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 8313c76f..8c488573 100644
--- a/frontend-web/src/open-canvas/components/ai-writing/AIWritingPanel.tsx
+++ b/frontend-web/src/open-canvas/components/ai-writing/AIWritingPanel.tsx
@@ -17,6 +17,24 @@ import { useCanvasGraphContext } from '../../contexts/CanvasGraphContext';
 import { useAIWriting } from '../../hooks/useAIWritingStream';
 import { useAIWritingSessions } from '../../hooks/useAIWritingSessions';
 import { adminCleanupAIWritingSessions } from '../../api/ai-writing-sessions';
+
+/**
+ * 暂停 / 继续功能开关（默认 false = 关）。
+ *
+ * 历史背景：legacy 路径下，为了让用户点"停止"后还能"继续"，引入了 per-worker
+ * 内存 dict（``_event_queues`` / ``_session_pending_pause``）。多 worker 部署下
+ * 这些 dict 互相看不见 → SSE 重连失败 / "会话丢失" → 触发了整个 ai-writing-优化.md
+ * 的迁移计划。
+ *
+ * 当前 LangGraph Server 路径下架构已无 per-worker 状态，但 ``stream.stop()`` 仅
+ * 取消客户端 SSE 订阅，**不会真正暂停后端 run**（LLM 继续烧 token）；``resume``
+ * 也无法重新订阅同一个 run。这是个产品语义没想清楚的功能，先关掉。
+ *
+ * 后期想清楚怎么做"真·暂停"（``client.runs.cancel(run_id, action='interrupt')``
+ * 保留 checkpoint）+ "真·继续"（``client.runs.joinStream`` 续上原 run）时，
+ * 把这个 flag 翻成 true 即可，UI / hook 全套代码都已就位。
+ */
+const ENABLE_PAUSE_RESUME_UI = false;
 import { AIWritingForm } from './AIWritingForm';
 import { AIWritingTimeline } from './AIWritingTimeline';
 import { InterventionCard } from './InterventionCard';
@@ -48,8 +66,7 @@ const STATUS_LABEL: Record<string, string> = {
 export function AIWritingPanel() {
   const navigate = useNavigate();
   const { setArtifact } = useCanvasGraphContext();
-  // pauseWriting / resumeWriting 当前 UI 不暴露（见 header 注释），从解构里省略避免 unused 警告
-  const { state, startWriting, submitIntervention, reset } = useAIWriting();
+  const { state, startWriting, submitIntervention, pauseWriting, resumeWriting, reset } = useAIWriting();
   const { sessions, loading: sessionsLoading, refresh, rename, remove } = useAIWritingSessions();
   const { articleTypes, refresh: refreshArticleTypes } = useArticleTypes();
 
@@ -270,12 +287,36 @@ export function AIWritingPanel() {
               新建
             </Button>
           )}
-          {/* ⏸️/▶️ "停止 / 继续" 按钮暂时隐藏 ——
-              当前实现是 stream.stop() 断 SSE 订阅而非真正取消后端 run，
-              点了停止后端继续烧 token，点继续 useStream 也不会自动重连，
-              用户体验是半残废的。等后期想清楚怎么"真·暂停"（cancel run +
-              保留 checkpoint）+ "真·继续"（重新订阅同一 run）再加回来。
-              hook 里的 pauseWriting / resumeWriting 代码保留以便后期复用。 */}
+          {/* ⏸️/▶️ "停止 / 继续" 由 ENABLE_PAUSE_RESUME_UI 控制（默认 false 隐藏）。
+              后期把 flag 翻成 true 即可恢复，UI / hook 全套代码都已就位。 */}
+          {ENABLE_PAUSE_RESUME_UI && isRunning && !state.isPaused && (
+            <Button
+              variant="ghost"
+              size="sm"
+              className="h-7 px-2 text-xs gap-1 animate-breathe-stop text-red-400 hover:text-red-500 hover:bg-red-50 dark:hover:bg-red-950/30"
+              onClick={pauseWriting}
+            >
+              <Square className="size-3 fill-current" />
+              停止
+            </Button>
+          )}
+          {ENABLE_PAUSE_RESUME_UI && state.isPaused && (
+            <>
+              <Button
+                variant="ghost"
+                size="sm"
+                className="h-7 px-2 text-xs gap-1 text-primary hover:text-primary hover:bg-primary/8"
+                onClick={resumeWriting}
+              >
+                <Play className="size-3 fill-current" />
+                继续
+              </Button>
+              <Button variant="ghost" size="sm" className="h-7 px-2 text-xs gap-1" onClick={reset}>
+                <RotateCcw className="size-3" />
+                重置
+              </Button>
+            </>
+          )}
           {(isDone || isError) && (
             <Button variant="ghost" size="sm" className="h-7 px-2 text-xs gap-1" onClick={reset}>
               <RotateCcw className="size-3" />
@@ -394,7 +435,7 @@ export function AIWritingPanel() {
             isError ? 'text-destructive' : 'text-muted-foreground',
           )}
         >
-          {state.isPaused ? (
+          {ENABLE_PAUSE_RESUME_UI && state.isPaused ? (
             <>
               <span className="inline-block size-1.5 rounded-full bg-amber-400" />
               已暂停
diff --git a/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/researcher.py b/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/researcher.py
index 79c6c9c9..a3b4424a 100644
--- a/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/researcher.py
+++ b/offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/researcher.py
@@ -19,8 +19,8 @@ from ._utils import llm_json, safe_get_stream_writer
 
 logger = logging.getLogger(__name__)
 
-# 素材总量上限边界：config.yaml 的 ai_writing.max_materials，非法值回退默认 20
-_DEFAULT_MAX_MATERIALS = 20
+# 素材总量上限边界：config.yaml 的 ai_writing.max_materials，非法值回退默认 50
+_DEFAULT_MAX_MATERIALS = 50
 _MIN_MAX_MATERIALS = 1
 _MAX_MAX_MATERIALS = 100
 
@@ -93,12 +93,18 @@ async def researcher_node(state: AIWritingState, config: RunnableConfig) -> dict
     article_type = state.get("article_type", "analysis")
     target_audience = state.get("target_audience", "普通读者")
 
-    # 检查是否是补充搜索（PAUSE-1 re_search action）
+    # 检查干预模式（PAUSE-1 素材确认 / 素材不足求助）
     intervention = state.get("last_user_intervention") or {}
     extra_keywords: list[str] = intervention.get("_extra_keywords") or intervention.get("extra_keywords") or []
+    action = intervention.get("action")
 
     # 素材不足求助点触发的「补充检索」：保留原大纲，检索完直接回写作、不走素材确认/大纲
-    is_supplement = intervention.get("action") == "supplement_search"
+    is_supplement = action == "supplement_search"
+    # PAUSE-1 素材确认卡用户点「追加检索」并填了新关键词：
+    #   - 跳过 LLM 重新生成检索词（节省一次调用 + 尊重用户意图）
+    #   - 只检索 extra_keywords 这几个新词，不重检索之前 LLM 出过的词
+    #   - 保留 existing material_package，按 URL 去重合并
+    is_append_only = action == "re_search" and bool(extra_keywords)
 
     # 用户指定的检索关键词数量（1~10），决定生成多少个检索词、检索多少个方向
     keyword_count = _resolve_keyword_count(state.get("keyword_count"))
@@ -119,32 +125,43 @@ async def researcher_node(state: AIWritingState, config: RunnableConfig) -> dict
         writer({"type": "research_progress", "message": message})
 
     # Step 1: 提取关键词
-    _emit("正在生成检索词…")
-    kw_result = await llm_json(
-        EXTRACT_KEYWORDS_SYSTEM,
-        EXTRACT_KEYWORDS_USER.format(
-            user_intent=intent,
-            article_type=article_type,
-            target_audience=target_audience,
-            keyword_count=keyword_count,
-        ),
-        fallback={"keywords": [intent]},
-        config=config,
-        model_name=keyword_model,
-    )
-    # 清洗关键词：剔除占位符 / 空值，全空时回退写作意图
-    keywords: list[str] = _clean_keywords(kw_result.get("keywords"), intent)
-    if extra_keywords:
-        keywords = list(dict.fromkeys(keywords + extra_keywords))
+    kw_result: dict = {}  # 走追加分支时跳过 LLM 调用，但下方 notice 拼接需要兜底值
+    if is_append_only:
+        # 追加模式：直接用用户填的关键词，不再调 LLM
+        keywords: list[str] = _clean_keywords(extra_keywords, intent)
+        _emit(f"追加检索 {len(keywords)} 个关键词…")
+    else:
+        _emit("正在生成检索词…")
+        kw_result = await llm_json(
+            EXTRACT_KEYWORDS_SYSTEM,
+            EXTRACT_KEYWORDS_USER.format(
+                user_intent=intent,
+                article_type=article_type,
+                target_audience=target_audience,
+                keyword_count=keyword_count,
+            ),
+            fallback={"keywords": [intent]},
+            config=config,
+            model_name=keyword_model,
+        )
+        # 清洗关键词：剔除占位符 / 空值，全空时回退写作意图
+        keywords = _clean_keywords(kw_result.get("keywords"), intent)
+        if extra_keywords:
+            keywords = list(dict.fromkeys(keywords + extra_keywords))
 
     # Step 2: 并发搜索
     provider = get_search_provider(ai_writing_config)
 
-    # 按用户指定的关键词数量截断；补充搜索时不截断用户额外添加的关键词
-    search_limit = keyword_count
-    if extra_keywords:
-        search_limit = max(search_limit, len(keywords))
-    search_keywords = keywords[:search_limit]
+    # 按用户指定的关键词数量截断。
+    # 追加模式：keywords 本身就是用户填的，全部检索不截断（用户填 N 个就搜 N 个）。
+    # 其他模式：受 keyword_count 限制，但 extra_keywords 在内时上限放宽到 keywords 总数。
+    if is_append_only:
+        search_keywords = keywords
+    else:
+        search_limit = keyword_count
+        if extra_keywords:
+            search_limit = max(search_limit, len(keywords))
+        search_keywords = keywords[:search_limit]
     total = len(search_keywords)
     # 每个检索词分到的检索条数 = 总量上限 / 检索词数（向上取整）
     per_keyword = _per_keyword_limit(max_materials, total)
@@ -168,10 +185,14 @@ async def researcher_node(state: AIWritingState, config: RunnableConfig) -> dict
     all_results_nested: list[list[SearchResult]] = await asyncio.gather(*search_tasks, return_exceptions=True)
 
     # 合并去重
+    # - 首次检索 / 普通 re_search（不带 extra_keywords，用户只想 LLM 重出）：旧素材丢弃
+    # - 素材不足求助 supplement_search：保留旧素材 + 追加新检索结果
+    # - PAUSE-1 追加检索 is_append_only：保留旧素材 + 仅检索新关键词
     seen_urls: set[str] = set()
     raw_materials: list[Material] = []
     existing_pkg = state.get("material_package")
-    if existing_pkg and intervention.get("action") != "re_search":
+    preserve_existing = existing_pkg and (is_supplement or is_append_only)
+    if preserve_existing:
         for m in existing_pkg.get("materials", []):
             seen_urls.add(m.get("url", ""))
             raw_materials.append(m)
@@ -235,7 +256,13 @@ async def researcher_node(state: AIWritingState, config: RunnableConfig) -> dict
     if not materials:
         summary = "未检索到相关素材，可重新搜索或在无素材的情况下继续写作"
 
-    pkg = MaterialPackage(keywords=keywords, materials=materials, summary=summary)
+    # 追加模式下，pkg.keywords 体现到目前为止累积的检索词（旧 + 新追加），
+    # 让前端列表 / 历史回显能看到完整的搜索路径。
+    pkg_keywords = keywords
+    if is_append_only and existing_pkg:
+        old_kws = list(existing_pkg.get("keywords") or [])
+        pkg_keywords = list(dict.fromkeys(old_kws + keywords))
+    pkg = MaterialPackage(keywords=pkg_keywords, materials=materials, summary=summary)
 
     # 容错告警：检索接口 / 模型调用出错时以非致命 notice 提示用户，不打断写作流程
     notices: list[dict] = []
@@ -253,12 +280,18 @@ async def researcher_node(state: AIWritingState, config: RunnableConfig) -> dict
             raw="\n\n----------\n\n".join(raw_dumps[:3]),
         ))
 
+    # 进度事件文案：追加模式下让用户看到只搜了哪几个新词
+    step_start_msg = (
+        f"追加检索关键词：{', '.join(keywords[:3])}"
+        if is_append_only
+        else f"搜索关键词：{', '.join(keywords[:3])}"
+    )
     return {
         "material_package": pkg,
         # 补充检索：直接回写作（沿用原大纲）；常规检索：进入素材确认暂停点
         "status": "writing_draft" if is_supplement else "awaiting_material_confirm",
         "progress_events": [
-            {"type": "step_start", "agent_name": "素材收集专家", "message": f"搜索关键词：{', '.join(keywords[:3])}"},
+            {"type": "step_start", "agent_name": "素材收集专家", "message": step_start_msg},
             *notices,
             {"type": "materials_ready", "agent_name": "素材收集专家", "message": summary},
         ],