chore: commit local changes

offline-backend-20260512/backend/packages/harness/deerflow/persistence/llm_metrics/base.py

@@ -34,6 +34,9 @@ class LlmMetricsQuery:
     """Filter set for the admin list / export endpoints."""
 
     user_id: str | None = None
+    # Multiple user ids — set when a fuzzy username search resolves to several
+    # accounts. Takes precedence over ``user_id`` when non-empty.
+    user_ids: list[str] | None = None
     model_name: str | None = None
     status: str | None = None
     since: datetime | None = None

offline-backend-20260512/backend/packages/harness/deerflow/persistence/llm_metrics/sql.py

@@ -58,7 +58,9 @@ class SqlLlmMetricsStore(LlmMetricsStore):
             logger.warning("Failed to persist llm_call_metrics row %s", metric.id, exc_info=True)
 
     def _apply_filters(self, stmt, query: LlmMetricsQuery):
-        if query.user_id:
+        if query.user_ids:
+            stmt = stmt.where(LlmCallMetricRow.user_id.in_(query.user_ids))
+        elif query.user_id:
             stmt = stmt.where(LlmCallMetricRow.user_id == query.user_id)
         if query.model_name:
             stmt = stmt.where(LlmCallMetricRow.model_name == query.model_name)
@@ -92,6 +94,7 @@ class SqlLlmMetricsStore(LlmMetricsStore):
         while True:
             page_query = LlmMetricsQuery(
                 user_id=query.user_id,
+                user_ids=query.user_ids,
                 model_name=query.model_name,
                 status=query.status,
                 since=query.since,

offline-backend-20260512/backend/packages/harness/deerflow/persistence/run/sql.py

@@ -205,6 +205,21 @@ class RunRepository(RunStore):
             await session.execute(update(RunRow).where(RunRow.run_id == run_id).values(**values))
             await session.commit()
 
+    async def count_by_user(self) -> dict[str, int]:
+        """Return ``{user_id: run_count}`` via a single GROUP BY query.
+
+        Each run is one user question, so this is the source for the admin
+        user center's 问答次数 column. Runs without a ``user_id`` are skipped.
+        """
+        stmt = (
+            select(RunRow.user_id, func.count().label("cnt"))
+            .where(RunRow.user_id.is_not(None))
+            .group_by(RunRow.user_id)
+        )
+        async with self._sf() as session:
+            rows = (await session.execute(stmt)).all()
+        return {r.user_id: r.cnt for r in rows}
+
     async def aggregate_tokens_by_thread(self, thread_id: str) -> dict[str, Any]:
         """Aggregate token usage via a single SQL GROUP BY query."""
         _completed = RunRow.status.in_(("success", "error"))

offline-backend-20260512/backend/packages/harness/deerflow/runtime/runs/store/base.py

@@ -93,3 +93,12 @@ class RunStore(abc.ABC):
         by_caller ({lead_agent, subagent, middleware}).
         """
         pass
+
+    async def count_by_user(self) -> dict[str, int]:
+        """Return ``{user_id: run_count}`` across all runs.
+
+        One run corresponds to one user submission, so this backs the admin
+        user center's 问答次数 (Q&A count) column — each question counts once.
+        Non-overriding stores return an empty mapping (treated as 0 per user).
+        """
+        return {}

offline-backend-20260512/backend/packages/harness/deerflow/runtime/scheduler/html_page.py

@@ -38,6 +38,11 @@ _DANGEROUS_URL_RE = re.compile(r"^\s*(javascript|vbscript|data:text/html)", re.I
 # Attributes carrying a URL we must scheme-check.
 _URL_ATTRS = frozenset({"href", "src", "action", "formaction", "xlink:href", "background", "poster"})
 
+_LOCAL_DOC_LINK_RE = re.compile(
+    r"^(?![a-z][a-z0-9+.-]*:)(?!//)(?P<path>[^#?]*\.(?:html?|md|markdown))(?P<query>\?[^#]*)?(?P<fragment>#.+)?$",
+    re.IGNORECASE,
+)
+
 
 def parse_frontmatter(text: str) -> tuple[dict[str, Any], str]:
     """Split a ``---`` YAML frontmatter block from the body of a SKILL.md.
@@ -225,6 +230,38 @@ def sanitize_html(html: str) -> tuple[str, list[str]]:
     return str(soup), removed
 
 
+def constrain_same_page_anchor_links(html: str) -> tuple[str, list[str]]:
+    """Keep generated HTML document jumps inside the current HTML file."""
+    if not html or not isinstance(html, str):
+        return "", []
+    try:
+        from bs4 import BeautifulSoup
+    except Exception:  # pragma: no cover - bs4 is a declared dependency
+        logger.warning("BeautifulSoup unavailable; returning HTML links unconstrained")
+        return html, []
+
+    changed: list[str] = []
+    soup = BeautifulSoup(html, "html.parser")
+    for tag in soup.find_all("a"):
+        href = tag.get("href")
+        if not isinstance(href, str):
+            continue
+        value = href.strip()
+        if not value or value.startswith("#"):
+            continue
+        match = _LOCAL_DOC_LINK_RE.match(value)
+        if not match:
+            continue
+        fragment = match.group("fragment")
+        if fragment:
+            tag["href"] = fragment
+            changed.append(f"{value}->{fragment}")
+        else:
+            del tag.attrs["href"]
+            changed.append(f"{value}->removed")
+    return str(soup), changed
+
+
 def validate_html(html: str) -> list[str]:
     """Return a list of structural problems with ``html`` (empty == valid).
 
@@ -299,6 +336,7 @@ _GENERATION_TEMPLATE = """你是一个内网页面生成助手。请根据输入
 6. 不要包含 <script>、外链脚本、iframe、表单提交、on* 事件属性。
 7. 参考页面只用于风格和布局，页面内容必须来自“本次数据”，不要复用参考页旧数据。
 8. 页面使用中文，排版适合在浏览器中直接预览。
+9. HTML 内的目录/锚点跳转必须只跳转到当前这个 HTML 文件内部：只使用 href="#section-id"，不要使用 href="index.html#section-id"、".md#..."、其它文件路径或任何跨文件锚点；目标区块必须有匹配的 id。
 """
 
 _REPAIR_TEMPLATE = """你是一个 HTML 修复助手。下面的 HTML 存在问题，请修复后重新输出完整页面。
@@ -317,6 +355,7 @@ _REPAIR_TEMPLATE = """你是一个 HTML 修复助手。下面的 HTML 存在问
 2. fixed_html 必须是完整 HTML 文档，包含 <!doctype html>、html、head、body。
 3. 移除所有 <script>、iframe、外链脚本、on* 事件属性、表单提交。
 4. 保留原有内容与数据，只修复结构与安全问题。
+5. HTML 内的目录/锚点跳转必须只跳转到当前这个 HTML 文件内部：只使用 href="#section-id"，不要链接到 .html/.md 文件作为锚点。
 """
 
 _STYLE_EXTRACT_TEMPLATE = """你是一个页面风格分析助手。请分析下面的 HTML 页面，提取可复用的风格与布局特征。

offline-backend-20260512/backend/packages/harness/deerflow/runtime/scheduler/service.py

@@ -53,8 +53,18 @@ _MARKDOWN_MIN_CHARS = 100
 _MARKDOWN_INSTRUCTION = (
     "\n\n【输出要求】请将本次任务的最终结果生成为一个 Markdown(.md) 格式的文件，"
     "保存到 /mnt/user-data/outputs 目录，并使用 present_files 展示该文件。"
+    "Markdown 内的目录/锚点跳转必须只跳转到当前这个 Markdown 文件内部："
+    "只使用 [标题](#section-id)，不要使用 [标题](xxx.md#section-id)、"
+    "[标题](xxx.html#section-id) 或任何跨文件锚点。"
 )
 
+_LOCAL_DOC_MARKDOWN_LINK_RE = re.compile(
+    r"^(?![a-z][a-z0-9+.-]*:)(?!//)(?P<path>[^#?]*\.(?:html?|md|markdown))(?P<query>\?[^#]*)?(?P<fragment>#.+)?$",
+    re.IGNORECASE,
+)
+_MARKDOWN_INLINE_LINK_RE = re.compile(r"(?<!!)\[([^\]]+)\]\(([^)\s]+)(?:\s+\"[^\"]*\")?\)")
+_MARKDOWN_REFERENCE_LINK_RE = re.compile(r"^(\s*\[[^\]]+\]:\s*)(\S+)(.*)$", re.MULTILINE)
+
 
 class MarkdownRequirementError(Exception):
     """Raised when a ``require_markdown`` task did not produce a valid .md file.
@@ -262,6 +272,43 @@ def _has_valid_markdown_file(result_files: list[dict[str, Any]] | None) -> bool:
     return False
 
 
+def _normalize_same_page_markdown_target(target: str) -> str | None:
+    value = (target or "").strip()
+    if not value or value.startswith("#"):
+        return value
+    match = _LOCAL_DOC_MARKDOWN_LINK_RE.match(value)
+    if not match:
+        return value
+    fragment = match.group("fragment")
+    return fragment or None
+
+
+def _constrain_markdown_same_page_anchor_links(content: str) -> str:
+    """Keep generated Markdown document jumps inside the current Markdown file."""
+    if not content or not isinstance(content, str):
+        return content or ""
+
+    def replace_inline(match: re.Match[str]) -> str:
+        label = match.group(1)
+        target = match.group(2)
+        normalized = _normalize_same_page_markdown_target(target)
+        if normalized is None:
+            return label
+        return f"[{label}]({normalized})"
+
+    def replace_reference(match: re.Match[str]) -> str:
+        prefix = match.group(1)
+        target = match.group(2)
+        suffix = match.group(3)
+        normalized = _normalize_same_page_markdown_target(target)
+        if normalized is None:
+            return f"{prefix}#"
+        return f"{prefix}{normalized}{suffix}"
+
+    content = _MARKDOWN_INLINE_LINK_RE.sub(replace_inline, content)
+    return _MARKDOWN_REFERENCE_LINK_RE.sub(replace_reference, content)
+
+
 def _build_result(
     task: dict[str, Any],
     *,
@@ -1169,6 +1216,7 @@ class ScheduledTaskService:
 
         # 4) Sanitize + validate, with one repair pass on failure.
         html, _removed = html_page_utils.sanitize_html(html)
+        html, _link_changes = html_page_utils.constrain_same_page_anchor_links(html)
         errors = html_page_utils.validate_html(html)
         if errors:
             logger.info("Generated HTML failed validation (%s); attempting repair", errors)
@@ -1181,6 +1229,7 @@ class ScheduledTaskService:
                 fixed = repaired.get("fixed_html")
                 if isinstance(fixed, str) and "<" in fixed:
                     fixed_clean, _ = html_page_utils.sanitize_html(fixed)
+                    fixed_clean, _ = html_page_utils.constrain_same_page_anchor_links(fixed_clean)
                     if not html_page_utils.validate_html(fixed_clean):
                         html = fixed_clean
             except Exception:
@@ -1540,6 +1589,14 @@ class ScheduledTaskService:
                 index += 1
 
         shutil.copy2(path, candidate)
+        if candidate.suffix.lower() in {".md", ".markdown"}:
+            try:
+                original = candidate.read_text(encoding="utf-8", errors="ignore")
+                constrained = _constrain_markdown_same_page_anchor_links(original)
+                if constrained != original:
+                    candidate.write_text(constrained, encoding="utf-8", newline="")
+            except Exception:
+                logger.debug("Failed to constrain scheduled markdown links in %s", candidate, exc_info=True)
         return candidate
 
     def _build_file_result(self, *, virtual_path: str, source_path: Path, archive_path: Path) -> dict[str, Any]: