feat: 提问导航悬浮索引 + 知识库召回会话级开关 + 智能体技能白名单收紧

- 新增 QuestionNavigator：聊天/智能体页右侧悬浮提问索引，点击滚动定位到对应消息（message-list 为 human 消息生成锚点 id）
- 知识库召回开关从单会话头部按钮迁移到侧边栏，改为会话级内存开关（刷新重置，不再写 localStorage），开启时所有对话默认携带 knowledge_rag_enabled
- 智能体可配置技能收紧为「内置 + 自己的 + 已发布到广场的」（新增 core/skills/utils.ts 的 isAgentSkillCandidate，NewAgentPage 与 agent-detail-panel 共用）
- 智能体详情面板在 artifact 预览展开/新文件生成时自动收起让位
- 启用 html-multipage 技能并更新后端 CLAUDE.md 文档

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

offline-backend-20260512/backend/CLAUDE.md

@@ -281,7 +281,7 @@ Two on-demand/automatic refresh paths layer on top:
 
 **Per-user UI preferences** (`deerflow/persistence/user_preferences/`, wired as `app.state.user_preferences_store`; migration `20260603_03`, also auto-created by `create_all`): one JSON-blob row per user so new toggles need no schema change. Router `/api/user-preferences` (`GET`/`PUT`, account-scoped). Current key: `hide_admin_prompt_prefix` — when true the chat input hides the admin-configured prompt-prefix switch and never applies the prefix (the user uses only their own prompts; enforced in `ChatPage` for both `shouldApplyPrefix` and `promptPrefixProp`). Toggled from the "管理常用提示词" dialog. Tests: `tests/test_user_preferences.py`.
 
-**HTML skills (tools the render step uses).** Two offline skills under `skills/public/` shape the generated HTML and are toggled in the skills settings page (enabled by default): `html-page-builder` (a self-contained inline-CSS design system — colors/cards/tables/KPI/responsive) and `html-charts` (inline-SVG charts that render in the no-`allow-scripts` preview iframe; optional local ECharts documented in its `assets/README.md`). A skill opts into HTML injection via `metadata.kind: html_page` in its SKILL.md frontmatter; `_collect_html_skill_instructions()` reads the bodies of **enabled** such skills and `build_generation_prompt(skill_instructions=…)` injects them, so the (raw-model) render step honors the user's skill selection. The data-gathering step and the render/repair/style passes all use the task's selected `model_name` (no silent fallback to the default model — and there is **no hardcoded GLM-5**; an unset selection falls back to config `models[0]`). To stop auxiliary LLM calls from leaking onto the default model, **`TitleMiddleware` is skipped when `is_scheduled_run` is set** (`_build_middlewares` in `lead_agent/agent.py`) — scheduled/HTML runs use throwaway threads whose titles are never shown, and `config.title.model_name` is usually unset (→ would call `models[0]`, not the chosen model). The completeness gate (`validate_html`) also requires a non-empty `<title>` and forbids external stylesheets, on top of the doctype/html/head/body + non-empty-body + no-script checks.
+**HTML skills (tools the render step uses).** Two offline skills under `skills/public/` shape the generated HTML and are toggled in the skills settings page (enabled by default): `html-page-builder` (a self-contained inline-CSS design system — colors/cards/tables/KPI/responsive), `html-charts` (inline-SVG charts that render in the no-`allow-scripts` preview iframe; optional local ECharts documented in its `assets/README.md`), and `html-multipage` (multi-"page" navigation **inside the single generated HTML file**: each page is a `<section class="page" id="page-…">` toggled by anchor links + CSS `:target` — no JS, so it survives `sanitize_html`, and `constrain_same_page_anchor_links` keeps `#page-…` hrefs untouched; the frontend's `html-preview-frame.tsx` renders via a `blob:` URL precisely so in-page anchors work; cross-file `*.html` links remain forbidden/stripped). A skill opts into HTML injection via `metadata.kind: html_page` in its SKILL.md frontmatter; `_collect_html_skill_instructions()` reads the bodies of **enabled** such skills and `build_generation_prompt(skill_instructions=…)` injects them, so the (raw-model) render step honors the user's skill selection. The data-gathering step and the render/repair/style passes all use the task's selected `model_name` (no silent fallback to the default model — and there is **no hardcoded GLM-5**; an unset selection falls back to config `models[0]`). To stop auxiliary LLM calls from leaking onto the default model, **`TitleMiddleware` is skipped when `is_scheduled_run` is set** (`_build_middlewares` in `lead_agent/agent.py`) — scheduled/HTML runs use throwaway threads whose titles are never shown, and `config.title.model_name` is usually unset (→ would call `models[0]`, not the chosen model). The completeness gate (`validate_html`) also requires a non-empty `<title>` and forbids external stylesheets, on top of the doctype/html/head/body + non-empty-body + no-script checks.
 
 ### Knowledge Base (`packages/harness/deerflow/knowledge/`)
 

offline-backend-20260512/backend/extensions_config.json

@@ -8,6 +8,7 @@
     "workspace-search": { "enabled": true },
     "policy-knowledge": { "enabled": false },
     "html-page-builder": { "enabled": true },
-    "html-charts": { "enabled": true }
+    "html-charts": { "enabled": true },
+    "html-multipage": { "enabled": true }
   }
 }