diff --git a/.gitignore b/.gitignore
index ba42257b..663c0f2e 100644
--- a/.gitignore
+++ b/.gitignore
@@ -84,5 +84,12 @@ offline-backend-20260512/backend/tests/*
 deploy
 .history
 
+# Repo-root offline dependency pack (local only)
+/dist/
+
+# Frontend offline pack working directories
+frontend-web/.offline-pack-build/
+frontend-web/.offline-extract-test/
+
 # Frontend skills components must be tracked (excluded from the broad skills/ ignore above)
 !frontend-web/src/components/workspace/skills/
diff --git a/frontend-web/src/components/workspace/messages/markdown-content.tsx b/frontend-web/src/components/workspace/messages/markdown-content.tsx
index dc0888c5..739bcd77 100644
--- a/frontend-web/src/components/workspace/messages/markdown-content.tsx
+++ b/frontend-web/src/components/workspace/messages/markdown-content.tsx
@@ -8,7 +8,6 @@ import {
   type MessageResponseProps,
 } from "@/components/ai-elements/message";
 import { streamdownPlugins } from "@/core/streamdown";
-import { desensitize } from "@/lib/desensitize";
 import { cn } from "@/lib/utils";
 
 import { CitationLink } from "../citations/citation-link";
@@ -36,13 +35,13 @@ export function MarkdownContent({
 }: MarkdownContentProps) {
   const components = useMemo(() => {
     return {
-      a: (props: AnchorHTMLAttributes<HTMLAnchorElement>) => {
-        if (props.href?.startsWith("#reference-")) {
-          return <CitationLink {...props} />;
-        }
-        if (typeof props.children === "string") {
-          const match = /^citation:(.+)$/.exec(props.children);
-          if (match) {
+      a: (props: AnchorHTMLAttributes<HTMLAnchorElement>) => {
+        if (props.href?.startsWith("#reference-")) {
+          return <CitationLink {...props} />;
+        }
+        if (typeof props.children === "string") {
+          const match = /^citation:(.+)$/.exec(props.children);
+          if (match) {
             const [, text] = match;
             return <CitationLink {...props}>{text}</CitationLink>;
           }
@@ -67,8 +66,6 @@ export function MarkdownContent({
 
   if (!content) return null;
 
-  const displayContent = desensitize(content);
-
   return (
     <MessageResponse
       className={className}
@@ -76,7 +73,7 @@ export function MarkdownContent({
       rehypePlugins={rehypePlugins}
       components={components}
     >
-      {displayContent}
+      {content}
     </MessageResponse>
   );
 }
diff --git a/frontend-web/src/components/workspace/messages/message-list-item.tsx b/frontend-web/src/components/workspace/messages/message-list-item.tsx
index 4daa04ce..3c35df6f 100644
--- a/frontend-web/src/components/workspace/messages/message-list-item.tsx
+++ b/frontend-web/src/components/workspace/messages/message-list-item.tsx
@@ -42,7 +42,6 @@ import {
   type RuntimeDisplaySource,
 } from "@/core/skill-display/runtime";
 import { humanMessagePlugins } from "@/core/streamdown";
-import { desensitize } from "@/lib/desensitize";
 import { cn } from "@/lib/utils";
 
 import { CitationLink } from "../citations/citation-link";
@@ -259,7 +258,7 @@ function MessageContent_({
           <TaskTrigger title="">
             <div className="text-muted-foreground flex w-full cursor-default items-center gap-2 text-sm select-none">
               <Loader className="size-4" />
-              <span>{desensitize(contentToDisplay)}</span>
+              <span>{contentToDisplay}</span>
             </div>
           </TaskTrigger>
         </Task>
@@ -273,7 +272,7 @@ function MessageContent_({
       <AIElementMessageContent className={className}>
         <Reasoning isStreaming={isLoading}>
           <ReasoningTrigger />
-          <ReasoningContent>{desensitize(reasoningContent)}</ReasoningContent>
+          <ReasoningContent>{reasoningContent}</ReasoningContent>
         </Reasoning>
         <MessageTokenUsage
           enabled={tokenUsageEnabled}
@@ -292,7 +291,7 @@ function MessageContent_({
         components={components}
         parseIncompleteMarkdown={false}
       >
-        {desensitize(contentToDisplay)}
+        {contentToDisplay}
       </AIElementMessageResponse>
     ) : null;
     return (
diff --git a/frontend-web/src/roundtable-planning/hooks/useDraftPersistence.ts b/frontend-web/src/roundtable-planning/hooks/useDraftPersistence.ts
index c9b61a34..4f7f262d 100644
--- a/frontend-web/src/roundtable-planning/hooks/useDraftPersistence.ts
+++ b/frontend-web/src/roundtable-planning/hooks/useDraftPersistence.ts
@@ -57,6 +57,16 @@ export interface UseDraftPersistenceOptions {
    * 否则保持原 false 让进入 Step 2 时正常初始化。
    */
   setStep2InitGate: (initialized: boolean) => void;
+  /**
+   * Step 1 是否正在等待任务理解回复。由 true→false 视为「本轮模型回答完」,
+   * 触发一次自动保存(详见 autoSave 的两个 effect)。
+   */
+  intentSending: boolean;
+  /**
+   * Step 2 已收尾(streaming===false)的对话气泡数量。每次递增意味着总控/某个
+   * 子智能体刚交付完一轮,触发一次自动保存。流式增量不改变该计数,因此不会刷屏式保存。
+   */
+  step2FinalizedCount: number;
 }
 
 export interface UseDraftPersistenceReturn {
@@ -96,6 +106,8 @@ export function useDraftPersistence(
     hydrateStep2,
     cancelAllStreams,
     setStep2InitGate,
+    intentSending,
+    step2FinalizedCount,
   } = options;
 
   const [drafts, setDrafts] = useState<DraftListItem[]>([]);
@@ -112,6 +124,11 @@ export function useDraftPersistence(
   const currentDraftIdRef = useRef<string | null>(null);
   // 串行化所有写操作，避免并发 create。
   const saveQueueRef = useRef<Promise<unknown>>(Promise.resolve());
+  // 当前步骤镜像 —— 自动保存(不带 furthestStep)落到 create 分支时,用它兜底建档步骤。
+  const currentStepRef = useRef<number>(currentStep);
+  useEffect(() => {
+    currentStepRef.current = currentStep;
+  }, [currentStep]);
 
   // 包一层 setter：同时更新 state 与同步 ref。主页面通过它 setCurrentDraftId(null)。
   const setCurrentDraftId = useCallback<React.Dispatch<React.SetStateAction<string | null>>>(
@@ -157,9 +174,15 @@ export function useDraftPersistence(
    * 串行执行一次保存：有 id 走 update（不带 title，保护重命名）；无 id 走 create
    * （写一次派生标题）。create/update 成功后同步 currentDraftIdRef。
    * 返回 draft id（无可保存内容返回 null）。
+   *
+   * `furthestStep` 可选：
+   *  - 传值（前进切换 / 手动保存 / ensureDraft）→ update 时一并推进 furthest_step;
+   *  - 省略（每轮模型回答完的自动保存）→ update **不带** furthest_step,只回写
+   *    step1/step2 内容,避免把「回到上一步继续研讨」误判成 furthest 回退。create
+   *    分支用 currentStepRef 兜底建档步骤。
    */
   const enqueueSave = useCallback(
-    (furthestStep: 1 | 2 | 3): Promise<string | null> => {
+    (furthestStep?: 1 | 2 | 3): Promise<string | null> => {
       const run = async (): Promise<string | null> => {
         const step1 = getStep1Snapshot();
         const step2 = getStep2Snapshot();
@@ -167,7 +190,11 @@ export function useDraftPersistence(
 
         const id = currentDraftIdRef.current;
         if (id) {
-          const updated = await updateDraft(id, { furthestStep, step1, step2 });
+          const updated = await updateDraft(id, {
+            ...(furthestStep !== undefined ? { furthestStep } : {}),
+            step1,
+            step2,
+          });
           if (updated) return updated.id;
           // 服务端已不存在（被删/换账号）→ 落回新建。
         }
@@ -176,7 +203,7 @@ export function useDraftPersistence(
             intentReady: step1.intentReady,
             interactiveMessages: step1.interactiveMessages,
           }),
-          furthestStep,
+          furthestStep: furthestStep ?? ((currentStepRef.current as 1 | 2 | 3) || 1),
           step1,
           step2,
         });
@@ -227,6 +254,41 @@ export function useDraftPersistence(
     // eslint-disable-next-line react-hooks/exhaustive-deps
   }, [currentStep]);
 
+  /**
+   * 每轮模型回答完成后落一次草稿。不带 furthestStep(见 enqueueSave),只回写内容。
+   * 仅当本次保存是「新建」时才刷新列表 —— 普通增量回写无需让下拉重排/多发一个 GET。
+   */
+  const autoSave = useCallback(async (): Promise<void> => {
+    const hadId = currentDraftIdRef.current;
+    const id = await enqueueSave();
+    if (id && !hadId) await refreshList();
+  }, [enqueueSave, refreshList]);
+
+  // ── 触发器 1:Step 1 任务理解一轮回复结束(isIntentSending true→false)。 ──
+  const prevIntentSendingRef = useRef<boolean>(intentSending);
+  useEffect(() => {
+    const was = prevIntentSendingRef.current;
+    prevIntentSendingRef.current = intentSending;
+    // 此刻 React 已 flush 完终态 setState(气泡内容/INTENT_READY),
+    // 快照 getter 读到的是收尾后的最新状态,不会丢最后一段。
+    if (was && !intentSending) void autoSave();
+  }, [intentSending, autoSave]);
+
+  // ── 触发器 2:Step 2 某个气泡刚收尾(已收尾计数递增)。 ──
+  // 加载草稿会把历史对话整批 hydrate 进来,计数从 0 跳到 N —— 那不是一次真实交付,
+  // 用 suppress latch 跳过这一跳变(handleLoadDraft 在 hydrate 前置位)。
+  const prevFinalizedCountRef = useRef<number>(step2FinalizedCount);
+  const suppressStep2AutoSaveRef = useRef<boolean>(false);
+  useEffect(() => {
+    const prev = prevFinalizedCountRef.current;
+    prevFinalizedCountRef.current = step2FinalizedCount;
+    if (suppressStep2AutoSaveRef.current) {
+      suppressStep2AutoSaveRef.current = false;
+      return;
+    }
+    if (step2FinalizedCount > prev) void autoSave();
+  }, [step2FinalizedCount, autoSave]);
+
   const handleLoadDraft = useCallback(
     (id: string) => {
       void (async () => {
@@ -248,6 +310,9 @@ export function useDraftPersistence(
         // 先置 latch,再 setCurrentStep —— 顺序反过来会让 step1/step2 init effect
         // 把 loadedFromDraftRef.current 当成 false 处理,导致角色面板卡在「正在上线」。
         loadedFromDraftRef.current = true;
+        // hydrate 会让 step2 已收尾计数从 0 跳到 N,抑制由此触发的「假交付」自动保存,
+        // 避免刚加载就把同样内容原样回写一遍、并把该草稿顶到列表最前。
+        suppressStep2AutoSaveRef.current = true;
 
         if (draft.step1) hydrateStep1(draft.step1 as Step1Snapshot);
         if (draft.step2) hydrateStep2(draft.step2 as Step2Snapshot);
diff --git a/frontend-web/src/roundtable-planning/pages/RoundtablePlanningPage.tsx b/frontend-web/src/roundtable-planning/pages/RoundtablePlanningPage.tsx
index 91832475..71e990b1 100644
--- a/frontend-web/src/roundtable-planning/pages/RoundtablePlanningPage.tsx
+++ b/frontend-web/src/roundtable-planning/pages/RoundtablePlanningPage.tsx
@@ -17,11 +17,12 @@
  *  1) special 顺序派活,**不要**改回 `Promise.all`(详见 §5.4)
  *  2) step1InitGenRef / step2InitRef 闸门用于规避 StrictMode 双 mount 副作用
  *  3) 离开 Step 2 不销毁 threadIds,允许返回继续看
- *  4) 草稿自动保存仅在前进切换(1→2、2→3)触发,后退不保存
+ *  4) 草稿自动保存:前进切换(1→2、2→3)推进 furthestStep;此外每轮模型回答完
+ *     (Step 1 一轮任务理解结束 / Step 2 每个气泡收尾)也增量回写一次,后者不动 furthestStep
  *  5) Persona 抽屉改的 model 不会立即重启编排,下一轮才生效
  *  6) 模型选择全局默认 `availableModels[0].name`,Persona 覆盖更高优先级
  */
-import { useCallback, useEffect, useRef, useState } from "react";
+import { useCallback, useEffect, useMemo, useRef, useState } from "react";
 import {
   Activity,
   Check,
@@ -173,11 +174,21 @@ export default function RoundtablePlanningPage() {
     };
   }, [step1.intentReady, step1.interactiveMessages]);
 
+  // Step 2 已收尾(非 streaming)的气泡数量 —— 递增即「某轮模型刚交付完」,
+  // 作为草稿自动保存的触发信号(详见 useDraftPersistence)。流式增量不计入,
+  // 所以不会在 token 级别刷屏式保存。
+  const step2FinalizedCount = useMemo(
+    () => step2.step2RoundtableDialogues.filter((d) => !d.streaming).length,
+    [step2.step2RoundtableDialogues],
+  );
+
   // 草稿 hook —— 同时依赖 step1 + step2 的 snapshot / hydrate。
   const drafts = useDraftPersistence({
     currentStep,
     setCurrentStep,
     triggerToast,
+    intentSending: step1.isIntentSending,
+    step2FinalizedCount,
     getStep1Snapshot: () => ({
       intentThreadId: step1.intentThreadId,
       intentReady: step1.intentReady,