览上下文隔离，iframe 即便同源也有独立一份）而非 `localStorage`（跨标签共享）。开关：(1) **`?embed=1`（可带 `&theme=`）** — 去全局 chrome（`PageLayout` 的 `pageSidebarAllowed && !isEmbedActive()` → 隐藏 Header + PageSidebar）+ 按外部主题色**强制**明/暗：走 next-themes 的 `forcedTheme`（`App.tsx` `AppProviders` 调 `resolveForcedEmbedTheme`，**不写** localStorage `strategy-theme`），离散映射 `dark-blue`/`dark`→深、其余→浅；嵌入侧**绝不**调 `setTheme`/`persistSidebarEnabled`（那俩落 localStorage 跨标签污染，`LoginPage` 据 `readEmbedParam` 跳过）。(2) **`?hideJump=1`** — 隐藏深链底部「快捷跳转」栏 `TaskJumpBar` 的两部分（底部「下一步」等按钮 + 右侧垂直居中悬浮菜单），单智能体问答 + 会商共用该栏，`isJumpBarHidden()` 为真时整条 `return null`。(3) **登录用户隔离** — 嵌入态下鉴权/身份键（`deerflow.auth` token、`userInfo`、`login.*`、`tools_token`、maxkey/userDetail token）全部经 `core/auth/scoped-storage.ts` 的 `authStorage()`（嵌入→sessionStorage / 正常→localStorage）读写，使 iframe 里登录的用户与正常登录用户**完全隔离**、互不覆盖；所有 token 读取都过 `getStoredAuth()`/`getAuthorizationHeaderValue()`（API 客户端 `core/api/*` 全走这俩），改中心函数即让全体调用方自动隔离；跨标签账号变更监听 `AuthSyncWatcher`（`core/auth/sync.tsx`）在嵌入态禁用。嵌入会话在 `main.tsx` 渲染前 `syncEmbedSessionFromCurrentUrl()` 同步启用（早于任何鉴权写入，避免首帧 effect 顺序漏写 localStorage）；`AppProviders` 的 effect 再按 SPA 导航同步。`embed=0`/`hideJump=0` 可显式关闭。正常（非嵌入）标签页所有 `isXxx()` 恒 false，行为**零变化**。
 
 ### Configuration
diff --git a/frontend-web/src/core/auth/task-deeplink.ts b/frontend-web/src/core/auth/task-deeplink.ts
index c8a2322c..4574ee53 100644
--- a/frontend-web/src/core/auth/task-deeplink.ts
+++ b/frontend-web/src/core/auth/task-deeplink.ts
@@ -187,12 +187,16 @@ export interface TaskDeeplinkTarget {
   state?: Record<string, unknown>;
 }
 
-/** 用 taskName / taskContent 拼出开场信息文本；末尾附上任务 id（task_id），供后端识别任务。 */
-function buildOpeningText(detail: CopTaskDetail, taskId: string): string {
+/**
+ * 用 taskName / taskContent 拼出开场信息文本；末尾附上任务 id（task_id），供后端识别任务。
+ * xdfx（行动分析）额外附上「行动id（action_id）」——传入 actionId 时追加该行。
+ */
+function buildOpeningText(detail: CopTaskDetail, taskId: string, actionId?: string): string {
   return [
     detail.taskName ? `任务名称：${detail.taskName}` : "",
     detail.taskContent ? `任务内容：${detail.taskContent}` : "",
     taskId ? `任务id（task_id）为${taskId}` : "",
+    actionId ? `行动id（action_id）为${actionId}` : "",
   ]
     .filter(Boolean)
     .join("\n");
@@ -255,11 +259,15 @@ export async function resolveTaskDeeplinkTarget(
   // 暂存任务上下文（taskId + agentId + 开场信息），跳到 chats 索引，由工作区按 taskId 决定
   // 开最新历史对话 / 新建。rwfx 在工作区里额外过「目的树闸门」；3qfx/xdfx 仅 1:N 列表，
   // 新建对话直接带开场信息自动发出（见 task-purpose / AgentChatPage 的 taskGate / taskBind）。
+  // xdfx（行动分析）：聊天记录按「行动id」划分（partitionByAction），且开场信息追加 action_id。
+  const isXdfx = parsed.businessCode === XDFX_BUSINESS_CODE;
+  const actionId = rawId || taskId;
   persistTaskContext({
     taskId,
     agentId,
-    openingText: buildOpeningText(detail, taskId),
-    actionId: rawId || taskId,
+    openingText: buildOpeningText(detail, taskId, isXdfx ? actionId : undefined),
+    actionId,
+    partitionByAction: isXdfx,
   });
   return { path: `${taskBase}/agent/${agentId}/chats` };
 }
diff --git a/frontend-web/src/core/auth/task-purpose.ts b/frontend-web/src/core/auth/task-purpose.ts
index 5fe3c5f2..375e3c51 100644
--- a/frontend-web/src/core/auth/task-purpose.ts
+++ b/frontend-web/src/core/auth/task-purpose.ts
@@ -41,6 +41,12 @@ export interface TaskContext {
   openingText: string;
   /** 深链原始 id（xdfx 为「行动id」，需追加行动id 的按钮用；其它业务等于 taskId）。 */
   actionId?: string;
+  /**
+   * xdfx（行动分析）专属：聊天记录按「行动id」(actionId) 而非「任务id」划分。
+   * 为 true 时，新建对话额外打 `metadata.actionId` 标签、对话列表按 actionId 过滤
+   * （仍保留 taskId 标签以维持任务深链对话「全局共享、不分用户」的后端语义）。
+   */
+  partitionByAction?: boolean;
 }
 
 const TASK_CONTEXT_STORAGE_KEY = "task.rwfxContext";
@@ -311,12 +317,29 @@ export function readTaskContext(): TaskContext | null {
       agentId: String(parsed.agentId),
       openingText: typeof parsed.openingText === "string" ? parsed.openingText : "",
       actionId: typeof parsed.actionId === "string" ? parsed.actionId : undefined,
+      partitionByAction: parsed.partitionByAction === true,
     };
   } catch {
     return null;
   }
 }
 
+/**
+ * 任务深链对话的 thread metadata 标签（既用于「新建对话」打标签，也用于对话列表过滤）。
+ *   - 普通业务（rwfx/3qfx）：`{ taskId }` —— 按任务划分（1:N）。
+ *   - xdfx（partitionByAction）：`{ taskId, actionId }` —— 额外按「行动id」划分聊天记录；
+ *     仍带 taskId 以保留后端「任务深链对话全局共享」的语义（后端据 metadata.taskId 跳过分用户过滤）。
+ * 无有效 taskId 时返回 undefined（调用方据此不打标签）。
+ */
+export function taskThreadMetadata(
+  ctx: TaskContext | null | undefined,
+): Record<string, unknown> | undefined {
+  if (!ctx?.taskId) return undefined;
+  return ctx.partitionByAction && ctx.actionId
+    ? { taskId: ctx.taskId, actionId: ctx.actionId }
+    : { taskId: ctx.taskId };
+}
+
 export function clearTaskContext(): void {
   try {
     sessionStorage.removeItem(TASK_CONTEXT_STORAGE_KEY);
diff --git a/frontend-web/src/core/threads/use-task-threads.ts b/frontend-web/src/core/threads/use-task-threads.ts
index b9101dc4..f9a7cd19 100644
--- a/frontend-web/src/core/threads/use-task-threads.ts
+++ b/frontend-web/src/core/threads/use-task-threads.ts
@@ -8,6 +8,10 @@
  * 同一个 taskId 可能挂在多条业务链上（rwfx / 3qfx / xdfx 共用任务 id），每条业务用各自的
  * 智能体。传入 ``agentId`` 时额外按 ``metadata.agent_id`` 过滤（后端 metadata 过滤是 AND/子集
  * 匹配），把不同业务的对话列表隔离开——否则三种业务的对话会混在一起。
+ *
+ * xdfx（行动分析）：聊天记录按「行动id」划分——传入 ``actionId`` 时额外按 ``metadata.actionId``
+ * 过滤（同一任务下的不同行动各自一份对话列表）；仍保留 taskId 过滤以命中后端「任务深链对话
+ * 全局共享」的判定（后端据 metadata.taskId 跳过分用户过滤）。
  */
 
 import { useThreadsPage } from "@/core/threads/hooks";
@@ -16,13 +20,16 @@ import type { AgentThread } from "@/core/threads/types";
 export function useTaskThreads(
   taskId: string | null | undefined,
   agentId?: string | null | undefined,
+  actionId?: string | null | undefined,
 ) {
   const enabled = !!taskId;
   // 任务工作区只在有 taskId 时渲染；无 taskId 时传一个不可能命中的过滤，返回空列表。
   const metadata = enabled
-    ? agentId
-      ? { taskId, agent_id: agentId }
-      : { taskId }
+    ? {
+        taskId,
+        ...(actionId ? { actionId } : {}),
+        ...(agentId ? { agent_id: agentId } : {}),
+      }
     : { taskId: "__none__" };
   const query = useThreadsPage({ page: 1, pageSize: 100, metadata });
   const threads: AgentThread[] = enabled ? (query.data?.threads ?? []) : [];
diff --git a/frontend-web/src/pages/AgentChatPage.tsx b/frontend-web/src/pages/AgentChatPage.tsx
index 659507ba..e59f7e09 100644
--- a/frontend-web/src/pages/AgentChatPage.tsx
+++ b/frontend-web/src/pages/AgentChatPage.tsx
@@ -31,7 +31,7 @@ import { saveLastAgentId, useThreadSettings } from "@/core/settings";
 import { stopThreadRun, useThreadStream } from "@/core/threads/hooks";
 import { textOfMessage } from "@/core/threads/utils";
 import { env } from "@/env";
-import { readTaskContext } from "@/core/auth/task-purpose";
+import { readTaskContext, taskThreadMetadata } from "@/core/auth/task-purpose";
 import { TaskPurposePanel } from "@/strategy-components/components/task-components/TaskPurposePanel";
 import { desensitize } from "@/lib/desensitize";
 
@@ -143,8 +143,9 @@ export default function AgentChatPage({
     threadId: isNewThread ? undefined : threadId,
     assistantId: agent_id,
     context: { ...settings.context, agent_id },
-    // 任务闸门：新建 thread 打 metadata.taskId 标签，供任务对话列表(1:N)按任务过滤。
-    threadMetadata: taskId ? { taskId } : undefined,
+    // 任务闸门：新建 thread 打 metadata.taskId 标签，供任务对话列表(1:N)按任务过滤；
+    // xdfx 额外打 metadata.actionId 标签，按「行动id」划分聊天记录（见 taskThreadMetadata）。
+    threadMetadata: taskId ? taskThreadMetadata(taskContext) : undefined,
     onStart: (createdThreadId) => {
       setThreadId(createdThreadId);
       setIsNewThread(false);
diff --git a/frontend-web/src/strategy-components/components/task-components/TaskFxWorkspace.tsx b/frontend-web/src/strategy-components/components/task-components/TaskFxWorkspace.tsx
index f21ba0b1..ebee077c 100644
--- a/frontend-web/src/strategy-components/components/task-components/TaskFxWorkspace.tsx
+++ b/frontend-web/src/strategy-components/components/task-components/TaskFxWorkspace.tsx
@@ -46,11 +46,19 @@ import AgentChatPage from "@/pages/AgentChatPage";
 import { ChatRuntime } from "@/pages/WorkspaceRoutes";
 import { TaskJumpBar } from "@/strategy-components/components/task-components/TaskJumpBar";
 
-function useChatsBase(taskBase: string): { taskId: string; agentId: string; chatsBase: string } {
+function useChatsBase(taskBase: string): {
+  taskId: string;
+  agentId: string;
+  /** xdfx（partitionByAction）：按「行动id」划分聊天记录用；其它业务为 ""（仍按 taskId 划分）。 */
+  actionId: string;
+  chatsBase: string;
+} {
   const { agent_id } = useParams<{ agent_id: string }>();
   const agentId = agent_id ?? "";
-  const taskId = readTaskContext()?.taskId ?? "";
-  return { taskId, agentId, chatsBase: `${taskBase}/agent/${agentId}/chats` };
+  const ctx = readTaskContext();
+  const taskId = ctx?.taskId ?? "";
+  const actionId = ctx?.partitionByAction ? (ctx.actionId ?? "") : "";
+  return { taskId, agentId, actionId, chatsBase: `${taskBase}/agent/${agentId}/chats` };
 }
 
 /**
@@ -142,19 +150,22 @@ function parseThreadId(pathname: string, chatsBase: string): string | null {
 function TaskConversationsList({
   taskId,
   agentId,
+  actionId,
   chatsBase,
   business,
   lite,
 }: {
   taskId: string;
   agentId: string;
+  /** xdfx：按「行动id」划分聊天记录；其它业务为 ""。 */
+  actionId: string;
   chatsBase: string;
   business: TaskBusiness;
   lite?: boolean;
 }) {
   const { pathname } = useLocation();
   const navigate = useNavigate();
-  const { threads, isLoading } = useTaskThreads(taskId, agentId);
+  const { threads, isLoading } = useTaskThreads(taskId, agentId, actionId);
   const currentThreadId = useMemo(() => parseThreadId(pathname, chatsBase), [pathname, chatsBase]);
   const businessName = desensitize(TASK_BUSINESSES.find((b) => b.code === business)?.name ?? "");
 
@@ -371,13 +382,14 @@ export function TaskFxWorkspace({
   business: TaskBusiness;
   lite?: boolean;
 }) {
-  const { taskId, agentId, chatsBase } = useChatsBase(taskBase);
+  const { taskId, agentId, actionId, chatsBase } = useChatsBase(taskBase);
   return (
     <div className="flex size-full min-h-0 flex-col overflow-hidden">
       <div className="flex min-h-0 flex-1 overflow-hidden">
         <TaskConversationsList
           taskId={taskId}
           agentId={agentId}
+          actionId={actionId}
           chatsBase={chatsBase}
           business={business}
           lite={lite}
@@ -395,9 +407,9 @@ export function TaskFxWorkspace({
 
 /** chats 索引：有历史对话 → 跳最新；否则 → 跳 new（rwfx 走闸门 / 3qfx/xdfx 直接带开场信息）。 */
 export function TaskFxIndexResolver({ taskBase }: { taskBase: string }) {
-  const { taskId, agentId, chatsBase } = useChatsBase(taskBase);
+  const { taskId, agentId, actionId, chatsBase } = useChatsBase(taskBase);
   const navigate = useNavigate();
-  const { isLoading, hasThreads, latestThreadId } = useTaskThreads(taskId, agentId);
+  const { isLoading, hasThreads, latestThreadId } = useTaskThreads(taskId, agentId, actionId);
 
   useEffect(() => {
     if (isLoading) return;