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/appearance-controls.tsx b/frontend-web/src/components/workspace/appearance-controls.tsx
new file mode 100644
index 00000000..02efa7c8
--- /dev/null
+++ b/frontend-web/src/components/workspace/appearance-controls.tsx
@@ -0,0 +1,186 @@
+"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。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,
+  fallback = "compact",
+}: {
+  tone: AppearanceControlTone;
+  fallback?: "full" | "compact";
+}) {
+  const [mode, setLayoutMode] = useLayoutMode();
+  const current = mode ?? fallback;
+  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/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/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/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/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/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/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/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>
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 };
 }
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/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/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 4b9cf882..3fb86805 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}
@@ -3111,13 +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/frontend-web/src/strategy-components/Header.tsx b/frontend-web/src/strategy-components/Header.tsx
index 4d75ccbd..0bcb8334 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 { 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";
@@ -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 +124,12 @@ 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"} />
+
+      <LayoutModeSwitcher
+        tone={isDark ? "banner-dark" : "banner-light"}
+        fallback="full"
+      />
 
       <Button
         type="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/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"},
         ],
     }
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,
         }