feat: standardize Word export and Office conversion

offline-backend-20260512/backend/CLAUDE.md

@@ -324,7 +324,7 @@ FastAPI application on port 8001 with health checks at `GET|HEAD /health`, `/api
 | **Threads** (`/api/threads/{id}`) | `DELETE /` - remove DeerFlow-managed local thread data after LangGraph thread deletion; unexpected failures are logged server-side and return a generic 500 detail |
 | **Threads search** (`POST /threads/search`, LangGraph-compat) | Conversation listing. **System threads are excluded by default**: rows whose `metadata.system` is truthy (scheduler runs, roundtable workers) never crowd the user's list; a caller whose `metadata` filter explicitly references `thread_type`/`system` (e.g. studio notebooks) opts out of the exclusion. New `query` field = case-insensitive title (`display_name`) substring search, used by the chats page's server-side pagination. `POST /threads/count` takes the same request body (limit/offset ignored) and returns `{total}` under identical filtering/exclusion semantics ? powers the chats page's ??/??? display without fetching rows (`ThreadMetaStore.count`). `ThreadMetaStore.search` gained `exclude_system`/`query` params; JSON-blob filters batch-scan until the requested `limit`/`offset` window is full (so pagination over a heavily-filtered set never under-fills), with a `thread_id` tie-breaker on the `updated_at DESC` ordering for stable OFFSET paging. Tests: `tests/test_thread_meta_search.py` |
 | **Artifacts** (`/api/threads/{id}/artifacts`) | `GET /{path}` - serve artifacts; active content types (`text/html`, `application/xhtml+xml`, `image/svg+xml`) are always forced as download attachments to reduce XSS risk; `?download=true` still forces download for other file types |
-| **Word export** (`POST /api/writing/export/docx`) | Shared Markdown → DOCX generator used by conversation, artifact, scheduled-task, Deep Research, and AI-writing download surfaces. `app/gateway/word_export.py` builds OOXML with no Python conversion dependency: A4 portrait, 3.7/3.5/2.8/2.6 cm margins, real multilevel heading numbering, formal body/table styles, page numbers beginning at 1 with odd pages left-aligned and even pages right-aligned, and embedded `app/gateway/assets/fonts/方正小标宋简体.ttf`. The font loader validates its internal family, SHA-256, and OpenType embedding permission (`OS/2.fsType`) before export. Tests: `tests/test_word_export.py`. |
+| **Word export** (`POST /api/writing/export/docx`) | Shared Markdown → DOCX generator used by conversation, artifact, scheduled-task, Deep Research, and AI-writing download surfaces. `app/gateway/word_export.py` builds OOXML with no Python conversion dependency: A4 portrait, 3.7/3.5/2.8/2.6 cm margins, real multilevel heading numbering (`一、` / `(一)` / `1.`), two-character first-line indent on heading levels 1–3, formal body/table styles, page numbers beginning at 1 with odd pages left-aligned and even pages right-aligned, and embedded `app/gateway/assets/fonts/方正小标宋简体.ttf`. Before numbering it removes complete manual heading prefixes (including `2.1` / `2.1.1`) and drops Markdown horizontal rules from Word output. The default ordinary-Q&A Lead Agent prompt emits the matching Markdown heading forms and avoids unordered bullet lists; writing setup, writing mode, notebook, scheduled runs, and SOUL-driven custom agents keep their independent prompts. The font loader validates its internal family, SHA-256, and OpenType embedding permission (`OS/2.fsType`) before export. Tests: `tests/test_word_export.py`, `tests/test_es_query_routing_prompt.py`. |
 | **Deep Research** (`/api/deep-research`) | Durable, user-owned research sessions/jobs with replayable SSE events and hidden-thread artifacts. `basic`/`quick` use the compact runner; `detailed` creates bounded subtopics, searches/writes sections concurrently, and assembles a report; `deep` recursively searches follow-up questions under a deterministic breadth/depth query budget. `multi_agent` is a LangGraph 1.x editor/researcher/writer/reviewer graph: its initial plan emits `awaiting_input`, the recorder stores the interrupt and releases the lease, and `POST /jobs/{id}/resume` rebuilds the graph from the persisted approved/revised plan. A bounded `custom_outline` is frozen into the session; headings/list entries deterministically supply the detailed and multi-agent plan, and the shared writer receives it only as a guarded content constraint. Completed reports support owner-scoped follow-up (`GET /sessions/{id}/messages`, `POST /sessions/{id}/chat`): context is only the current report, selected sources and recent messages; source-id citations are persisted, while `allow_new_research=true` is explicitly rejected. Owners may change that evidence set only after the job stops via `PATCH /sessions/{id}/sources/{source_id}`; this applies to future follow-ups rather than rewriting a completed report. Optional report illustrations are gated by `config.yaml -> deep_research.image_generation`: the Harness `OpenAICompatibleImageProvider` requires base64 output, the shared runner step writes only bounded `research-image-01..04.(png|jpg|webp)` artifacts and Markdown uses private `deep-research://` links. `GET /sessions/{id}/artifacts/{name}` checks session ownership before serving an image; an unavailable provider yields a warning but never fails the text report. Runners live in `packages/harness/deerflow/agents/deep_research/runners/`, may never import `app.*`, and use only the `AdapterBundle`. |
 | **Gateway auth** | Owner-scoped Deep Research routes await the platform's asynchronous `get_current_user` resolver before they read or write persistence, ensuring each session and job operation receives a concrete user ID. Deep Research artifacts are always resolved from a whitelisted filename via `/mnt/user-data/outputs/<name>` before mapping to a host path, preserving the sandbox guard on native Windows and Linux hosts. |
 | **Suggestions** (`/api/threads/{id}/suggestions`) | `POST /` - generate follow-up questions; rich list/block model content is normalized before JSON parsing |
@@ -1611,3 +1611,16 @@ write, log the exception, complete the rewrite with
 `versionSnapshotFailed: true`, and do not emit a user-facing failure or restore
 the original. The sandbox uses that signal only to enable its ordinary manual
 save fallback.
+
+### Offline local-sandbox Office runtime
+
+The DeerFlow-only offline deployment uses `LocalSandboxProvider`, so document
+dependencies belong in the backend image rather than a separate sandbox image.
+`deploy/linux-docker-offline/Dockerfile.backend` installs python-docx,
+LibreOffice Writer/Calc/Impress, Pandoc, Poppler, and CJK fonts in addition to
+the MarkItDown Office extras. `file_conversion.py` normalizes legacy
+DOC/XLS/PPT and ODT/ODS/ODP through a per-conversion LibreOffice user profile
+before MarkItDown extraction, avoiding concurrent profile locks. The image
+build and `OFFLINE_INSTALL.sh --office-check` both run
+`scripts/office_runtime_selfcheck.py`; that smoke test creates and re-parses
+minimal DOCX/XLSX/PPTX files and does not need the application database.

offline-backend-20260512/backend/Dockerfile.backend-offline

@@ -60,7 +60,10 @@ RUN python -m venv /opt/venv \
     && /opt/venv/bin/python -m pip install --no-cache-dir --upgrade -i "${UV_INDEX_URL}" pip setuptools wheel \
     && sed -E '/--hash=/d; s/[[:space:]]+\\$//; s|^-e[[:space:]]+./packages/harness$|./packages/harness|' requirements-offline.txt > /tmp/requirements-docker.txt \
     && /opt/venv/bin/pip install --no-cache-dir -i "${UV_INDEX_URL}" -r /tmp/requirements-docker.txt \
-    && /opt/venv/bin/pip install --no-cache-dir -i "${UV_INDEX_URL}" "asyncmy>=0.2.11"
+    && /opt/venv/bin/pip install --no-cache-dir -i "${UV_INDEX_URL}" \
+        "asyncmy>=0.2.11" \
+        "python-docx==1.2.0" \
+    && /opt/venv/bin/python -c 'import docx, lxml, mammoth, markitdown, openpyxl, pandas, pptx, xlrd; print("Office Python dependencies OK")'
 
 COPY . .
 
@@ -92,9 +95,17 @@ RUN if [ -n "${APT_MIRROR}" ]; then \
         ca-certificates \
         curl \
         docker.io \
+        file \
+        fontconfig \
+        fonts-liberation \
+        fonts-noto-cjk \
         libmagic1 \
+        libreoffice-calc \
+        libreoffice-impress \
+        libreoffice-writer \
         nodejs \
         npm \
+        pandoc \
         poppler-utils \
         tini \
     && rm -rf /var/lib/apt/lists/*
@@ -102,6 +113,11 @@ RUN if [ -n "${APT_MIRROR}" ]; then \
 COPY --from=builder /opt/venv /opt/venv
 COPY --from=builder /build/backend /app/backend
 
+RUN /opt/venv/bin/python -c 'import docx, lxml, mammoth, markitdown, openpyxl, pandas, pptx, xlrd' \
+    && command -v soffice >/dev/null \
+    && command -v pandoc >/dev/null \
+    && fc-list :lang=zh | grep -q .
+
 EXPOSE 8001
 
 ENTRYPOINT ["tini", "--"]

offline-backend-20260512/backend/README.md

@@ -156,7 +156,7 @@ FastAPI application providing REST endpoints for frontend integration:
 | `GET/POST /api/deep-research/*` | Durable Deep Research sessions and jobs. `quick`/`basic` run the compact report flow; `detailed` plans subtopics and writes independent sections; `deep` recursively investigates bounded follow-up branches. All modes reuse DeerFlow's configured material provider, LLM factory, hidden-thread artifacts, and SSE event log. Deep Research session/job/event/source/message writes serialise their rows in the writer transaction, rather than doing a post-commit ORM refresh or readback; this prevents MySQL read/write-splitting replica lag from failing “start writing”, dispatcher claim, report persistence, or follow-up history. Whole-document artifact/report rewrites likewise create their reversible version snapshots from writer-local ORM data and never refresh after commit, so replica lag cannot restore an otherwise successful rewrite. A report is written to the session and confirmed before its Job becomes `completed`; an unconfirmed write is retried and never exposed as a false completion. Retrying an already-created queued job re-nudges the dispatcher, recovering an interrupted HTTP response without creating a second report. If progress-event persistence, durable replay, or live fan-out is briefly unavailable, the worker/stream logs the degradation and continues or reconnects rather than failing the whole job. Whole-report rewrite accepts an optional `reportOutline` Markdown template; the durable job persists it and applies it as the chapter structure constraint. Optional report illustrations use a separately configured OpenAI-compatible image endpoint and are served only through the owning session's authenticated artifact route. |
 | `/api/enterprise-research/*` | Temporarily disabled. Its router and persistence initialization are not registered, so these endpoints return 404 and no enterprise-research tables are created or migrated during startup. |
 | `POST /api/ai-writing/sample/extract` | Extract text from Word/PDF/Markdown/TXT samples for AI-writing imitation mode |
-| `POST /api/writing/export/docx` | Generate the shared formal Word download from Markdown. Uses the standard-library OOXML builder in `app/gateway/word_export.py`, applies the prescribed A4/margin/heading/body/table/footer profile, and embeds the server-bundled `方正小标宋简体.ttf`; the client computer does not need that font installed. |
+| `POST /api/writing/export/docx` | Generate the shared formal Word download from Markdown. Uses the standard-library OOXML builder in `app/gateway/word_export.py`, removes manual/compound heading prefixes and Markdown horizontal rules, applies the prescribed A4/margin/heading/body/table/footer profile, and embeds the server-bundled `方正小标宋简体.ttf`; the client computer does not need that font installed. |
 | `POST /cop/saveSuperiorTask` | TaskCOP compatibility API (DeerFlow Bearer/session authentication required): create a situation-overview task with a sequential four-digit id (`0001`, `0002`, ...) and return the legacy `{state, msg, data}` envelope |
 | `POST /api/taskcop/import/tasks` | TaskCOP compatibility API (authenticated): import tasks from a JSON array or `tasks`/`records` envelope; every row's supplied `id`/`taskId` is authoritative and a duplicate id completely replaces the stored task |
 | `DELETE /cop/tasks/{task_id}` | TaskCOP compatibility API (authenticated): delete one task and its task-scoped situation-report document |
@@ -943,3 +943,13 @@ The final rewrite commit uses a deliberately short sibling temporary filename,
 which keeps long Windows artifact paths below the path-length limit. It also
 recreates a missing `outputs` parent before retrying, so transient
 sandbox-directory recreation cannot discard a completed rewrite.
+
+## Offline local-sandbox Office runtime
+
+The pure DeerFlow offline image treats the backend container as the default
+`LocalSandboxProvider` runtime. It bundles MarkItDown, python-docx, openpyxl,
+python-pptx, LibreOffice Headless, Pandoc, Poppler, and CJK fonts. Modern OOXML
+uploads are parsed directly; legacy DOC/XLS/PPT and ODT/ODS/ODP files are
+normalized through an isolated LibreOffice profile before Markdown extraction.
+Run `./OFFLINE_INSTALL.sh --office-check` from the offline bundle to generate
+and re-parse Word, Excel, and PowerPoint samples without requiring a database.

offline-backend-20260512/backend/app/gateway/word_export.py

@@ -214,7 +214,13 @@ def _strip_inline_markdown(text: str) -> str:
 
 def _strip_manual_heading_number(text: str) -> str:
     return re.sub(
-        r"^\s*(?:[一二三四五六七八九十百千]+、|[（(][一二三四五六七八九十百千]+[）)]|\d+[.、]|[（(]\d+[）)]|\d+[）)])\s*",
+        r"^\s*(?:"
+        r"[一二三四五六七八九十百千]+[、.．]"
+        r"|[（(]\s*[一二三四五六七八九十百千]+\s*[）)]"
+        r"|[（(]\s*\d+(?:[.．]\d+)*\s*[）)]"
+        r"|\d+(?:[.．]\d+)+(?:[.．、])?"
+        r"|\d+[.．、)）]"
+        r")\s*",
         "",
         text,
     ).strip()
@@ -338,7 +344,6 @@ def _parse_markdown(markdown: str, explicit_title: str | None) -> tuple[str | No
 
         if re.fullmatch(r"\s*(?:---|\*\*\*|___)\s*", raw):
             flush_paragraph()
-            blocks.append(ParagraphBlock("horizontal_rule", "────────"))
             index += 1
             continue
 
@@ -584,8 +589,6 @@ def _blocks_xml(blocks: list[Block], relationships: _DocumentRelationships) -> s
             output.append(_paragraph_xml("QuoteText", _inline_xml(block.text, relationships)))
         elif block.kind == "image":
             output.append(_image_paragraph_xml(block.text, relationships))
-        elif block.kind == "horizontal_rule":
-            output.append(_paragraph_xml("BodyText", _run_xml(block.text), alignment="center"))
         else:
             output.append(_paragraph_xml("BodyText", _inline_xml(block.text, relationships)))
     return "".join(output)
@@ -677,6 +680,7 @@ def _styles_xml() -> str:
             font=HEADING_1_FONT_FAMILY,
             size=BODY_SIZE_HALF_POINTS,
             alignment="left",
+            first_line_chars=200,
             keep_next=True,
             keep_lines=True,
             outline_level=0,
@@ -687,6 +691,7 @@ def _styles_xml() -> str:
             font=HEADING_2_FONT_FAMILY,
             size=BODY_SIZE_HALF_POINTS,
             alignment="left",
+            first_line_chars=200,
             keep_next=True,
             keep_lines=True,
             outline_level=1,
@@ -700,6 +705,7 @@ def _styles_xml() -> str:
                 font=BODY_FONT_FAMILY,
                 size=BODY_SIZE_HALF_POINTS,
                 alignment="left",
+                first_line_chars=200 if level == 3 else 0,
                 keep_next=True,
                 keep_lines=True,
                 outline_level=level - 1,
@@ -769,10 +775,11 @@ def _styles_xml() -> str:
 
 def _heading_level_xml(level: int, format_value: str, text: str, font: str) -> str:
     restart = f'<w:lvlRestart w:val="{level}"/>' if level > 0 else ""
+    first_line_chars = ' w:firstLineChars="200"' if level <= 2 else ""
     return (
         f'<w:lvl w:ilvl="{level}"><w:start w:val="1"/>{restart}<w:numFmt w:val="{format_value}"/>'
         f'<w:lvlText w:val="{_xml_attr(text)}"/><w:lvlJc w:val="left"/>'
-        '<w:pPr><w:ind w:left="0" w:hanging="0"/></w:pPr>'
+        f'<w:pPr><w:ind w:left="0" w:hanging="0"{first_line_chars}/></w:pPr>'
         f'<w:rPr>{_font_run_properties(font, BODY_SIZE_HALF_POINTS)}</w:rPr></w:lvl>'
     )
 
@@ -798,7 +805,7 @@ def _numbering_xml() -> str:
     heading_levels = "".join(
         [
             _heading_level_xml(0, "chineseCounting", "%1、", HEADING_1_FONT_FAMILY),
-            _heading_level_xml(1, "chineseCounting", "（%2）", HEADING_2_FONT_FAMILY),
+            _heading_level_xml(1, "chineseCounting", "(%2)", HEADING_2_FONT_FAMILY),
             _heading_level_xml(2, "decimal", "%3.", BODY_FONT_FAMILY),
             _heading_level_xml(3, "decimal", "（%4）", BODY_FONT_FAMILY),
             _heading_level_xml(4, "decimal", "%5）", BODY_FONT_FAMILY),

offline-backend-20260512/backend/docs/FILE_UPLOAD.md

@@ -106,13 +106,14 @@ DELETE /api/threads/{thread_id}/uploads/{filename}
 
 以下格式在显式启用 `uploads.auto_convert_documents: true` 时会自动转换为 Markdown：
 - PDF (`.pdf`)
-- PowerPoint (`.ppt`, `.pptx`)
-- Excel (`.xls`, `.xlsx`)
-- Word (`.doc`, `.docx`)
+- PowerPoint (`.ppt`, `.pptx`)
+- Excel (`.xls`, `.xlsx`)
+- Word (`.doc`, `.docx`)
+- OpenDocument (`.odt`, `.ods`, `.odp`)
 
 转换后的 Markdown 文件会保存在同一目录下，文件名为原文件名 + `.md` 扩展名。
 
-默认情况下，自动转换是关闭的，以避免在网关主机上对不受信任的 Office/PDF 上传执行解析。只有在受信任部署中明确接受此风险时，才应将 `uploads.auto_convert_documents` 设置为 `true`。
+通用部署可以关闭自动转换，以避免在网关主机上对不受信任的 Office/PDF 上传执行解析。本仓库当前 `config.yaml` 和纯 DeerFlow 离线部署包默认设置为 `uploads.auto_convert_documents: true`，用于受信任的内网部署。
 
 ## Agent 集成
 
@@ -247,8 +248,15 @@ backend/.deer-flow/threads/
 
 ### 依赖
 
-- `markitdown>=0.0.1a2` - 文档转换
-- `python-multipart>=0.0.20` - 文件上传处理
+- `markitdown>=0.0.1a2` - 文档转换
+- `python-multipart>=0.0.20` - 文件上传处理
+- 纯 DeerFlow 离线镜像还内置 `python-docx`、`openpyxl`、`python-pptx`、LibreOffice Headless、Pandoc、Poppler 和中文字体。旧版 DOC/XLS/PPT 及 ODT/ODS/ODP 会先经 LibreOffice 规范化，再由 MarkItDown 提取。
+
+离线部署包可在不连接数据库的情况下运行依赖自检：
+
+```bash
+./OFFLINE_INSTALL.sh --office-check
+```
 
 ## 故障排查
 

offline-backend-20260512/backend/packages/harness/deerflow/agents/lead_agent/agent.py

@@ -686,10 +686,11 @@ def _make_lead_agent(config: RunnableConfig, *, app_config: AppConfig):
             available_skills=prompt_available_skills,
             app_config=resolved_app_config,
             memory_injection_enabled=memory_injection_enabled,
-            rag_mode_enabled=rag_mode_enabled,
-            is_notebook_mode=is_notebook_mode,
-            notebook_source_ids=notebook_source_ids,
-        )
+            rag_mode_enabled=rag_mode_enabled,
+            is_notebook_mode=is_notebook_mode,
+            notebook_source_ids=notebook_source_ids,
+            is_scheduled_run=bool(cfg.get("is_scheduled_run", False)),
+        )
 
     return create_agent(
         model=create_chat_model(name=model_name, thinking_enabled=thinking_enabled, reasoning_effort=reasoning_effort, force_disable_thinking=thinking_force_disabled, app_config=resolved_app_config),

offline-backend-20260512/backend/packages/harness/deerflow/agents/lead_agent/prompt.py

@@ -486,7 +486,19 @@ task(description="Oracle Cloud analysis", prompt="...", subagent_type="general-p
 </subagent_system>"""
 
 
-SYSTEM_PROMPT_TEMPLATE = """
+_ORDINARY_QA_MARKDOWN_FORMAT_SECTION = """
+<ordinary_qa_markdown_format>
+当普通问答需要输出带章节结构的较长内容时，必须采用以下 Markdown 规范，以便下载为 Word 后得到正确的自动编号和公文标题格式：
+- 文档主标题使用 `# 主标题`；主标题只写标题文字，不添加章节编号。
+- 一级章节使用 `## 一、标题`，二级章节使用 `### (一) 标题`，三级章节使用 `#### 1. 标题`。二级编号必须使用半角括号 `(一)`。
+- 不要在标题中使用 `2.1`、`2.1.1` 等多级阿拉伯数字编号；章节编号直接按上述三种形式书写。
+- 默认禁止使用以 `-`、`*`、`+` 或 `•` 开头的无序圆点列表；并列内容优先改写为完整自然段，确需逐项表达时使用有序列表。只有用户明确要求无序列表时才使用。
+- 简短问答不必为了套用格式而强行添加标题；本规范只约束实际使用的标题和列表。
+</ordinary_qa_markdown_format>
+"""
+
+
+SYSTEM_PROMPT_TEMPLATE = """
 <role>
 You are {agent_name}, an internal super agent.
 </role>
@@ -612,13 +624,15 @@ You: "Deploying to staging..." [proceed]
 {acp_section}
 </working_directory>
 
-<response_style>
-- Clear and Concise: Avoid over-formatting unless requested
-- Natural Tone: Use paragraphs and prose, not bullet points by default
-- Action-Oriented: Focus on delivering results, not explaining processes
-</response_style>
-
-<citations>
+<response_style>
+- Clear and Concise: Avoid over-formatting unless requested
+- Natural Tone: Use paragraphs and prose, not bullet points by default
+- Action-Oriented: Focus on delivering results, not explaining processes
+</response_style>
+
+{ordinary_qa_format_section}
+
+<citations>
 **CRITICAL: Always include citations when using web search results**
 
 - **When to Use**: MANDATORY after web_search, web_fetch, or any external information source
@@ -1374,7 +1388,7 @@ def build_report_structure_setup_prompt(
 """
 
 
-def apply_prompt_template(
+def apply_prompt_template(
     subagent_enabled: bool = False,
     max_concurrent_subagents: int = 3,
     *,
@@ -1386,10 +1400,11 @@ def apply_prompt_template(
     is_bootstrap: bool = False,
     app_config: AppConfig | None = None,
     memory_injection_enabled: bool = True,
-    rag_mode_enabled: bool = False,
-    is_notebook_mode: bool = False,
-    notebook_source_ids: list[int] | None = None,
-) -> str:
+    rag_mode_enabled: bool = False,
+    is_notebook_mode: bool = False,
+    notebook_source_ids: list[int] | None = None,
+    is_scheduled_run: bool = False,
+) -> str:
     storage_agent_id = agent_id if agent_id is not None else agent_name
 
     # Get memory context
@@ -1467,7 +1482,7 @@ Writing mode is enabled for this run.
 
     # Build the offline-network policy block (empty unless DEER_FLOW_OFFLINE_MODE is set).
     network_policy_section = _build_network_policy_section()
-    rag_mode_section = (
+    rag_mode_section = (
         """
 <rag_citation_mode>
 RAG/reference display mode is enabled.
@@ -1480,10 +1495,15 @@ When a search, knowledge, skill, or MCP tool returns a result list:
 </rag_citation_mode>
 """
         if rag_mode_enabled
-        else ""
-    )
-
-    # Format the prompt with dynamic skills and memory
+        else ""
+    )
+    ordinary_qa_format_section = (
+        _ORDINARY_QA_MARKDOWN_FORMAT_SECTION
+        if not is_bootstrap and not writing_mode and not is_notebook_mode and not is_scheduled_run
+        else ""
+    )
+
+    # Format the prompt with dynamic skills and memory
     prompt = SYSTEM_PROMPT_TEMPLATE.format(
         agent_name=agent_name or storage_agent_id or "CM 助手",
         soul=get_agent_soul(storage_agent_id),
@@ -1496,9 +1516,10 @@ When a search, knowledge, skill, or MCP tool returns a result list:
         network_policy_section=network_policy_section,
         subagent_section=subagent_section,
         subagent_reminder=subagent_reminder,
-        subagent_thinking=subagent_thinking,
-        acp_section=acp_and_mounts_section,
-    )
+        subagent_thinking=subagent_thinking,
+        acp_section=acp_and_mounts_section,
+        ordinary_qa_format_section=ordinary_qa_format_section,
+    )
 
     prompt = prompt + rag_mode_section + _current_time_context()
 

offline-backend-20260512/backend/packages/harness/deerflow/utils/file_conversion.py

@@ -14,25 +14,46 @@ asyncio.to_thread() to avoid blocking the event loop (fixes #1569).
 No FastAPI or HTTP dependencies — pure utility functions.
 """
 
-import asyncio
-import logging
-import re
-from pathlib import Path
+import asyncio
+import logging
+import os
+import re
+import shutil
+import subprocess
+import tempfile
+from pathlib import Path
 
 from deerflow.config.app_config import get_app_config
 
 logger = logging.getLogger(__name__)
 
 # File extensions that should be converted to markdown
-CONVERTIBLE_EXTENSIONS = {
+CONVERTIBLE_EXTENSIONS = {
     ".pdf",
     ".ppt",
     ".pptx",
     ".xls",
     ".xlsx",
     ".doc",
-    ".docx",
-}
+    ".docx",
+    ".odt",
+    ".ods",
+    ".odp",
+}
+
+# MarkItDown reads modern OOXML files directly.  LibreOffice is used as a
+# compatibility bridge for legacy binary Office files and OpenDocument files,
+# which are common in offline/intranet deployments but are not handled
+# consistently by the pure-Python readers.
+_LIBREOFFICE_TARGETS = {
+    ".doc": ".docx",
+    ".ppt": ".pptx",
+    ".xls": ".xlsx",
+    ".odt": ".docx",
+    ".odp": ".pptx",
+    ".ods": ".xlsx",
+}
+_LIBREOFFICE_TIMEOUT_SECONDS = 120
 
 # Files larger than this threshold are converted in a background thread.
 # Small files complete in < 1s synchronously; spawning a thread adds unnecessary
@@ -94,24 +115,92 @@ def _convert_pdf_with_pymupdf4llm(file_path: Path) -> str | None:
         return None
 
 
-def _convert_with_markitdown(file_path: Path) -> str:
+def _convert_with_markitdown(file_path: Path) -> str:
     """Convert any supported file to markdown text using MarkItDown."""
     from markitdown import MarkItDown
 
     md = MarkItDown()
-    return md.convert(str(file_path)).text_content
-
-
-def _do_convert(file_path: Path, pdf_converter: str) -> str:
+    return md.convert(str(file_path)).text_content
+
+
+def _convert_with_libreoffice_then_markitdown(file_path: Path) -> str | None:
+    """Normalize a legacy Office/OpenDocument file, then extract its text.
+
+    Each conversion gets an isolated LibreOffice user profile.  This avoids
+    the profile-lock conflict that otherwise appears when several uploads are
+    converted concurrently in the same sandbox container.
+    """
+    target_suffix = _LIBREOFFICE_TARGETS.get(file_path.suffix.lower())
+    if target_suffix is None:
+        return None
+
+    executable = shutil.which("soffice") or shutil.which("libreoffice")
+    if executable is None:
+        logger.warning(
+            "LibreOffice is unavailable; trying MarkItDown directly for %s",
+            file_path.name,
+        )
+        return None
+
+    with tempfile.TemporaryDirectory(prefix="deerflow-office-") as temporary:
+        temp_root = Path(temporary).resolve()
+        output_dir = temp_root / "output"
+        profile_dir = temp_root / "profile"
+        output_dir.mkdir()
+        profile_dir.mkdir()
+        command = [
+            executable,
+            "--headless",
+            "--nologo",
+            "--nodefault",
+            "--nolockcheck",
+            "--nofirststartwizard",
+            f"-env:UserInstallation={profile_dir.as_uri()}",
+            "--convert-to",
+            target_suffix.lstrip("."),
+            "--outdir",
+            str(output_dir),
+            str(file_path.resolve()),
+        ]
+        try:
+            completed = subprocess.run(
+                command,
+                check=False,
+                capture_output=True,
+                text=True,
+                timeout=_LIBREOFFICE_TIMEOUT_SECONDS,
+                env={**os.environ, "HOME": str(temp_root)},
+            )
+        except (OSError, subprocess.TimeoutExpired):
+            logger.exception("LibreOffice failed to normalize %s", file_path.name)
+            return None
+
+        converted_path = output_dir / f"{file_path.stem}{target_suffix}"
+        if completed.returncode != 0 or not converted_path.is_file():
+            logger.error(
+                "LibreOffice could not normalize %s (exit=%s, stderr=%s)",
+                file_path.name,
+                completed.returncode,
+                completed.stderr.strip()[-1000:],
+            )
+            return None
+        return _convert_with_markitdown(converted_path)
+
+
+def _do_convert(file_path: Path, pdf_converter: str) -> str:
     """Synchronous conversion — called directly or via asyncio.to_thread.
 
     Args:
         file_path: Path to the file.
         pdf_converter: "auto" | "pymupdf4llm" | "markitdown"
     """
-    is_pdf = file_path.suffix.lower() == ".pdf"
-
-    if is_pdf and pdf_converter != "markitdown":
+    is_pdf = file_path.suffix.lower() == ".pdf"
+
+    normalized_office_text = _convert_with_libreoffice_then_markitdown(file_path)
+    if normalized_office_text is not None:
+        return normalized_office_text
+
+    if is_pdf and pdf_converter != "markitdown":
         # Try pymupdf4llm first (auto or explicit)
         pymupdf_text = _convert_pdf_with_pymupdf4llm(file_path)
 

offline-backend-20260512/backend/scripts/office_runtime_selfcheck.py

@@ -0,0 +1,78 @@
+#!/usr/bin/env python3
+"""Offline runtime smoke test for the LocalSandbox Office toolchain."""
+
+from __future__ import annotations
+
+import shutil
+import subprocess
+import tempfile
+from pathlib import Path
+
+
+def require_command(name: str) -> str:
+    path = shutil.which(name)
+    if path is None:
+        raise RuntimeError(f"missing command: {name}")
+    return path
+
+
+def assert_contains(text: str, expected: str, label: str) -> None:
+    normalized = text.replace("\\_", "_")
+    if expected not in normalized:
+        raise RuntimeError(f"{label} conversion lost test text: {text[:300]!r}")
+
+
+def main() -> int:
+    from docx import Document
+    from markitdown import MarkItDown
+    from openpyxl import Workbook
+    from pptx import Presentation
+
+    soffice = require_command("soffice")
+    require_command("pandoc")
+    require_command("pdftotext")
+    require_command("file")
+
+    subprocess.run(
+        [soffice, "--headless", "--version"],
+        check=True,
+        capture_output=True,
+        text=True,
+        timeout=30,
+    )
+
+    with tempfile.TemporaryDirectory(prefix="deerflow-office-selfcheck-") as temporary:
+        root = Path(temporary)
+
+        word_path = root / "word-test.docx"
+        document = Document()
+        document.add_heading("Word 自检", level=1)
+        document.add_paragraph("DEERFLOW_WORD_OK")
+        document.save(word_path)
+
+        excel_path = root / "excel-test.xlsx"
+        workbook = Workbook()
+        worksheet = workbook.active
+        worksheet.title = "自检"
+        worksheet.append(["状态", "DEERFLOW_EXCEL_OK"])
+        workbook.save(excel_path)
+
+        ppt_path = root / "ppt-test.pptx"
+        presentation = Presentation()
+        slide = presentation.slides.add_slide(presentation.slide_layouts[5])
+        slide.shapes.title.text = "DEERFLOW_PPT_OK"
+        presentation.save(ppt_path)
+
+        converter = MarkItDown()
+        assert_contains(converter.convert(word_path).text_content, "DEERFLOW_WORD_OK", "Word")
+        assert_contains(converter.convert(excel_path).text_content, "DEERFLOW_EXCEL_OK", "Excel")
+        assert_contains(converter.convert(ppt_path).text_content, "DEERFLOW_PPT_OK", "PowerPoint")
+
+    print("OFFICE_RUNTIME_SELF_CHECK=PASS")
+    print("支持：DOC/DOCX、XLS/XLSX、PPT/PPTX、ODT/ODS/ODP、PDF")
+    print("组件：MarkItDown、python-docx、openpyxl、python-pptx、LibreOffice、Pandoc、中文字体")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

offline-backend-20260512/backend/tests/test_es_query_routing_prompt.py

@@ -68,3 +68,29 @@ def test_es_query_routing_is_injected_into_lead_and_allowed_custom_agent_prompts
     assert "自定义 ES 路由规则" in lead_prompt
     assert '<skill_routing skill="es_query">' in allowed_custom_prompt
     assert '<skill_routing skill="es_query">' not in denied_custom_prompt
+
+
+def test_formal_markdown_rules_are_scoped_to_ordinary_qa(monkeypatch) -> None:
+    config = _app_config(enabled=False)
+    monkeypatch.setattr(prompt_module, "_get_memory_context", lambda *args, **kwargs: "")
+    monkeypatch.setattr(prompt_module, "get_agent_soul", lambda *args, **kwargs: "")
+    monkeypatch.setattr(prompt_module, "get_skills_prompt_section", lambda *args, **kwargs: "")
+    monkeypatch.setattr(prompt_module, "get_deferred_tools_prompt_section", lambda *args, **kwargs: "")
+    monkeypatch.setattr(prompt_module, "_build_acp_section", lambda *args, **kwargs: "")
+    monkeypatch.setattr(prompt_module, "_build_custom_mounts_section", lambda *args, **kwargs: "")
+    monkeypatch.setattr(prompt_module, "_current_time_context", lambda: "")
+
+    ordinary_prompt = prompt_module.apply_prompt_template(app_config=config)
+    writing_prompt = prompt_module.apply_prompt_template(app_config=config, writing_mode=True)
+    notebook_prompt = prompt_module.apply_prompt_template(app_config=config, is_notebook_mode=True)
+    scheduled_prompt = prompt_module.apply_prompt_template(app_config=config, is_scheduled_run=True)
+
+    assert '<ordinary_qa_markdown_format>' in ordinary_prompt
+    assert "## 一、标题" in ordinary_prompt
+    assert "### (一) 标题" in ordinary_prompt
+    assert "#### 1. 标题" in ordinary_prompt
+    assert "2.1.1" in ordinary_prompt
+    assert "无序圆点列表" in ordinary_prompt
+    assert '<ordinary_qa_markdown_format>' not in writing_prompt
+    assert '<ordinary_qa_markdown_format>' not in notebook_prompt
+    assert '<ordinary_qa_markdown_format>' not in scheduled_prompt

offline-backend-20260512/backend/tests/test_file_conversion_office.py

@@ -0,0 +1,57 @@
+from __future__ import annotations
+
+from pathlib import Path
+from types import SimpleNamespace
+
+from deerflow.utils import file_conversion
+
+
+def test_legacy_word_is_normalized_before_markitdown(monkeypatch, tmp_path: Path) -> None:
+    source = tmp_path / "legacy.doc"
+    source.write_bytes(b"legacy")
+
+    monkeypatch.setattr(file_conversion.shutil, "which", lambda _name: "/usr/bin/soffice")
+    monkeypatch.setattr(
+        file_conversion,
+        "_convert_with_markitdown",
+        lambda path: f"converted:{path.suffix}",
+    )
+
+    def fake_run(command, **_kwargs):
+        output_dir = Path(command[command.index("--outdir") + 1])
+        (output_dir / "legacy.docx").write_bytes(b"normalized")
+        return SimpleNamespace(returncode=0, stderr="")
+
+    monkeypatch.setattr(file_conversion.subprocess, "run", fake_run)
+
+    assert file_conversion._do_convert(source, "auto") == "converted:.docx"
+
+
+def test_modern_office_file_does_not_start_libreoffice(monkeypatch, tmp_path: Path) -> None:
+    source = tmp_path / "modern.pptx"
+    source.write_bytes(b"modern")
+    monkeypatch.setattr(
+        file_conversion,
+        "_convert_with_markitdown",
+        lambda path: f"direct:{path.suffix}",
+    )
+    monkeypatch.setattr(
+        file_conversion.subprocess,
+        "run",
+        lambda *_args, **_kwargs: (_ for _ in ()).throw(AssertionError("unexpected LibreOffice call")),
+    )
+
+    assert file_conversion._do_convert(source, "auto") == "direct:.pptx"
+
+
+def test_legacy_office_falls_back_when_libreoffice_is_missing(monkeypatch, tmp_path: Path) -> None:
+    source = tmp_path / "legacy.xls"
+    source.write_bytes(b"legacy")
+    monkeypatch.setattr(file_conversion.shutil, "which", lambda _name: None)
+    monkeypatch.setattr(
+        file_conversion,
+        "_convert_with_markitdown",
+        lambda path: f"fallback:{path.suffix}",
+    )
+
+    assert file_conversion._do_convert(source, "auto") == "fallback:.xls"

offline-backend-20260512/backend/tests/test_word_export.py

@@ -41,8 +41,10 @@ def test_formal_docx_contains_required_layout_numbering_and_embedded_font(tmp_pa
 ## 一、一级标题
 这是正文，用于验证首行缩进、两端对齐和固定行距。
 
-### （一）二级标题
-#### 1. 三级标题
+### 2.1 二级标题
+#### 2.1.1 三级标题
+
+---
 
 表1 测试表格
 | 项目 | 数值 |
@@ -77,8 +79,11 @@ def test_formal_docx_contains_required_layout_numbering_and_embedded_font(tmp_pa
         assert '<w:pStyle w:val="DocumentTitle"/>' in document
         assert "方正小标宋简体" in document
         assert "一、一级标题" not in document
-        assert "（一）二级标题" not in document
-        assert "1. 三级标题" not in document
+        assert "2.1 二级标题" not in document
+        assert "2.1.1 三级标题" not in document
+        assert "────────" not in document
+        assert "二级标题" in document
+        assert "三级标题" in document
 
         assert 'w:line="579" w:lineRule="exact"' in styles
         assert '<w:ind w:left="0" w:hanging="0" w:firstLineChars="200"/>' in styles
@@ -86,11 +91,13 @@ def test_formal_docx_contains_required_layout_numbering_and_embedded_font(tmp_pa
         assert 'w:val="44"' in styles
         assert "仿宋" in styles and 'w:val="32"' in styles
         assert "宋体" in styles and 'w:val="28"' in styles
+        assert styles.count('w:firstLineChars="200"') >= 6
 
         assert '<w:numFmt w:val="chineseCounting"/>' in numbering
         assert '<w:lvlText w:val="%1、"/>' in numbering
-        assert '<w:lvlText w:val="（%2）"/>' in numbering
+        assert '<w:lvlText w:val="(%2)"/>' in numbering
         assert '<w:lvlText w:val="%3."/>' in numbering
+        assert numbering.count('w:firstLineChars="200"') == 3
         assert '<w:evenAndOddHeaders/>' in settings
         assert '<w:updateFields w:val="true"/>' in settings
 
@@ -114,3 +121,20 @@ def test_export_rejects_wrong_font(tmp_path: Path) -> None:
 
     with pytest.raises(word_export.WordExportFontError):
         word_export.build_markdown_docx("正文", font_path=wrong_font)
+
+
+@pytest.mark.parametrize(
+    ("source", "expected"),
+    [
+        ("一、一级标题", "一级标题"),
+        ("(一) 二级标题", "二级标题"),
+        ("（一）二级标题", "二级标题"),
+        ("1. 三级标题", "三级标题"),
+        ("2.1 二级标题", "二级标题"),
+        ("2.1.1 三级标题", "三级标题"),
+        ("2.1.1. 三级标题", "三级标题"),
+        ("正文标题", "正文标题"),
+    ],
+)
+def test_manual_heading_number_is_removed_as_one_complete_prefix(source: str, expected: str) -> None:
+    assert word_export._strip_manual_heading_number(source) == expected