feat(ai写作): 草稿局部修改只改命中章节，支持改章节标题，修改时右侧流式重写

用户在草稿确认填修改意见(user_revise)时，按章节定位只重写命中章节、
其余字节级保留；新增范围识别区分「改标题」与「改正文」，改标题同步回写
outline 使其持久生效（修复改章节标题不生效）；整体性/结构性意见自动回退
整篇重写。前端新增 isRevising：修改时右侧先恢复初始占位（清空旧稿），
重写流到达后实时展示被改章节的流式重写。

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

frontend-web/src/open-canvas/components/ai-writing/AIWritingDraftPanel.tsx

@@ -41,10 +41,23 @@ export function AIWritingDraftPanel() {
   const showCompletedDraftView = isDraftEditable || isHistoryMode;
 
   const writingStatus = writing?.state.status;
+  const streamingSectionCount = writing?.state.streamingDraftSections.length ?? 0;
+
+  // 按用户意见修改中、且重写流还没开始（无任何流式章节）：右侧先恢复初始空占位，
+  // 把旧稿清掉；一旦重写流式 chunk 到达（streamingDraftSections 非空）就走下面的
+  // 流式预览，实时展示重写过程；draft_ready 后 isRevising 置回 false 显示新成稿。
+  if (writing?.state.isRevising && streamingSectionCount === 0) {
+    return (
+      <div className="h-full bg-muted/10 border-l">
+        <DraftWaitingPlaceholder />
+      </div>
+    );
+  }
+
   // 仅当章节仍在累积时走流式 Markdown；draft_ready 清空 sections 后即使 status 尚未切走也进画布
   const isActivelyStreamingDraft =
     (writingStatus === 'writing_draft' || writingStatus === 'revising') &&
-    (writing?.state.streamingDraftSections.length ?? 0) > 0;
+    streamingSectionCount > 0;
 
   const streamingMarkdown = isActivelyStreamingDraft && writing
     ? assembleStreamingDraftMarkdown(

frontend-web/src/open-canvas/hooks/useAIWritingStream.ts

@@ -70,6 +70,7 @@ const INITIAL_STATE: AIWritingState = {
   streamingReviewItems: [],
   revisionSummary: null,
   pendingRevisionContext: null,
+  isRevising: false,
 };
 
 export { INITIAL_STATE };
@@ -118,7 +119,8 @@ export function reducer(prev: AIWritingState, action: Action): AIWritingState {
         ...(ns === 'writing_outline' ? { outlineStreamingText: '', outlineThinkingText: '' } : {}),
         ...(ns === 'writing_draft' ? { streamingDraftSections: [], streamingDraftTitle: '' } : {}),
         ...(ns === 'reviewing' ? { streamingReviewItems: [] } : {}),
-        ...(ns === 'revising' ? { streamingDraftSections: [], streamingDraftTitle: '' } : {}),
+        // 进入「修改」：右侧画布恢复初始空占位（isRevising），不再显示旧稿/流式重写
+        ...(ns === 'revising' ? { streamingDraftSections: [], streamingDraftTitle: '', isRevising: true } : {}),
         completedInterventions: [...prev.completedInterventions, action.record],
         ...(action.pendingRevisionContext ? { pendingRevisionContext: action.pendingRevisionContext } : {}),
       };
@@ -357,6 +359,8 @@ function handleEvent(prev: AIWritingState, event: ProgressEvent): AIWritingState
         streamingDraftTitle: '',
         revisionSummary: summary,
         pendingRevisionContext: null,
+        // 修改完成：解除右侧空占位，展示新成稿
+        isRevising: false,
       };
     }
 

frontend-web/src/open-canvas/types/ai-writing.ts

@@ -389,4 +389,7 @@ export interface AIWritingState {
     skippedIssues: ReviewIssue[];
     oldWordCount: number;
   } | null;
+  /** 是否正按用户意见修改中（revising 提交起 → draft_ready 止）。
+   *  为 true 时右侧画布恢复初始空占位，不展示旧稿 / 流式重写过程。 */
+  isRevising: boolean;
 }

offline-backend-20260512/backend/CLAUDE.md

@@ -278,7 +278,7 @@ Seeding mirrors the roundtable pattern: verified `config.yaml` + `SOUL.md` ship
 
 **Agentic skill execution (`ai_writing.researcher_skill_agent`, default false).** By default a configured non-native skill is *not* run agent-style — the node just turns its `description` into a query for the generic search provider. With this flag on, the directed-skill branch instead runs `nodes/_skill_agent.py::run_skill_search_agent`: it injects the skill's **SKILL.md** (read from `skill.skill_file`) alongside the researcher SOUL as the system prompt, binds a `search` tool to the researcher model, and runs a bounded (`_MAX_TURNS=4`) tool-call loop so the model drives the retrieval per the skill's instructions; the `SearchResult`s the tool returns are harvested directly as materials (no fuzzy text extraction). Requires a tool-calling model; if the model can't tool-call (returns no `tool_calls`) or the agent yields nothing, `_execute_skill` falls back to the legacy description→query search so materials are never empty. The built-in `knowledge-base-search` native concurrent path is unchanged. Tests: `tests/test_ai_writing_skill_agent.py`.
 
-**Targeted user revision (局部修改 — 只改命中章节).** When the user submits a free-text 修改意见 on the draft-confirm card (`action == "user_revise"`), `writer_draft_node` no longer always rewrites the whole article section-by-section. Before the full-rewrite loop it tries `_targeted_user_revision` (`nodes/writer_draft.py`): `_split_draft_sections` parses the previous `current_draft.full_markdown` into `(article_title, [(section_title, body)], references_block)` (level-1 `#` title, level-2 `##` sections, trailing `## 参考文献` pulled out and kept verbatim), then `_detect_revision_scope` runs a small `llm_json` classifier (prompts `DRAFT_REVISE_SCOPE_{SYSTEM,USER}`) that maps the note to the section indices it targets ("第一段/开头/引言"→0, "结尾/结论"→last, by title/topic, etc.). **Only the hit sections are rewritten** via `_stream_revise_section` (prompt `DRAFT_REVISE_SECTION_USER`, streamed as `draft_chunk` so the user watches the edit); **every unhit section is preserved byte-for-byte** and emitted to the stream as one full-text `draft_chunk` + `is_section_done` so the frontend's `streamingDraftSections` rebuilds the complete article (it was reset to `[]` on entering `revising`). The classifier returns `whole_article=true` (整体性意见 like 「整体语气更正式」/「全文都改」, or any structure-changing ask) → `_detect_revision_scope` yields `None` → `_targeted_user_revision` returns `None` → the node **falls back to the existing whole-article rewrite** (which still runs `_revise_outline_with_notes` for title/structure changes). Also falls back when there is no prior draft, an empty note, or scope detection errors. Editor-review revisions (`accept_review`) are unaffected — they keep the holistic rewrite. Reassembly preserves the article title + section structure and re-appends the original references block; word count is computed on the body (excluding references), matching the main path. Tests: `tests/test_ai_writing_targeted_revision.py`.
+**Targeted user revision (局部修改 — 只改命中章节).** When the user submits a free-text 修改意见 on the draft-confirm card (`action == "user_revise"`), `writer_draft_node` no longer always rewrites the whole article section-by-section. Before the full-rewrite loop it tries `_targeted_user_revision` (`nodes/writer_draft.py`): `_split_draft_sections` parses the previous `current_draft.full_markdown` into `(article_title, [(section_title, body)], references_block)` (level-1 `#` title, level-2 `##` sections, trailing `## 参考文献` pulled out and kept verbatim), then `_detect_revision_scope` runs a small `llm_json` classifier (prompts `DRAFT_REVISE_SCOPE_{SYSTEM,USER}`) returning **per-section targets** `[{index, new_title, rewrite_body}]`: it maps the note to the section indices it targets ("第一段/开头/引言"→0, "结尾/结论"→last, by title/topic) **and** distinguishes a **章节标题改名** (`new_title` set, `rewrite_body=false` → rename only, body untouched) from a **正文修改** (`rewrite_body=true`). Title-only renames apply directly to the assembled `## {new_title}` heading **and are written back into `current_outline.sections[i].section_title`** (so the renamed title persists through confirm/finalize/history — this fixes "改章节标题不生效"). Body rewrites run `_stream_revise_section` (prompt `DRAFT_REVISE_SECTION_USER`, streamed as `draft_chunk`). **Only hit sections are touched; unhit sections are preserved byte-for-byte and are NOT re-emitted to the stream** — the frontend clears the right canvas to its initial placeholder on entering `revising` and shows only the live rewrite of the changed section(s), so the user sees「清空 → 流式重写 → 终稿」rather than the whole article rebuilding. The classifier returns `whole_article=true` (整体性意见 like 「整体语气更正式」/「全文都改」, **structure changes** 加章/删章/调整顺序, or a **文章总标题** change) → `_detect_revision_scope` yields `None` → `_targeted_user_revision` returns `None` → the node **falls back to the whole-article rewrite** (which still runs `_revise_outline_with_notes` for title/structure changes). Also falls back when there is no prior draft, an empty note, or scope-detection errors. Editor-review revisions (`accept_review`) keep the holistic rewrite. Reassembly preserves the article title + structure and re-appends the original references block; word count is computed on the body (excluding references). Frontend: `isRevising` in `AIWritingState` (`types/ai-writing.ts`) is set in `useAIWritingStream.ts`'s `intervention_submitted` (`ns === 'revising'`) and cleared on `draft_ready`; `AIWritingDraftPanel.tsx` shows the initial `DraftWaitingPlaceholder` while `isRevising && streamingDraftSections.length === 0`, then the streaming preview once chunks arrive. Tests: `tests/test_ai_writing_targeted_revision.py`.
 
 **History draft editing (历史回看可编辑成稿).** The right-side draft in 历史回看 is no longer read-only — the user can edit it like the live 写作完成态 (Tiptap + 目录 + 格式工具栏 + AI 润色) and **persist** the change back to the session. `PUT /api/ai-writing/sessions/{id}/draft` `{draft_markdown, draft_title?}` (owner-checked → 404 otherwise; `draft_title` omitted ⇒ keep原标题) updates the row via the existing `AIWritingSessionRepository.update`. Frontend: `AIWritingDraftPanel` now passes `readOnly={false}` + an `onSaveDraft` handler (only in history mode) → `TextArtifactRenderer` → `WritingToolbar` 「保存」button (loading + toast); the handler is `AIWritingHistoryContext.saveHistoryDraft(markdown, title?)` → `updateAIWritingDraft` (`api/ai-writing-sessions.ts`), which write-backs the returned summary to `historySession`. Edits to the **live** completed draft stay ephemeral (export-only) as before — only history mode shows 保存. Tests: `tests/test_ai_writing_draft_update.py`.
 

offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/nodes/writer_draft.py

@@ -406,15 +406,20 @@ def _split_draft_sections(markdown: str) -> tuple[str, list[tuple[str, str]], st
 async def _detect_revision_scope(
     state: AIWritingState, config: RunnableConfig,
     parsed_sections: list[tuple[str, str]], note: str,
-) -> set[int] | None:
-    """LLM 识别修改意见指向哪些章节编号；整体性/无法定位/异常一律返回 None（回退整篇重写）。"""
+) -> list[dict] | None:
+    """LLM 识别意见指向哪些章节、各自怎么改（改标题 / 改正文）。
+
+    返回 ``[{"index": int, "new_title": str | None, "rewrite_body": bool}, ...]``，
+    整体性 / 改变结构 / 无法定位 / 异常一律返回 None（回退整篇重写）。
+    """
     listing = "\n".join(f"{i}. {t}" for i, (t, _b) in enumerate(parsed_sections))
     model_name = (state.get("model_name") or "").strip() or None
+    n = len(parsed_sections)
     try:
         result = await llm_json(
             DRAFT_REVISE_SCOPE_SYSTEM,
-            DRAFT_REVISE_SCOPE_USER.format(n=len(parsed_sections), sections=listing, note=note),
-            {"whole_article": True, "target_indexes": []},
+            DRAFT_REVISE_SCOPE_USER.format(n=n, sections=listing, note=note),
+            {"whole_article": True, "targets": []},
             config,
             model_name=model_name,
         )
@@ -423,18 +428,41 @@ async def _detect_revision_scope(
         return None
     if not isinstance(result, dict) or result.get("whole_article"):
         return None
-    raw = result.get("target_indexes") or result.get("targetIndexes") or []
-    if not isinstance(raw, list):
-        return None
-    idxs: set[int] = set()
-    for x in raw:
-        try:
-            k = int(x)
-        except (ValueError, TypeError):
-            continue
-        if 0 <= k < len(parsed_sections):
-            idxs.add(k)
-    return idxs or None
+
+    by_idx: dict[int, dict] = {}
+    raw_targets = result.get("targets")
+    if isinstance(raw_targets, list) and raw_targets:
+        for item in raw_targets:
+            if not isinstance(item, dict):
+                continue
+            try:
+                idx = int(item.get("index"))
+            except (ValueError, TypeError):
+                continue
+            if not (0 <= idx < n):
+                continue
+            nt = item.get("new_title")
+            new_title = nt.strip() if isinstance(nt, str) and nt.strip() else None
+            rewrite_body = item.get("rewrite_body")
+            # 缺省：模型没给 rewrite_body 时——有新标题就当「只改标题」，否则当「改正文」
+            if rewrite_body is None:
+                rewrite_body = new_title is None
+            by_idx[idx] = {"index": idx, "new_title": new_title, "rewrite_body": bool(rewrite_body)}
+    else:
+        # 兼容旧结构 target_indexes：纯正文重写
+        raw_idx = result.get("target_indexes") or result.get("targetIndexes") or []
+        if isinstance(raw_idx, list):
+            for x in raw_idx:
+                try:
+                    idx = int(x)
+                except (ValueError, TypeError):
+                    continue
+                if 0 <= idx < n:
+                    by_idx[idx] = {"index": idx, "new_title": None, "rewrite_body": True}
+
+    # 丢弃既不改标题也不改正文的空操作项
+    targets = [t for t in by_idx.values() if t["new_title"] is not None or t["rewrite_body"]]
+    return targets or None
 
 
 async def _stream_revise_section(
@@ -516,14 +544,27 @@ async def _targeted_user_revision(
     targets = await _detect_revision_scope(state, config, parsed, note)
     if not targets:
         return None
+    by_idx = {t["index"]: t for t in targets}
 
     outline = state.get("current_outline") or {"title": article_title, "sections": []}
     out_sections = outline.get("sections", [])
     revision_count = state.get("revision_count", 0)
 
+    new_titles: list[str] = []
     new_bodies: list[str] = []
+    changed_labels: list[str] = []
+    title_changed = False
     for idx, (sec_title, body) in enumerate(parsed):
-        if idx in targets:
+        t = by_idx.get(idx)
+        if t is None:
+            # 未命中章节：标题、正文字节级原样保留（不重写、不下发流式）
+            new_titles.append(sec_title)
+            new_bodies.append(body)
+            continue
+        final_title = t["new_title"] or sec_title
+        if t["new_title"] and t["new_title"] != sec_title:
+            title_changed = True
+        if t["rewrite_body"]:
             out_sec = out_sections[idx] if idx < len(out_sections) else {}
             key_points_str = _render_key_points_for_prompt(
                 out_sec.get("key_points") or out_sec.get("keyPoints") or []
@@ -532,49 +573,54 @@ async def _targeted_user_revision(
                 out_sec.get("material_ids") or out_sec.get("materialIds") or [], pkg, cite_map
             )
             new_body = await _stream_revise_section(
-                state, config, article_title=article_title, idx=idx, section_title=sec_title, existing_text=body,
-                key_points_str=key_points_str, section_materials=section_materials,
+                state, config, article_title=article_title, idx=idx, section_title=final_title,
+                existing_text=body, key_points_str=key_points_str, section_materials=section_materials,
                 note=note, add_references=add_references, writer=writer,
                 use_agent=use_agent, legacy_model_name=legacy_model_name,
             )
-            new_bodies.append(new_body)
         else:
-            # 未改动章节：整段瞬时下发 + 收尾，让前端草稿视图完整重建（其余字节级原样保留）
-            writer({
-                "type": "draft_chunk", "section_index": idx, "section_title": sec_title,
-                "text": body, "is_thinking": False, "is_section_done": False,
-            })
-            writer({
-                "type": "draft_chunk", "section_index": idx, "section_title": sec_title,
-                "text": "", "is_section_done": True,
-            })
-            new_bodies.append(body)
+            # 只改标题、不动正文：正文原样保留
+            new_body = body
+        new_titles.append(final_title)
+        new_bodies.append(new_body)
+        changed_labels.append(final_title)
 
     body_md = "\n\n".join(
         f"## {t}\n\n{_normalize_section_headings(b)}"
-        for (t, _old), b in zip(parsed, new_bodies)
+        for t, b in zip(new_titles, new_bodies)
     )
     full_markdown = f"# {article_title}\n\n{body_md}"
     word_count = count_words(full_markdown)
     if references_block:
         full_markdown = f"{full_markdown}\n\n{references_block}"
 
+    # 章节标题被改 → 同步回 outline（按 index 改 section_title），使后续确认/审核/历史结构一致
+    updated_outline = outline
+    if title_changed and out_sections:
+        updated_outline = {
+            **outline,
+            "sections": [
+                ({**s, "section_title": new_titles[i]} if i < len(new_titles) else s)
+                for i, s in enumerate(out_sections)
+            ],
+        }
+
     draft = DraftArticle(
-        title=article_title or outline.get("title", "文章"),
-        outline=outline,
+        title=article_title or updated_outline.get("title", "文章"),
+        outline=updated_outline,
         full_markdown=full_markdown,
         word_count=word_count,
         revision_count=revision_count,
     )
     await _save_draft_to_repo(config, draft)
 
-    changed_names = "、".join(parsed[i][0] for i in sorted(targets))
+    changed_names = "、".join(changed_labels) if changed_labels else "相关章节"
     notices: list[dict] = list(extra_notices or [])
     notices.append({
         "type": "notice", "agent_name": agent_label, "title": "局部修改",
-        "detail": f"已按你的意见只修改了 {len(targets)} 个章节（{changed_names}），其余内容原样保留。",
+        "detail": f"已按你的意见只改动了 {len(by_idx)} 个章节（{changed_names}），其余内容原样保留。",
     })
-    return {
+    updates: dict = {
         "current_draft": draft,
         "drafts": [draft],
         "status": "awaiting_draft_confirm",
@@ -584,12 +630,15 @@ async def _targeted_user_revision(
             {
                 "type": "draft_ready", "agent_name": agent_label,
                 "message": f"修改完成，共 {word_count} 字",
-                "draft_title": article_title or outline.get("title", "文章"),
+                "draft_title": article_title or updated_outline.get("title", "文章"),
                 "draft_markdown": full_markdown,
                 "revision_count": revision_count,
             },
         ],
     }
+    if title_changed:
+        updates["current_outline"] = updated_outline
+    return updates
 
 
 async def writer_draft_node(state: AIWritingState, config: RunnableConfig) -> dict:

offline-backend-20260512/backend/packages/harness/deerflow/agents/ai_writing/prompts/writer_prompts.py

@@ -231,32 +231,37 @@ DRAFT_SECTION_USER = """文章标题：{article_title}
 
 # ── 用户局部修改：识别意见指向的章节范围 ───────────────────────────────────────
 
-DRAFT_REVISE_SCOPE_SYSTEM = """你是一名编辑助理。用户对一篇已成稿的文章提出了一句修改意见，你要判断这条意见究竟要改文章里的哪些「章节」。
+DRAFT_REVISE_SCOPE_SYSTEM = """你是一名编辑助理。用户对一篇已成稿的文章提出了一句修改意见，你要判断这条意见要改文章里的哪些「章节」，以及每个章节具体改什么。
 文章由若干带编号的章节组成（编号从 0 开始）。
 
 【判断规则】（最高优先级）
-- 仔细理解意见指向，可能是：
-  - 序号：「第一段 / 第一部分 / 第1章」→ 编号 0，「第二段」→ 编号 1，依此类推；
-  - 位置：「开头 / 引言 / 导语」→ 第一个章节，「结尾 / 结论 / 最后一段」→ 最后一个章节；
-  - 直接点名某个章节的标题或主题。
-- 若意见只针对个别章节，输出这些章节的编号列表，并令 whole_article=false。
-- 若意见是**整体性**的（如「全文都改」「整体语气更正式」「每一部分都精简」「通篇润色」），
-  或会**改变文章结构**（如「标题改成…」「加一章 / 删一章 / 调整章节顺序」），令 whole_article=true、target_indexes 留空。
-- 拿不准、无法明确定位到具体章节时，一律 whole_article=true（交给整篇重写更安全）。
+1. 定位意见指向的章节：
+   - 序号：「第一段 / 第一部分 / 第1章」→ 编号 0，「第二段」→ 编号 1，依此类推；
+   - 位置：「开头 / 引言 / 导语」→ 第一个章节，「结尾 / 结论 / 最后一段」→ 最后一个章节；
+   - 直接点名某个章节的标题或主题。
+2. 对每个命中的章节，判断用户要改什么：
+   - 若要改这一章节的**标题**（如「把第一章的标题改为简介」「第二部分改名叫…」），把新标题文字放进 new_title；否则 new_title 写 null。
+   - 若要改这一章节的**正文内容**（润色 / 补充 / 精简 / 重写等），rewrite_body 写 true；若只改标题、不动正文，rewrite_body 写 false。
+3. 下列情况属于整体性、或会改变文章结构，一律令 whole_article=true、targets 留空：
+   - 整体性：「全文都改」「通篇润色」「整体语气更正式」「每一部分都精简」；
+   - 结构性：新增章节、删除章节、调整章节顺序、合并 / 拆分章节；
+   - 改**文章总标题**（注意：是整篇文章的标题，不是某个章节的标题）。
+4. 拿不准、无法明确定位到具体章节时，一律 whole_article=true（交给整篇重写更安全）。
 
 【严禁思考与表态】
 - 不要输出任何思考、解释或对用户说的话；收到输入后立即直接输出最终 JSON。
 
 【输出要求】
 - 只输出一个 JSON 对象，不要解释文字，不要 markdown 代码块，必须用英文双引号。
-- 结构：{"whole_article": false, "target_indexes": [0]}"""
+- new_title 没有就写 null；targets 为空数组表示走整篇重写。
+- 结构：{"whole_article": false, "targets": [{"index": 0, "new_title": "简介", "rewrite_body": false}]}"""
 
 DRAFT_REVISE_SCOPE_USER = """文章共有 {n} 个章节，编号与标题如下：
 {sections}
 
 用户的修改意见：{note}
 
-请判断这条意见要改哪些章节，按要求只输出 JSON。"""
+请判断这条意见要改哪些章节、各自怎么改（标题 / 正文），按要求只输出 JSON。"""
 
 # ── 用户局部修改：只重写命中的单个章节（保留其余内容）─────────────────────────