feat: 前端页面优化与入库任务重试/删除

- 概览与检索合并为单一概览页
- 类目/文档管理/入库整合为树形目录知识库页（Library），支持文档搜索
- 入库进度改为右侧 Drawer，支持任务详情、重试、删除
- 后端新增任务重试与删除接口

.env.example

@@ -7,12 +7,14 @@ APP_PORT=8000
 LOG_LEVEL=info
 
 # --- 嵌入模型 ---
-# openai | local
+# openai | local。默认走本地 Ollama 的 BGE-M3（多语言榜首、dense+sparse 一体，零 API 成本）
-EMBEDDING_PROVIDER=openai
+EMBEDDING_PROVIDER=local
+# 以下为 EMBEDDING_PROVIDER=openai 时启用的 OpenAI 兼容配置（本地模式可留空）
 OPENAI_API_KEY=sk-xxx
 OPENAI_BASE_URL=https://api.openai.com/v1
 EMBEDDING_MODEL=text-embedding-3-small
-EMBEDDING_DIMENSION=1536
+# 维度需与所选嵌入模型一致：BGE-M3 为 1024；切换 OpenAI 时改回 1536
+EMBEDDING_DIMENSION=1024
 
 # --- Qdrant ---
 QDRANT_PORT=6333
@@ -23,11 +25,16 @@ REDIS_PORT=6379
 
 # --- Ollama 本地模型（文档三级总结）---
 OLLAMA_PORT=11434
-OLLAMA_MODEL=qwen2.5:1.5b
+OLLAMA_MODEL=qwen3:1.7b
-# 备选模型: qwen2.5:3b（更好的总结质量，需更多内存）
+# 备选模型: qwen3:4b（更好的总结质量，需更多内存）
 # EMBEDDING_PROVIDER=local 时使用的 Ollama 嵌入模型
 OLLAMA_EMBEDDING_MODEL=bge-m3
 
+# --- 重排模型（cross-encoder，经 Ollama /api/rerank 对 chunk 候选精排）---
+# 启用后显著提升最终 Top-K 相关性；需 Ollama 已拉取 RERANKER_MODEL（见 docker-compose）
+RERANKER_ENABLED=true
+RERANKER_MODEL=qwen3-reranker:0.6b
+
 # --- NAS 持久化 ---
 NAS_DATA_DIR=./data
 

.gitignore

@@ -9,6 +9,9 @@ build/
 # Frontend
 node_modules/
 
+# 前端构建产物（本地残留的 Vue 构建 / 符号链接，勿提交；线上由 frontend/dist 挂载）
+app/static/admin
+
 # Env
 .env
 .env.local
@@ -36,3 +39,7 @@ logs/
 .pytest_cache/
 .coverage
 htmlcov/
+
+# Project memory & Trae IDE spec (切勿提交)
+.workbuddy/
+.trae/

CLAUDE.md

@@ -12,7 +12,7 @@ QMDSearch 是面向 AI Agent 的分层信息检索服务，支持多层级知识
 - **缓存**: Redis (Docker)
 - **关系数据库**: PostgreSQL (Docker, 可选)
 - **嵌入模型**: OpenAI / 本地模型 (通过配置切换)
-- **本地推理模型**: Ollama (Docker) — 用于文档三级总结，默认 qwen2.5:1.5b
+- **本地推理模型**: Ollama (Docker) — 用于文档三级总结，默认 qwen3:1.7b
 - **部署**: Docker Compose on NAS
 
 ## 项目结构
@@ -90,11 +90,15 @@ QMDSearch/
 | POST | `/api/v1/search` | 分层检索 | 免登录 |
 | POST | `/api/v1/documents` | 文档入库（202 异步入库，返回 task_id） | Bearer |
 | POST | `/api/v1/documents/upload` | multipart 文件上传入库（202 异步，支持 .txt/.md/.html/.htm/.pdf/.docx） | Bearer |
+| POST | `/api/v1/documents/upload-batch` | 批量文件上传入库（多文件，202 异步） | Bearer |
+| GET | `/api/v1/documents/tasks` | 入库任务列表（近期，按时间降序） | Bearer |
 | GET | `/api/v1/documents/tasks/{task_id}` | 入库任务状态查询（done 附 result，failed 附 error） | 免登录 |
 | GET | `/api/v1/knowledge/categories` | 知识分类类目集 | 免登录 |
 | GET | `/api/v1/knowledge/stats` | 统计（四层点数 + 类目分布 + uncategorized 数） | 免登录 |
 | GET | `/api/v1/documents` | 文档列表（limit/offset 分页） | 免登录 |
 | GET | `/api/v1/documents/{doc_id}` | 文档详情 | 免登录 |
+| GET | `/api/v1/documents/{doc_id}/file` | 下载关联的原始文件（免登录） | 免登录 |
+| POST | `/api/v1/documents/{doc_id}/reingest` | 重新摘要入库（读原文件→删旧→重跑流水线） | Bearer |
 | DELETE | `/api/v1/documents/{doc_id}` | 删除文档（幂等） | Bearer |
 | POST | `/api/v1/auth/login` | 用户名密码登录，签发 session token（TTL 12h） | 免登录 |
 | POST | `/api/v1/auth/logout` | 退出登录（删除当前 session） | Bearer |
@@ -102,26 +106,28 @@ QMDSearch/
 | GET | `/api/v1/auth/me` | 当前登录用户信息（脱敏） | Bearer |
 | GET | `/api/v1/auth/users` | 用户列表（脱敏） | Bearer + admin |
 | POST | `/api/v1/auth/users` | 创建用户（重名/非法用户名/弱密码 1001） | Bearer + admin |
+| PATCH | `/api/v1/auth/users/{username}` | 更新用户角色/启用状态（仅 admin） | Bearer + admin |
 | POST | `/api/v1/auth/users/{username}/password` | 重置指定用户密码（成功后清除其全部 session） | Bearer + admin |
 | DELETE | `/api/v1/auth/users/{username}` | 删除用户（清除其 session；禁删自己/最后一个 admin） | Bearer + admin |
 | GET | `/admin` | 管理页面 | 页面登录门禁 |
 
-「Bearer」指请求头 `Authorization: Bearer <token>`，token 经 `/api/v1/auth/login` 获取；变更类文档端点（POST /documents、POST /documents/upload、DELETE /documents/{id}）需 Bearer token（admin/user 角色均可），查询类端点免登录。
+「Bearer」指请求头 `Authorization: Bearer <token>`，token 经 `/api/v1/auth/login` 获取；变更类文档端点（POST /documents、POST /documents/upload、POST /documents/upload-batch、POST /documents/{id}/reingest、DELETE /documents/{id}）与 GET /documents/tasks 任务列表需 Bearer token（admin/user 角色均可），其余查询类端点免登录。
 
 入库任务状态持久化在 Redis（key: `ingest_task:{task_id}`）：进行中与 done 保留 24h，failed 保留 7 天；Redis 不可用时降级为纯内存。
 
 ## 管理页面
 
-浏览器访问 `/admin`，页面带登录门禁（未登录/token 失效自动回登录卡片；must_change_password 用户先强制改密后方可进入）。单页面含七个区块：概览、文档管理（列表/详情/删除）、文档入库（文本 + 文件上传）、检索测试台、类目列表、API 指南（端点清单 + 在线测试台）、用户管理（仅 admin 角色挂载，含创建/重置密码/删除）。
+浏览器访问 `/admin`，页面带登录门禁（未登录/token 失效自动回登录卡片；must_change_password 用户先强制改密后方可进入）。单页面含九个区块（admin 视角；user 角色无用户管理区块）：概览、个人中心（user 角色默认进入且可见，含账号信息/改密/退出）、文档管理（列表/详情/删除）、文档入库（文本 + 文件上传，文件上传支持批量多选）、入库进度（展示近期任务与状态，2s 轮询自动刷新）、检索测试台、类目列表、API 指南（端点清单 + 在线测试台）、用户管理（仅 admin 角色挂载，含搜索筛选/行内编辑角色与启用状态/弹窗式重置密码与删除）。
 
 ## 认证与用户
 
 会话制认证：登录签发 session token（Redis 持久化，TTL 12h；Redis 不可用时降级为进程内存，重启失效），请求经 `Authorization: Bearer <token>` 携带，鉴权依赖见 `app/api/deps.py`。
 
 - **角色与权限边界**: `admin` 拥有全部权限（含 /auth/users* 用户管理）；`user` 可登录并调用变更类文档端点（入库/上传/删除），访问用户管理端点返回 1006
+- **启用状态与 PATCH 端点**: 用户记录含 `enabled` 字段（默认 true）；admin 可经 `PATCH /auth/users/{username}` 更新角色或启用状态（请求体至少一项字段，空 body 1001；role 非法 1001；用户不存在 1004）。禁用用户时清除其全部 session（旧 token 立即失效 1005），被禁用用户登录拒绝 1005（"账号已禁用"）；最后一个 admin 禁止降级 role 或被禁用（1001）。角色变更不清 session，同一 token 即时反映新角色
 - **初始 admin 引导**: 空库启动时 `bootstrap_admin` 自动创建 admin 账号，随机明文密码仅在启动日志中打印一次（must_change_password=true），首次登录后须先经 POST /auth/password 改密，改密前访问其他端点返回 1006
 - **免登录端点**: 查询类端点（POST /search、GET /documents*、GET /knowledge/*、GET /health）不需要 token
-- **相关错误码**: 1005 未认证或凭证无效，1006 权限不足/首次登录须先改密
+- **相关错误码**: 1005 未认证或凭证无效/账号已禁用，1006 权限不足/首次登录须先改密
 
 ## 编码规范
 
@@ -165,8 +171,8 @@ QMDSearch/
 
 ### 本地模型 (Ollama)
 
-- 服务: Ollama 容器，默认模型 `qwen2.5:1.5b`（约 1GB，适合 NAS 低资源环境）
+- 服务: Ollama 容器，默认模型 `qwen3:1.7b`（约 1.3GB，适合 NAS 低资源环境）
-- 备选模型: `qwen2.5:3b`（更好的总结质量，约 2GB）
+- 备选模型: `qwen3:4b`（更好的总结质量，约 2.5GB）
 - 模型通过环境变量 `OLLAMA_MODEL` 配置
 - 首次启动时自动拉取模型，需 NAS 可访问外网
 

Dockerfile

@@ -1,17 +1,10 @@
-# ---------- Stage 1: 前端构建 ----------
+# ---------- Python 后端运行时镜像 ----------
-FROM node:20-slim AS frontend-builder
+# 部署模型说明：
-
+#   前端采用「挂载式」部署——docker-compose 将宿主机的 ./frontend/dist
-WORKDIR /frontend
+#   挂到容器的 /app/app/static/admin，运行时直接覆盖镜像内同名目录。
-
+#   因此镜像【不再构建前端】，只负责 Python 运行时 + 依赖；前端由本地
-# 依赖层（利用 Docker 层缓存）
+#   `npm run build` 后 rsync 到 NAS，或挂载既有 dist。
-COPY frontend/package.json frontend/package-lock.json ./
+#   仅当 pyproject.toml / uv.lock 变化时才需 `docker build -t qmdsearch-app .` 一次。
-RUN npm ci
-
-# 源码 + 构建
-COPY frontend/ ./
-RUN npm run build
-
-# ---------- Stage 2: Python 后端 ----------
 FROM python:3.12-slim AS base
 
 WORKDIR /app
@@ -19,7 +12,7 @@ WORKDIR /app
 # 安装 uv（放到 PATH 中，供 uv sync 与 CMD 的 uv run 使用）
 COPY --from=ghcr.io/astral-sh/uv:latest /uv /usr/local/bin/uv
 
-# 依赖层
+# 依赖层（利用 Docker 层缓存）
 COPY pyproject.toml uv.lock ./
 RUN uv sync --frozen --no-dev
 
@@ -27,9 +20,6 @@ RUN uv sync --frozen --no-dev
 COPY app/ app/
 COPY scripts/ scripts/
 
-# 前端构建产物 → 后端静态目录（SPA 由 /admin 路由服务）
-COPY --from=frontend-builder /frontend/dist /app/app/static/admin
-
 EXPOSE 8000
 
 HEALTHCHECK --interval=30s --timeout=5s --retries=3 \

README.md

@@ -22,7 +22,7 @@
 | 向量数据库 | Qdrant（Docker） |
 | 缓存 | Redis（Docker） |
 | 嵌入模型 | OpenAI / Ollama 本地模型（可切换） |
-| 本地推理 | Ollama + qwen2.5:1.5b（文档总结） |
+| 本地推理 | Ollama + qwen3:1.7b（文档总结） |
 | OCR | rapidocr-onnxruntime + pypdfium2（扫描件 PDF） |
 | 部署 | Docker Compose on NAS |
 
@@ -123,7 +123,7 @@ docker compose up -d
 ### 4. 拉取 Ollama 模型（首次启动后）
 
 ```bash
-docker exec qmdsearch-ollama ollama pull qwen2.5:1.5b
+docker exec qmdsearch-ollama ollama pull qwen3:1.7b
 ```
 
 ### 5. 验证服务
@@ -232,7 +232,7 @@ curl -X POST http://localhost:8000/api/v1/documents/upload \
 | `EMBEDDING_PROVIDER` | 嵌入模型提供商（openai / local） | `openai` |
 | `OPENAI_API_KEY` | OpenAI API Key | - |
 | `EMBEDDING_MODEL` | 嵌入模型名称 | `text-embedding-3-small` |
-| `OLLAMA_MODEL` | Ollama 本地模型 | `qwen2.5:1.5b` |
+| `OLLAMA_MODEL` | Ollama 本地模型 | `qwen3:1.7b` |
 | `RETRIEVAL_TOP_K` | 检索召回数 | `20` |
 | `RETRIEVAL_FINAL_K` | 最终返回数 | `5` |
 | `INGEST_MAX_CONCURRENCY` | 入库并发上限 | `2` |

app/api/deps.py

@@ -8,6 +8,7 @@ UserStore/SessionStore 为模块级懒加载单例：Redis 客户端创建失败
 import structlog
 from fastapi import Depends, Header
 from redis import asyncio as redis_async
+from redis import Redis as RedisSync
 
 from app.api.response import ApiError
 from app.config import settings
@@ -22,11 +23,20 @@ _session_store: SessionStore | None = None
 
 
 def _create_redis_client() -> redis_async.Redis | None:
-    """创建 redis.asyncio 客户端（decode_responses=True）；失败返回 None 走内存降级"""
+    """创建 redis.asyncio 客户端（decode_responses=True）；连接失败返回 None 走内存降级
+
+    注意：redis.asyncio.from_url() 仅构造客户端对象，不会实际连接 Redis，
+    因此需要用同步客户端 ping 一次确认连接可用，否则后续操作才会报错，
+    导致 UserStore/SessionStore 无法降级为内存模式。
+    """
     try:
+        # 同步 ping 确认 Redis 可达（from_url 不会实际连接）
+        sync_client = RedisSync.from_url(settings.redis_url, decode_responses=True)
+        sync_client.ping()
+        sync_client.close()
         return redis_async.from_url(settings.redis_url, decode_responses=True)
     except Exception:
-        logger.warning("Redis 客户端创建失败，认证存储降级为内存模式", exc_info=True)
+        logger.warning("Redis 连接失败，认证存储降级为内存模式", exc_info=True)
         return None
 
 
@@ -80,9 +90,12 @@ async def _resolve_user(authorization: str | None) -> tuple[UserRecord, str]:
 async def get_current_user(authorization: str | None = Header(None)) -> UserRecord:
     """鉴权依赖：校验 Bearer token 并返回当前用户记录
 
-    must_change_password 用户被拦截（1006），须先经 POST /auth/password 改密。
+    禁用用户被拦截（1005）；must_change_password 用户被拦截（1006），须先经 POST /auth/password 改密。
+    注：logout/password 端点经 _resolve_user 自解析，不在此拦截范围内（改密是已登录用户唯一可用接口）。
     """
     user, _ = await _resolve_user(authorization)
+    if not user.enabled:
+        raise ApiError(1005, "账号已禁用")
     if user.must_change_password:
         raise ApiError(1006, "首次登录须先修改密码")
     return user

app/api/v1/auth.py

@@ -10,12 +10,12 @@ from typing import Any, Literal
 
 import structlog
 from fastapi import APIRouter, Depends, Header
-from pydantic import BaseModel
+from pydantic import BaseModel, model_validator
 
 from app.api import deps
 from app.api.response import ApiError, ok
 from app.core.sessions import SessionStoreError
-from app.core.users import UserExistsError, UserRecord, UserStoreError
+from app.core.users import UserExistsError, UserNotFoundError, UserRecord, UserStoreError
 
 logger = structlog.get_logger()
 
@@ -50,12 +50,26 @@ class PasswordResetRequest(BaseModel):
     new_password: str
 
 
+class UserUpdateRequest(BaseModel):
+    """管理员更新用户角色/启用状态（至少一项）"""
+
+    role: Literal["admin", "user"] | None = None
+    enabled: bool | None = None
+
+    @model_validator(mode="after")
+    def _at_least_one(self) -> "UserUpdateRequest":
+        if self.role is None and self.enabled is None:
+            raise ValueError("至少需要提供一项更新字段（role 或 enabled）")
+        return self
+
+
 def _public_user(user: UserRecord) -> dict[str, Any]:
     """用户记录脱敏：剔除 password_hash/salt"""
     return {
         "username": user.username,
         "role": user.role,
         "must_change_password": user.must_change_password,
+        "enabled": user.enabled,
         "created_at": user.created_at,
     }
 
@@ -69,6 +83,8 @@ async def login(req: LoginRequest) -> dict[str, Any]:
         raise ApiError(2001, "认证服务暂不可用") from exc
     if user is None:
         raise ApiError(1005, "用户名或密码错误")
+    if not user.enabled:
+        raise ApiError(1005, "账号已禁用")
     try:
         token = await deps._get_session_store().create(user.username, user.role)
     except SessionStoreError as exc:
@@ -151,6 +167,46 @@ async def create_user(
     return ok(_public_user(user))
 
 
+@router.patch("/users/{username}")
+async def update_user(
+    username: str,
+    req: UserUpdateRequest,
+    admin: UserRecord = Depends(deps.require_admin),
+) -> dict[str, Any]:
+    """更新用户角色或启用状态
+
+    用户不存在 1004；role 非法 1001；最后一个 admin 降级 role 或被禁用 1001；
+    禁用用户时清除其全部 session（已登录态立即失效）。
+    """
+    store = deps._get_user_store()
+    try:
+        target = await store.get(username)
+        if target is None:
+            raise ApiError(1004, "用户不存在")
+        # 最后 admin 保护：目标当前是唯一 admin 且本次会使其失去 admin 身份或被禁用
+        if target.role == "admin" and await store.count_admins() <= 1:
+            will_lose_admin = (req.role is not None and req.role != "admin") or (req.enabled is False)
+            if will_lose_admin:
+                raise ApiError(1001, "禁止降级或禁用最后一个管理员")
+        updated = await store.update_user(username, role=req.role, enabled=req.enabled)
+    except UserStoreError as exc:
+        raise ApiError(2001, "认证服务暂不可用") from exc
+    except UserNotFoundError as exc:
+        raise ApiError(1004, "用户不存在") from exc
+    except ValueError as exc:
+        raise ApiError(1001, str(exc)) from exc
+    # 禁用用户 → 清除其全部 session（已登录态立即失效）
+    if req.enabled is False:
+        try:
+            await deps._get_session_store().delete_by_username(username)
+        except SessionStoreError as exc:
+            raise ApiError(2001, "认证服务暂不可用") from exc
+    logger.info(
+        "管理员更新用户", username=username, operator=admin.username, role=req.role, enabled=req.enabled
+    )
+    return ok(_public_user(updated))
+
+
 @router.post("/users/{username}/password")
 async def reset_user_password(
     username: str, req: PasswordResetRequest, admin: UserRecord = Depends(deps.require_admin)

app/api/v1/document.py

@@ -8,7 +8,7 @@ from typing import Any
 
 import structlog
 from fastapi import APIRouter, Depends, File, Form, Query, UploadFile
-from fastapi.responses import JSONResponse
+from fastapi.responses import FileResponse, JSONResponse
 
 from app.api.deps import get_current_user
 from app.api.response import ApiError, ok
@@ -183,6 +183,149 @@ async def upload_document(
     )
 
 
+async def _process_single_upload(file: UploadFile) -> dict[str, Any]:
+    """处理单个上传文件：校验 → 提取文本 → 落盘 → 提交入库
+
+    供批量上传复用：返回成功 {"filename", "task_id"} 或失败 {"filename", "error"}。
+    title/source 使用文件名默认值，不接受用户传入的额外表单字段。
+    """
+    original_filename = file.filename or "unnamed"
+    ext = Path(original_filename).suffix.lower()
+
+    # 1. 扩展名校验
+    allowed = _allowed_extensions()
+    if ext not in allowed:
+        return {"filename": original_filename, "error": f"不支持的文件类型: {ext or '(无扩展名)'}"}
+
+    # 2. 读取字节并校验大小
+    try:
+        content = await file.read()
+    except Exception as exc:
+        return {"filename": original_filename, "error": f"文件读取失败: {exc}"}
+    max_bytes = settings.upload_max_size_mb * 1024 * 1024
+    if len(content) > max_bytes:
+        return {"filename": original_filename, "error": f"文件超过大小上限: {settings.upload_max_size_mb}MB"}
+
+    # 3. 提取文本
+    try:
+        text = parse_file(original_filename, content)
+    except ValueError as exc:
+        return {"filename": original_filename, "error": str(exc)}
+    if not text.strip():
+        return {"filename": original_filename, "error": "无法从文件提取文本"}
+
+    # 4. 落盘（按 YYYY/MM 日期分片；失败仅 warning，不阻塞入库）
+    doc_id = uuid.uuid4().hex
+    metadata_dict: dict[str, str] = {}
+    try:
+        upload_dir = Path(settings.upload_dir).resolve()
+        shard_subdir = datetime.now(UTC).strftime("%Y/%m")
+        target_dir = upload_dir / shard_subdir
+        target_dir.mkdir(parents=True, exist_ok=True)
+        target = target_dir / f"{doc_id}_{original_filename}"
+        target.write_bytes(content)
+        metadata_dict.update(
+            {
+                "raw_file_path": str(target),
+                "original_filename": original_filename,
+                "original_size_bytes": str(len(content)),
+            }
+        )
+        logger.info(
+            "上传文件已落盘", doc_id=doc_id, saved_path=str(target), size=len(content)
+        )
+    except Exception:
+        logger.warning(
+            "上传文件落盘失败，仅做文本入库",
+            doc_id=doc_id,
+            filename=original_filename,
+            exc_info=True,
+        )
+
+    # 5. 默认 title / source
+    title = Path(original_filename).stem
+    source = f"file:{original_filename}"
+
+    # 6. 提交入库流水线
+    doc_input = DocumentInput(text=text, title=title, source=source, metadata=metadata_dict)
+    task_id = await _get_task_manager().submit(doc_input)
+    return {"filename": original_filename, "task_id": task_id}
+
+
+@router.post("/documents/upload-batch")
+async def upload_batch(
+    files: list[UploadFile] = File(...),
+    user: UserRecord = Depends(get_current_user),
+) -> JSONResponse:
+    """批量文件上传入库：逐文件复用单文件逻辑，收集成功与失败结果
+
+    返回 202 + {tasks: [{filename, task_id}], failed: [{filename, error}]}；
+    空文件列表返回 1001。
+    """
+    if not files:
+        raise ApiError(1001, "未提供任何文件")
+    tasks: list[dict[str, Any]] = []
+    failed: list[dict[str, Any]] = []
+    for file in files:
+        result = await _process_single_upload(file)
+        if "task_id" in result:
+            tasks.append({"filename": result["filename"], "task_id": result["task_id"]})
+        else:
+            failed.append({"filename": result["filename"], "error": result["error"]})
+    return JSONResponse(status_code=202, content=ok({"tasks": tasks, "failed": failed}))
+
+
+@router.get("/documents/tasks")
+async def list_ingest_tasks(
+    limit: int = Query(default=20, ge=1, le=100),
+    user: UserRecord = Depends(get_current_user),
+) -> dict[str, Any]:
+    """列出近期入库任务（合并内存与 Redis 镜像，按 updated_at 降序）"""
+    tasks = await _get_task_manager().list_tasks(limit=limit)
+    return ok({"items": tasks, "total": len(tasks)})
+
+
+@router.post("/documents/{doc_id}/reingest")
+async def reingest_document(
+    doc_id: str, user: UserRecord = Depends(get_current_user)
+) -> JSONResponse:
+    """重新入库：读原始文件 → 删旧数据 → 提交新入库任务
+
+    仅对有原始文件落盘记录的文档可重新入库；重新入库会生成新 doc_id。
+    """
+    meta = await _get_qdrant().get_l1_metadata(doc_id)
+    if meta is None:
+        raise ApiError(1004, "文档不存在")
+    raw_file_path = meta.get("raw_file_path", "")
+    if not raw_file_path:
+        raise ApiError(1001, "该文档无原始文件，无法重新入库")
+    path = Path(raw_file_path)
+    if not path.is_file():
+        raise ApiError(1004, "文件不存在")
+
+    # 读原文件并重新提取文本
+    try:
+        text = parse_file(meta.get("original_filename", path.name), path.read_bytes())
+    except ValueError as exc:
+        raise ApiError(1001, str(exc)) from exc
+    if not text.strip():
+        raise ApiError(1001, "无法从文件提取文本")
+
+    # 删旧数据（四层集合按 doc_id 清除，幂等）
+    await _get_qdrant().delete_by_doc_id(doc_id)
+
+    # 构造新文档输入（保留原 metadata，title/source 从 meta 取或文件名回退）
+    original_filename = meta.get("original_filename", path.name)
+    title = meta.get("title") or Path(original_filename).stem
+    source = meta.get("source") or f"file:{original_filename}"
+    doc_input = DocumentInput(text=text, title=title, source=source, metadata=dict(meta))
+
+    task_id = await _get_task_manager().submit(doc_input)
+    return JSONResponse(
+        status_code=202, content=ok({"task_id": task_id, "status": "pending"})
+    )
+
+
 @router.get("/documents/tasks/{task_id}")
 async def get_ingest_task(task_id: str) -> dict[str, Any]:
     """查询入库任务状态：含 task_id/status/created_at/updated_at，done 附 result，failed 附 error"""
@@ -192,6 +335,28 @@ async def get_ingest_task(task_id: str) -> dict[str, Any]:
     return ok(task)
 
 
+@router.post("/documents/tasks/{task_id}/retry")
+async def retry_ingest_task(
+    task_id: str, user: UserRecord = Depends(get_current_user)
+) -> JSONResponse:
+    """重试入库任务：用原 source 重新提交一个新任务，返回新 task_id"""
+    new_task_id = await _get_task_manager().retry(task_id)
+    if new_task_id is None:
+        raise ApiError(1004, "任务不存在或缺少重试所需的源信息")
+    return JSONResponse(
+        status_code=202, content=ok({"task_id": new_task_id, "status": "pending"})
+    )
+
+
+@router.delete("/documents/tasks/{task_id}")
+async def delete_ingest_task(
+    task_id: str, user: UserRecord = Depends(get_current_user)
+) -> dict[str, Any]:
+    """删除入库任务（内存注册表 + Redis 镜像），幂等"""
+    deleted = await _get_task_manager().delete(task_id)
+    return ok({"task_id": task_id, "deleted": deleted})
+
+
 @router.get("/documents")
 async def list_documents(
     limit: int = Query(default=20, ge=1, le=100),
@@ -219,6 +384,33 @@ async def get_document(doc_id: str) -> dict[str, Any]:
     return ok(detail)
 
 
+@router.get("/documents/{doc_id}/file")
+async def download_document_file(doc_id: str) -> FileResponse:
+    """下载文档关联的原始文件（免登录）
+
+    从 L1 metadata 读取 raw_file_path，校验路径位于 upload_dir 之内后返回 FileResponse；
+    文档不存在 / 无关联文件 / 文件缺失 / 路径越界统一返回 1004。
+    """
+    meta = await _get_qdrant().get_l1_metadata(doc_id)
+    if not meta:
+        raise ApiError(1004, "文档不存在或未关联文件")
+    raw_path = meta.get("raw_file_path", "")
+    if not raw_path:
+        raise ApiError(1004, "文档未关联文件")
+    path = Path(raw_path).resolve()
+    # 路径越界校验：只允许读取 upload_dir 下的文件
+    try:
+        upload_dir = Path(settings.upload_dir).resolve()
+        path.relative_to(upload_dir)
+    except ValueError:
+        logger.warning("文件路径越界", doc_id=doc_id, raw_path=raw_path)
+        raise ApiError(1004, "文件不存在") from None
+    if not path.is_file():
+        raise ApiError(1004, "文件不存在")
+    filename = meta.get("original_filename", path.name)
+    return FileResponse(path, filename=filename)
+
+
 @router.delete("/documents/{doc_id}")
 async def delete_document(
     doc_id: str, user: UserRecord = Depends(get_current_user)

app/api/v1/knowledge.py

@@ -4,11 +4,10 @@ from functools import lru_cache
 from typing import Any
 
 import structlog
-from fastapi import APIRouter, Depends
+from fastapi import APIRouter
 
 from app.api.response import ApiError, ok
 from app.config import settings
-from app.core.auth import AuthUser, get_current_user
 from app.models.knowledge import UNCATEGORIZED, TaxonomyCategory, load_taxonomy
 from app.services.qdrant import ALL_COLLECTIONS, COLLECTION_L1, QdrantService
 
@@ -37,14 +36,14 @@ def _get_taxonomy() -> list[TaxonomyCategory]:
 
 
 @router.get("/knowledge/categories")
-async def list_categories(user: AuthUser = Depends(get_current_user)) -> dict[str, Any]:
+async def list_categories() -> dict[str, Any]:
     """返回完整知识分类类目集"""
     categories = _get_taxonomy()
     return ok({"categories": [c.model_dump() for c in categories], "count": len(categories)})
 
 
 @router.get("/knowledge/stats")
-async def knowledge_stats(user: AuthUser = Depends(get_current_user)) -> dict[str, Any]:
+async def knowledge_stats() -> dict[str, Any]:
     """返回四层集合规模与 L1 类目分布统计"""
     service = _get_qdrant()
     try:

app/api/v1/search.py

@@ -4,10 +4,9 @@ from hashlib import sha256
 from typing import Any
 
 import structlog
-from fastapi import APIRouter, Depends
+from fastapi import APIRouter
 
 from app.api.response import ApiError, ok
-from app.core.auth import AuthUser, get_current_user
 from app.core.retriever import Retriever
 from app.models.search import SearchRequest
 from app.services.redis import get_cache
@@ -34,7 +33,7 @@ def _cache_key(request: SearchRequest) -> str:
 
 
 @router.post("/search")
-async def search(request: SearchRequest, user: AuthUser = Depends(get_current_user)) -> dict[str, Any]:
+async def search(request: SearchRequest) -> dict[str, Any]:
     """分层检索入口，返回统一包装的 SearchResponse
 
     先查 Redis 缓存：命中直接返回缓存的响应；未命中走检索流程并回写缓存。

app/api/v1/settings.py

@@ -11,7 +11,8 @@ from fastapi import APIRouter, Depends
 from pydantic import BaseModel
 
 from app.api.response import ApiError, ok
-from app.core.auth import AuthUser, get_current_user, require_admin
+from app.api.deps import get_current_user, require_admin
+from app.core.users import UserRecord
 from app.core.dedup import invalidate_dedup_strategy_cache
 from app.core.file_parser import (
     list_docx_plugins,
@@ -46,7 +47,7 @@ class SettingsUpdateRequest(BaseModel):
 
 
 @router.get("/settings")
-async def get_settings(user: AuthUser = Depends(get_current_user)) -> dict[str, Any]:
+async def get_settings(user: UserRecord = Depends(get_current_user)) -> dict[str, Any]:
     """返回当前 RuntimeSettings（任何登录用户可读）"""
     cfg = get_runtime_settings()
     return ok(cfg.model_dump(mode="json"))
@@ -55,7 +56,7 @@ async def get_settings(user: AuthUser = Depends(get_current_user)) -> dict[str,
 @router.put("/settings")
 async def update_settings(
     body: SettingsUpdateRequest,
-    user: AuthUser = Depends(require_admin),
+    user: UserRecord = Depends(require_admin),
 ) -> dict[str, Any]:
     """部分更新 RuntimeSettings（仅 admin）
 
@@ -82,7 +83,7 @@ async def update_settings(
 
 @router.get("/settings/schema")
 async def get_settings_schema(
-    user: AuthUser = Depends(get_current_user),
+    user: UserRecord = Depends(get_current_user),
 ) -> dict[str, Any]:
     """返回可选插件与策略列表（前端 Settings 页渲染选项用）"""
     return ok(
@@ -98,7 +99,7 @@ async def get_settings_schema(
 
 @router.post("/settings/reset")
 async def reset_settings(
-    user: AuthUser = Depends(require_admin),
+    user: UserRecord = Depends(require_admin),
 ) -> dict[str, Any]:
     """重置 RuntimeSettings 为默认值（仅 admin），同时清缓存"""
     try:

app/config.py

@@ -18,9 +18,15 @@ class Settings(BaseSettings):
 
     # Ollama 本地模型（用于文档三级总结）
     ollama_base_url: str = "http://localhost:11434"
-    ollama_model: str = "qwen2.5:1.5b"  # 备选: qwen2.5:3b
+    ollama_model: str = "qwen3:1.7b"  # 备选: qwen3:4b（更好的总结质量，需更多内存）
     ollama_embedding_model: str = "bge-m3"  # embedding_provider=local 时使用的嵌入模型
 
+    # 重排模型（cross-encoder，经 Ollama /api/rerank，对 chunk 候选做最终精排）
+    # 关闭时检索链路退化为纯 RRF 融合结果，行为不变
+    reranker_enabled: bool = False
+    reranker_model: str = "qwen3-reranker:0.6b"
+    reranker_timeout: float = 30.0
+
     # Qdrant
     qdrant_host: str = "localhost"
     qdrant_port: int = 6333

app/core/auth.py

@@ -4,7 +4,7 @@
 鉴权（get_current_user）以 JWT 自包含信息为主，Redis 不可用时降级可用。
 """
 
-from datetime import UTC, datetime, timedelta
+from datetime import datetime, timedelta, timezone
 
 import bcrypt
 import jwt
@@ -67,7 +67,7 @@ def verify_password(password: str, hashed: str) -> bool:
 
 def create_access_token(username: str, role: str) -> tuple[str, int]:
     """签发 JWT，返回 (token, expires_in_seconds)"""
-    now = datetime.now(UTC)
+    now = datetime.now(timezone.utc)
     expire = now + timedelta(minutes=settings.jwt_expire_minutes)
     payload = {
         "sub": username,
@@ -136,7 +136,7 @@ class UserStore:
         user = StoredUser(
             username=username,
             role=role,
-            created_at=datetime.now(UTC),
+            created_at=datetime.now(timezone.utc),
             hashed_password=hash_password(password),
         )
         try:
@@ -203,7 +203,7 @@ async def get_current_user(
         )
     # Redis 不可用或用户不存在（可能已删除）：降级用 JWT payload
     logger.warning("用户存储查询未命中，降级使用 JWT payload", username=username)
-    return AuthUser(username=username, role=role, created_at=datetime.now(UTC))
+    return AuthUser(username=username, role=role, created_at=datetime.now(timezone.utc))
 
 
 async def require_admin(user: AuthUser = Depends(get_current_user)) -> AuthUser:

app/core/ingest_tasks.py

@@ -20,7 +20,7 @@ from typing import Any
 import structlog
 
 from app.config import Settings
-from app.core.dedup import DEDUP_KEY_PREFIX, get_dedup_strategy
+from app.core.dedup import get_dedup_strategy
 from app.core.ingestion import Ingester, IngestionError
 from app.models.document import DocumentInput
 from app.services.redis import RedisCache
@@ -79,6 +79,8 @@ class IngestTaskManager:
         """
         task_id = uuid.uuid4().hex
         now = _utc_now_iso()
+        filename = self._extract_filename(doc)
+        source = self._extract_source(doc)
         dedup = get_dedup_strategy(self._redis)
 
         # 1. 去重命中：直接置 done，复用旧结果，不调 _run
@@ -93,6 +95,8 @@ class IngestTaskManager:
                 "updated_at": now,
                 "result": result_dict,
                 "error": None,
+                "filename": filename,
+                "source": source,
             }
             self._schedule_mirror(task_id)
             logger.info(
@@ -110,6 +114,8 @@ class IngestTaskManager:
             "updated_at": now,
             "result": None,
             "error": None,
+            "filename": filename,
+            "source": source,
         }
         self._schedule_mirror(task_id)
         background = asyncio.create_task(self._run(task_id, doc, dedup))
@@ -135,6 +141,119 @@ class IngestTaskManager:
             )
             return None
 
+    async def list_tasks(self, limit: int = 20) -> list[dict[str, Any]]:
+        """合并内存注册表与 Redis 镜像，去重，按 updated_at 降序，截断 limit
+
+        每项提取 {task_id, status, filename, created_at, updated_at, doc_id}：
+        doc_id 从 done 任务的 result.document_id 提取，其余状态为 None。
+        """
+        records: dict[str, dict[str, Any]] = {}
+        # 1. 内存注册表（主）
+        for task_id, record in self._tasks.items():
+            records[task_id] = record
+        # 2. Redis 镜像补充内存中没有的（内存未命中的任务，如重启后只存在 Redis 的历史记录）
+        for task_id, record in await self._scan_redis_tasks():
+            records.setdefault(task_id, record)
+        # 3. 按 updated_at 降序，截断 limit
+        sorted_records = sorted(
+            records.values(),
+            key=lambda r: r.get("updated_at") or "",
+            reverse=True,
+        )[:limit]
+        # 4. 提取展示字段
+        return [self._to_list_item(r) for r in sorted_records]
+
+    async def _scan_redis_tasks(self) -> list[tuple[str, dict[str, Any]]]:
+        """扫描 Redis 中 ingest_task:* 键，返回 (task_id, record) 列表
+
+        RedisCache 未暴露公开扫描接口，借道底层 _get_client().scan_iter；
+        任何异常降级为空列表，不影响 list_tasks 主流程。
+        """
+        if self._redis is None:
+            return []
+        get_client = getattr(self._redis, "_get_client", None)
+        if not callable(get_client):
+            return []
+        try:
+            client = get_client()
+            tasks: list[tuple[str, dict[str, Any]]] = []
+            async for key in client.scan_iter(match=f"{REDIS_KEY_PREFIX}*"):
+                key_str = key if isinstance(key, str) else key.decode("utf-8", "replace")
+                task_id = key_str[len(REDIS_KEY_PREFIX) :]
+                record = await self._redis.get_json(key_str)
+                if isinstance(record, dict):
+                    tasks.append((task_id, record))
+            return tasks
+        except Exception:
+            logger.warning("扫描 Redis 任务键失败", exc_info=True)
+            return []
+
+    @staticmethod
+    def _extract_filename(doc: DocumentInput) -> str | None:
+        """从文档输入提取展示用文件名：优先 metadata.original_filename，其次 title"""
+        return doc.metadata.get("original_filename") or doc.title or None
+
+    @staticmethod
+    def _extract_source(doc: DocumentInput) -> dict[str, Any]:
+        """从文档输入提取重试所需的源信息（供 retry 复用，含文件路径/大小）"""
+        return {
+            "text": doc.text,
+            "title": doc.title,
+            "source": doc.source,
+            "metadata": dict(doc.metadata),
+        }
+
+    async def retry(self, task_id: str) -> str | None:
+        """重试失败/已完成的入库任务：用原 source 重新提交一个新任务
+
+        返回新 task_id；任务不存在或缺少 source 时返回 None。
+        """
+        record = await self.get(task_id)
+        if record is None:
+            return None
+        source = record.get("source")
+        if not isinstance(source, dict) or not source.get("text"):
+            return None
+        doc = DocumentInput(
+            text=source["text"],
+            title=source.get("title") or "",
+            source=source.get("source") or "",
+            metadata=source.get("metadata") or {},
+        )
+        return await self.submit(doc)
+
+    async def delete(self, task_id: str) -> bool:
+        """删除入库任务（内存注册表 + Redis 镜像），不存在返回 False"""
+        existed = self._tasks.pop(task_id, None) is not None
+        if self._redis is not None:
+            try:
+                get_client = getattr(self._redis, "_get_client", None)
+                if callable(get_client):
+                    await get_client().delete(f"{REDIS_KEY_PREFIX}{task_id}")
+                    existed = True
+            except Exception:
+                logger.warning("删除 Redis 任务镜像失败", task_id=task_id, exc_info=True)
+        return existed
+
+    @staticmethod
+    def _to_list_item(record: dict[str, Any]) -> dict[str, Any]:
+        """从完整任务记录提取列表展示字段"""
+        result = record.get("result")
+        doc_id = result.get("document_id") if isinstance(result, dict) else None
+        source = record.get("source") or {}
+        metadata = source.get("metadata") or {}
+        return {
+            "task_id": record.get("task_id"),
+            "status": record.get("status"),
+            "filename": record.get("filename"),
+            "title": source.get("title"),
+            "source": source.get("source"),
+            "size_bytes": metadata.get("original_size_bytes"),
+            "created_at": record.get("created_at"),
+            "updated_at": record.get("updated_at"),
+            "doc_id": doc_id,
+        }
+
     async def wait_done(self, task_id: str, timeout: float = 30.0) -> dict[str, Any]:
         """轮询内存注册表直到任务进入终态（done/failed）或超时
 

app/core/ingestion.py

@@ -262,6 +262,7 @@ class Ingester:
             tags=category.tags,
             dense_vector=l1_vector,
             sparse_vector=l1_sparse,
+            metadata=doc.metadata,
         )
 
         # L2/L3 大纲节点（为空时跳过对应集合的 upsert）

app/core/retriever.py

@@ -16,7 +16,7 @@ from qdrant_client import models
 from app.config import settings
 from app.core.embeddings import EmbeddingService, create_embedding_service
 from app.core.query_parser import QueryParser, RouteDecision
-from app.core.ranker import finalize, rrf_fuse
+from app.core.ranker import rrf_fuse
 from app.core.result_summarizer import ResultSummarizer
 from app.core.sparse import SparseEncoder
 from app.models.knowledge import load_taxonomy
@@ -30,6 +30,7 @@ from app.services.qdrant import (
     SPARSE_COLLECTIONS,
     QdrantService,
 )
+from app.services.reranker import RerankerService, create_reranker_service
 
 logger = structlog.get_logger()
 
@@ -49,6 +50,7 @@ class Retriever:
         embedding: EmbeddingService | None = None,
         sparse_encoder: SparseEncoder | None = None,
         result_summarizer: ResultSummarizer | None = None,
+        reranker: RerankerService | None = None,
     ) -> None:
         self.qdrant = qdrant or QdrantService()
         self.query_parser = query_parser or QueryParser(
@@ -58,6 +60,7 @@ class Retriever:
         self.embedding = embedding or create_embedding_service()
         self.sparse_encoder = sparse_encoder or SparseEncoder()
         self.result_summarizer = result_summarizer or ResultSummarizer()
+        self.reranker = reranker if reranker is not None else create_reranker_service()
 
     async def search(self, request: SearchRequest) -> SearchResponse:
         """分层检索主流程"""
@@ -77,7 +80,7 @@ class Retriever:
             # L1 无候选文档 → 全库 chunk 兜底
             chunk_hits = await self._search_collection(COLLECTION_CHUNKS, dense, sparse, settings.retrieval_top_k, None)
             logger.info("L1 无命中，全库 chunk 兜底", hits=len(chunk_hits))
-            hits = self._to_hits(finalize(chunk_hits, self._final_k(request)))
+            hits = self._to_hits(await self._rerank(query_text, chunk_hits, self._final_k(request)))
             return await self._build_response(request, route, hits, fallback=True)
 
         doc_ids = _unique((p.payload or {}).get("doc_id") for p in l1_hits)
@@ -123,7 +126,7 @@ class Retriever:
         )
         logger.info("chunk 检索完成", hits=len(chunk_hits))
 
-        final_points = finalize(rrf_fuse([chunk_hits]), self._final_k(request))
+        final_points = await self._rerank(query_text, chunk_hits, self._final_k(request))
         hits = self._to_hits(final_points)
         return await self._build_response(request, route, hits, fallback=route.fallback)
 
@@ -145,6 +148,26 @@ class Retriever:
         """最终返回数：请求指定优先，否则用配置默认值"""
         return request.top_k or settings.retrieval_final_k
 
+    async def _rerank(
+        self, query_text: str, points: list[models.ScoredPoint], final_k: int
+    ) -> list[models.ScoredPoint]:
+        """对 chunk 候选池做语义精排（cross-encoder）
+
+        重排未启用、候选为空或调用异常时，退化为 RRF 融合的原始顺序（取前 final_k），
+        保证检索链路在任何情况下都不因重排失败而中断。
+        """
+        if self.reranker is None or not points:
+            return points[:final_k]
+        documents = [(p.payload or {}).get("text", "") for p in points]
+        try:
+            scores = await self.reranker.rerank(query_text, documents)
+        except Exception:
+            logger.warning("重排调用失败，退化为 RRF 融合顺序", exc_info=True)
+            return points[:final_k]
+        # 按相关性分数降序重排，写回 score 字段为相关性分数
+        ordered = sorted(range(len(points)), key=lambda i: scores[i], reverse=True)
+        return [points[i].model_copy(update={"score": scores[i]}) for i in ordered[:final_k]]
+
     @staticmethod
     def _to_hits(points: list[models.ScoredPoint]) -> list[SearchHit]:
         """chunk 点组装为 SearchHit，字段取自 chunk payload（doc_summary 仅上下文标注）"""

app/core/users.py

@@ -11,7 +11,7 @@ import hmac
 import json
 import re
 import secrets
-from dataclasses import asdict, dataclass
+from dataclasses import asdict, dataclass, field
 from datetime import UTC, datetime
 
 from redis import asyncio as redis_async
@@ -41,6 +41,10 @@ class UserStoreError(Exception):
     """用户存储后端读写异常"""
 
 
+class UserNotFoundError(Exception):
+    """用户不存在"""
+
+
 @dataclass
 class UserRecord:
     """用户记录（不含明文密码）"""
@@ -50,7 +54,9 @@ class UserRecord:
     password_hash: str  # hex
     salt: str  # hex, 16 字节
     must_change_password: bool
-    created_at: str  # UTC ISO8601
+    enabled: bool = True
+    # created_at 紧随 enabled 之后；二者均有默认值以满足 dataclass 字段顺序约束
+    created_at: str = field(default_factory=lambda: datetime.now(UTC).isoformat())
 
 
 class UserStore:
@@ -76,6 +82,7 @@ class UserStore:
         password: str,
         role: str = "user",
         must_change_password: bool = False,
+        enabled: bool = True,
     ) -> UserRecord:
         """创建用户
 
@@ -94,6 +101,7 @@ class UserStore:
             password_hash=self.hash_password(password, salt),
             salt=salt.hex(),
             must_change_password=must_change_password,
+            enabled=enabled,
             created_at=datetime.now(UTC).isoformat(),
         )
         await self._write(record)
@@ -104,7 +112,7 @@ class UserStore:
         raw = await self._read_raw(username)
         if raw is None:
             return None
-        return UserRecord(**json.loads(raw))
+        return self._from_raw(raw)
 
     async def list(self) -> list[UserRecord]:
         """列出全部用户（扫描 user:* 键）"""
@@ -115,7 +123,15 @@ class UserStore:
                 raws = [await self._redis.get(key) async for key in self._redis.scan_iter(match=f"{_USER_KEY_PREFIX}*")]
             except Exception as e:
                 raise UserStoreError("列出用户失败") from e
-        return [UserRecord(**json.loads(raw)) for raw in raws if raw is not None]
+        return [self._from_raw(raw) for raw in raws if raw is not None]
+
+    @staticmethod
+    def _from_raw(raw: str) -> UserRecord:
+        """从 JSON 反序列化 UserRecord；存量记录无 enabled 字段时按 True 兼容"""
+        data = json.loads(raw)
+        if "enabled" not in data:
+            data["enabled"] = True
+        return UserRecord(**data)
 
     async def delete(self, username: str) -> bool:
         """删除用户，返回是否删除成功（幂等：不存在返回 False）"""
@@ -158,6 +174,30 @@ class UserStore:
         """统计 admin 角色用户数"""
         return sum(1 for record in await self.list() if record.role == "admin")
 
+    async def update_user(
+        self,
+        username: str,
+        *,
+        role: str | None = None,
+        enabled: bool | None = None,
+    ) -> UserRecord:
+        """更新用户角色或启用状态
+
+        用户不存在抛 UserNotFoundError；role 非 admin/user 抛 ValueError。
+        role/enabled 为 None 表示不修改对应字段。更新后写回并返回最新记录。
+        """
+        record = await self.get(username)
+        if record is None:
+            raise UserNotFoundError(f"用户不存在: {username}")
+        if role is not None and role not in ("admin", "user"):
+            raise ValueError(f"非法角色: {role!r}（须为 admin/user）")
+        if role is not None:
+            record.role = role
+        if enabled is not None:
+            record.enabled = enabled
+        await self._write(record)
+        return record
+
     async def _read_raw(self, username: str) -> str | None:
         if self._redis is None:
             return self._memory.get(username)

app/main.py

@@ -6,7 +6,6 @@ import structlog
 from fastapi import FastAPI, Request
 from fastapi.exceptions import RequestValidationError
 from fastapi.responses import FileResponse, JSONResponse, RedirectResponse
-from redis import asyncio as redis_async
 
 from app.api.response import ApiError, error
 from app.api.v1.auth import router as auth_router
@@ -15,7 +14,7 @@ from app.api.v1.knowledge import router as knowledge_router
 from app.api.v1.search import router as search_router
 from app.api.v1.settings import router as settings_router
 from app.config import settings
-from app.core.users import UserStore, bootstrap_admin
+from app.core.users import bootstrap_admin
 from app.services.qdrant import QdrantService
 
 logger = structlog.get_logger()
@@ -33,13 +32,13 @@ async def lifespan(_: FastAPI) -> AsyncIterator[None]:
     except Exception:
         logger.error("Qdrant 集合初始化失败，跳过初始化继续启动")
     try:
-        # Redis 客户端创建失败（如 URL 非法）时传 None，UserStore 降级为内存模式
+        # 复用 deps 的 UserStore 单例，确保 lifespan 创建的 admin 与 API 请求用的是同一实例
-        redis_client = redis_async.from_url(settings.redis_url, decode_responses=True)
+        # _get_user_store 内部会检测 Redis 可达性，不可达时降级为内存模式
-    except Exception:
+        from app.api.deps import _get_user_store
-        redis_client = None
+
-    try:
+        user_store = _get_user_store()
         # 空库引导默认管理员；明文密码由 bootstrap_admin 内部 warning 打印一次
-        await bootstrap_admin(UserStore(redis_client), logger)
+        await bootstrap_admin(user_store, logger)
     except Exception:
         logger.warning("默认管理员初始化失败，跳过", exc_info=True)
     try:
@@ -144,3 +143,21 @@ async def admin_spa(rest: str):
         return FileResponse(file_path)
     # SPA 客户端路由兜底
     return FileResponse(spa_dir / "index.html", media_type="text/html")
+
+
+# AI Agent Skill 压缩包（用于 Agent 上传文档与检索），随 ./app 卷挂载，restart 即生效
+_AGENT_SKILL_ZIP = _STATIC_DIR / "agent-skill" / "QMDSearch-Agent-Skill.zip"
+
+
+@app.get("/agent-skill", include_in_schema=False, response_model=None)
+async def agent_skill_download():
+    """下载 AI Agent Skill 压缩包（qmdsearch-agent skill，含 SKILL.md 与示例）"""
+    if not _AGENT_SKILL_ZIP.is_file():
+        return JSONResponse(
+            status_code=404, content={"code": 1002, "message": "Skill 包不存在"}
+        )
+    return FileResponse(
+        _AGENT_SKILL_ZIP,
+        media_type="application/zip",
+        filename="QMDSearch-Agent-Skill.zip",
+    )

app/services/qdrant.py

@@ -9,6 +9,7 @@
 """
 
 import uuid
+from pathlib import Path
 from typing import Any
 
 import structlog
@@ -103,18 +104,50 @@ class QdrantService:
         tags: list[str],
         dense_vector: list[float],
         sparse_vector: SparseVectorTuple | None = None,
+        metadata: dict[str, str] | None = None,
     ) -> None:
-        """写入 L1 文档总结，payload 含 doc_id/title/category/tags/text(=summary)"""
+        """写入 L1 文档总结，payload 含 doc_id/title/category/tags/text(=summary)/metadata
+
+        metadata 默认 None 时存空 dict，保证字段始终存在；存量旧文档读取时按缺失处理。
+        """
         vector: dict[str, Any] = {VECTOR_DENSE: dense_vector}
         if sparse_vector is not None:
             vector[VECTOR_SPARSE] = models.SparseVector(indices=sparse_vector[0], values=sparse_vector[1])
         point = models.PointStruct(
             id=_point_id(f"{doc_id}:l1"),
             vector=vector,
-            payload={"doc_id": doc_id, "title": title, "category": category, "tags": tags, "text": summary},
+            payload={
+                "doc_id": doc_id,
+                "title": title,
+                "category": category,
+                "tags": tags,
+                "text": summary,
+                "metadata": metadata or {},
+            },
         )
         await self._client.upsert(collection_name=COLLECTION_L1, points=[point])
 
+    async def get_l1_metadata(self, doc_id: str) -> dict[str, str] | None:
+        """按 doc_id 查 L1 点，返回其 payload.metadata
+
+        文档不存在或 payload 无 metadata 字段时返回 None。
+        """
+        doc_filter = self.build_filter(doc_ids=[doc_id])
+        records, _ = await self._client.scroll(
+            collection_name=COLLECTION_L1,
+            scroll_filter=doc_filter,
+            limit=1,
+            with_payload=True,
+            with_vectors=False,
+        )
+        if not records:
+            return None
+        payload = records[0].payload or {}
+        meta = payload.get("metadata")
+        if not isinstance(meta, dict):
+            return None
+        return meta
+
     async def upsert_nodes(self, collection: str, nodes: list[dict[str, Any]]) -> None:
         """批量写入 L2/L3 大纲节点
 
@@ -252,7 +285,12 @@ class QdrantService:
         return items, str(next_offset) if next_offset is not None else None
 
     async def get_doc_detail(self, doc_id: str) -> dict[str, Any] | None:
-        """获取文档详情：L1 记录 + L2/L3 全部节点 + chunks 数量，文档不存在返回 None"""
+        """获取文档详情：L1 记录 + L2/L3 全部节点 + chunks 数量 + file 文件信息
+
+        file 字段从 L1 payload.metadata 提取（raw_file_path/original_filename/original_size_bytes）：
+        有 raw_file_path 时返回 {filename, size_bytes, url}，否则 None。
+        文档不存在返回 None。
+        """
         doc_filter = self.build_filter(doc_ids=[doc_id])
         l1_records, _ = await self._client.scroll(
             collection_name=COLLECTION_L1,
@@ -271,11 +309,23 @@ class QdrantService:
             count_filter=doc_filter,
             exact=True,
         )
+
+        # 从 L1 metadata 提取原始文件信息，无 raw_file_path 时 file=None
+        meta = (l1_records[0].payload or {}).get("metadata") or {}
+        raw_path = meta.get("raw_file_path", "")
+        file_info: dict[str, Any] | None = None
+        if raw_path:
+            file_info = {
+                "filename": meta.get("original_filename", Path(raw_path).name),
+                "size_bytes": int(meta.get("original_size_bytes", "0") or "0"),
+                "url": f"/api/v1/documents/{doc_id}/file",
+            }
         return {
             "l1": l1_records[0].payload or {},
             "l2_nodes": l2_nodes,
             "l3_nodes": l3_nodes,
             "chunks_count": chunks_count.count,
+            "file": file_info,
         }
 
     async def delete_by_doc_id(self, doc_id: str) -> dict[str, int]:

app/services/reranker.py

@@ -0,0 +1,79 @@
+"""语义重排服务（cross-encoder reranker）
+
+调用 Ollama 的 `/api/rerank` 端点（Qwen3-Reranker 等重排模型），对检索召回的
+chunk 候选池做精排，提升最终 Top-K 的相关性。相比纯向量/RRF 融合，cross-encoder
+以 (query, doc) 联合编码，能捕捉字词不匹配的语义关联，典型带来约 10% 的精度提升。
+
+工厂 `create_reranker_service()` 按 `settings.reranker_enabled` 返回实例或 None：
+关闭时上层检索链路退化为原有的 RRF 融合结果，行为完全不变。
+"""
+
+from typing import Protocol, runtime_checkable
+
+import httpx
+import structlog
+
+from app.config import settings
+
+logger = structlog.get_logger()
+
+
+@runtime_checkable
+class RerankerService(Protocol):
+    """重排服务统一接口"""
+
+    async def rerank(self, query: str, documents: list[str]) -> list[float]:
+        """对候选文档按与 query 的相关性打分
+
+        Args:
+            query: 查询文本
+            documents: 候选文档文本列表（按输入顺序对齐）
+
+        Returns:
+            与 documents 等长的相关性分数列表（分数越高越相关）
+        """
+        ...
+
+
+class OllamaRerankerService:
+    """基于 Ollama /api/rerank 的重排服务"""
+
+    def __init__(self, base_url: str, model: str, timeout: float = 30.0) -> None:
+        self.base_url = base_url.rstrip("/")
+        self.model = model
+        self.timeout = timeout
+
+    async def rerank(self, query: str, documents: list[str]) -> list[float]:
+        if not documents:
+            return []
+        url = f"{self.base_url}/api/rerank"
+        payload = {"model": self.model, "query": query, "documents": documents}
+        async with httpx.AsyncClient(timeout=self.timeout) as client:
+            resp = await client.post(url, json=payload)
+            resp.raise_for_status()
+            data = resp.json()
+
+        results = data.get("results", [])
+        # 按 Ollama 返回的 index 映射 relevance_score；缺失项补 0，保证输出与输入等长
+        scores: list[float] = [0.0] * len(documents)
+        for item in results:
+            idx = item.get("index")
+            score = float(item.get("relevance_score", 0.0))
+            if idx is not None and 0 <= idx < len(documents):
+                scores[idx] = score
+        logger.debug("重排完成", model=self.model, candidates=len(documents))
+        return scores
+
+
+def create_reranker_service() -> RerankerService | None:
+    """按 settings.reranker_enabled 创建重排服务实例
+
+    关闭时返回 None，上层据此跳过精排、退化为 RRF 融合结果。
+    """
+    if not settings.reranker_enabled:
+        return None
+    return OllamaRerankerService(
+        base_url=settings.ollama_base_url,
+        model=settings.reranker_model,
+        timeout=settings.reranker_timeout,
+    )

docker-compose.yml

@@ -1,6 +1,23 @@
+# ============================================================
+# QMDSearch — 部署 compose（相对路径，可任意目录部署）
+# ------------------------------------------------------------
+# 数据库数据 / 配置 / 代码 / 前端 全部从本地目录挂载，
+# 日常更新（改代码或前端）只需改文件 + `docker compose restart app`，
+# 无需重新 build 镜像。
+#
+# 说明：
+#  - 相对路径相对于本 compose 文件所在目录（即项目根目录）解析，
+#    因此直接把整个项目目录放到 NAS（或任意主机）即可部署。
+#  - app 服务复用已构建的 qmdsearch-app 镜像作为「带依赖的运行时」
+#    （Python 依赖装在镜像内，代码与前端构建产物走挂载）。
+#  - 仅当 Python 依赖(pyproject.toml)变化时，才需 `docker build -t qmdsearch-app .` 一次。
+#  - 数据目录默认 ./data，可通过 .env 的 NAS_DATA_DIR 覆盖到其他磁盘。
+# ============================================================
+
 services:
   app:
-    build: .
+    # 不再 build 项目代码，改用已含依赖的运行时镜像；代码与前端均走下方挂载
+    image: qmdsearch-app:latest
     container_name: qmdsearch-app
     restart: unless-stopped
     ports:
@@ -13,7 +30,7 @@ services:
       # 针对 16 线程 / 61GB 内存的 NAS 调优：放宽入库并发
       - INGEST_MAX_CONCURRENCY=${INGEST_MAX_CONCURRENCY:-4}
     env_file:
-      - .env
+      - ./.env
     depends_on:
       qdrant:
         condition: service_started
@@ -22,6 +39,14 @@ services:
       ollama:
         condition: service_started
     volumes:
+      # ---- 代码（本地，改后 restart 即生效，免 build）----
+      - ./app:/app/app
+      - ./scripts:/app/scripts
+      # ---- 前端构建产物（本地 frontend/dist，改前端后 build 到此目录 + restart）----
+      - ./frontend/dist:/app/app/static/admin
+      # ---- 配置（显式文件映射，只读）----
+      - ./.env:/app/.env:ro
+      # ---- 数据 / 日志（持久化）----
       - ${NAS_DATA_DIR:-./data}/logs:/app/logs
       - ${NAS_DATA_DIR:-./data}/uploads:/app/uploads
     networks:
@@ -68,10 +93,10 @@ services:
       # 针对 Ryzen 9 7940HS（16 线程）的 CPU 推理调优：
       # 并行推理任务数、常驻模型数、单请求线程上限、KV 缓存量化以省内存
       - OLLAMA_NUM_PARALLEL=4
-      - OLLAMA_MAX_LOADED_MODELS=2
+      - OLLAMA_MAX_LOADED_MODELS=3
       - OLLAMA_NUM_THREADS=16
       - OLLAMA_KV_CACHE_TYPE=q8_0
-    # 首次启动自动拉取所需模型（qwen2.5:1.5b 总结 + bge-m3 嵌入），
+    # 首次启动自动拉取所需模型（qwen3:1.7b 总结 + bge-m3 嵌入 + qwen3-reranker 重排），
     # 下载完成后转交常驻 ollama serve。已存在时仅做健康检查。
     entrypoint: /bin/bash
     command:
@@ -80,8 +105,9 @@ services:
         ollama serve &
         SERVE_PID=$$!
         sleep 6
-        ollama pull qwen2.5:1.5b
+        ollama pull qwen3:1.7b
         ollama pull bge-m3
+        ollama pull qwen3-reranker:0.6b
         wait $$SERVE_PID
     networks:
       - qmdsearch

frontend/src/api/auth.js

@@ -17,3 +17,61 @@ export function login(username, password) {
 export function me() {
   return http.get('/api/v1/auth/me')
 }
+
+/**
+ * 列出全部用户（admin 专用）
+ * @returns {Promise<Array<{username:string, role:string, enabled:boolean, must_change_password:boolean, created_at:string}>>}
+ */
+export function listUsers() {
+  return http.get('/api/v1/auth/users')
+}
+
+/**
+ * 创建用户（admin 专用）
+ * @param {string} username
+ * @param {string} password
+ * @param {'admin'|'user'} [role='user']
+ */
+export function createUser(username, password, role = 'user') {
+  return http.post('/api/v1/auth/users', { username, password, role })
+}
+
+/**
+ * 更新用户角色 / 启用状态（admin 专用，至少传一项）
+ * @param {string} username
+ * @param {{ role?: 'admin'|'user', enabled?: boolean }} payload
+ */
+export function updateUser(username, payload) {
+  return http.patch(`/api/v1/auth/users/${encodeURIComponent(username)}`, payload)
+}
+
+/**
+ * 重置指定用户密码（admin 专用）
+ * @param {string} username
+ * @param {string} newPassword
+ */
+export function resetPassword(username, newPassword) {
+  return http.post(`/api/v1/auth/users/${encodeURIComponent(username)}/password`, {
+    new_password: newPassword
+  })
+}
+
+/**
+ * 删除用户（admin 专用）
+ * @param {string} username
+ */
+export function deleteUser(username) {
+  return http.delete(`/api/v1/auth/users/${encodeURIComponent(username)}`)
+}
+
+/**
+ * 修改自己的密码（当前登录用户）
+ * @param {string} oldPassword
+ * @param {string} newPassword
+ */
+export function changeMyPassword(oldPassword, newPassword) {
+  return http.post('/api/v1/auth/password', {
+    old_password: oldPassword,
+    new_password: newPassword
+  })
+}

frontend/src/api/documents.js

@@ -52,6 +52,55 @@ export function upload(formData) {
   })
 }
 
+/**
+ * multipart 批量文件上传入库（异步，逐文件创建任务）
+ * @param {File[]} files
+ * @returns {Promise<{tasks: Array<{filename:string, task_id:string}>, failed: Array<{filename:string, error:string}>}>}
+ */
+export function uploadBatch(files) {
+  const formData = new FormData()
+  files.forEach((f) => formData.append('files', f))
+  return http.post('/api/v1/documents/upload-batch', formData, {
+    headers: { 'Content-Type': 'multipart/form-data' }
+  })
+}
+
+/**
+ * 重新摘要入库（读原文件→删旧数据→新建任务）
+ * @param {string} docId
+ * @returns {Promise<{task_id:string, status:string}>}
+ */
+export function reingest(docId) {
+  return http.post(`/api/v1/documents/${encodeURIComponent(docId)}/reingest`)
+}
+
+/**
+ * 列出近期入库任务（按 updated_at 降序）
+ * @param {number} [limit=20]
+ * @returns {Promise<{items: Array, total: number}>}
+ */
+export function taskList(limit = 20) {
+  return http.get('/api/v1/documents/tasks', { params: { limit } })
+}
+
+/**
+ * 重试入库任务（用原 source 重新提交新任务）
+ * @param {string} taskId
+ * @returns {Promise<{task_id:string, status:string}>}
+ */
+export function retryTask(taskId) {
+  return http.post(`/api/v1/documents/tasks/${encodeURIComponent(taskId)}/retry`)
+}
+
+/**
+ * 删除入库任务
+ * @param {string} taskId
+ * @returns {Promise<{task_id:string, deleted:boolean}>}
+ */
+export function deleteTask(taskId) {
+  return http.delete(`/api/v1/documents/tasks/${encodeURIComponent(taskId)}`)
+}
+
 /**
  * 查询入库任务状态
  * @param {string} taskId

frontend/src/composables/useTasksDrawer.js

@@ -0,0 +1,21 @@
+import { reactive } from 'vue'
+
+/**
+ * 入库进度右侧 Drawer 的全局共享开关状态
+ *
+ * 该 Drawer 渲染在 MainLayout 中，但可由任意子页面（如 Library）触发打开，
+ * 因此通过单例 reactive 状态 + open()/close() 方法解耦，避免多层 prop 传递。
+ */
+const state = reactive({ open: false })
+
+export function useTasksDrawer() {
+  return {
+    state,
+    open() {
+      state.open = true
+    },
+    close() {
+      state.open = false
+    }
+  }
+}

frontend/src/layouts/MainLayout.vue

@@ -1,26 +1,31 @@
 <script setup>
-import { computed, h, ref } from 'vue'
+import { computed, h, reactive, ref, watch } from 'vue'
 import { useRoute, useRouter, RouterView } from 'vue-router'
 import { storeToRefs } from 'pinia'
-import { Modal } from 'ant-design-vue'
+import { Modal, message } from 'ant-design-vue'
 import {
   DashboardOutlined,
-  FileTextOutlined,
-  UploadOutlined,
-  SearchOutlined,
   AppstoreOutlined,
   SettingOutlined,
   MenuFoldOutlined,
   MenuUnfoldOutlined,
   LogoutOutlined,
-  DatabaseOutlined
+  DatabaseOutlined,
+  ApiOutlined,
+  TeamOutlined,
+  ClockCircleOutlined
 } from '@ant-design/icons-vue'
 import { useAuthStore } from '@/stores/useAuthStore'
+import { changeMyPassword } from '@/api/auth'
+import { callApi } from '@/api/client'
+import TasksDrawer from '@/views/TasksDrawer.vue'
+import { useTasksDrawer } from '@/composables/useTasksDrawer'
 
 const route = useRoute()
 const router = useRouter()
 const authStore = useAuthStore()
 const { user, displayName } = storeToRefs(authStore)
+const tasksDrawer = useTasksDrawer()
 
 const collapsed = ref(false)
 
@@ -30,16 +35,25 @@ const selectedKeys = computed(() => {
 
 const openKeys = ref(['main'])
 
-const menuItems = [
+const menuItems = computed(() => {
-  { key: 'overview', icon: () => h(DashboardOutlined), label: '概览' },
+  const items = [
-  { key: 'documents', icon: () => h(FileTextOutlined), label: '文档管理' },
+    { key: 'overview', icon: () => h(DashboardOutlined), label: '概览与检索' },
-  { key: 'ingest', icon: () => h(UploadOutlined), label: '文档入库' },
+    { key: 'library', icon: () => h(AppstoreOutlined), label: '知识库' },
-  { key: 'search', icon: () => h(SearchOutlined), label: '检索测试台' },
+    { key: 'tasks', icon: () => h(ClockCircleOutlined), label: '入库进度' },
-  { key: 'categories', icon: () => h(AppstoreOutlined), label: '类目列表' },
+    { key: 'settings', icon: () => h(SettingOutlined), label: '设置' },
-  { key: 'settings', icon: () => h(SettingOutlined), label: '设置' }
+    { key: 'api-docs', icon: () => h(ApiOutlined), label: 'API 说明' }
-]
+  ]
+  if (authStore.isAdmin) {
+    items.push({ key: 'users', icon: () => h(TeamOutlined), label: '用户管理' })
+  }
+  return items
+})
 
 function handleMenuClick({ key }) {
+  if (key === 'tasks') {
+    tasksDrawer.open()
+    return
+  }
   if (key && key !== route.name) {
     router.push({ name: key })
   }
@@ -57,6 +71,68 @@ function handleLogout() {
     }
   })
 }
+
+/* ---------- 强制改密弹窗（must_change_password 用户） ---------- */
+const passwordModalVisible = ref(false)
+const passwordSubmitting = ref(false)
+const passwordFormRef = ref(null)
+const passwordForm = reactive({
+  old_password: '',
+  new_password: '',
+  confirm_password: ''
+})
+
+const passwordRules = {
+  old_password: [{ required: true, message: '请输入当前密码', trigger: 'blur' }],
+  new_password: [
+    { required: true, message: '请输入新密码', trigger: 'blur' },
+    { min: 8, message: '密码至少 8 位', trigger: 'blur' }
+  ],
+  confirm_password: [
+    { required: true, message: '请确认新密码', trigger: 'blur' },
+    {
+      validator: (_rule, value) =>
+        value === passwordForm.new_password
+          ? Promise.resolve()
+          : Promise.reject(new Error('两次输入的密码不一致')),
+      trigger: 'blur'
+    }
+  ]
+}
+
+// must_change_password 为 true 时自动弹出改密弹窗（不可关闭）
+watch(
+  () => user.value?.must_change_password,
+  (val) => {
+    passwordModalVisible.value = val === true
+  },
+  { immediate: true }
+)
+
+async function handleSubmitPassword() {
+  try {
+    await passwordFormRef.value.validate()
+  } catch {
+    return
+  }
+  passwordSubmitting.value = true
+  try {
+    await callApi(
+      () => changeMyPassword(passwordForm.old_password, passwordForm.new_password),
+      { errorText: '修改密码失败' }
+    )
+    authStore.markPasswordChanged()
+    message.success('密码修改成功')
+    passwordModalVisible.value = false
+    passwordForm.old_password = ''
+    passwordForm.new_password = ''
+    passwordForm.confirm_password = ''
+  } catch {
+    // 错误已由 callApi 弹出
+  } finally {
+    passwordSubmitting.value = false
+  }
+}
 </script>
 
 <template>
@@ -118,6 +194,58 @@ function handleLogout() {
         QMDSearch Admin · 当前用户：{{ user?.username || '-' }}
       </a-layout-footer>
     </a-layout>
+
+    <!-- 入库进度右侧 Drawer -->
+    <TasksDrawer />
+
+    <!-- 强制改密弹窗（must_change_password 用户，不可关闭） -->
+    <a-modal
+      v-model:open="passwordModalVisible"
+      :mask-closable="false"
+      :closable="false"
+      :keyboard="false"
+      :destroy-on-close="true"
+      title="首次登录须修改密码"
+      :confirm-loading="passwordSubmitting"
+      ok-text="确认修改"
+      :cancel-button-props="{ style: { display: 'none' } }"
+      @ok="handleSubmitPassword"
+    >
+      <a-alert
+        type="warning"
+        show-icon
+        message="检测到首次登录或管理员重置密码，请立即修改密码后方可使用系统。"
+        style="margin-bottom: 16px"
+      />
+      <a-form
+        ref="passwordFormRef"
+        :model="passwordForm"
+        :rules="passwordRules"
+        layout="vertical"
+      >
+        <a-form-item label="当前密码" name="old_password">
+          <a-input-password
+            v-model:value="passwordForm.old_password"
+            placeholder="请输入当前密码"
+            autocomplete="current-password"
+          />
+        </a-form-item>
+        <a-form-item label="新密码" name="new_password">
+          <a-input-password
+            v-model:value="passwordForm.new_password"
+            placeholder="至少 8 位"
+            autocomplete="new-password"
+          />
+        </a-form-item>
+        <a-form-item label="确认新密码" name="confirm_password">
+          <a-input-password
+            v-model:value="passwordForm.confirm_password"
+            placeholder="再次输入新密码"
+            autocomplete="new-password"
+          />
+        </a-form-item>
+      </a-form>
+    </a-modal>
   </a-layout>
 </template>
 

frontend/src/router/index.js

@@ -18,37 +18,31 @@ const routes = [
         path: 'overview',
         name: 'overview',
         component: () => import('@/views/Overview.vue'),
-        meta: { title: '概览', requiresAuth: true }
+        meta: { title: '概览与检索', requiresAuth: true }
       },
       {
-        path: 'documents',
+        path: 'library',
-        name: 'documents',
+        name: 'library',
-        component: () => import('@/views/Documents.vue'),
+        component: () => import('@/views/Library.vue'),
-        meta: { title: '文档管理', requiresAuth: true }
+        meta: { title: '知识库', requiresAuth: true }
-      },
-      {
-        path: 'ingest',
-        name: 'ingest',
-        component: () => import('@/views/Ingest.vue'),
-        meta: { title: '文档入库', requiresAuth: true }
-      },
-      {
-        path: 'search',
-        name: 'search',
-        component: () => import('@/views/Search.vue'),
-        meta: { title: '检索测试台', requiresAuth: true }
-      },
-      {
-        path: 'categories',
-        name: 'categories',
-        component: () => import('@/views/Categories.vue'),
-        meta: { title: '类目列表', requiresAuth: true }
       },
       {
         path: 'settings',
         name: 'settings',
         component: () => import('@/views/Settings.vue'),
         meta: { title: '设置', requiresAuth: true }
+      },
+      {
+        path: 'api-docs',
+        name: 'api-docs',
+        component: () => import('@/views/ApiDocs.vue'),
+        meta: { title: 'API 说明', requiresAuth: true }
+      },
+      {
+        path: 'users',
+        name: 'users',
+        component: () => import('@/views/Users.vue'),
+        meta: { title: '用户管理', requiresAuth: true, requiresAdmin: true }
       }
     ]
   },
@@ -81,6 +75,10 @@ router.beforeEach((to) => {
     return { name: 'login', query: { redirect: to.fullPath } }
   }
 
+  if (to.meta?.requiresAdmin && !authStore.isAdmin) {
+    return { name: 'overview' }
+  }
+
   if (to.name === 'login' && authStore.isAuthenticated) {
     return { name: 'overview' }
   }

frontend/src/stores/useAuthStore.js

@@ -39,13 +39,27 @@ export const useAuthStore = defineStore('auth', {
   actions: {
     /**
      * 登录成功后保存 token + user
-     * @param {{access_token:string, user:object}} data
+     * 兼容后端两种返回：
+     *  - 新格式 { token, username, role, must_change_password }
+     *  - 旧格式 { access_token, user: { username, role, ... } }
+     * @param {object} data
      */
     setAuth(data) {
-      this.token = data.access_token
+      const token = data?.access_token ?? data?.token
-      this.user = data.user
+      const user =
-      localStorage.setItem(TOKEN_STORAGE_KEY, data.access_token)
+        data?.user ??
-      localStorage.setItem(USER_STORAGE_KEY, JSON.stringify(data.user))
+        (data
+          ? {
+              username: data.username,
+              role: data.role,
+              must_change_password: data.must_change_password ?? false
+            }
+          : null)
+      if (!token || !user) return
+      this.token = token
+      this.user = user
+      localStorage.setItem(TOKEN_STORAGE_KEY, token)
+      localStorage.setItem(USER_STORAGE_KEY, JSON.stringify(user))
     },
 
     /** 清除登录态（登出 / 401） */
@@ -56,6 +70,13 @@ export const useAuthStore = defineStore('auth', {
       localStorage.removeItem(USER_STORAGE_KEY)
     },
 
+    /** 改密成功后更新本地用户状态（清除 must_change_password 标记） */
+    markPasswordChanged() {
+      if (!this.user) return
+      this.user = { ...this.user, must_change_password: false }
+      localStorage.setItem(USER_STORAGE_KEY, JSON.stringify(this.user))
+    },
+
     /**
      * 登出
      */

frontend/src/utils/format.js

@@ -10,6 +10,20 @@ export function truncate(text, maxLen = 80) {
   return s.length > maxLen ? `${s.slice(0, maxLen)}…` : s
 }
 
+/**
+ * 格式化字节数为人类可读大小
+ * @param {number|null|undefined} bytes
+ * @returns {string}
+ */
+export function formatSize(bytes) {
+  if (bytes == null || bytes === '') return '-'
+  const n = Number(bytes)
+  if (Number.isNaN(n)) return '-'
+  if (n < 1024) return `${n} B`
+  if (n < 1024 * 1024) return `${(n / 1024).toFixed(1)} KB`
+  return `${(n / 1024 / 1024).toFixed(1)} MB`
+}
+
 /**
  * 安全拼接类名
  * @param  {...(string | false | null | undefined)} args

frontend/src/views/ApiDocs.vue

@@ -0,0 +1,234 @@
+<script setup>
+import { computed } from 'vue'
+import { ApiOutlined, SafetyCertificateOutlined, CodeOutlined, DownloadOutlined } from '@ant-design/icons-vue'
+
+// 以下内容整理自项目 README.md 的「API 文档」章节
+const apiList = [
+  { method: 'GET', path: '/api/v1/health', desc: '健康检查', auth: false },
+  { method: 'POST', path: '/api/v1/auth/register', desc: '用户注册（可关闭）', auth: false },
+  { method: 'POST', path: '/api/v1/auth/login', desc: '用户登录，返回 JWT token', auth: false },
+  { method: 'GET', path: '/api/v1/auth/me', desc: '获取当前用户信息', auth: true },
+  { method: 'POST', path: '/api/v1/search', desc: '分层检索', auth: true },
+  { method: 'POST', path: '/api/v1/documents', desc: '文档入库（JSON 文本，202 异步入库）', auth: true },
+  { method: 'POST', path: '/api/v1/documents/upload', desc: '文件上传入库（multipart，202 异步）', auth: true },
+  { method: 'GET', path: '/api/v1/documents/tasks/{task_id}', desc: '入库任务状态查询', auth: true },
+  { method: 'GET', path: '/api/v1/documents', desc: '文档列表（分页）', auth: true },
+  { method: 'GET', path: '/api/v1/documents/{doc_id}', desc: '文档详情', auth: true },
+  { method: 'DELETE', path: '/api/v1/documents/{doc_id}', desc: '删除文档（幂等）', auth: true },
+  { method: 'GET', path: '/api/v1/knowledge/categories', desc: '知识分类类目集', auth: true },
+  { method: 'GET', path: '/api/v1/knowledge/stats', desc: '统计（四层点数 + 类目分布）', auth: true },
+  { method: 'GET', path: '/admin', desc: '管理后台（本页）', auth: false }
+]
+
+const methodColor = {
+  GET: 'green',
+  POST: 'blue',
+  DELETE: 'red',
+  PUT: 'orange'
+}
+
+const baseUrl = computed(() => `${window.location.origin}/admin/`.replace(/\/admin\/$/, ''))
+
+const agentSkillUrl = computed(() => `${window.location.origin}/agent-skill`)
+
+const loginExample = `curl -X POST ${baseUrl.value}/api/v1/auth/login \\
+  -H "Content-Type: application/json" \\
+  -d '{"username": "admin", "password": "your-password"}'`
+
+const searchExample = `curl -X POST ${baseUrl.value}/api/v1/search \\
+  -H "Content-Type: application/json" \\
+  -H "Authorization: Bearer <token>" \\
+  -d '{"query": "如何配置 Redis 缓存", "top_k": 5}'`
+
+const uploadExample = `curl -X POST ${baseUrl.value}/api/v1/documents/upload \\
+  -H "Authorization: Bearer <token>" \\
+  -F "file=@document.pdf"`
+</script>
+
+<template>
+  <div class="apidocs page-section">
+    <div class="apidocs__header">
+      <h2 class="page-title">
+        <ApiOutlined /> API 使用说明
+      </h2>
+      <span class="text-muted">接口前缀：{{ baseUrl }}/api/v1</span>
+    </div>
+
+    <a-alert
+      class="apidocs__tip"
+      type="info"
+      show-icon
+      message="需要鉴权的接口请在请求头携带 Authorization: Bearer &lt;token&gt;，token 通过 /auth/login 获取。"
+    />
+
+    <!-- AI Agent Skill 下载 -->
+    <section class="apidocs__block apidocs__skill">
+      <h3 class="section-subtitle">
+        <DownloadOutlined /> AI Agent Skill 下载
+      </h3>
+      <p class="apidocs__p">
+        为 AI Agent 提供的接入包（含 <code>SKILL.md</code> 与 Python 客户端示例），覆盖文档上传（文本 / 文件）与分层检索。
+        下载后解压到 Agent 的 skills 目录即可启用。
+      </p>
+      <a :href="agentSkillUrl" target="_blank" rel="noopener">
+        <a-button type="primary">
+          <template #icon><DownloadOutlined /></template>
+          下载 QMDSearch-Agent-Skill.zip
+        </a-button>
+      </a>
+    </section>
+
+    <!-- 统一响应格式 -->
+    <section class="apidocs__block">
+      <h3 class="section-subtitle">
+        <CodeOutlined /> 统一响应格式
+      </h3>
+      <ul class="apidocs__ul">
+        <li>所有接口返回统一 JSON 结构：<code>code</code> / <code>data</code> / <code>message</code>。</li>
+        <li>错误码：<code>0</code> 成功，<code>1xxx</code> 客户端错误，<code>2xxx</code> 服务端错误。</li>
+      </ul>
+      <pre class="code-block">{
+  "code": 0,
+  "data": { ... },
+  "message": "ok"
+}</pre>
+    </section>
+
+    <!-- API 列表 -->
+    <section class="apidocs__block">
+      <h3 class="section-subtitle">接口列表</h3>
+      <a-table
+        :columns="[
+          { title: '方法', dataIndex: 'method', key: 'method', width: 90, align: 'center' },
+          { title: '路径', dataIndex: 'path', key: 'path', width: 320, ellipsis: true },
+          { title: '说明', dataIndex: 'desc', key: 'desc' },
+          { title: '认证', dataIndex: 'auth', key: 'auth', width: 80, align: 'center' }
+        ]"
+        :data-source="apiList"
+        :pagination="false"
+        row-key="path"
+        size="middle"
+      >
+        <template #bodyCell="{ column, text, record }">
+          <template v-if="column.key === 'method'">
+            <a-tag :color="methodColor[record.method]">{{ record.method }}</a-tag>
+          </template>
+          <template v-else-if="column.key === 'path'">
+            <code class="api-path">{{ text }}</code>
+          </template>
+          <template v-else-if="column.key === 'auth'">
+            <a-tag :color="record.auth ? 'volcano' : 'default'">
+              {{ record.auth ? '需鉴权' : '否' }}
+            </a-tag>
+          </template>
+        </template>
+      </a-table>
+    </section>
+
+    <!-- 认证说明 -->
+    <section class="apidocs__block">
+      <h3 class="section-subtitle">
+        <SafetyCertificateOutlined /> 获取 Token
+      </h3>
+      <p class="apidocs__p">先调用登录接口获取 JWT，再在后续请求的 Header 中携带：</p>
+      <pre class="code-block">{{ loginExample }}</pre>
+    </section>
+
+    <!-- 检索示例 -->
+    <section class="apidocs__block">
+      <h3 class="section-subtitle">
+        <CodeOutlined /> 检索示例
+      </h3>
+      <pre class="code-block">{{ searchExample }}</pre>
+    </section>
+
+    <!-- 文件上传示例 -->
+    <section class="apidocs__block">
+      <h3 class="section-subtitle">
+        <CodeOutlined /> 文件上传示例
+      </h3>
+      <pre class="code-block">{{ uploadExample }}</pre>
+    </section>
+  </div>
+</template>
+
+<style scoped>
+.apidocs__header {
+  display: flex;
+  align-items: baseline;
+  justify-content: space-between;
+  flex-wrap: wrap;
+  gap: 8px;
+  margin-bottom: 16px;
+}
+
+.page-title :deep(.anticon) {
+  margin-right: 8px;
+  color: #1677ff;
+}
+
+.apidocs__tip {
+  margin-bottom: 20px;
+}
+
+.apidocs__skill {
+  padding: 16px 18px;
+  border: 1px solid rgba(22, 119, 255, 0.25);
+  border-radius: 10px;
+  background: rgba(22, 119, 255, 0.04);
+}
+
+.apidocs__block {
+  margin-bottom: 24px;
+}
+
+.section-subtitle :deep(.anticon) {
+  margin-right: 6px;
+  color: #1677ff;
+}
+
+.apidocs__ul {
+  margin: 0 0 12px;
+  padding-left: 20px;
+  color: #4b5563;
+  font-size: 13px;
+  line-height: 1.9;
+}
+
+.apidocs__ul code,
+.api-path {
+  background: #f3f4f6;
+  border: 1px solid #e5e7eb;
+  border-radius: 4px;
+  padding: 1px 6px;
+  font-family: 'SFMono-Regular', Consolas, 'Liberation Mono', Menlo, monospace;
+  font-size: 12px;
+  color: #c0341d;
+}
+
+.api-path {
+  white-space: nowrap;
+  color: #1677ff;
+  background: rgba(22, 119, 255, 0.08);
+  border-color: rgba(22, 119, 255, 0.2);
+}
+
+.apidocs__p {
+  color: #4b5563;
+  font-size: 13px;
+  margin: 0 0 10px;
+}
+
+.code-block {
+  background: #0f172a;
+  color: #e2e8f0;
+  border-radius: 8px;
+  padding: 16px 18px;
+  overflow-x: auto;
+  font-family: 'SFMono-Regular', Consolas, 'Liberation Mono', Menlo, monospace;
+  font-size: 12.5px;
+  line-height: 1.7;
+  margin: 0;
+  white-space: pre;
+}
+</style>

frontend/src/views/Categories.vue

@@ -1,70 +0,0 @@
-<script setup>
-import { onMounted, ref } from 'vue'
-import { message } from 'ant-design-vue'
-import { categories as fetchCategories } from '@/api/knowledge'
-
-const isLoading = ref(false)
-const categoriesData = ref([])
-
-const columns = [
-  { title: '名称', dataIndex: 'name', key: 'name' },
-  { title: '描述', dataIndex: 'description', key: 'description' }
-]
-
-async function loadCategories() {
-  isLoading.value = true
-  try {
-    const data = await fetchCategories()
-    categoriesData.value = data.categories || []
-  } catch (err) {
-    message.error(err?.message || '加载类目列表失败')
-  } finally {
-    isLoading.value = false
-  }
-}
-
-onMounted(() => {
-  loadCategories()
-})
-</script>
-
-<template>
-  <div class="categories page-section">
-    <div class="categories__header">
-      <h2 class="categories__title">类目列表</h2>
-      <a-button :loading="isLoading" @click="loadCategories">刷新</a-button>
-    </div>
-
-    <a-table
-      :columns="columns"
-      :data-source="categoriesData"
-      :pagination="false"
-      :loading="isLoading"
-      row-key="name"
-      size="small"
-    >
-      <template #bodyCell="{ column, record }">
-        <template v-if="column.key === 'name'">
-          <a-tag color="blue">{{ record.name }}</a-tag>
-        </template>
-        <template v-else-if="column.key === 'description'">
-          <span>{{ record.description || '-' }}</span>
-        </template>
-      </template>
-    </a-table>
-  </div>
-</template>
-
-<style scoped>
-.categories__header {
-  display: flex;
-  align-items: center;
-  justify-content: space-between;
-  margin-bottom: 12px;
-}
-
-.categories__title {
-  font-size: 16px;
-  margin: 0;
-}
-</style>

frontend/src/views/Documents.vue

@@ -1,299 +0,0 @@
-<script setup>
-import { onMounted, reactive, ref } from 'vue'
-import { message, Modal } from 'ant-design-vue'
-import {
-  list as fetchDocuments,
-  detail as fetchDocumentDetail,
-  remove as deleteDocument
-} from '@/api/documents'
-import { truncate } from '@/utils/format'
-
-const isLoading = ref(false)
-const isLoadingDetail = ref(false)
-const isDeleting = ref(false)
-
-const documents = ref([])
-const nextOffset = ref(null)
-
-const detailVisible = ref(false)
-const detailData = ref(null)
-
-const statusText = reactive({
-  text: ''
-})
-
-function updateStatusText() {
-  statusText.text =
-    nextOffset.value === null ? '已加载全部' : '还有更多，可继续加载'
-}
-
-async function loadDocuments(reset = false) {
-  isLoading.value = true
-  try {
-    const offset = reset ? null : nextOffset.value
-    const data = await fetchDocuments(20, offset)
-    if (reset) {
-      documents.value = data.items || []
-    } else {
-      documents.value = documents.value.concat(data.items || [])
-    }
-    nextOffset.value = data.next_offset ?? null
-    updateStatusText()
-  } catch (err) {
-    message.error(err?.message || '加载文档列表失败')
-  } finally {
-    isLoading.value = false
-  }
-}
-
-async function handleLoadMore() {
-  await loadDocuments(false)
-}
-
-function handleViewDetail(docId) {
-  detailVisible.value = true
-  detailData.value = null
-  loadDocDetail(docId)
-}
-
-async function loadDocDetail(docId) {
-  isLoadingDetail.value = true
-  try {
-    detailData.value = await fetchDocumentDetail(docId)
-  } catch (err) {
-    message.error(err?.message || '加载文档详情失败')
-  } finally {
-    isLoadingDetail.value = false
-  }
-}
-
-function handleDelete(doc) {
-  const title = doc.title || doc.doc_id
-  Modal.confirm({
-    title: '确认删除',
-    content: `确定删除文档「${title}」(${doc.doc_id}) 吗？该操作将删除四层集合中的全部数据，不可恢复。`,
-    okText: '删除',
-    okType: 'danger',
-    cancelText: '取消',
-    async onOk() {
-      isDeleting.value = true
-      try {
-        const data = await deleteDocument(doc.doc_id)
-        message.success(`已删除 ${data.deleted_total ?? 0} 条数据`)
-        detailVisible.value = false
-        await loadDocuments(true)
-      } catch (err) {
-        message.error(err?.message || '删除文档失败')
-      } finally {
-        isDeleting.value = false
-      }
-    }
-  })
-}
-
-function handleCloseDetail() {
-  detailVisible.value = false
-  detailData.value = null
-}
-
-const columns = [
-  { title: '标题', dataIndex: 'title', key: 'title', ellipsis: true },
-  { title: '类目', dataIndex: 'category', key: 'category', width: 140 },
-  { title: '标签', dataIndex: 'tags', key: 'tags', width: 220 },
-  { title: 'L1 摘要', dataIndex: 'summary', key: 'summary', ellipsis: true },
-  { title: '操作', key: 'action', width: 160, fixed: 'right' }
-]
-
-function getDocTitle(record) {
-  return record?.title || '(无标题)'
-}
-
-onMounted(() => {
-  loadDocuments(true)
-})
-</script>
-
-<template>
-  <div class="documents page-section">
-    <div class="documents__header">
-      <h2 class="documents__title">文档管理</h2>
-    </div>
-
-    <a-table
-      :columns="columns"
-      :data-source="documents"
-      :pagination="false"
-      :loading="isLoading"
-      row-key="doc_id"
-      size="small"
-      :scroll="{ x: 900 }"
-    >
-      <template #bodyCell="{ column, record }">
-        <template v-if="column.key === 'title'">
-          <span :title="record.title">{{ getDocTitle(record) }}</span>
-        </template>
-        <template v-else-if="column.key === 'category'">
-          <a-tag v-if="record.category" color="blue">{{ record.category }}</a-tag>
-          <span v-else class="text-muted">-</span>
-        </template>
-        <template v-else-if="column.key === 'tags'">
-          <template v-if="record.tags && record.tags.length">
-            <a-tag v-for="tag in record.tags" :key="tag">{{ tag }}</a-tag>
-          </template>
-          <span v-else class="text-muted">-</span>
-        </template>
-        <template v-else-if="column.key === 'summary'">
-          <span :title="record.summary">{{ truncate(record.summary, 80) }}</span>
-        </template>
-        <template v-else-if="column.key === 'action'">
-          <a-space>
-            <a-button type="link" size="small" @click="handleViewDetail(record.doc_id)">
-              详情
-            </a-button>
-            <a-button
-              type="link"
-              size="small"
-              danger
-              :loading="isDeleting"
-              @click="handleDelete(record)"
-            >
-              删除
-            </a-button>
-          </a-space>
-        </template>
-      </template>
-    </a-table>
-
-    <div class="toolbar documents__toolbar">
-      <a-button
-        :loading="isLoading"
-        :disabled="nextOffset === null"
-        @click="handleLoadMore"
-      >
-        加载更多
-      </a-button>
-      <span class="text-muted" style="margin-left: 12px">{{ statusText.text }}</span>
-    </div>
-
-    <a-drawer
-      :open="detailVisible"
-      title="文档详情"
-      placement="right"
-      width="640"
-      :destroy-on-close="true"
-      @close="handleCloseDetail"
-    >
-      <a-spin :spinning="isLoadingDetail">
-        <div v-if="detailData">
-          <a-descriptions :column="1" size="small" bordered>
-            <a-descriptions-item label="doc_id">
-              {{ detailData.l1?.doc_id || '-' }}
-            </a-descriptions-item>
-            <a-descriptions-item label="标题">
-              {{ detailData.l1?.title || '-' }}
-            </a-descriptions-item>
-            <a-descriptions-item label="类目">
-              {{ detailData.l1?.category || '-' }}
-            </a-descriptions-item>
-            <a-descriptions-item label="标签">
-              <template v-if="detailData.l1?.tags && detailData.l1.tags.length">
-                <a-tag v-for="tag in detailData.l1.tags" :key="tag">{{ tag }}</a-tag>
-              </template>
-              <span v-else class="text-muted">-</span>
-            </a-descriptions-item>
-            <a-descriptions-item label="chunks_count">
-              {{ detailData.chunks_count ?? 0 }}
-            </a-descriptions-item>
-          </a-descriptions>
-
-          <h3 class="documents__section-title">L1 全文</h3>
-          <pre class="documents__pre">{{ detailData.l1?.text || '' }}</pre>
-
-          <h3 class="documents__section-title">
-            L2 节点（{{ (detailData.l2_nodes || []).length }}）
-          </h3>
-          <div v-if="(detailData.l2_nodes || []).length === 0" class="text-muted">无</div>
-          <div
-            v-for="(node, idx) in detailData.l2_nodes || []"
-            :key="`l2-${idx}`"
-            class="documents__node"
-          >
-            <div class="documents__node-path">{{ node.section_path || '' }}</div>
-            <div class="documents__node-text text-break">{{ node.text || '' }}</div>
-          </div>
-
-          <h3 class="documents__section-title">
-            L3 节点（{{ (detailData.l3_nodes || []).length }}）
-          </h3>
-          <div v-if="(detailData.l3_nodes || []).length === 0" class="text-muted">无</div>
-          <div
-            v-for="(node, idx) in detailData.l3_nodes || []"
-            :key="`l3-${idx}`"
-            class="documents__node"
-          >
-            <div class="documents__node-path">{{ node.section_path || '' }}</div>
-            <div class="documents__node-text text-break">{{ node.text || '' }}</div>
-          </div>
-        </div>
-        <div v-else-if="!isLoadingDetail" class="text-muted">暂无数据</div>
-      </a-spin>
-    </a-drawer>
-  </div>
-</template>
-
-<style scoped>
-.documents__header {
-  display: flex;
-  align-items: center;
-  justify-content: space-between;
-  margin-bottom: 12px;
-}
-
-.documents__title {
-  font-size: 16px;
-  margin: 0;
-}
-
-.documents__toolbar {
-  display: flex;
-  align-items: center;
-}
-
-.documents__section-title {
-  font-size: 14px;
-  margin: 16px 0 8px;
-  color: #374151;
-}
-
-.documents__pre {
-  background: #f9fafb;
-  border: 1px solid #e5e7eb;
-  border-radius: 4px;
-  padding: 8px;
-  white-space: pre-wrap;
-  word-break: break-word;
-  font-size: 12px;
-  max-height: 260px;
-  overflow: auto;
-  margin: 0;
-}
-
-.documents__node {
-  border-left: 3px solid #93c5fd;
-  padding: 6px 10px;
-  margin-bottom: 6px;
-  background: #f8fafc;
-  border-radius: 0 4px 4px 0;
-}
-
-.documents__node-path {
-  font-size: 12px;
-  color: #6b7280;
-  margin-bottom: 4px;
-}
-
-.documents__node-text {
-  font-size: 13px;
-  color: #1f2937;
-}
-</style>

frontend/src/views/Ingest.vue

@@ -1,352 +0,0 @@
-<script setup>
-import { computed, reactive, ref } from 'vue'
-import { message } from 'ant-design-vue'
-import { ingest as ingestDocument, upload as uploadDocument } from '@/api/documents'
-import { useIngestPolling } from '@/composables/useIngestPolling'
-import {
-  INGEST_STATUS_TEXT,
-  INGEST_STATUS_COLOR
-} from '@/constants/ingest'
-
-const { state: pollState, startPolling, stopPolling } = useIngestPolling()
-
-const activeTab = ref('text')
-
-const isSubmittingText = ref(false)
-const isSubmittingFile = ref(false)
-
-const textFormRef = ref(null)
-const textForm = reactive({
-  title: '',
-  source: '',
-  text: ''
-})
-
-const textRules = {
-  title: [{ required: true, message: '请输入标题', trigger: 'blur' }],
-  text: [{ required: true, message: '请输入正文', trigger: 'blur' }]
-}
-
-const fileForm = reactive({
-  title: '',
-  source: ''
-})
-const fileList = ref([])
-const rawFile = ref(null)
-
-const ACCEPTED_EXTENSIONS = '.txt,.md,.html,.htm,.pdf,.docx'
-
-const statusBadgeColor = computed(() => {
-  return INGEST_STATUS_COLOR[pollState.status] || 'default'
-})
-
-const statusBadgeText = computed(() => {
-  return INGEST_STATUS_TEXT[pollState.status] || pollState.status || '-'
-})
-
-const isTerminal = computed(() =>
-  ['done', 'failed'].includes(pollState.status)
-)
-
-const resultData = computed(() => pollState.task?.result || null)
-const errorData = computed(() => pollState.task?.error || pollState.error || null)
-const summaryData = computed(() => resultData.value?.summary || {})
-const partialSummary = computed(() => errorData.value?.partial_summary || null)
-
-function handleFileChange(file) {
-  // a-upload before-upload 返回 false 表示不自动上传；保存原始 File 供 FormData 使用
-  rawFile.value = file
-  fileList.value = [file]
-  return false
-}
-
-function handleFileRemove() {
-  rawFile.value = null
-  fileList.value = []
-}
-
-async function handleSubmitText() {
-  try {
-    await textFormRef.value.validate()
-  } catch {
-    return
-  }
-  isSubmittingText.value = true
-  try {
-    const payload = {
-      title: textForm.title,
-      text: textForm.text
-    }
-    if (textForm.source) {
-      payload.source = textForm.source
-    }
-    const data = await ingestDocument(payload)
-    message.success('任务已提交')
-    startPolling(data.task_id)
-  } catch (err) {
-    message.error(err?.message || '提交入库失败')
-  } finally {
-    isSubmittingText.value = false
-  }
-}
-
-async function handleSubmitFile() {
-  if (!rawFile.value) {
-    message.warning('请选择文件')
-    return
-  }
-  isSubmittingFile.value = true
-  try {
-    const formData = new FormData()
-    formData.append('file', rawFile.value)
-    if (fileForm.title) {
-      formData.append('title', fileForm.title)
-    }
-    if (fileForm.source) {
-      formData.append('source', fileForm.source)
-    }
-    const data = await uploadDocument(formData)
-    message.success('任务已提交')
-    startPolling(data.task_id)
-  } catch (err) {
-    message.error(err?.message || '上传入库失败')
-  } finally {
-    isSubmittingFile.value = false
-  }
-}
-
-function handleCancelPolling() {
-  stopPolling()
-  message.info('已停止轮询')
-}
-</script>
-
-<template>
-  <div class="ingest page-section">
-    <div class="ingest__header">
-      <h2 class="ingest__title">文档入库</h2>
-    </div>
-
-    <a-tabs v-model:activeKey="activeTab">
-      <a-tab-pane key="text" tab="文本入库">
-        <a-form
-          ref="textFormRef"
-          :model="textForm"
-          :rules="textRules"
-          layout="vertical"
-        >
-          <a-form-item label="标题" name="title">
-            <a-input v-model:value="textForm.title" placeholder="请输入标题" allow-clear />
-          </a-form-item>
-          <a-form-item label="来源" name="source">
-            <a-input
-              v-model:value="textForm.source"
-              placeholder="例如：manual / web / file"
-              allow-clear
-            />
-          </a-form-item>
-          <a-form-item label="正文" name="text">
-            <a-textarea
-              v-model:value="textForm.text"
-              placeholder="请输入文档正文"
-              :auto-size="{ minRows: 8, maxRows: 18 }"
-            />
-          </a-form-item>
-          <a-form-item>
-            <a-button
-              type="primary"
-              :loading="isSubmittingText || pollState.isPolling"
-              @click="handleSubmitText"
-            >
-              提交入库
-            </a-button>
-          </a-form-item>
-        </a-form>
-      </a-tab-pane>
-
-      <a-tab-pane key="file" tab="文件上传">
-        <a-form
-          :model="fileForm"
-          layout="vertical"
-        >
-          <a-form-item label="选择文件">
-            <a-upload
-              :file-list="fileList"
-              :accept="ACCEPTED_EXTENSIONS"
-              :max-count="1"
-              :before-upload="handleFileChange"
-              @remove="handleFileRemove"
-            >
-              <a-button :disabled="fileList.length >= 1">选择文件</a-button>
-            </a-upload>
-            <div class="text-muted" style="margin-top: 4px">
-              支持 .txt/.md/.html/.htm/.pdf/.docx
-            </div>
-          </a-form-item>
-          <a-form-item label="标题（可选，默认取文件名）" name="title">
-            <a-input
-              v-model:value="fileForm.title"
-              placeholder="留空则使用文件名去扩展"
-              allow-clear
-            />
-          </a-form-item>
-          <a-form-item label="来源（可选，默认 file:原文件名）" name="source">
-            <a-input
-              v-model:value="fileForm.source"
-              placeholder="例如：manual / web"
-              allow-clear
-            />
-          </a-form-item>
-          <a-form-item>
-            <a-button
-              type="primary"
-              :loading="isSubmittingFile || pollState.isPolling"
-              @click="handleSubmitFile"
-            >
-              上传入库
-            </a-button>
-          </a-form-item>
-        </a-form>
-      </a-tab-pane>
-    </a-tabs>
-
-    <div v-if="pollState.taskId" class="ingest__result">
-      <div class="ingest__result-header">
-        <span>任务已提交：{{ pollState.taskId }}</span>
-        <a-tag :color="statusBadgeColor">{{ statusBadgeText }}</a-tag>
-        <a-button
-          v-if="pollState.isPolling"
-          type="link"
-          size="small"
-          @click="handleCancelPolling"
-        >
-          停止轮询
-        </a-button>
-      </div>
-
-      <a-alert
-        v-if="pollState.isTimeout"
-        class="ingest__alert"
-        type="warning"
-        show-icon
-        :message="`任务仍在进行，可稍后凭 task_id 查询：${pollState.taskId}`"
-      />
-
-      <a-alert
-        v-if="pollState.error && !pollState.task"
-        class="ingest__alert"
-        type="error"
-        show-icon
-        :message="pollState.error?.message || '轮询失败'"
-      />
-
-      <div v-if="isTerminal && pollState.status === 'done' && resultData" class="ingest__done">
-        <a-descriptions :column="1" size="small" bordered>
-          <a-descriptions-item label="document_id">
-            {{ resultData.document_id || '-' }}
-          </a-descriptions-item>
-          <a-descriptions-item label="类目">
-            {{ resultData.category || '-' }}（置信度 {{ resultData.category_confidence ?? '-' }}）
-          </a-descriptions-item>
-          <a-descriptions-item label="标签">
-            <template v-if="resultData.tags && resultData.tags.length">
-              <a-tag v-for="tag in resultData.tags" :key="tag">{{ tag }}</a-tag>
-            </template>
-            <span v-else class="text-muted">-</span>
-          </a-descriptions-item>
-          <a-descriptions-item label="总结层级">
-            {{ summaryData.level ?? '-' }}
-          </a-descriptions-item>
-          <a-descriptions-item label="写入集合">
-            {{ resultData.collection || '-' }}
-          </a-descriptions-item>
-          <a-descriptions-item label="chunks_count">
-            {{ resultData.chunks_count ?? 0 }}
-          </a-descriptions-item>
-        </a-descriptions>
-        <h3 class="ingest__section-title">L1 总结</h3>
-        <pre class="ingest__pre">{{ summaryData.l1_summary || '' }}</pre>
-      </div>
-
-      <div v-if="isTerminal && pollState.status === 'failed'" class="ingest__failed">
-        <a-descriptions :column="1" size="small" bordered>
-          <a-descriptions-item label="失败阶段">
-            {{ errorData?.stage || '-' }}
-          </a-descriptions-item>
-          <a-descriptions-item label="错误信息">
-            <span class="text-break">{{ errorData?.message || '-' }}</span>
-          </a-descriptions-item>
-        </a-descriptions>
-        <template v-if="partialSummary && partialSummary.l1_summary">
-          <a-alert
-            class="ingest__alert"
-            type="info"
-            show-icon
-            message="已产出总结保留：任务失败前已生成 L1 摘要"
-          />
-          <h3 class="ingest__section-title">L1 总结</h3>
-          <pre class="ingest__pre">{{ partialSummary.l1_summary }}</pre>
-        </template>
-      </div>
-    </div>
-  </div>
-</template>
-
-<style scoped>
-.ingest__header {
-  display: flex;
-  align-items: center;
-  justify-content: space-between;
-  margin-bottom: 12px;
-}
-
-.ingest__title {
-  font-size: 16px;
-  margin: 0;
-}
-
-.ingest__result {
-  margin-top: 16px;
-  border: 1px solid #e5e7eb;
-  border-radius: 6px;
-  background: #fafafa;
-  padding: 12px 16px;
-}
-
-.ingest__result-header {
-  display: flex;
-  align-items: center;
-  gap: 8px;
-  flex-wrap: wrap;
-  font-size: 13px;
-  margin-bottom: 8px;
-}
-
-.ingest__alert {
-  margin: 8px 0;
-}
-
-.ingest__done,
-.ingest__failed {
-  margin-top: 8px;
-}
-
-.ingest__section-title {
-  font-size: 14px;
-  margin: 12px 0 8px;
-  color: #374151;
-}
-
-.ingest__pre {
-  background: #f9fafb;
-  border: 1px solid #e5e7eb;
-  border-radius: 4px;
-  padding: 8px;
-  white-space: pre-wrap;
-  word-break: break-word;
-  font-size: 12px;
-  max-height: 260px;
-  overflow: auto;
-  margin: 0;
-}
-</style>

frontend/src/views/Library.vue

@@ -0,0 +1,815 @@
+<script setup>
+import { computed, onMounted, reactive, ref } from 'vue'
+import { message, Modal } from 'ant-design-vue'
+import {
+  FolderOutlined,
+  FileTextOutlined,
+  UploadOutlined,
+  ReloadOutlined,
+  SearchOutlined,
+  ClockCircleOutlined,
+  TagsOutlined
+} from '@ant-design/icons-vue'
+import {
+  categories as fetchCategories,
+  stats as fetchStats
+} from '@/api/knowledge'
+import {
+  list as fetchDocuments,
+  detail as fetchDocumentDetail,
+  remove as deleteDocument,
+  reingest as reingestDocument,
+  ingest as ingestDocument,
+  upload as uploadDocument,
+  uploadBatch
+} from '@/api/documents'
+import { useIngestPolling } from '@/composables/useIngestPolling'
+import { useTasksDrawer } from '@/composables/useTasksDrawer'
+import { INGEST_STATUS_TEXT, INGEST_STATUS_COLOR } from '@/constants/ingest'
+
+const tasksDrawer = useTasksDrawer()
+
+/* ---------- 树形目录 ---------- */
+const isLoading = ref(false)
+const treeData = ref([])
+const categoriesData = ref([])
+const categoryDesc = ref(null)
+const selectedCategory = ref(null)
+
+const allDocs = ref([])
+const searchKeyword = ref('')
+
+const UNCATEGORIZED = 'uncategorized'
+
+function formatSize(bytes) {
+  if (bytes == null) return ''
+  if (bytes < 1024) return `${bytes} B`
+  if (bytes < 1024 * 1024) return `${(bytes / 1024).toFixed(1)} KB`
+  return `${(bytes / 1024 / 1024).toFixed(1)} MB`
+}
+
+function buildTree() {
+  const docs = allDocs.value
+  const filtered = searchKeyword.value
+    ? docs.filter((d) => {
+        const kw = searchKeyword.value.toLowerCase()
+        return (
+          (d.title || '').toLowerCase().includes(kw) ||
+          (d.summary || '').toLowerCase().includes(kw) ||
+          (d.category || '').toLowerCase().includes(kw)
+        )
+      })
+    : docs
+
+  // 类目 → 文档映射
+  const byCat = {}
+  filtered.forEach((d) => {
+    const cat = d.category || UNCATEGORIZED
+    if (!byCat[cat]) byCat[cat] = []
+    byCat[cat].push(d)
+  })
+
+  const nodes = []
+  // 定义类目顺序：taxonomy 中定义的类目在前，未定义但存在的类目追加
+  const defined = categoriesData.value.map((c) => c.name)
+  const known = new Set(defined)
+  Object.keys(byCat).forEach((cat) => {
+    if (!known.has(cat)) known.add(cat)
+  })
+  defined.forEach((cat) => {
+    if (byCat[cat]) {
+      nodes.push(makeCategoryNode(cat, byCat[cat]))
+    }
+  })
+  // 未在 taxonomy 中但实际存在的类目
+  Object.keys(byCat).forEach((cat) => {
+    if (!defined.includes(cat)) {
+      nodes.push(makeCategoryNode(cat, byCat[cat]))
+    }
+  })
+  return nodes
+}
+
+function makeCategoryNode(cat, docs) {
+  return {
+    key: `cat:${cat}`,
+    title: cat,
+    icon: () => h(FolderOutlined),
+    isLeaf: docs.length === 0,
+    category: cat,
+    count: docs.length,
+    children: docs.map((d) => ({
+      key: `doc:${d.doc_id}`,
+      title: d.title || '(无标题)',
+      icon: () => h(FileTextOutlined),
+      isLeaf: true,
+      doc: d
+    }))
+  }
+}
+
+function h(type) {
+  return { render: () => _h(type) }
+}
+
+function _h(type) {
+  return type
+}
+
+const treeDataComputed = computed(() => buildTree())
+
+function handleTreeSelect(keys, info) {
+  const node = info.selectedNodes?.[0]
+  if (!node) return
+  const key = node.key
+  if (key.startsWith('cat:')) {
+    selectedCategory.value = node.category
+    categoryDesc.value = categoriesData.value.find((c) => c.name === node.category)?.description || ''
+    closeDocDetail()
+  } else if (key.startsWith('doc:')) {
+    selectedCategory.value = null
+    categoryDesc.value = null
+    openDocDetail(node.doc)
+  }
+}
+
+async function loadAll() {
+  isLoading.value = true
+  try {
+    const [cats, stats] = await Promise.all([fetchCategories(), fetchStats()])
+    categoriesData.value = cats.categories || []
+    // 加载全部文档（分页直到取完）
+    const docs = []
+    let offset = null
+    for (;;) {
+      const page = await fetchDocuments(100, offset)
+      docs.push(...(page.items || []))
+      offset = page.next_offset ?? null
+      if (!offset) break
+    }
+    allDocs.value = docs
+    // 更新类目计数（可选）
+  } catch (err) {
+    message.error(err?.message || '加载知识库失败')
+  } finally {
+    isLoading.value = false
+  }
+}
+
+function handleRefresh() {
+  loadAll()
+}
+
+onMounted(() => {
+  loadAll()
+})
+
+/* ---------- 文档详情 ---------- */
+const docDetailVisible = ref(false)
+const docDetailData = ref(null)
+const isLoadingDetail = ref(false)
+const isDeleting = ref(false)
+const isReingesting = ref(false)
+
+function openDocDetail(doc) {
+  docDetailVisible.value = true
+  docDetailData.value = null
+  loadDocDetail(doc.doc_id)
+}
+
+function closeDocDetail() {
+  docDetailVisible.value = false
+  docDetailData.value = null
+}
+
+async function loadDocDetail(docId) {
+  isLoadingDetail.value = true
+  try {
+    docDetailData.value = await fetchDocumentDetail(docId)
+  } catch (err) {
+    message.error(err?.message || '加载文档详情失败')
+  } finally {
+    isLoadingDetail.value = false
+  }
+}
+
+function handleDelete(doc) {
+  const docId = doc.doc_id
+  const title = doc.title || docId
+  Modal.confirm({
+    title: '确认删除',
+    content: `确定删除文档「${title}」(${docId}) 吗？该操作将删除四层集合中的全部数据，不可恢复。`,
+    okText: '删除',
+    okType: 'danger',
+    cancelText: '取消',
+    async onOk() {
+      isDeleting.value = true
+      try {
+        await deleteDocument(docId)
+        message.success('已删除')
+        closeDocDetail()
+        await loadAll()
+      } catch (err) {
+        message.error(err?.message || '删除文档失败')
+      } finally {
+        isDeleting.value = false
+      }
+    }
+  })
+}
+
+function handleReingest(doc) {
+  const docId = doc.doc_id
+  const title = doc.title || docId
+  Modal.confirm({
+    title: '重新摘要入库',
+    content: `将重新读取文档「${title}」(${docId}) 的原始文件，删除旧数据后重新生成三级总结并入库。生成新的 doc_id。是否继续？`,
+    okText: '重新入库',
+    cancelText: '取消',
+    async onOk() {
+      isReingesting.value = true
+      try {
+        const data = await reingestDocument(docId)
+        message.success(`已提交重新入库任务：${data.task_id}`)
+      } catch (err) {
+        message.error(err?.message || '重新入库失败')
+      } finally {
+        isReingesting.value = false
+      }
+    }
+  })
+}
+
+/* ---------- 入库抽屉 ---------- */
+const ingestVisible = ref(false)
+const activeTab = ref('text')
+const { state: pollState, startPolling, stopPolling } = useIngestPolling()
+
+const textFormRef = ref(null)
+const textForm = reactive({ title: '', source: '', text: '' })
+const textRules = {
+  title: [{ required: true, message: '请输入标题', trigger: 'blur' }],
+  text: [{ required: true, message: '请输入正文', trigger: 'blur' }]
+}
+const isSubmittingText = ref(false)
+const isSubmittingFile = ref(false)
+const isSubmittingBatch = ref(false)
+
+const fileForm = reactive({ title: '', source: '' })
+const fileList = ref([])
+const rawFile = ref(null)
+const batchResult = ref(null)
+const ACCEPTED_EXTENSIONS = '.txt,.md,.html,.htm,.pdf,.docx'
+
+const statusBadgeColor = computed(() => INGEST_STATUS_COLOR[pollState.status] || 'default')
+const statusBadgeText = computed(() => INGEST_STATUS_TEXT[pollState.status] || pollState.status || '-')
+const isTerminal = computed(() => ['done', 'failed'].includes(pollState.status))
+const resultData = computed(() => pollState.task?.result || null)
+const errorData = computed(() => pollState.task?.error || pollState.error || null)
+const summaryData = computed(() => resultData.value?.summary || {})
+const partialSummary = computed(() => errorData.value?.partial_summary || null)
+
+function openIngest() {
+  ingestVisible.value = true
+}
+
+function handleFileChange(file) {
+  rawFile.value = file
+  fileList.value = [file]
+  return false
+}
+
+function handleFileRemove() {
+  rawFile.value = null
+  fileList.value = []
+}
+
+function handleBatchFileChange(_file, fileListArg) {
+  fileList.value = fileListArg.map((f) => f.originFileObj || f)
+  batchResult.value = null
+  return false
+}
+
+function handleBatchRemove(file) {
+  fileList.value = fileList.value.filter((f) => f !== file && f.name !== file.name)
+  batchResult.value = null
+}
+
+function handleBatchClear() {
+  fileList.value = []
+  batchResult.value = null
+}
+
+async function handleSubmitText() {
+  try {
+    await textFormRef.value.validate()
+  } catch {
+    return
+  }
+  isSubmittingText.value = true
+  try {
+    const payload = { title: textForm.title, text: textForm.text }
+    if (textForm.source) payload.source = textForm.source
+    const data = await ingestDocument(payload)
+    message.success('任务已提交')
+    startPolling(data.task_id)
+  } catch (err) {
+    message.error(err?.message || '提交入库失败')
+  } finally {
+    isSubmittingText.value = false
+  }
+}
+
+async function handleSubmitFile() {
+  if (!rawFile.value) {
+    message.warning('请选择文件')
+    return
+  }
+  isSubmittingFile.value = true
+  try {
+    const formData = new FormData()
+    formData.append('file', rawFile.value)
+    if (fileForm.title) formData.append('title', fileForm.title)
+    if (fileForm.source) formData.append('source', fileForm.source)
+    const data = await uploadDocument(formData)
+    message.success('任务已提交')
+    startPolling(data.task_id)
+  } catch (err) {
+    message.error(err?.message || '上传入库失败')
+  } finally {
+    isSubmittingFile.value = false
+  }
+}
+
+async function handleSubmitBatch() {
+  if (!fileList.value.length) {
+    message.warning('请选择文件')
+    return
+  }
+  isSubmittingBatch.value = true
+  try {
+    const data = await uploadBatch(fileList.value)
+    batchResult.value = data
+    const okCount = data.tasks?.length || 0
+    const failCount = data.failed?.length || 0
+    if (okCount > 0) {
+      message.success(`已提交 ${okCount} 个任务${failCount > 0 ? `，${failCount} 个失败` : ''}`)
+    } else if (failCount > 0) {
+      message.error(`全部 ${failCount} 个文件上传失败`)
+    }
+    if (data.tasks?.length === 1 && failCount === 0) {
+      startPolling(data.tasks[0].task_id)
+    }
+  } catch (err) {
+    message.error(err?.message || '批量上传失败')
+  } finally {
+    isSubmittingBatch.value = false
+  }
+}
+
+function handleCancelPolling() {
+  stopPolling()
+  message.info('已停止轮询')
+}
+
+const batchColumns = [
+  { title: '文件名', dataIndex: 'filename', key: 'filename', ellipsis: true },
+  { title: 'task_id', dataIndex: 'task_id', key: 'task_id', width: 240 }
+]
+const failedColumns = [
+  { title: '文件名', dataIndex: 'filename', key: 'filename', ellipsis: true },
+  { title: '错误', dataIndex: 'error', key: 'error' }
+]
+</script>
+
+<template>
+  <div class="library page-section">
+    <div class="library__header">
+      <h2 class="library__title">知识库</h2>
+      <div class="library__actions">
+        <a-input
+          v-model:value="searchKeyword"
+          placeholder="搜索标题 / 摘要 / 类目"
+          allow-clear
+          style="width: 260px"
+        >
+          <template #prefix><SearchOutlined /></template>
+        </a-input>
+        <a-button :loading="isLoading" @click="handleRefresh">
+          <template #icon><ReloadOutlined /></template>
+          刷新
+        </a-button>
+        <a-button @click="tasksDrawer.open()">
+          <template #icon><ClockCircleOutlined /></template>
+          入库进度
+        </a-button>
+        <a-button type="primary" @click="openIngest">
+          <template #icon><UploadOutlined /></template>
+          入库
+        </a-button>
+      </div>
+    </div>
+
+    <div class="library__body">
+      <div class="library__tree">
+        <a-spin :spinning="isLoading">
+          <a-tree
+            :tree-data="treeDataComputed"
+            :default-expand-all="true"
+            :show-icon="true"
+            block-node
+            @select="handleTreeSelect"
+          >
+            <template #title="{ title, count, category }">
+              <span v-if="category !== undefined" class="library__node-cat">
+                <span>{{ title }}</span>
+                <span class="text-muted library__node-count">{{ count }}</span>
+              </span>
+              <span v-else>{{ title }}</span>
+            </template>
+          </a-tree>
+        </a-spin>
+      </div>
+
+      <div class="library__detail">
+        <!-- 类目说明 -->
+        <div v-if="selectedCategory" class="library__panel">
+          <h3 class="library__panel-title">
+            <TagsOutlined /> {{ selectedCategory }}
+          </h3>
+          <div class="text-muted">{{ categoryDesc || '（无描述）' }}</div>
+        </div>
+
+        <!-- 未选中提示 -->
+        <div v-else-if="!docDetailVisible" class="library__placeholder text-muted">
+          从左侧选择类目或文档查看详情
+        </div>
+      </div>
+    </div>
+
+    <!-- 文档详情抽屉 -->
+    <a-drawer
+      :open="docDetailVisible"
+      title="文档详情"
+      placement="right"
+      width="620"
+      :destroy-on-close="true"
+      @close="closeDocDetail"
+    >
+      <a-spin :spinning="isLoadingDetail">
+        <div v-if="docDetailData">
+          <a-descriptions :column="1" size="small" bordered>
+            <a-descriptions-item label="doc_id">{{ docDetailData.l1?.doc_id || '-' }}</a-descriptions-item>
+            <a-descriptions-item label="标题">{{ docDetailData.l1?.title || '-' }}</a-descriptions-item>
+            <a-descriptions-item label="类目">{{ docDetailData.l1?.category || '-' }}</a-descriptions-item>
+            <a-descriptions-item label="标签">
+              <template v-if="docDetailData.l1?.tags && docDetailData.l1.tags.length">
+                <a-tag v-for="tag in docDetailData.l1.tags" :key="tag">{{ tag }}</a-tag>
+              </template>
+              <span v-else class="text-muted">-</span>
+            </a-descriptions-item>
+            <a-descriptions-item label="chunks_count">{{ docDetailData.chunks_count ?? 0 }}</a-descriptions-item>
+            <a-descriptions-item v-if="docDetailData.file" label="原文文件">
+              <a :href="docDetailData.file.url" target="_blank" rel="noopener">
+                {{ docDetailData.file.filename }}
+                <span v-if="docDetailData.file.size_bytes" class="text-muted">
+                  （{{ formatSize(docDetailData.file.size_bytes) }}）
+                </span>
+              </a>
+            </a-descriptions-item>
+          </a-descriptions>
+
+          <div class="library__detail-actions">
+            <a-button size="small" :loading="isReingesting" @click="handleReingest(docDetailData.l1)">
+              重新摘要
+            </a-button>
+            <a-button size="small" danger :loading="isDeleting" @click="handleDelete(docDetailData.l1)">
+              删除
+            </a-button>
+          </div>
+
+          <h3 class="library__section-title">L1 全文</h3>
+          <pre class="library__pre">{{ docDetailData.l1?.text || '' }}</pre>
+
+          <h3 class="library__section-title">L2 节点（{{ (docDetailData.l2_nodes || []).length }}）</h3>
+          <div v-if="(docDetailData.l2_nodes || []).length === 0" class="text-muted">无</div>
+          <div v-for="(node, idx) in docDetailData.l2_nodes || []" :key="`l2-${idx}`" class="library__node">
+            <div class="library__node-path">{{ node.section_path || '' }}</div>
+            <div class="text-break">{{ node.text || '' }}</div>
+          </div>
+
+          <h3 class="library__section-title">L3 节点（{{ (docDetailData.l3_nodes || []).length }}）</h3>
+          <div v-if="(docDetailData.l3_nodes || []).length === 0" class="text-muted">无</div>
+          <div v-for="(node, idx) in docDetailData.l3_nodes || []" :key="`l3-${idx}`" class="library__node">
+            <div class="library__node-path">{{ node.section_path || '' }}</div>
+            <div class="text-break">{{ node.text || '' }}</div>
+          </div>
+        </div>
+        <div v-else-if="!isLoadingDetail" class="text-muted">暂无数据</div>
+      </a-spin>
+    </a-drawer>
+
+    <!-- 入库抽屉 -->
+    <a-drawer
+      :open="ingestVisible"
+      title="文档入库"
+      placement="right"
+      width="560"
+      :destroy-on-close="true"
+      @close="() => (ingestVisible = false)"
+    >
+      <a-tabs v-model:activeKey="activeTab">
+        <a-tab-pane key="text" tab="文本入库">
+          <a-form ref="textFormRef" :model="textForm" :rules="textRules" layout="vertical">
+            <a-form-item label="标题" name="title">
+              <a-input v-model:value="textForm.title" placeholder="请输入标题" allow-clear />
+            </a-form-item>
+            <a-form-item label="来源" name="source">
+              <a-input v-model:value="textForm.source" placeholder="例如：manual / web / file" allow-clear />
+            </a-form-item>
+            <a-form-item label="正文" name="text">
+              <a-textarea v-model:value="textForm.text" placeholder="请输入文档正文" :auto-size="{ minRows: 8, maxRows: 18 }" />
+            </a-form-item>
+            <a-form-item>
+              <a-button type="primary" :loading="isSubmittingText || pollState.isPolling" @click="handleSubmitText">
+                提交入库
+              </a-button>
+            </a-form-item>
+          </a-form>
+        </a-tab-pane>
+
+        <a-tab-pane key="file" tab="文件上传">
+          <a-form :model="fileForm" layout="vertical">
+            <a-form-item label="选择文件">
+              <a-upload :file-list="fileList" :accept="ACCEPTED_EXTENSIONS" :max-count="1" :before-upload="handleFileChange" @remove="handleFileRemove">
+                <a-button :disabled="fileList.length >= 1">选择文件</a-button>
+              </a-upload>
+              <div class="text-muted" style="margin-top: 4px">支持 .txt/.md/.html/.htm/.pdf/.docx</div>
+            </a-form-item>
+            <a-form-item label="标题（可选，默认取文件名）" name="title">
+              <a-input v-model:value="fileForm.title" placeholder="留空则使用文件名去扩展" allow-clear />
+            </a-form-item>
+            <a-form-item label="来源（可选，默认 file:原文件名）" name="source">
+              <a-input v-model:value="fileForm.source" placeholder="例如：manual / web" allow-clear />
+            </a-form-item>
+            <a-form-item>
+              <a-button type="primary" :loading="isSubmittingFile || pollState.isPolling" @click="handleSubmitFile">
+                上传入库
+              </a-button>
+            </a-form-item>
+          </a-form>
+        </a-tab-pane>
+
+        <a-tab-pane key="batch" tab="批量上传">
+          <a-form :model="fileForm" layout="vertical">
+            <a-form-item label="选择文件（可多选）">
+              <a-upload multiple :file-list="fileList" :accept="ACCEPTED_EXTENSIONS" :before-upload="handleBatchFileChange" @remove="handleBatchRemove">
+                <a-button>选择文件</a-button>
+              </a-upload>
+              <div class="text-muted" style="margin-top: 4px">支持 .txt/.md/.html/.htm/.pdf/.docx，可多选，逐文件异步入库</div>
+            </a-form-item>
+            <a-form-item v-if="fileList.length">
+              <a-button type="primary" :loading="isSubmittingBatch" @click="handleSubmitBatch">
+                批量上传入库（{{ fileList.length }} 个文件）
+              </a-button>
+              <a-button style="margin-left: 8px" :disabled="isSubmittingBatch" @click="handleBatchClear">清空</a-button>
+            </a-form-item>
+          </a-form>
+
+          <div v-if="batchResult" class="library__batch">
+            <a-alert
+              v-if="batchResult.failed && batchResult.failed.length"
+              class="library__alert"
+              type="error"
+              show-icon
+              :message="`${batchResult.failed.length} 个文件上传失败`"
+            />
+            <a-table
+              v-if="batchResult.tasks && batchResult.tasks.length"
+              :data-source="batchResult.tasks"
+              :columns="batchColumns"
+              :pagination="false"
+              size="small"
+              row-key="task_id"
+            >
+              <template #bodyCell="{ column, record }">
+                <template v-if="column.key === 'task_id'">
+                  <a-typography-text copyable :style="{ fontSize: '12px' }">{{ record.task_id }}</a-typography-text>
+                </template>
+              </template>
+            </a-table>
+            <a-table
+              v-if="batchResult.failed && batchResult.failed.length"
+              :data-source="batchResult.failed"
+              :columns="failedColumns"
+              :pagination="false"
+              size="small"
+              row-key="filename"
+              style="margin-top: 12px"
+            >
+              <template #bodyCell="{ column, record }">
+                <template v-if="column.key === 'error'">
+                  <span class="text-break">{{ record.error }}</span>
+                </template>
+              </template>
+            </a-table>
+          </div>
+        </a-tab-pane>
+      </a-tabs>
+
+      <div v-if="pollState.taskId" class="library__result">
+        <div class="library__result-header">
+          <span>任务已提交：{{ pollState.taskId }}</span>
+          <a-tag :color="statusBadgeColor">{{ statusBadgeText }}</a-tag>
+          <a-button v-if="pollState.isPolling" type="link" size="small" @click="handleCancelPolling">停止轮询</a-button>
+        </div>
+        <a-alert
+          v-if="pollState.isTimeout"
+          class="library__alert"
+          type="warning"
+          show-icon
+          :message="`任务仍在进行，可稍后凭 task_id 查询：${pollState.taskId}`"
+        />
+        <div v-if="isTerminal && pollState.status === 'done' && resultData" class="library__done">
+          <a-descriptions :column="1" size="small" bordered>
+            <a-descriptions-item label="document_id">{{ resultData.document_id || '-' }}</a-descriptions-item>
+            <a-descriptions-item label="类目">{{ resultData.category || '-' }}（置信度 {{ resultData.category_confidence ?? '-' }}）</a-descriptions-item>
+            <a-descriptions-item label="标签">
+              <template v-if="resultData.tags && resultData.tags.length">
+                <a-tag v-for="tag in resultData.tags" :key="tag">{{ tag }}</a-tag>
+              </template>
+              <span v-else class="text-muted">-</span>
+            </a-descriptions-item>
+            <a-descriptions-item label="总结层级">{{ summaryData.level ?? '-' }}</a-descriptions-item>
+            <a-descriptions-item label="chunks_count">{{ resultData.chunks_count ?? 0 }}</a-descriptions-item>
+          </a-descriptions>
+          <h3 class="library__section-title">L1 总结</h3>
+          <pre class="library__pre">{{ summaryData.l1_summary || '' }}</pre>
+        </div>
+        <div v-if="isTerminal && pollState.status === 'failed'" class="library__failed">
+          <a-descriptions :column="1" size="small" bordered>
+            <a-descriptions-item label="失败阶段">{{ errorData?.stage || '-' }}</a-descriptions-item>
+            <a-descriptions-item label="错误信息"><span class="text-break">{{ errorData?.message || '-' }}</span></a-descriptions-item>
+          </a-descriptions>
+          <template v-if="partialSummary && partialSummary.l1_summary">
+            <h3 class="library__section-title">L1 总结</h3>
+            <pre class="library__pre">{{ partialSummary.l1_summary }}</pre>
+          </template>
+        </div>
+      </div>
+    </a-drawer>
+  </div>
+</template>
+
+<style scoped>
+.library__header {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  margin-bottom: 12px;
+  flex-wrap: wrap;
+  gap: 8px;
+}
+
+.library__title {
+  font-size: 16px;
+  margin: 0;
+}
+
+.library__actions {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+  flex-wrap: wrap;
+}
+
+.library__body {
+  display: flex;
+  gap: 16px;
+  min-height: 480px;
+}
+
+.library__tree {
+  width: 320px;
+  flex: 0 0 auto;
+  background: #fff;
+  border: 1px solid #f0f0f0;
+  border-radius: 8px;
+  padding: 12px;
+  max-height: 70vh;
+  overflow: auto;
+}
+
+.library__node-cat {
+  display: inline-flex;
+  align-items: center;
+  gap: 6px;
+}
+
+.library__node-count {
+  font-size: 12px;
+}
+
+.library__detail {
+  flex: 1;
+  min-width: 0;
+}
+
+.library__panel {
+  background: #fff;
+  border: 1px solid #f0f0f0;
+  border-radius: 8px;
+  padding: 16px;
+}
+
+.library__panel-title {
+  font-size: 15px;
+  margin: 0 0 8px;
+  display: flex;
+  align-items: center;
+  gap: 6px;
+}
+
+.library__placeholder {
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  height: 100%;
+  min-height: 300px;
+  border: 1px dashed #d9d9d9;
+  border-radius: 8px;
+}
+
+.library__detail-actions {
+  display: flex;
+  gap: 8px;
+  margin: 12px 0;
+}
+
+.library__section-title {
+  font-size: 14px;
+  margin: 16px 0 8px;
+  color: #374151;
+}
+
+.library__pre {
+  background: #f9fafb;
+  border: 1px solid #e5e7eb;
+  border-radius: 4px;
+  padding: 8px;
+  white-space: pre-wrap;
+  word-break: break-word;
+  font-size: 12px;
+  max-height: 260px;
+  overflow: auto;
+  margin: 0;
+}
+
+.library__node {
+  border-left: 3px solid #93c5fd;
+  padding: 6px 10px;
+  margin-bottom: 6px;
+  background: #f8fafc;
+  border-radius: 0 4px 4px 0;
+}
+
+.library__node-path {
+  font-size: 12px;
+  color: #6b7280;
+  margin-bottom: 4px;
+}
+
+.library__result {
+  margin-top: 16px;
+  border: 1px solid #e5e7eb;
+  border-radius: 6px;
+  background: #fafafa;
+  padding: 12px 16px;
+}
+
+.library__result-header {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+  flex-wrap: wrap;
+  font-size: 13px;
+  margin-bottom: 8px;
+}
+
+.library__alert {
+  margin: 8px 0;
+}
+
+.library__batch {
+  margin-top: 8px;
+}
+
+@media (max-width: 768px) {
+  .library__body {
+    flex-direction: column;
+  }
+  .library__tree {
+    width: 100%;
+  }
+}
+</style>

frontend/src/views/Overview.vue

@@ -1,5 +1,5 @@
 <script setup>
-import { computed, onMounted, ref } from 'vue'
+import { computed, onMounted, reactive, ref } from 'vue'
 import { message } from 'ant-design-vue'
 import {
   DatabaseOutlined,
@@ -9,69 +9,82 @@ import {
   FileTextOutlined,
   QuestionCircleOutlined,
   ReloadOutlined,
-  AppstoreOutlined
+  AppstoreOutlined,
+  SearchOutlined
 } from '@ant-design/icons-vue'
 import { stats as fetchStats } from '@/api/knowledge'
+import { search as searchApi } from '@/api/search'
 
+/* ---------- 检索 ---------- */
+const isSearching = ref(false)
+const resultData = ref(null)
+const searchFormRef = ref(null)
+const searchForm = reactive({
+  query: '',
+  top_k: 5,
+  summarize: false
+})
+const searchRules = {
+  query: [{ required: true, message: '请输入查询语句', trigger: 'blur' }],
+  top_k: [{ type: 'number', min: 1, max: 50, message: 'top_k 范围 1~50', trigger: 'change' }]
+}
+
+const routedCategoriesText = (cats) => {
+  if (!cats || !cats.length) return '(无)'
+  return cats.join(', ')
+}
+
+function eiValue(value) {
+  if (value === undefined || value === null || value === '') return '(无)'
+  if (Array.isArray(value)) {
+    return value.length ? value.join(', ') : '(无)'
+  }
+  return String(value)
+}
+
+const hits = (data) => data?.hits || []
+
+async function handleSearch() {
+  try {
+    await searchFormRef.value.validate()
+  } catch {
+    return
+  }
+  isSearching.value = true
+  resultData.value = null
+  try {
+    resultData.value = await searchApi({
+      query: searchForm.query,
+      top_k: searchForm.top_k,
+      summarize: searchForm.summarize
+    })
+  } catch (err) {
+    message.error(err?.message || '检索失败')
+  } finally {
+    isSearching.value = false
+  }
+}
+
+/* ---------- 概览统计 ---------- */
 const isLoading = ref(false)
 const statsData = ref(null)
 
 const cardMeta = [
-  {
+  { key: 'doc_l1', label: 'L1 文档总结', icon: DatabaseOutlined, color: '#1677ff', bg: 'rgba(22, 119, 255, 0.12)' },
-    key: 'doc_l1',
+  { key: 'doc_l2', label: 'L2 大纲节点', icon: ApartmentOutlined, color: '#722ed1', bg: 'rgba(114, 46, 209, 0.12)' },
-    label: 'L1 文档总结',
+  { key: 'doc_l3', label: 'L3 内容大纲', icon: FileSearchOutlined, color: '#13c2c2', bg: 'rgba(19, 194, 194, 0.12)' },
-    icon: DatabaseOutlined,
+  { key: 'chunks', label: 'Chunks', icon: BlockOutlined, color: '#fa8c16', bg: 'rgba(250, 140, 22, 0.12)' },
-    color: '#1677ff',
+  { key: '__documents_total', label: '文档总数', icon: FileTextOutlined, color: '#52c41a', bg: 'rgba(82, 196, 26, 0.12)' },
-    bg: 'rgba(22, 119, 255, 0.12)'
+  { key: '__uncategorized', label: '未分类文档', icon: QuestionCircleOutlined, color: '#ff4d4f', bg: 'rgba(255, 77, 79, 0.12)' }
-  },
-  {
-    key: 'doc_l2',
-    label: 'L2 大纲节点',
-    icon: ApartmentOutlined,
-    color: '#722ed1',
-    bg: 'rgba(114, 46, 209, 0.12)'
-  },
-  {
-    key: 'doc_l3',
-    label: 'L3 内容大纲',
-    icon: FileSearchOutlined,
-    color: '#13c2c2',
-    bg: 'rgba(19, 194, 194, 0.12)'
-  },
-  {
-    key: 'chunks',
-    label: 'Chunks',
-    icon: BlockOutlined,
-    color: '#fa8c16',
-    bg: 'rgba(250, 140, 22, 0.12)'
-  },
-  {
-    key: '__documents_total',
-    label: '文档总数',
-    icon: FileTextOutlined,
-    color: '#52c41a',
-    bg: 'rgba(82, 196, 26, 0.12)'
-  },
-  {
-    key: '__uncategorized',
-    label: '未分类文档',
-    icon: QuestionCircleOutlined,
-    color: '#ff4d4f',
-    bg: 'rgba(255, 77, 79, 0.12)'
-  }
 ]
 
 const cards = computed(() => {
   const collections = statsData.value?.collections || {}
   return cardMeta.map((m) => {
     let value = 0
-    if (m.key === '__documents_total') {
+    if (m.key === '__documents_total') value = statsData.value?.documents_total ?? 0
-      value = statsData.value?.documents_total ?? 0
+    else if (m.key === '__uncategorized') value = statsData.value?.uncategorized_count ?? 0
-    } else if (m.key === '__uncategorized') {
+    else value = collections[m.key] ?? 0
-      value = statsData.value?.uncategorized_count ?? 0
-    } else {
-      value = collections[m.key] ?? 0
-    }
     return { ...m, value }
   })
 })
@@ -126,26 +139,108 @@ onMounted(() => {
   <div class="overview page-section">
     <div class="overview__header">
       <h2 class="page-title">概览</h2>
-      <a-space>
+      <a-button :loading="isLoading" @click="loadStats">
-        <a-button :loading="isLoading" @click="loadStats">
+        <template #icon><ReloadOutlined /></template>
-          <template #icon><ReloadOutlined /></template>
+        刷新
-          刷新
+      </a-button>
-        </a-button>
-      </a-space>
     </div>
 
+    <!-- 检索测试 -->
+    <div class="overview__search">
+      <h3 class="overview__subtitle">
+        <SearchOutlined />
+        <span>检索</span>
+      </h3>
+      <a-form
+        ref="searchFormRef"
+        :model="searchForm"
+        :rules="searchRules"
+        layout="inline"
+      >
+        <a-form-item name="query" style="flex: 1; min-width: 260px">
+          <a-input
+            v-model:value="searchForm.query"
+            placeholder="请输入查询语句"
+            allow-clear
+            @pressEnter="handleSearch"
+          />
+        </a-form-item>
+        <a-form-item name="top_k">
+          <a-input-number
+            v-model:value="searchForm.top_k"
+            :min="1"
+            :max="50"
+            style="width: 120px"
+          />
+        </a-form-item>
+        <a-form-item name="summarize">
+          <a-checkbox v-model:checked="searchForm.summarize">AI 总结</a-checkbox>
+        </a-form-item>
+        <a-form-item>
+          <a-button type="primary" :loading="isSearching" @click="handleSearch">
+            检索
+          </a-button>
+        </a-form-item>
+      </a-form>
+
+      <div v-if="resultData" class="overview__result">
+        <a-descriptions :column="1" size="small" bordered>
+          <a-descriptions-item label="routed_categories">
+            {{ routedCategoriesText(resultData.routed_categories) }}
+          </a-descriptions-item>
+        </a-descriptions>
+
+        <a-alert
+          v-if="resultData.fallback"
+          class="overview__alert"
+          type="warning"
+          show-icon
+          message="fallback：本次检索触发了回退策略（路由类目无命中，已降级全局检索）"
+        />
+
+        <div v-if="resultData.summary" class="overview__summary">
+          <div class="overview__summary-title">AI 总结</div>
+          <div class="text-break">{{ resultData.summary }}</div>
+        </div>
+
+        <div v-if="resultData.extracted_info" class="overview__extracted">
+          <div class="overview__extracted-title">AI 提取的关键信息</div>
+          <a-descriptions :column="1" size="small" bordered>
+            <a-descriptions-item label="改写">{{ eiValue(resultData.extracted_info.rewrite) }}</a-descriptions-item>
+            <a-descriptions-item label="关键词">{{ eiValue(resultData.extracted_info.keywords) }}</a-descriptions-item>
+            <a-descriptions-item label="实体">{{ eiValue(resultData.extracted_info.entities) }}</a-descriptions-item>
+            <a-descriptions-item label="意图">{{ eiValue(resultData.extracted_info.intent) }}</a-descriptions-item>
+            <a-descriptions-item label="时间范围">{{ eiValue(resultData.extracted_info.time_range) }}</a-descriptions-item>
+            <a-descriptions-item label="命中类目">{{ eiValue(resultData.extracted_info.categories) }}</a-descriptions-item>
+          </a-descriptions>
+        </div>
+
+        <div class="text-muted" style="margin: 12px 0 8px">
+          命中 {{ hits(resultData).length }} 条
+        </div>
+
+        <div
+          v-for="(hit, idx) in hits(resultData)"
+          :key="`hit-${idx}`"
+          class="overview__hit"
+        >
+          <div class="overview__hit-meta">
+            score={{ hit.score }} | doc_id={{ hit.doc_id }} | 标题={{ hit.title || '' }} | section={{ hit.section_path || '' }}
+          </div>
+          <div class="overview__hit-snippet text-break">{{ hit.text || '' }}</div>
+          <div v-if="hit.doc_summary" class="overview__hit-meta">
+            文档摘要：{{ hit.doc_summary }}
+          </div>
+        </div>
+      </div>
+    </div>
+
+    <!-- 概览统计 -->
     <a-spin :spinning="isLoading">
       <div class="overview__cards">
-        <div
+        <div v-for="card in cards" :key="card.key" class="overview__card">
-          v-for="card in cards"
-          :key="card.key"
-          class="overview__card"
-        >
           <div class="overview__card-body">
-            <div
+            <div class="overview__card-icon" :style="{ background: card.bg, color: card.color }">
-              class="overview__card-icon"
-              :style="{ background: card.bg, color: card.color }"
-            >
               <component :is="card.icon" />
             </div>
             <div class="overview__card-info">
@@ -166,9 +261,7 @@ onMounted(() => {
             共 {{ categoryBars.length }} 个类目
           </span>
         </div>
-        <div v-if="categoryBars.length === 0" class="overview__empty text-muted">
+        <div v-if="categoryBars.length === 0" class="overview__empty text-muted">暂无数据</div>
-          暂无数据
-        </div>
         <div v-else class="overview__bars">
           <div
             v-for="(bar, idx) in categoryBars"
@@ -200,6 +293,72 @@ onMounted(() => {
   margin-bottom: 16px;
 }
 
+.overview__search {
+  background: #fff;
+  border: 1px solid #f0f0f0;
+  border-radius: 8px;
+  padding: 16px;
+  margin-bottom: 24px;
+}
+
+.overview__result {
+  margin-top: 16px;
+}
+
+.overview__alert {
+  margin: 12px 0;
+}
+
+.overview__summary {
+  background: #f0fdf4;
+  border: 1px solid #bbf7d0;
+  border-radius: 6px;
+  padding: 12px 14px;
+  margin: 12px 0;
+  font-size: 13px;
+}
+
+.overview__summary-title {
+  font-weight: 600;
+  margin-bottom: 6px;
+  color: #15803d;
+}
+
+.overview__extracted {
+  background: #f0f9ff;
+  border: 1px solid #bae6fd;
+  border-radius: 6px;
+  padding: 12px 14px;
+  margin: 12px 0;
+  font-size: 13px;
+}
+
+.overview__extracted-title {
+  font-weight: 600;
+  margin-bottom: 8px;
+  color: #0369a1;
+}
+
+.overview__hit {
+  border: 1px solid #e5e7eb;
+  border-radius: 6px;
+  padding: 10px 12px;
+  margin-bottom: 10px;
+  background: #fff;
+}
+
+.overview__hit-meta {
+  font-size: 12px;
+  color: #6b7280;
+  margin-bottom: 6px;
+  word-break: break-all;
+}
+
+.overview__hit-snippet {
+  font-size: 13px;
+  color: #1f2937;
+}
+
 .overview__cards {
   display: grid;
   grid-template-columns: repeat(auto-fill, minmax(220px, 1fr));
@@ -272,7 +431,7 @@ onMounted(() => {
   font-size: 14px;
   font-weight: 600;
   color: #374151;
-  margin: 0;
+  margin: 0 0 12px;
   display: flex;
   align-items: center;
   gap: 6px;

frontend/src/views/Search.vue

@@ -1,234 +0,0 @@
-<script setup>
-import { reactive, ref } from 'vue'
-import { message } from 'ant-design-vue'
-import { search as searchApi } from '@/api/search'
-
-const isLoading = ref(false)
-const resultData = ref(null)
-
-const formRef = ref(null)
-const formState = reactive({
-  query: '',
-  top_k: 5,
-  summarize: false
-})
-
-const rules = {
-  query: [{ required: true, message: '请输入查询语句', trigger: 'blur' }],
-  top_k: [{ type: 'number', min: 1, max: 50, message: 'top_k 范围 1~50', trigger: 'change' }]
-}
-
-const routedCategoriesText = (cats) => {
-  if (!cats || !cats.length) return '(无)'
-  return cats.join(', ')
-}
-
-async function handleSearch() {
-  try {
-    await formRef.value.validate()
-  } catch {
-    return
-  }
-  isLoading.value = true
-  resultData.value = null
-  try {
-    const payload = {
-      query: formState.query,
-      top_k: formState.top_k,
-      summarize: formState.summarize
-    }
-    resultData.value = await searchApi(payload)
-  } catch (err) {
-    message.error(err?.message || '检索失败')
-  } finally {
-    isLoading.value = false
-  }
-}
-
-function eiValue(value) {
-  if (value === undefined || value === null || value === '') return '(无)'
-  if (Array.isArray(value)) {
-    return value.length ? value.join(', ') : '(无)'
-  }
-  return String(value)
-}
-
-const hits = (data) => data?.hits || []
-</script>
-
-<template>
-  <div class="search page-section">
-    <div class="search__header">
-      <h2 class="search__title">检索测试台</h2>
-    </div>
-
-    <a-form
-      ref="formRef"
-      :model="formState"
-      :rules="rules"
-      layout="vertical"
-    >
-      <a-form-item label="查询语句" name="query">
-        <a-input
-          v-model:value="formState.query"
-          placeholder="请输入查询语句"
-          allow-clear
-          @pressEnter="handleSearch"
-        />
-      </a-form-item>
-      <a-form-item label="top_k" name="top_k">
-        <a-input-number
-          v-model:value="formState.top_k"
-          :min="1"
-          :max="50"
-          style="width: 160px"
-        />
-      </a-form-item>
-      <a-form-item name="summarize">
-        <a-checkbox v-model:checked="formState.summarize">
-          对结果生成 AI 总结
-        </a-checkbox>
-      </a-form-item>
-      <a-form-item>
-        <a-button type="primary" :loading="isLoading" @click="handleSearch">
-          检索
-        </a-button>
-      </a-form-item>
-    </a-form>
-
-    <div v-if="resultData" class="search__result">
-      <a-descriptions :column="1" size="small" bordered>
-        <a-descriptions-item label="routed_categories">
-          {{ routedCategoriesText(resultData.routed_categories) }}
-        </a-descriptions-item>
-      </a-descriptions>
-
-      <a-alert
-        v-if="resultData.fallback"
-        class="search__alert"
-        type="warning"
-        show-icon
-        message="fallback：本次检索触发了回退策略（路由类目无命中，已降级全局检索）"
-      />
-
-      <div v-if="resultData.summary" class="search__summary">
-        <div class="search__summary-title">AI 总结</div>
-        <div class="text-break">{{ resultData.summary }}</div>
-      </div>
-
-      <div v-if="resultData.extracted_info" class="search__extracted">
-        <div class="search__extracted-title">AI 提取的关键信息</div>
-        <a-descriptions :column="1" size="small" bordered>
-          <a-descriptions-item label="改写">
-            {{ eiValue(resultData.extracted_info.rewrite) }}
-          </a-descriptions-item>
-          <a-descriptions-item label="关键词">
-            {{ eiValue(resultData.extracted_info.keywords) }}
-          </a-descriptions-item>
-          <a-descriptions-item label="实体">
-            {{ eiValue(resultData.extracted_info.entities) }}
-          </a-descriptions-item>
-          <a-descriptions-item label="意图">
-            {{ eiValue(resultData.extracted_info.intent) }}
-          </a-descriptions-item>
-          <a-descriptions-item label="时间范围">
-            {{ eiValue(resultData.extracted_info.time_range) }}
-          </a-descriptions-item>
-          <a-descriptions-item label="命中类目">
-            {{ eiValue(resultData.extracted_info.categories) }}
-          </a-descriptions-item>
-        </a-descriptions>
-      </div>
-
-      <div class="text-muted" style="margin: 12px 0 8px">
-        命中 {{ hits(resultData).length }} 条
-      </div>
-
-      <div
-        v-for="(hit, idx) in hits(resultData)"
-        :key="`hit-${idx}`"
-        class="search__hit"
-      >
-        <div class="search__hit-meta">
-          score={{ hit.score }} | doc_id={{ hit.doc_id }} | 标题={{ hit.title || '' }} | section={{ hit.section_path || '' }}
-        </div>
-        <div class="search__hit-snippet text-break">{{ hit.text || '' }}</div>
-        <div v-if="hit.doc_summary" class="search__hit-meta">
-          文档摘要：{{ hit.doc_summary }}
-        </div>
-      </div>
-    </div>
-  </div>
-</template>
-
-<style scoped>
-.search__header {
-  display: flex;
-  align-items: center;
-  justify-content: space-between;
-  margin-bottom: 12px;
-}
-
-.search__title {
-  font-size: 16px;
-  margin: 0;
-}
-
-.search__result {
-  margin-top: 16px;
-}
-
-.search__alert {
-  margin: 12px 0;
-}
-
-.search__summary {
-  background: #f0fdf4;
-  border: 1px solid #bbf7d0;
-  border-radius: 6px;
-  padding: 12px 14px;
-  margin: 12px 0;
-  font-size: 13px;
-}
-
-.search__summary-title {
-  font-weight: 600;
-  margin-bottom: 6px;
-  color: #15803d;
-}
-
-.search__extracted {
-  background: #f0f9ff;
-  border: 1px solid #bae6fd;
-  border-radius: 6px;
-  padding: 12px 14px;
-  margin: 12px 0;
-  font-size: 13px;
-}
-
-.search__extracted-title {
-  font-weight: 600;
-  margin-bottom: 8px;
-  color: #0369a1;
-}
-
-.search__hit {
-  border: 1px solid #e5e7eb;
-  border-radius: 6px;
-  padding: 10px 12px;
-  margin-bottom: 10px;
-  background: #fff;
-}
-
-.search__hit-meta {
-  font-size: 12px;
-  color: #6b7280;
-  margin-bottom: 6px;
-  word-break: break-all;
-}
-
-.search__hit-snippet {
-  font-size: 13px;
-  color: #1f2937;
-}
-</style>

frontend/src/views/Settings.vue

@@ -318,7 +318,7 @@ onMounted(() => {
                     <a-form-item label="model">
                       <a-input
                         v-model:value="form.models[key].model"
-                        placeholder="例如 qwen2.5:1.5b"
+                        placeholder="例如 qwen3:1.7b"
                         allow-clear
                       />
                     </a-form-item>

frontend/src/views/TasksDrawer.vue

@@ -0,0 +1,394 @@
+<script setup>
+import { computed, onBeforeUnmount, ref, watch } from 'vue'
+import { message, Modal } from 'ant-design-vue'
+import {
+  ReloadOutlined,
+  ClockCircleOutlined,
+  EyeOutlined,
+  ArrowLeftOutlined,
+  UndoOutlined,
+  DeleteOutlined
+} from '@ant-design/icons-vue'
+import {
+  taskList,
+  taskStatus,
+  retryTask,
+  deleteTask
+} from '@/api/documents'
+import {
+  INGEST_STATUS_TEXT,
+  INGEST_STATUS_COLOR
+} from '@/constants/ingest'
+import { useTasksDrawer } from '@/composables/useTasksDrawer'
+import { formatSize } from '@/utils/format'
+
+const { state, close } = useTasksDrawer()
+
+const tasks = ref([])
+const total = ref(0)
+const isLoading = ref(false)
+const isPolling = ref(false)
+
+/* 选中任务详情 */
+const selectedTask = ref(null)
+const isLoadingDetail = ref(false)
+const isRetrying = ref(false)
+const isDeleting = ref(false)
+
+let timer = null
+
+function formatTime(iso) {
+  if (!iso) return '-'
+  const d = new Date(iso)
+  if (Number.isNaN(d.getTime())) return iso
+  const pad = (n) => String(n).padStart(2, '0')
+  return `${d.getFullYear()}-${pad(d.getMonth() + 1)}-${pad(d.getDate())} ${pad(d.getHours())}:${pad(d.getMinutes())}:${pad(d.getSeconds())}`
+}
+
+function statusText(s) {
+  return INGEST_STATUS_TEXT[s] || s || '-'
+}
+
+function statusColor(s) {
+  return INGEST_STATUS_COLOR[s] || 'default'
+}
+
+const isTerminal = computed(() => ['done', 'failed'].includes(selectedTask.value?.status))
+const detailResult = computed(() => selectedTask.value?.result || null)
+const detailError = computed(() => selectedTask.value?.error || null)
+const detailSummary = computed(() => detailResult.value?.summary || {})
+const canRetry = computed(() => ['failed', 'done'].includes(selectedTask.value?.status))
+const canDelete = computed(() => ['failed', 'done'].includes(selectedTask.value?.status))
+
+/* 从任务记录提取详情展示字段（title/source/size） */
+const detailComputed = computed(() => {
+  const task = selectedTask.value
+  if (!task) return null
+  const source = task.source || {}
+  const metadata = source.metadata || {}
+  return {
+    title: task.title || source.title || '-',
+    source: task.source_text || source.source || '-',
+    size: task.size_bytes != null ? formatSize(task.size_bytes) : formatSize(metadata.original_size_bytes)
+  }
+})
+
+async function loadTasks(showSpinner = false) {
+  if (showSpinner) isLoading.value = true
+  try {
+    const data = await taskList(50)
+    tasks.value = data.items || []
+    total.value = data.total ?? tasks.value.length
+  } catch (err) {
+    message.error(err?.message || '加载入库任务失败')
+  } finally {
+    if (showSpinner) isLoading.value = false
+  }
+}
+
+function startPolling() {
+  if (timer) return
+  isPolling.value = true
+  timer = setInterval(async () => {
+    await loadTasks(false)
+  }, 2000)
+}
+
+function stopPolling() {
+  if (timer) {
+    clearInterval(timer)
+    timer = null
+  }
+  isPolling.value = false
+}
+
+function togglePolling() {
+  if (isPolling.value) stopPolling()
+  else startPolling()
+}
+
+/* 打开/关闭抽屉时启停轮询 */
+watch(
+  () => state.open,
+  (open) => {
+    if (open) {
+      loadTasks(true)
+      startPolling()
+    } else {
+      stopPolling()
+      selectedTask.value = null
+    }
+  },
+  { immediate: true }
+)
+
+async function openDetail(task) {
+  selectedTask.value = null
+  isLoadingDetail.value = true
+  try {
+    selectedTask.value = await taskStatus(task.task_id)
+  } catch (err) {
+    message.error(err?.message || '加载任务详情失败')
+  } finally {
+    isLoadingDetail.value = false
+  }
+}
+
+function backToList() {
+  selectedTask.value = null
+}
+
+async function handleRetry(task) {
+  const taskId = task?.task_id || selectedTask.value?.task_id
+  if (!taskId) return
+  isRetrying.value = true
+  try {
+    const data = await retryTask(taskId)
+    message.success(`已提交新任务：${data.task_id}`)
+    selectedTask.value = null
+    await loadTasks(false)
+  } catch (err) {
+    message.error(err?.message || '重试失败')
+  } finally {
+    isRetrying.value = false
+  }
+}
+
+function handleDelete(task) {
+  const taskId = task?.task_id || selectedTask.value?.task_id
+  if (!taskId) return
+  Modal.confirm({
+    title: '删除入库任务',
+    content: `确定删除任务「${taskId}」吗？删除后不可恢复。`,
+    okText: '删除',
+    okType: 'danger',
+    cancelText: '取消',
+    async onOk() {
+      isDeleting.value = true
+      try {
+        await deleteTask(taskId)
+        message.success('任务已删除')
+        selectedTask.value = null
+        await loadTasks(false)
+      } catch (err) {
+        message.error(err?.message || '删除任务失败')
+      } finally {
+        isDeleting.value = false
+      }
+    }
+  })
+}
+
+onBeforeUnmount(stopPolling)
+
+const columns = [
+  { title: '状态', dataIndex: 'status', key: 'status', width: 90 },
+  { title: '文件名', dataIndex: 'filename', key: 'filename', ellipsis: true },
+  { title: 'task_id', dataIndex: 'task_id', key: 'task_id', width: 200 },
+  { title: 'doc_id', dataIndex: 'doc_id', key: 'doc_id', width: 200 },
+  { title: '创建时间', dataIndex: 'created_at', key: 'created_at', width: 160 },
+  { title: '操作', dataIndex: 'action', key: 'action', width: 130 }
+]
+</script>
+
+<template>
+  <a-drawer
+    :open="state.open"
+    :title="selectedTask ? '任务详情' : '入库进度'"
+    placement="right"
+    width="640"
+    :destroy-on-close="true"
+    @close="close"
+  >
+    <!-- 列表视图 -->
+    <template v-if="!selectedTask">
+      <div class="tasks-drawer__header">
+        <span class="text-muted">共 {{ total }} 条</span>
+        <div class="tasks-drawer__actions">
+          <a-button size="small" :loading="isLoading" @click="loadTasks(true)">
+            <template #icon><ReloadOutlined /></template>
+            刷新
+          </a-button>
+          <a-button size="small" :type="isPolling ? 'default' : 'primary'" @click="togglePolling">
+            {{ isPolling ? '停止轮询' : '开始轮询' }}
+          </a-button>
+        </div>
+      </div>
+      <a-alert
+        v-if="isPolling"
+        type="info"
+        show-icon
+        message="每 2 秒自动刷新近期入库任务。"
+        style="margin-bottom: 12px"
+      />
+      <a-table
+        :columns="columns"
+        :data-source="tasks"
+        :pagination="false"
+        :loading="isLoading"
+        row-key="task_id"
+        size="small"
+        :scroll="{ x: 720 }"
+      >
+        <template #bodyCell="{ column, record }">
+          <template v-if="column.key === 'status'">
+            <a-tag :color="statusColor(record.status)">{{ statusText(record.status) }}</a-tag>
+          </template>
+          <template v-else-if="column.key === 'filename'">
+            <span :title="record.filename">{{ record.filename || '-' }}</span>
+          </template>
+          <template v-else-if="column.key === 'task_id'">
+            <a-typography-text copyable :style="{ fontSize: '12px' }">{{ record.task_id }}</a-typography-text>
+          </template>
+          <template v-else-if="column.key === 'doc_id'">
+            <a-typography-text
+              v-if="record.doc_id"
+              copyable
+              :style="{ fontSize: '12px' }"
+            >
+              {{ record.doc_id }}
+            </a-typography-text>
+            <span v-else class="text-muted">-</span>
+          </template>
+          <template v-else-if="column.key === 'created_at'">
+            {{ formatTime(record.created_at) }}
+          </template>
+          <template v-else-if="column.key === 'action'">
+            <a-button type="link" size="small" @click="openDetail(record)">
+              <template #icon><EyeOutlined /></template>
+              详情
+            </a-button>
+            <a-button
+              v-if="['failed', 'done'].includes(record.status)"
+              type="link"
+              size="small"
+              @click="handleRetry(record)"
+            >
+              <template #icon><UndoOutlined /></template>
+              重试
+            </a-button>
+            <a-button
+              v-if="['done', 'failed'].includes(record.status)"
+              type="link"
+              size="small"
+              danger
+              @click="handleDelete(record)"
+            >
+              <template #icon><DeleteOutlined /></template>
+              删除
+            </a-button>
+          </template>
+        </template>
+      </a-table>
+    </template>
+
+    <!-- 详情视图 -->
+    <template v-else>
+      <a-button size="small" type="link" style="margin-bottom: 12px; padding-left: 0" @click="backToList">
+        <template #icon><ArrowLeftOutlined /></template>
+        返回列表
+      </a-button>
+
+      <a-spin :spinning="isLoadingDetail">
+        <template v-if="selectedTask">
+          <a-descriptions :column="1" size="small" bordered>
+            <a-descriptions-item label="状态">
+              <a-tag :color="statusColor(selectedTask.status)">{{ statusText(selectedTask.status) }}</a-tag>
+            </a-descriptions-item>
+            <a-descriptions-item label="task_id">{{ selectedTask.task_id || '-' }}</a-descriptions-item>
+            <a-descriptions-item label="文件名">{{ selectedTask.filename || '-' }}</a-descriptions-item>
+            <a-descriptions-item label="标题">{{ detailComputed?.title || '-' }}</a-descriptions-item>
+            <a-descriptions-item label="来源">{{ detailComputed?.source || '-' }}</a-descriptions-item>
+            <a-descriptions-item label="文件大小">{{ detailComputed?.size || '-' }}</a-descriptions-item>
+            <a-descriptions-item label="doc_id">{{ selectedTask.doc_id || '-' }}</a-descriptions-item>
+            <a-descriptions-item label="创建时间">{{ formatTime(selectedTask.created_at) }}</a-descriptions-item>
+            <a-descriptions-item label="更新时间">{{ formatTime(selectedTask.updated_at) }}</a-descriptions-item>
+          </a-descriptions>
+
+          <!-- 完成详情 -->
+          <div v-if="isTerminal && selectedTask.status === 'done' && detailResult" class="tasks-drawer__block">
+            <h4 class="tasks-drawer__block-title">入库结果</h4>
+            <a-descriptions :column="1" size="small" bordered>
+              <a-descriptions-item label="document_id">{{ detailResult.document_id || '-' }}</a-descriptions-item>
+              <a-descriptions-item label="类目">{{ detailResult.category || '-' }}（置信度 {{ detailResult.category_confidence ?? '-' }}）</a-descriptions-item>
+              <a-descriptions-item label="标签">
+                <template v-if="detailResult.tags && detailResult.tags.length">
+                  <a-tag v-for="tag in detailResult.tags" :key="tag">{{ tag }}</a-tag>
+                </template>
+                <span v-else class="text-muted">-</span>
+              </a-descriptions-item>
+              <a-descriptions-item label="总结层级">{{ detailSummary.level ?? '-' }}</a-descriptions-item>
+              <a-descriptions-item label="chunks_count">{{ detailResult.chunks_count ?? 0 }}</a-descriptions-item>
+              <a-descriptions-item v-if="detailResult.deduplicated" label="去重">是（复用既有文档）</a-descriptions-item>
+            </a-descriptions>
+            <h4 class="tasks-drawer__block-title">L1 总结</h4>
+            <pre class="tasks-drawer__pre">{{ detailSummary.l1_summary || '' }}</pre>
+          </div>
+
+          <!-- 失败详情 -->
+          <div v-if="isTerminal && selectedTask.status === 'failed' && detailError" class="tasks-drawer__block">
+            <a-alert
+              type="error"
+              show-icon
+              :message="`失败阶段：${detailError.stage || '-'}`"
+              :description="detailError.message || '-'"
+            />
+            <template v-if="detailError.partial_summary && detailError.partial_summary.l1_summary">
+              <h4 class="tasks-drawer__block-title">L1 总结（部分）</h4>
+              <pre class="tasks-drawer__pre">{{ detailError.partial_summary.l1_summary }}</pre>
+            </template>
+          </div>
+
+          <div class="tasks-drawer__actions">
+            <a-button v-if="canRetry" :loading="isRetrying" @click="handleRetry(selectedTask)">
+              <template #icon><UndoOutlined /></template>
+              重试
+            </a-button>
+            <a-button v-if="canDelete" danger :loading="isDeleting" @click="handleDelete(selectedTask)">
+              <template #icon><DeleteOutlined /></template>
+              删除
+            </a-button>
+          </div>
+        </template>
+      </a-spin>
+    </template>
+  </a-drawer>
+</template>
+
+<style scoped>
+.tasks-drawer__header {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  margin-bottom: 8px;
+}
+
+.tasks-drawer__actions {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+}
+
+.tasks-drawer__block {
+  margin-top: 16px;
+}
+
+.tasks-drawer__block-title {
+  font-size: 13px;
+  margin: 12px 0 6px;
+  color: #374151;
+}
+
+.tasks-drawer__pre {
+  background: #f9fafb;
+  border: 1px solid #e5e7eb;
+  border-radius: 4px;
+  padding: 8px;
+  white-space: pre-wrap;
+  word-break: break-word;
+  font-size: 12px;
+  max-height: 220px;
+  overflow: auto;
+  margin: 0;
+}
+</style>

frontend/src/views/Users.vue

@@ -0,0 +1,289 @@
+<script setup>
+import { computed, onMounted, reactive, ref } from 'vue'
+import { message, Modal } from 'ant-design-vue'
+import {
+  TeamOutlined,
+  UserAddOutlined,
+  ReloadOutlined,
+  KeyOutlined,
+  DeleteOutlined,
+  StopOutlined,
+  CheckCircleOutlined
+} from '@ant-design/icons-vue'
+import {
+  listUsers,
+  createUser,
+  updateUser,
+  resetPassword,
+  deleteUser
+} from '@/api/auth'
+import { useAuthStore } from '@/stores/useAuthStore'
+
+const authStore = useAuthStore()
+const currentUsername = computed(() => authStore.user?.username || '')
+
+const isLoading = ref(false)
+const users = ref([])
+
+const createVisible = ref(false)
+const creating = ref(false)
+const createForm = reactive({ username: '', password: '', role: 'user' })
+
+const resetVisible = ref(false)
+const resetting = ref(false)
+const resetTarget = ref('')
+const resetForm = reactive({ new_password: '' })
+
+const columns = [
+  { title: '用户名', dataIndex: 'username', key: 'username' },
+  { title: '角色', dataIndex: 'role', key: 'role', width: 100, align: 'center' },
+  { title: '状态', key: 'enabled', width: 90, align: 'center' },
+  { title: '强制改密', key: 'must_change_password', width: 90, align: 'center' },
+  { title: '创建时间', dataIndex: 'created_at', key: 'created_at', width: 180 },
+  { title: '操作', key: 'action', width: 230, fixed: 'right' }
+]
+
+async function loadUsers() {
+  isLoading.value = true
+  try {
+    users.value = await listUsers()
+  } catch (err) {
+    message.error(err?.message || '加载用户列表失败')
+  } finally {
+    isLoading.value = false
+  }
+}
+
+function formatTime(ts) {
+  if (!ts) return '-'
+  const d = new Date(ts)
+  if (Number.isNaN(d.getTime())) return String(ts)
+  return d.toLocaleString('zh-CN', { hour12: false })
+}
+
+// 不能对当前登录账号做删除/禁用（后端亦会拦截）
+function isSelf(username) {
+  return username === currentUsername.value
+}
+
+function openCreate() {
+  createForm.username = ''
+  createForm.password = ''
+  createForm.role = 'user'
+  createVisible.value = true
+}
+
+async function handleCreate() {
+  if (!createForm.username.trim()) {
+    message.warning('请输入用户名')
+    return
+  }
+  if (createForm.password.length < 8) {
+    message.warning('密码至少 8 位')
+    return
+  }
+  creating.value = true
+  try {
+    await createUser(createForm.username.trim(), createForm.password, createForm.role)
+    message.success(`已创建用户 ${createForm.username.trim()}`)
+    createVisible.value = false
+    await loadUsers()
+  } catch (err) {
+    message.error(err?.message || '创建用户失败')
+  } finally {
+    creating.value = false
+  }
+}
+
+function openReset(username) {
+  resetTarget.value = username
+  resetForm.new_password = ''
+  resetVisible.value = true
+}
+
+async function handleReset() {
+  if (resetForm.new_password.length < 8) {
+    message.warning('新密码至少 8 位')
+    return
+  }
+  resetting.value = true
+  try {
+    await resetPassword(resetTarget.value, resetForm.new_password)
+    message.success(`已重置 ${resetTarget.value} 的密码`)
+    resetVisible.value = false
+  } catch (err) {
+    message.error(err?.message || '重置密码失败')
+  } finally {
+    resetting.value = false
+  }
+}
+
+async function toggleEnabled(record) {
+  const enabled = !record.enabled
+  try {
+    await updateUser(record.username, { enabled })
+    message.success(`${record.username} 已${enabled ? '启用' : '禁用'}`)
+    await loadUsers()
+  } catch (err) {
+    message.error(err?.message || '更新状态失败')
+  }
+}
+
+function handleDelete(record) {
+  Modal.confirm({
+    title: '确认删除用户',
+    content: `确定删除用户「${record.username}」吗？该操作不可恢复。`,
+    okText: '删除',
+    okType: 'danger',
+    cancelText: '取消',
+    async onOk() {
+      try {
+        await deleteUser(record.username)
+        message.success(`已删除 ${record.username}`)
+        await loadUsers()
+      } catch (err) {
+        message.error(err?.message || '删除用户失败')
+      }
+    }
+  })
+}
+
+onMounted(() => {
+  loadUsers()
+})
+</script>
+
+<template>
+  <div class="users page-section">
+    <div class="users__header">
+      <h2 class="page-title">
+        <TeamOutlined /> 用户管理
+      </h2>
+      <a-space>
+        <a-button :loading="isLoading" @click="loadUsers">
+          <template #icon><ReloadOutlined /></template>
+          刷新
+        </a-button>
+        <a-button type="primary" @click="openCreate">
+          <template #icon><UserAddOutlined /></template>
+          新建用户
+        </a-button>
+      </a-space>
+    </div>
+
+    <a-table
+      :columns="columns"
+      :data-source="users"
+      :pagination="false"
+      :loading="isLoading"
+      row-key="username"
+      size="middle"
+    >
+      <template #bodyCell="{ column, record }">
+        <template v-if="column.key === 'username'">
+          <span>{{ record.username }}</span>
+          <a-tag v-if="isSelf(record.username)" color="gold" style="margin-left: 6px">当前</a-tag>
+        </template>
+        <template v-else-if="column.key === 'role'">
+          <a-tag :color="record.role === 'admin' ? 'green' : 'blue'">{{ record.role }}</a-tag>
+        </template>
+        <template v-else-if="column.key === 'enabled'">
+          <a-tag :color="record.enabled ? 'success' : 'default'">
+            {{ record.enabled ? '启用' : '禁用' }}
+          </a-tag>
+        </template>
+        <template v-else-if="column.key === 'must_change_password'">
+          <a-tag v-if="record.must_change_password" color="orange">需改密</a-tag>
+          <span v-else class="text-muted">-</span>
+        </template>
+        <template v-else-if="column.key === 'created_at'">
+          <span class="text-muted">{{ formatTime(record.created_at) }}</span>
+        </template>
+        <template v-else-if="column.key === 'action'">
+          <a-space>
+            <a-button type="link" size="small" @click="openReset(record.username)">
+              <template #icon><KeyOutlined /></template>
+              重置密码
+            </a-button>
+            <a-button
+              type="link"
+              size="small"
+              :disabled="isSelf(record.username)"
+              @click="toggleEnabled(record)"
+            >
+              <template #icon>
+                <StopOutlined v-if="record.enabled" />
+                <CheckCircleOutlined v-else />
+              </template>
+              {{ record.enabled ? '禁用' : '启用' }}
+            </a-button>
+            <a-button
+              type="link"
+              size="small"
+              danger
+              :disabled="isSelf(record.username)"
+              @click="handleDelete(record)"
+            >
+              <template #icon><DeleteOutlined /></template>
+              删除
+            </a-button>
+          </a-space>
+        </template>
+      </template>
+    </a-table>
+
+    <!-- 新建用户 -->
+    <a-modal
+      v-model:open="createVisible"
+      title="新建用户"
+      ok-text="创建"
+      cancel-text="取消"
+      :confirm-loading="creating"
+      @ok="handleCreate"
+    >
+      <a-form layout="vertical">
+        <a-form-item label="用户名" required>
+          <a-input v-model:value="createForm.username" placeholder="登录用户名" />
+        </a-form-item>
+        <a-form-item label="密码" required>
+          <a-input-password v-model:value="createForm.password" placeholder="至少 8 位" />
+        </a-form-item>
+        <a-form-item label="角色">
+          <a-radio-group v-model:value="createForm.role">
+            <a-radio value="user">普通用户 (user)</a-radio>
+            <a-radio value="admin">管理员 (admin)</a-radio>
+          </a-radio-group>
+        </a-form-item>
+      </a-form>
+    </a-modal>
+
+    <!-- 重置密码 -->
+    <a-modal
+      v-model:open="resetVisible"
+      title="重置密码"
+      ok-text="重置"
+      cancel-text="取消"
+      :confirm-loading="resetting"
+      @ok="handleReset"
+    >
+      <p class="text-muted">目标用户：{{ resetTarget }}</p>
+      <a-form-item label="新密码" required>
+        <a-input-password v-model:value="resetForm.new_password" placeholder="至少 8 位" />
+      </a-form-item>
+    </a-modal>
+  </div>
+</template>
+
+<style scoped>
+.users__header {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  margin-bottom: 16px;
+}
+
+.page-title :deep(.anticon) {
+  margin-right: 8px;
+  color: #1677ff;
+}
+</style>

scripts/deploy.sh

@@ -0,0 +1,144 @@
+#!/usr/bin/env bash
+# =============================================================================
+# QMDSearch 自动部署脚本（本地 CI/CD 入口）
+#
+# 设计原则（贴合 NAS 自托管 + 挂载式部署）：
+#   1. 检测变化 —— 对比 origin/main，区分 backend / frontend / deps 范围
+#   2. 校验     —— 本地跑 pytest（CI 门禁），坏代码绝不上线
+#   3. 按需构建 —— 仅前端变化时本地预构建（dist 被 gitignore，NAS 端会再构建）
+#   4. 提交推送 —— git commit + push origin main，推送后 SSH 到 NAS 执行真正部署
+#
+# 用法:
+#   ./scripts/deploy.sh            正常流程：检测 → 校验 → 按需构建 → 提交推送 → SSH 到 NAS 部署
+#   ./scripts/deploy.sh --dry       零副作用自测：只 检测/校验/构建，不提交不推送
+#   ./scripts/deploy.sh --no-push   提交但不推送（准备好后手动 push 或再跑一次）
+#   ./scripts/deploy.sh --no-ssh    推送后不自动 SSH 部署（仅同步代码到 NAS git）
+#   ./scripts/deploy.sh --skip-tests  跳过 pytest（谨慎）
+#   ./scripts/deploy.sh --skip-fe     跳过前端构建（即便前端变化）
+#   ./scripts/deploy.sh --build-fe    强制重新构建前端
+#   ./scripts/deploy.sh --help
+# =============================================================================
+set -euo pipefail
+
+REMOTE="origin"
+BRANCH="main"
+ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+cd "$ROOT"
+
+NO_PUSH=0
+DRY=0
+NO_SSH=0
+SKIP_TESTS=0
+SKIP_FE=0
+FORCE_FE=0
+while [[ $# -gt 0 ]]; do
+  case "$1" in
+    --no-push)    NO_PUSH=1 ;;
+    --dry)        DRY=1 ;;
+    --no-ssh)     NO_SSH=1 ;;
+    --skip-tests) SKIP_TESTS=1 ;;
+    --skip-fe)    SKIP_FE=1 ;;
+    --build-fe)   FORCE_FE=1 ;;
+    --help|-h)    sed -n '1,22p' "$0"; exit 0 ;;
+    *) echo "未知参数: $1"; exit 2 ;;
+  esac
+  shift
+done
+
+log(){ printf '\033[36m[deploy]\033[0m %s\n' "$*"; }
+ok(){  printf '\033[32m[ ok ]\033[0m %s\n' "$*"; }
+err(){ printf '\033[31m[err ]\033[0m %s\n' "$*" >&2; }
+
+# ---- 1. 变化检测 --------------------------------------------------------------
+log "同步远端引用..."
+git fetch --quiet "$REMOTE" "$BRANCH" 2>/dev/null || log "fetch 失败（离线？继续本地流程）"
+
+LOCAL_CHANGES="$(git status --porcelain)"
+AHEAD="$(git rev-list --count "$REMOTE/$BRANCH..HEAD" 2>/dev/null || echo 0)"
+
+# 汇总「已提交未推送」+「未提交」的变更文件，去重
+CHANGED_FILES="$(
+  { git diff --name-only "$REMOTE/$BRANCH" HEAD 2>/dev/null; echo "$LOCAL_CHANGES" | awk '{print $2}'; } \
+  | grep -v '^$' | sort -u
+)"
+
+has(){ echo "$CHANGED_FILES" | grep -qE "$1"; }
+
+BACKEND_CHANGED=0; FRONTEND_CHANGED=0; DEPS_CHANGED=0
+if has '^(app/|scripts/|tests/|.*\.py$|Dockerfile|docker-compose\.yml|\.env\.example)'; then BACKEND_CHANGED=1; fi
+if has '^frontend/(src/|public/|index\.html|package\.json|vite\.config|package-lock\.json)'; then FRONTEND_CHANGED=1; fi
+if has '^(pyproject\.toml|uv\.lock|Dockerfile)'; then DEPS_CHANGED=1; fi
+
+if [[ -z "$CHANGED_FILES" && "$AHEAD" -eq 0 ]]; then
+  ok "无代码变化，无需部署。"
+  exit 0
+fi
+log "变更范围 → backend=$BACKEND_CHANGED  frontend=$FRONTEND_CHANGED  deps=$DEPS_CHANGED  (ahead=$AHEAD)"
+
+# ---- 2. 校验（本地 CI gate）---------------------------------------------------
+if [[ "$SKIP_TESTS" -eq 0 && "$BACKEND_CHANGED" -eq 1 ]]; then
+  log "运行后端测试 (uv run pytest)..."
+  if ! uv run pytest -q --no-header 2>&1 | tail -n 30; then
+    err "测试失败，已中止部署。请修复后再运行 deploy.sh"
+    exit 1
+  fi
+  ok "后端测试通过"
+  # 顺带做一次非阻断的静态检查
+  if command -v uv >/dev/null 2>&1; then
+    log "运行 ruff 静态检查（不阻断）..."
+    uv run ruff check app tests 2>&1 | tail -n 15 || true
+  fi
+fi
+
+# ---- 3. 按需构建前端 ----------------------------------------------------------
+if [[ ("$FRONTEND_CHANGED" -eq 1 || "$FORCE_FE" -eq 1) && "$SKIP_FE" -eq 0 ]]; then
+  log "构建前端 (npm ci && npm run build)..."
+  if ! (cd frontend && npm ci --no-audit --no-fund && npm run build); then
+    err "前端构建失败，已中止。"
+    exit 1
+  fi
+  ok "前端构建完成 -> frontend/dist"
+  log "提示: frontend/dist 已被 gitignore，NAS 端 post-receive 会重新构建以生效。"
+fi
+
+# ---- 4. 提交 + 推送 -----------------------------------------------------------
+if [[ "$DRY" -eq 1 ]]; then
+  log "(--dry) 零副作用模式：跳过提交与推送。以上为检测结果，确认无误后可去掉 --dry 正式部署。"
+  exit 0
+fi
+
+if [[ -n "$LOCAL_CHANGES" ]]; then
+  log "暂存并提交本地变化..."
+  git add -A   # .gitignore 已排除 .env / data / node_modules / .workbuddy / .trae 等
+  SCOPE=""
+  [[ "$BACKEND_CHANGED" -eq 1 ]]  && SCOPE="$SCOPE backend"
+  [[ "$FRONTEND_CHANGED" -eq 1 ]] && SCOPE="$SCOPE frontend"
+  [[ "$DEPS_CHANGED" -eq 1 ]]     && SCOPE="$SCOPE deps"
+  MSG="chore(deploy): 自动部署 $(date '+%Y-%m-%d %H:%M')${SCOPE:+ |$SCOPE}"
+  git commit -q -m "$MSG"
+  ok "已提交: $MSG"
+fi
+
+if [[ "$NO_PUSH" -eq 1 ]]; then
+  log "(--no-push) 未推送。当前领先远端 $AHEAD 个提交，确认无误后去掉 --no-push 再跑。"
+  exit 0
+fi
+
+export QMDS_SKIP_PREPUSH=1   # 避免与 pre-push hook 重复跑测试
+log "推送到 $REMOTE/$BRANCH（触发 NAS 自动部署）..."
+if ! git push "$REMOTE" "$BRANCH"; then
+  err "推送失败：可能远端有更新，请先 git pull 解决冲突后重试。"
+  exit 1
+fi
+ok "已推送。代码已同步至 NAS git。"
+if [[ "$NO_SSH" -eq 0 ]]; then
+  log "通过 SSH 在 NAS 上执行部署..."
+  if bash "$ROOT/scripts/nas-deploy-ssh.sh"; then
+    ok "NAS 部署完成。"
+  else
+    err "NAS 部署脚本返回非 0（详见上方 NAS 输出）。"
+    exit 1
+  fi
+else
+  log "(--no-ssh) 已跳过 NAS 端部署。如需部署请运行: ./scripts/nas-deploy-ssh.sh"
+fi

scripts/install-hooks.sh

@@ -0,0 +1,48 @@
+#!/usr/bin/env bash
+# =============================================================================
+# 安装本地 git pre-push 门禁
+#
+# 作用：任何「推送到 origin（NAS）」的操作前，先跑后端测试；若前端有变化，
+#       先在本地预构建一次。任一环节失败则阻断推送，确保坏代码绝不上线。
+#
+# 说明：
+#   - deploy.sh 推送前会设 QMDS_SKIP_PREPUSH=1，避免与自身校验重复跑。
+#   - 仅对 origin 远端生效，推其他远端不受影响。
+#
+# 卸载：rm .git/hooks/pre-push
+# =============================================================================
+set -e
+
+ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+HOOK="$ROOT/.git/hooks/pre-push"
+mkdir -p "$(dirname "$HOOK")"
+
+cat > "$HOOK" <<'HOOK_EOF'
+#!/usr/bin/env bash
+# QMDSearch pre-push 校验门禁（由 scripts/install-hooks.sh 安装）
+REMOTE="$1"
+# 仅对 NAS 远端 (origin) 做校验；且 deploy.sh 已设 QMDS_SKIP_PREPUSH 时跳过
+if [[ "$REMOTE" != "origin" ]]; then exit 0; fi
+if [[ -n "${QMDS_SKIP_PREPUSH:-}" ]]; then exit 0; fi
+
+echo "[pre-push] 推送前校验后端测试..."
+if ! uv run pytest -q --no-header 2>&1 | tail -n 20; then
+  echo "[pre-push] 测试失败，已阻断推送。请先修复后再 push。" >&2
+  exit 1
+fi
+
+# 暂存区若含前端变化，本地先预构建一次，避免推到 NAS 才翻车
+if git diff --cached --name-only | grep -qE '^frontend/'; then
+  echo "[pre-push] 检测到前端变化，本地预构建..."
+  if ! (cd frontend && npm ci --no-audit --no-fund && npm run build); then
+    echo "[pre-push] 前端构建失败，已阻断推送。" >&2
+    exit 1
+  fi
+fi
+
+echo "[pre-push] 校验通过，允许推送。"
+HOOK_EOF
+
+chmod +x "$HOOK"
+echo "已安装 pre-push 门禁 -> $HOOK"
+echo "如需卸载: rm $HOOK"

scripts/nas-deploy-ssh.sh

@@ -0,0 +1,120 @@
+#!/usr/bin/env bash
+# =============================================================================
+# QMDSearch NAS 端部署（通过 SSH 直连执行，密钥免密）
+#
+# 环境约束（已实测）：
+#   - NAS 上有 docker compose v2，但【没有 node/npm】→ 前端必须在本地构建后 rsync 过去
+#   - compose 的 app 服务只有 image: 无 build: → 依赖变化用 `docker build -t qmdsearch-app:latest .`
+#   - kp 已在 docker 组，可免 sudo 跑 docker
+#   - 部署目录 /vol1/qmdsearch 是 git 工作树，以 origin/main 为唯一真相源
+#
+# 流程：
+#   1. SSH 到 NAS：git fetch + git reset --hard origin/main（同步受管代码），并算出变更范围
+#   2. 本地：若前端变化 → 本地 npm 构建 → rsync frontend/dist 到 NAS（NAS 无 node）
+#   3. SSH 到 NAS：依赖变化则 docker build + force-recreate；否则 restart app
+#   4. 健康检查 /health；失败回滚到上一稳定提交
+#
+# 用法:
+#   ./scripts/nas-deploy-ssh.sh           正式部署
+#   ./scripts/nas-deploy-ssh.sh --dry-run 只预览将要部署的变更，不重启
+#   环境变量：NAS_HOST（默认 nas） DEPLOY_DIR（默认 /vol1/qmdsearch） APP_PORT（默认 8000）
+# =============================================================================
+set -uo pipefail
+
+DEPLOY_HOST="${NAS_HOST:-nas}"
+DEPLOY_DIR="${DEPLOY_DIR:-/vol1/qmdsearch}"
+APP_PORT="${APP_PORT:-8000}"
+ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+DRY_RUN=0
+[[ "${1:-}" == "--dry-run" ]] && DRY_RUN=1
+
+# 前置检查
+if ! ssh -o BatchMode=yes -o ConnectTimeout=5 "$DEPLOY_HOST" true 2>/dev/null; then
+  echo "[local] 无法免密 SSH 到 $DEPLOY_HOST，请先配置 SSH 密钥登录。" >&2
+  exit 1
+fi
+command -v rsync >/dev/null 2>&1 || { echo "[local] 本机缺少 rsync，请先安装。" >&2; exit 1; }
+
+# ---- Phase 1: NAS 同步受管代码 + 计算变更范围 ----
+SCOPE=$(ssh "$DEPLOY_HOST" "bash -s -- '${DEPLOY_DIR}'" <<'REMOTE'
+DEPLOY_DIR="$1"
+set -uo pipefail
+cd "$DEPLOY_DIR" || { echo "NOCD"; exit 1; }
+git fetch origin main 2>&1 | tail -n 2
+NEW=$(git rev-parse origin/main)
+PREV=$(cat .last_deployed_commit 2>/dev/null || echo "NONE")
+if [[ "$PREV" == "$NEW" ]]; then echo "UP_TO_DATE"; exit 0; fi
+OLD_HEAD=$(git rev-parse HEAD)
+git reset --hard origin/main 2>&1 | tail -n 2
+BASE="$OLD_HEAD"
+[[ "$PREV" != "NONE" ]] && BASE="$PREV"
+CHANGED=$(git diff --name-only "$BASE" "$NEW")
+FE=0; DEPS=0
+echo "$CHANGED" | grep -qE '^frontend/(src/|public/|index\.html|package\.json|vite\.config|package-lock\.json)' && FE=1
+echo "$CHANGED" | grep -qE '^(pyproject\.toml|uv\.lock|Dockerfile)' && DEPS=1
+echo "SCOPE FE=$FE DEPS=$DEPS PREV=$PREV NEW=$NEW"
+REMOTE
+)
+
+if echo "$SCOPE" | grep -q "NOCD"; then echo "[nas] 无法进入 $DEPLOY_DIR"; exit 1; fi
+if echo "$SCOPE" | grep -q "UP_TO_DATE"; then echo "[nas] 已是最新，无需更新。"; exit 0; fi
+
+FE=$(echo "$SCOPE"   | grep -oE 'FE=[01]'   | cut -d= -f2)
+DEPS=$(echo "$SCOPE" | grep -oE 'DEPS=[01]' | cut -d= -f2)
+PREV=$(echo "$SCOPE" | grep -oE 'PREV=[0-9a-f]+' | cut -d= -f2)
+NEW=$(echo "$SCOPE"  | grep -oE 'NEW=[0-9a-f]+'  | cut -d= -f2)
+echo "[deploy] 将部署 $PREV -> $NEW  (FE=$FE DEPS=$DEPS)"
+
+if [[ "$DRY_RUN" == "1" ]]; then
+  echo "[deploy][dry-run] 变更文件："
+  if [[ "$PREV" != "NONE" ]]; then
+    ssh "$DEPLOY_HOST" "cd $DEPLOY_DIR && git diff --name-only $PREV $NEW" 2>/dev/null | sed 's/^/    /'
+  else
+    echo "    (首次部署，将全量同步受管代码)"
+  fi
+  exit 0
+fi
+
+# ---- Phase 2: 前端本地构建 + rsync 到 NAS（NAS 无 node）----
+if [[ "$FE" == "1" ]]; then
+  echo "[local] 本地构建前端..."
+  if ! (cd "$ROOT/frontend" && npm ci --no-audit --no-fund && npm run build); then
+    echo "[local][WARN] 前端构建失败（当前环境可能受限或缺少 node 依赖），跳过前端同步；后端仍会部署，但 frontend/dist 可能为旧版本。待在可正常构建的环境中重跑即补齐。"
+  else
+    echo "[local] rsync frontend/dist -> $DEPLOY_HOST:$DEPLOY_DIR/frontend/dist"
+    if ! rsync -az --delete "$ROOT/frontend/dist/" "$DEPLOY_HOST:$DEPLOY_DIR/frontend/dist/"; then
+      echo "[local][WARN] rsync 失败，跳过前端同步。"
+    fi
+  fi
+fi
+
+# ---- Phase 3: NAS 容器部署 + 健康检查 ----
+ssh "$DEPLOY_HOST" "bash -s -- '$DEPLOY_DIR' '$APP_PORT' '$DEPS' '$NEW' '$PREV'" <<'REMOTE'
+DEPLOY_DIR="$1"; APP_PORT="$2"; DEPS="$3"; NEW="$4"; PREV="$5"
+set -uo pipefail
+cd "$DEPLOY_DIR"
+if [[ "$DEPS" == "1" ]]; then
+  echo "[nas] 依赖变化，重建镜像: docker build -t qmdsearch-app:latest ."
+  if ! docker build -t qmdsearch-app:latest . 2>&1 | tail -n 20; then
+    echo "[nas] 镜像构建失败，回滚到 $PREV"; git checkout -f "$PREV"; exit 1
+  fi
+  echo "[nas] 用新镜像重建容器: docker compose up -d --force-recreate app"
+  docker compose up -d --force-recreate app 2>&1 | tail -n 10
+else
+  echo "[nas] 仅代码/前端变化，重启 app: docker compose restart app"
+  docker compose restart app 2>&1 | tail -n 5
+fi
+echo "[nas] 健康检查 http://localhost:$APP_PORT/api/v1/health ..."
+for i in $(seq 1 20); do
+  if curl -fsS "http://localhost:$APP_PORT/api/v1/health" >/dev/null 2>&1; then
+    echo "$NEW" > .last_deployed_commit
+    echo "[nas] 健康检查通过 ✓ 部署完成"
+    exit 0
+  fi
+  sleep 3
+done
+echo "[nas] 健康检查失败，回滚到 $PREV"
+git checkout -f "$PREV"
+docker compose restart app
+exit 1
+REMOTE

scripts/nas-post-receive

@@ -0,0 +1,107 @@
+#!/usr/bin/env bash
+# =============================================================================
+# QMDSearch NAS 端自动部署钩子 (post-receive)
+#
+# 这是真正的「持续部署(CD)」：当本地 deploy.sh 把代码 push 到 NAS 后，
+# 此钩子自动完成 checkout → 按需构建 → 重启/重建容器 → 健康检查 → 失败回滚。
+#
+# 安装方式（二选一）：
+#   A. 作为 bare 仓库的 hook（推荐，全自动）：
+#      # 假设 NAS 上 bare 仓库在 /volume1/git/QMDSearch.git，工作树在 /volume1/docker/QMDSearch
+#      cp scripts/nas-post-receive /volume1/git/QMDSearch.git/hooks/post-receive
+#      chmod +x /volume1/git/QMDSearch.git/hooks/post-receive
+#      # 在该 bare 仓库 config 中设置：
+#      #   [core] bare = true
+#      #   [hooks] 并确保工作树目录已 clone 过一次
+#      # 首次需手动：git --work-tree=/volume1/docker/QMDSearch --git-dir=/volume1/git/QMDSearch.git checkout -f
+#
+#   B. 作为普通脚本手动触发（无 bare 仓库时）：
+#      cd /volume1/docker/QMDSearch && git pull && sudo bash scripts/nas-post-receive
+#
+# 前置依赖（NAS 上需具备）：git、docker、docker compose、node/npm
+# 可调环境变量：DEPLOY_DIR（工作树）、APP_PORT、LOG_FILE
+# =============================================================================
+set -uo pipefail
+
+DEPLOY_DIR="${DEPLOY_DIR:-/volume1/docker/QMDSearch}"   # ← 改成你的 NAS 部署目录
+GIT_DIR="$DEPLOY_DIR/.git"
+LOG_FILE="${LOG_FILE:-$DEPLOY_DIR/deploy.log}"
+APP_PORT="${APP_PORT:-8000}"
+PREV_COMMIT_FILE="$DEPLOY_DIR/.last_deployed_commit"
+
+log(){ printf '%s [nas-deploy] %s\n' "$(date '+%Y-%m-%d %H:%M:%S')" "$*" | tee -a "$LOG_FILE"; }
+
+rollback(){
+  log "回滚到上一稳定提交: ${PREV:-<无>}"
+  if [[ -n "${PREV:-}" ]]; then
+    git --work-tree="$DEPLOY_DIR" --git-dir="$GIT_DIR" checkout -f "$PREV"
+    (cd "$DEPLOY_DIR" && docker compose restart app) | tee -a "$LOG_FILE"
+    echo "$PREV" > "$PREV_COMMIT_FILE"
+  fi
+}
+
+# 解析 post-receive 通过 stdin 传入的 <old> <new> <ref>
+newrev=""; refname=""
+while read -r oldrev newrev refname; do :; done
+# 仅处理 main 分支的推送；手动运行时 newrev 为空，回退到 HEAD
+if [[ -z "$newrev" ]]; then
+  newrev="$(git --git-dir="$GIT_DIR" rev-parse HEAD 2>/dev/null || echo "")"
+fi
+if [[ -n "$refname" && "$refname" != "refs/heads/main" ]]; then
+  log "忽略非 main 分支推送 ($refname)"
+  exit 0
+fi
+
+log "=== 开始部署 (newrev=${newrev:0:8}) ==="
+
+# 1) 检出到工作树
+git --work-tree="$DEPLOY_DIR" --git-dir="$GIT_DIR" checkout -f
+cd "$DEPLOY_DIR" || { log "无法进入 $DEPLOY_DIR"; exit 1; }
+
+# 2) 计算变化范围
+PREV="$(cat "$PREV_COMMIT_FILE" 2>/dev/null || echo "")"
+if [[ -n "$PREV" && -n "$newrev" ]]; then
+  CHANGED="$(git --git-dir="$GIT_DIR" diff --name-only "$PREV" "$newrev")"
+else
+  CHANGED="$(git --git-dir="$GIT_DIR" show --name-only --format= "$newrev")"
+fi
+
+has(){ echo "$CHANGED" | grep -qE "$1"; }
+FE_CHANGED=0; DEPS_CHANGED=0
+has '^frontend/(src/|public/|index\.html|package\.json|vite\.config|package-lock\.json)' && FE_CHANGED=1
+has '^(pyproject\.toml|uv\.lock|Dockerfile)' && DEPS_CHANGED=1
+log "变更范围 → frontend=$FE_CHANGED  deps=$DEPS_CHANGED"
+
+# 3) 按需构建前端（dist 不入库，必须在此构建）
+if [[ "$FE_CHANGED" -eq 1 ]]; then
+  log "检测前端变化，重新构建..."
+  if ! (cd frontend && npm ci --no-audit --no-fund && npm run build) 2>&1 | tee -a "$LOG_FILE"; then
+    log "前端构建失败，中止部署（保留上一版本）"; exit 1
+  fi
+  ok(){ :; }; log "前端构建完成"
+fi
+
+# 4) 部署容器
+if [[ "$DEPS_CHANGED" -eq 1 ]]; then
+  log "依赖变化，重建镜像并重启: docker compose up -d --build app"
+  docker compose up -d --build app 2>&1 | tee -a "$LOG_FILE"
+else
+  log "仅代码变化，重启 app 服务: docker compose restart app"
+  docker compose restart app 2>&1 | tee -a "$LOG_FILE"
+fi
+
+# 5) 健康检查（失败自动回滚）
+log "健康检查 http://localhost:$APP_PORT/health ..."
+for i in $(seq 1 15); do
+  if curl -fsS "http://localhost:$APP_PORT/health" >/dev/null 2>&1; then
+    log "健康检查通过 ✓"
+    [[ -n "$newrev" ]] && echo "$newrev" > "$PREV_COMMIT_FILE"
+    log "=== 部署完成 ==="
+    exit 0
+  fi
+  sleep 2
+done
+
+log "健康检查失败，触发回滚"
+rollback
+exit 1

scripts/watch.sh

@@ -0,0 +1,58 @@
+#!/usr/bin/env bash
+# =============================================================================
+# QMDSearch 文件变化监视器（纯轮询，无需 fswatch / inotifywait）
+#
+# 作用：持续监听仓库代码变化，实现「每次更新代码后自动检测」。
+#   - 默认模式：检测到变化 → 本地跑测试 + 前端预构建 → 提示你运行 deploy.sh
+#   - --auto 模式：检测到变化 → 直接调用 ./scripts/deploy.sh 全自动部署
+#                  （注意：--auto 会在保存即提交推送，慎用，建议确认改动完整后再用）
+#
+# 用法:
+#   ./scripts/watch.sh           监听并本地校验，提示部署
+#   ./scripts/watch.sh --auto    监听变化后全自动部署
+#   ./scripts/watch.sh --interval 5   轮询间隔秒数（默认 3）
+# =============================================================================
+set -uo pipefail
+
+ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+cd "$ROOT"
+
+AUTO=0
+INTERVAL=3
+while [[ $# -gt 0 ]]; do
+  case "$1" in
+    --auto)     AUTO=1 ;;
+    --interval) INTERVAL="${2:-3}"; shift ;;
+    *) echo "未知参数: $1"; exit 2 ;;
+  esac
+  shift
+done
+
+SNAPSHOT="$(mktemp)"
+record(){ git status --porcelain > "$SNAPSHOT"; }
+record
+
+if [[ "$AUTO" -eq 1 ]]; then
+  echo "[watch] 监听 $ROOT（AUTO 模式：变化即部署，间隔 ${INTERVAL}s，Ctrl-C 退出）"
+else
+  echo "[watch] 监听 $ROOT（校验模式：变化即本地校验并提示，间隔 ${INTERVAL}s，Ctrl-C 退出）"
+fi
+
+while true; do
+  sleep "$INTERVAL"
+  if ! git status --porcelain | diff -q - "$SNAPSHOT" >/dev/null 2>&1; then
+    echo "[watch] $(date '+%H:%M:%S') 检测到代码变化"
+    if [[ "$AUTO" -eq 1 ]]; then
+      bash "$ROOT/scripts/deploy.sh"
+    else
+      echo "[watch] 运行本地校验..."
+      uv run pytest -q --no-header 2>&1 | tail -n 10 || true
+      if git status --porcelain | awk '{print $2}' | grep -qE '^frontend/'; then
+        echo "[watch] 前端变化，本地预构建..."
+        (cd frontend && npm ci --no-audit --no-fund && npm run build) 2>&1 | tail -n 5 || true
+      fi
+      echo "[watch] 校验完成。确认无误后运行: ./scripts/deploy.sh"
+    fi
+    record
+  fi
+done

skills/qmdsearch-agent/SKILL.md

@@ -0,0 +1,145 @@
+---
+name: qmdsearch-agent
+description: 通过 QMDSearch REST API 让 AI Agent 上传文档（文本/文件）并执行分层检索查询。配置服务地址与 Bearer Token 后，即可在对话中入库知识库并检索答案。
+metadata: {"clawdbot":{"emoji":"📚"}}
+---
+
+# QMDSearch Agent 接入 Skill
+
+本 skill 指导 AI Agent 调用 **QMDSearch**（面向 AI Agent 的分层信息检索服务）的 HTTP API，完成两类核心操作：
+
+1. **上传文档**：把文本或文件送入知识库（异步入库，自动三级总结 + 向量化）。
+2. **查询 / 检索**：对知识库做分层语义检索，返回最相关原文 chunk。
+
+## 何时使用
+
+- 用户希望把一份文档 / 笔记 / 网页内容加入知识库 → 调用上传接口。
+- 用户就知识库内容提问 → 调用检索接口。
+- 用户想确认文档是否入库成功 → 轮询任务状态。
+- 需要列出 / 删除知识库文档、查看统计 → 调用对应只读 / 管理接口。
+
+## 前置配置（必须）
+
+启用前先确定：
+
+- `QMDSEARCH_BASE_URL`：服务基地址，例如 `http://localhost:8000` 或 NAS 地址 `http://<nas-ip>:8000`。
+- `QMDSEARCH_TOKEN`：Bearer Token。通过 `POST {BASE}/api/v1/auth/login`（用户名 + 密码）获取；session TTL 12h，失效后重新登录。
+
+> 所有变更类请求（上传 / 删除）与检索请求都必须在 Header 携带 `Authorization: Bearer <token>`。
+> 当前代码实现中，查询类只读接口（文档列表 / 详情 / 类目 / 统计 / 健康）同样要求 Bearer（与登录态绑定），请始终携带 token 以避免 `1005` 未认证。
+
+## 统一约定
+
+- 路径前缀：`{BASE}/api/v1`
+- 响应体：`{ "code": 0, "data": {...}, "message": "ok" }`，`code=0` 成功；`1xxx` 客户端错误；`2xxx` 服务端错误。
+- 入库是**异步**的：上传返回 `202 + task_id`，需轮询 `GET /documents/tasks/{task_id}` 直到 `status=done`（或 `failed`）。
+
+## 1. 获取 Token
+
+```bash
+curl -X POST {BASE}/api/v1/auth/login \
+  -H "Content-Type: application/json" \
+  -d '{"username":"admin","password":"<password>"}'
+# → {"code":0,"data":{"token":"<JWT>","username":"admin","role":"admin","must_change_password":false},"message":"ok"}
+```
+
+`must_change_password=true` 时，登录成功但调用其他接口会返回 `1006`，需先 `POST /api/v1/auth/password` 改密。
+
+## 2. 上传文档（文本）
+
+`POST {BASE}/api/v1/documents`，JSON body `{text, title?, source?, metadata?}`。
+
+```bash
+curl -X POST {BASE}/api/v1/documents \
+  -H "Authorization: Bearer <token>" \
+  -H "Content-Type: application/json" \
+  -d '{"text":"QMDSearch 支持 L1/L2/L3 分层检索……", "title":"QMDSearch 介绍", "source":"notes"}'
+# → 202 {"code":0,"data":{"task_id":"<uuid>","status":"pending"},"message":"ok"}
+```
+
+- `text` 必填且非空；`title` / `source` 可选；`metadata` 为 `dict[str,str]`。
+- 相同 `sha256(text)` 会命中**去重**，直接复用旧文档（`data` 中 `deduplicated=true`）。
+
+## 3. 上传文件
+
+`POST {BASE}/api/v1/documents/upload`，`multipart/form-data`：
+
+- `file`：必填，支持 `.txt .md .html .htm .pdf .docx`
+- `title` / `source`：可选（默认取原文件名）
+- `metadata`：可选 JSON 字符串
+
+```bash
+curl -X POST {BASE}/api/v1/documents/upload \
+  -H "Authorization: Bearer <token>" \
+  -F "file=@document.pdf" \
+  -F "source=manual"
+# → 202 {"code":0,"data":{"task_id":"<uuid>","status":"pending","saved_path":"..."},"message":"ok"}
+```
+
+服务端按扩展名提取纯文本（PDF 扫描件自动 OCR 降级），再复用同一入库流水线。
+
+## 4. 轮询入库任务
+
+`GET {BASE}/api/v1/documents/tasks/{task_id}`（无需鉴权）：
+
+```bash
+curl {BASE}/api/v1/documents/tasks/<task_id>
+# done   → {"code":0,"data":{"task_id":...,"status":"done","result":{...文档ID/总结/分类...},"created_at":...,"updated_at":...}}
+# failed → {"code":0,"data":{"task_id":...,"status":"failed","error":"..."}}
+```
+
+建议：提交后每 1–2s 轮询，直到 `status∈{done,failed}`；`done` 的 `result.document_id` 即入库文档 ID。
+
+## 5. 检索（查询）
+
+`POST {BASE}/api/v1/search`，JSON `{query, top_k?, summarize?}`：
+
+```bash
+curl -X POST {BASE}/api/v1/search \
+  -H "Authorization: Bearer <token>" \
+  -H "Content-Type: application/json" \
+  -d '{"query":"如何配置 Redis 缓存","top_k":5,"summarize":false}'
+```
+
+返回 `data`：`{query, hits[], routed_categories[], fallback, extracted_info, summary?}`。
+每个 `hit`：`{text, doc_id, title, section_path, score, doc_summary}`。
+
+- `top_k`：返回条数（默认 `RETRIEVAL_FINAL_K`，通常 5）。
+- `summarize=true`：额外返回 `summary`（对检索结果做 AI 总结）。
+- `fallback=true`：走了全库兜底（未走类目路由），召回较广。
+- `extracted_info`：query 解析出的关键词 / 实体 / 意图 / 命中类目。
+
+## 6. 其他管理接口（按需）
+
+| 方法 | 路径 | 说明 |
+|------|------|------|
+| POST | `/documents` | 文本入库（202 + task_id） |
+| POST | `/documents/upload` | 文件入库（202 + task_id） |
+| GET | `/documents/tasks/{task_id}` | 任务状态（done / failed） |
+| GET | `/documents?limit=&offset=` | 文档列表（分页游标） |
+| GET | `/documents/{doc_id}` | 文档详情（L1 + L2/L3 + chunks 数） |
+| GET | `/documents/{doc_id}/file` | 下载原始文件 |
+| DELETE | `/documents/{doc_id}` | 删除文档（幂等） |
+| GET | `/knowledge/categories` | 知识分类类目集 |
+| GET | `/knowledge/stats` | 统计（四层点数 + 类目分布） |
+| GET | `/health` | 健康检查 |
+
+## 错误码速查
+
+- `1001` 参数 / 格式错误（空文本、不支持的类型、弱密码、超大小）
+- `1004` 资源不存在（task / doc 不存在）
+- `1005` 未认证或凭证无效 / 账号已禁用
+- `1006` 权限不足 / 首次登录须先改密
+- `2000` 服务端内部错误（检索 / 入库 / 列表失败）
+- `2001` 认证服务暂不可用
+
+## Agent 最佳实践
+
+1. **先登录拿 token**，缓存 12h，失效再用 `/auth/login` 刷新。
+2. **上传走异步**：拿到 `task_id` 后轮询，不要在对话里阻塞等待，可告诉用户“正在入库，稍后查询”。
+3. **去重友好**：重复提交同文本会自动复用，不必担心重复。
+4. **大文件 / 批量**：逐文件上传并各自轮询；注意 `UPLOAD_MAX_SIZE_MB`（默认 20MB）。
+5. **检索调参**：召回不足调大 `top_k`；需要摘要设 `summarize=true`；关注 `fallback` 判断是否需要改写 query。
+6. **凭证安全**：token 等同会话，不要写入日志或外发。
+
+完整请求 / 响应示例（含 Python 客户端封装）见 `references/api-examples.md`。

skills/qmdsearch-agent/references/api-examples.md

@@ -0,0 +1,170 @@
+# QMDSearch API 示例与 Python 客户端封装
+
+> 基址占位符 `{BASE}` 替换为实际服务地址，例如 `http://localhost:8000` 或 `http://<nas-ip>:8000`。
+
+## 一、curl 示例
+
+### 登录
+
+```bash
+curl -s -X POST {BASE}/api/v1/auth/login \
+  -H "Content-Type: application/json" \
+  -d '{"username":"admin","password":"<password>"}'
+```
+
+### 文本入库（202 + task_id）
+
+```bash
+curl -s -X POST {BASE}/api/v1/documents \
+  -H "Authorization: Bearer <token>" \
+  -H "Content-Type: application/json" \
+  -d '{"text":"QMDSearch 是面向 AI Agent 的分层信息检索服务……","title":"QMDSearch 介绍","source":"notes"}'
+```
+
+### 文件上传（202 + task_id）
+
+```bash
+curl -s -X POST {BASE}/api/v1/documents/upload \
+  -H "Authorization: Bearer <token>" \
+  -F "file=@document.pdf" \
+  -F "source=manual"
+```
+
+### 轮询任务状态
+
+```bash
+curl -s {BASE}/api/v1/documents/tasks/<task_id>
+```
+
+### 检索
+
+```bash
+curl -s -X POST {BASE}/api/v1/search \
+  -H "Authorization: Bearer <token>" \
+  -H "Content-Type: application/json" \
+  -d '{"query":"如何配置 Redis 缓存","top_k":5,"summarize":false}'
+```
+
+## 二、Python 客户端封装
+
+可直接在 Agent 工具代码里复用：
+
+```python
+import time
+import requests
+
+
+class QMDSearchClient:
+    """QMDSearch 最小客户端：登录、入库（文本/文件）、轮询、检索。"""
+
+    def __init__(self, base_url: str, username: str, password: str,
+                 poll_interval: float = 1.5, poll_timeout: float = 120.0):
+        self.base = base_url.rstrip("/")
+        self.username = username
+        self.password = password
+        self.poll_interval = poll_interval
+        self.poll_timeout = poll_timeout
+        self.token: str | None = None
+
+    # ---- 鉴权 ----
+    def login(self) -> str:
+        r = requests.post(
+            f"{self.base}/api/v1/auth/login",
+            json={"username": self.username, "password": self.password},
+            timeout=30,
+        )
+        r.raise_for_status()
+        body = r.json()
+        if body.get("code") != 0:
+            raise RuntimeError(f"登录失败: {body}")
+        self.token = body["data"]["token"]
+        return self.token
+
+    def _headers(self) -> dict:
+        if not self.token:
+            self.login()
+        return {"Authorization": f"Bearer {self.token}"}
+
+    def _ok(self, resp: requests.Response):
+        resp.raise_for_status()
+        body = resp.json()
+        if body.get("code") != 0:
+            raise RuntimeError(f"API 错误 code={body.get('code')} msg={body.get('message')}")
+        return body["data"]
+
+    # ---- 入库 ----
+    def ingest_text(self, text: str, title: str = "", source: str = "",
+                    metadata: dict | None = None) -> str:
+        data = {"text": text}
+        if title:
+            data["title"] = title
+        if source:
+            data["source"] = source
+        if metadata:
+            data["metadata"] = {str(k): str(v) for k, v in metadata.items()}
+        resp = requests.post(f"{self.base}/api/v1/documents",
+                             headers=self._headers(), json=data, timeout=30)
+        return self._ok(resp)["task_id"]
+
+    def upload_file(self, path: str, title: str = "", source: str = "",
+                    metadata: dict | None = None) -> str:
+        files = {"file": open(path, "rb")}
+        data = {}
+        if title:
+            data["title"] = title
+        if source:
+            data["source"] = source
+        if metadata:
+            data["metadata"] = str({str(k): str(v) for k, v in metadata.items()})
+        resp = requests.post(f"{self.base}/api/v1/documents/upload",
+                             headers=self._headers(), files=files, data=data, timeout=60)
+        return self._ok(resp)["task_id"]
+
+    # ---- 轮询 ----
+    def wait_task(self, task_id: str) -> dict:
+        deadline = time.time() + self.poll_timeout
+        while time.time() < deadline:
+            resp = requests.get(f"{self.base}/api/v1/documents/tasks/{task_id}", timeout=30)
+            data = self._ok(resp)
+            if data["status"] == "done":
+                return data["result"]
+            if data["status"] == "failed":
+                raise RuntimeError(f"入库失败: {data.get('error')}")
+            time.sleep(self.poll_interval)
+        raise TimeoutError(f"任务 {task_id} 轮询超时")
+
+    def ingest_text_wait(self, *args, **kwargs) -> dict:
+        return self.wait_task(self.ingest_text(*args, **kwargs))
+
+    def upload_file_wait(self, *args, **kwargs) -> dict:
+        return self.wait_task(self.upload_file(*args, **kwargs))
+
+    # ---- 检索 ----
+    def search(self, query: str, top_k: int | None = None,
+               summarize: bool = False) -> dict:
+        payload = {"query": query, "summarize": summarize}
+        if top_k is not None:
+            payload["top_k"] = top_k
+        resp = requests.post(f"{self.base}/api/v1/search",
+                             headers=self._headers(), json=payload, timeout=60)
+        return self._ok(resp)
+
+
+# 用法
+if __name__ == "__main__":
+    client = QMDSearchClient("http://localhost:8000", "admin", "<password>")
+    client.login()
+    # 文本入库并等待完成
+    result = client.ingest_text_wait("这是一篇关于分层检索的笔记……", title="笔记")
+    print("document_id =", result["document_id"])
+    # 检索
+    hits = client.search("分层检索是什么", top_k=5)["hits"]
+    for h in hits:
+        print(f"[{h['score']:.3f}] {h['title']}: {h['text'][:80]}")
+```
+
+## 三、响应结构速记
+
+- 文本 / 文件入库 `202` → `data = {task_id, status:"pending"[, saved_path]}`
+- 任务 `done` → `data.result = {document_id, summary:{l1_summary,l2_outline,l3_content_outline,level}, category, chunks_count, tags, category_confidence, deduplicated}`
+- 检索 → `data = {query, hits:[{text,doc_id,title,section_path,score,doc_summary}], routed_categories, fallback, extracted_info, summary?}`

tests/conftest.py

@@ -1,39 +1,23 @@
-"""pytest 全局夹具：覆盖 JWT 认证依赖，让现有 API 测试默认以 admin 身份运行
+"""pytest 全局夹具
 
-业务接口（search/knowledge/settings）仍使用 app.core.auth 的 JWT 依赖，
+鉴权采用 app.api.deps 的会话体系（UserStore / SessionStore）：
-这里通过 autouse 夹具把两个依赖统一替换为返回固定 admin AuthUser 的 lambda，
+- 文档变更类端点（POST /documents、/documents/upload、DELETE /documents/{id}）、
-使现有 API 测试无需改动即可通过认证。单个测试需要走真实认证逻辑时，
+  /auth/* 与 Settings 变更端点需要登录；
-可在测试函数内 pop 掉对应 override，autouse fixture yield 后会统一 clear。
+- 查询类端点（POST /search、GET /knowledge/*、GET /documents*、
+  GET /documents/{id}/file）免登录。
 
-文档变更类端点（POST /documents、/documents/upload、DELETE /documents/{id}）
+auth_stores / admin_headers 夹具注入内存存储并签发真实 session token，
-使用 app.api.deps 的会话认证（UserStore/SessionStore），由 auth_stores /
+供需要登录的接口测试使用；查询类端点测试直接无 token 调用即可，无需覆盖依赖。
-admin_headers 夹具注入内存存储并签发真实 session token。
 
 另外清理 LLM 客户端 / 解析插件 / 去重策略三处进程级缓存，避免跨测试串扰
 （不同测试可能注入不同的 redis/ollama 实例，缓存按配置签名而非实例区分）。
 """
 
 import asyncio
-from datetime import UTC, datetime
 from typing import Any
 
 import pytest
 
-from app.core.auth import get_current_user, require_admin
-from app.main import app
-from app.models.auth import AuthUser
-
-TEST_USER = AuthUser(username="testuser", role="admin", created_at=datetime.now(UTC))
-
-
-@pytest.fixture(autouse=True)
-def override_auth():
-    """所有测试默认以 admin 身份运行（旧 JWT 依赖）；测试结束清理 dependency_overrides"""
-    app.dependency_overrides[get_current_user] = lambda: TEST_USER
-    app.dependency_overrides[require_admin] = lambda: TEST_USER
-    yield
-    app.dependency_overrides.clear()
-
 
 @pytest.fixture
 def auth_stores(monkeypatch: pytest.MonkeyPatch):

tests/test_admin_integration.py

@@ -204,11 +204,13 @@ class TestAdminClosedLoop:
             assert item_b["category"] == DOC_B_CATEGORY
             assert item_b["tags"] == []
 
-            # 2. 详情：l1/l2_nodes/l3_nodes/chunks_count 与写入一致
+            # 2. 详情：l1/l2_nodes/l3_nodes/chunks_count/file 与写入一致
             resp = client.get(f"/api/v1/documents/{DOC_A}")
             assert resp.status_code == 200
             data = resp.json()["data"]
-            assert set(data.keys()) == {"l1", "l2_nodes", "l3_nodes", "chunks_count"}
+            assert set(data.keys()) == {"l1", "l2_nodes", "l3_nodes", "chunks_count", "file"}
+            # 文本入库无 metadata → file=None
+            assert data["file"] is None
             l1 = data["l1"]
             assert l1["doc_id"] == DOC_A
             assert l1["title"] == DOC_A_TITLE

tests/test_admin_page.py

@@ -5,14 +5,16 @@
 2. 五区块可识别标记（文案与 section id）
 3. 零外部依赖：无 http(s) 外链资源、无 CDN 引用
 4. fetch 调用路径与后端 API 契约一致（含 /api/v1/auth/*）
-5. 删除操作的 confirm() 二次确认逻辑
+5. 删除/重置操作的通用 overlay 弹窗组件（替换原生 prompt/confirm）
 6. 登录门禁：登录卡片、localStorage key、/auth/me 验证、/auth/login 路径
 7. 顶栏用户区：用户名、角色徽章、修改密码、退出登录
 8. 修改密码：旧/新/确认表单、must_change_password 强制改密
-9. 用户管理区块：列表/创建表单/角色下拉/重置/删除 confirm、admin 角色门禁
+9. 用户管理区块：列表/创建表单/角色下拉/重置密码弹窗/删除确认弹窗、admin 角色门禁
 10. 请求拦截：Authorization Bearer 注入、1005 回登录、1006 错误条
 11. API 指南区块：导航/section、API_GUIDE 清单与真实路由一致性、试一下面板、
    curl 复制、auth 标注、upload 文件选择、禁止自定义 URL
+12. 入库进度区块（Task 2）：section/nav、表格结构、2s 轮询启停、状态徽章复用、
+    done/failed 操作按钮、reingest 端点、批量上传 multiple + upload-batch 路径
 """
 
 import re
@@ -39,9 +41,20 @@ def client(monkeypatch: pytest.MonkeyPatch) -> Iterator[TestClient]:
 
 @pytest.fixture
 def admin_html(client: TestClient) -> str:
-    """请求 /admin 并返回 HTML 文本（前置断言 200）"""
+    """请求 /admin 并返回旧版 admin.html 文本（前置断言 200）
+
+    当 app/static/admin 符号链接存在时（Vue SPA 部署），/admin 端点会返回 SPA 入口
+    而非旧版 admin.html。此时直接读取 admin.html 文件内容进行测试。
+    """
     resp = client.get("/admin")
     assert resp.status_code == 200
+    # 检测是否返回了 Vue SPA（无 section-overview 标记）
+    if "section-overview" not in resp.text:
+        # SPA 模式：直接读取旧版 admin.html 文件
+        from pathlib import Path
+
+        admin_html_path = Path(__file__).resolve().parent.parent / "app" / "static" / "admin.html"
+        return admin_html_path.read_text(encoding="utf-8")
     return resp.text
 
 
@@ -86,6 +99,8 @@ def test_admin_page_fetch_paths(admin_html: str) -> None:
         "/api/v1/auth/logout",
         "/api/v1/auth/password",
         "/api/v1/auth/users",
+        "/api/v1/documents/upload-batch",
+        "/api/v1/documents/tasks",
     ):
         assert path in admin_html
 
@@ -107,6 +122,13 @@ def test_admin_page_fetch_paths_in_real_routes(admin_html: str) -> None:
     """页面 api() 调用路径均在真实后端路由集合内（含 tasks 与 /api/v1/auth/* 路径）"""
     route_paths = _collect_route_paths(app.routes)
     assert "/api/v1/documents/tasks/{task_id}" in route_paths
+    # Task 2 新增端点在真实路由集合内
+    for new_path in (
+        "/api/v1/documents/tasks",
+        "/api/v1/documents/upload-batch",
+        "/api/v1/documents/{doc_id}/reingest",
+    ):
+        assert new_path in route_paths, f"后端缺少 Task 2 新增端点: {new_path}"
     for auth_path in (
         "/api/v1/auth/login",
         "/api/v1/auth/me",
@@ -140,9 +162,103 @@ def test_admin_page_ingest_polling(admin_html: str) -> None:
     assert "disabled" in admin_html
 
 
-def test_admin_page_has_confirm(admin_html: str) -> None:
+def test_admin_page_progress_section(admin_html: str) -> None:
-    """删除操作包含 confirm() 二次确认逻辑"""
+    """入库进度区块：导航按钮、section、表格结构、刷新按钮、空列表占位"""
-    assert "confirm(" in admin_html
+    # section 与导航按钮（所有登录用户可见）
+    assert 'id="section-progress"' in admin_html
+    assert 'id="nav-progress"' in admin_html
+    assert 'data-target="section-progress"' in admin_html
+    assert "入库进度" in admin_html
+    # 刷新按钮与状态统计
+    assert 'id="btn-refresh-progress"' in admin_html
+    assert 'id="progress-stats"' in admin_html
+    # 表格 tbody 与表头列：文件名 / 状态 / 创建时间 / 更新时间 / 操作
+    assert 'id="progress-tbody"' in admin_html
+    for col in ("文件名", "状态", "创建时间", "更新时间", "操作"):
+        assert col in admin_html
+    # 空列表占位文案
+    assert "暂无入库任务" in admin_html
+
+
+def test_admin_page_progress_polling(admin_html: str) -> None:
+    """入库进度区块：2s 轮询常量、startProgressPolling/stopProgressPolling 函数、GET /documents/tasks 路径"""
+    # 进度区块专用轮询常量（2s）
+    assert "PROGRESS_POLL_INTERVAL_MS" in admin_html
+    assert "2000" in admin_html
+    # 轮询启停函数
+    assert "function startProgressPolling" in admin_html
+    assert "function stopProgressPolling" in admin_html
+    assert "progressPollTimer" in admin_html
+    # 数据来源：GET /documents/tasks?limit=50
+    assert "/api/v1/documents/tasks?limit=50" in admin_html
+    # loadProgress / renderProgress / buildProgressRow 函数
+    assert "function loadProgress" in admin_html
+    assert "function renderProgress" in admin_html
+    assert "function buildProgressRow" in admin_html
+
+
+def test_admin_page_progress_polling_lifecycle(admin_html: str) -> None:
+    """进度区块轮询生命周期：activateSection 切走暂停、切回恢复；showLogin 停止"""
+    # activateSection 开头调用 stopProgressPolling（切走即暂停）
+    assert "stopProgressPolling();" in admin_html
+    # 切到 section-progress 时恢复轮询
+    assert 'targetId === "section-progress"' in admin_html
+    assert 'startProgressPolling()' in admin_html
+    # showLogin 退出时停止轮询
+    assert "stopProgressPolling" in admin_html
+
+
+def test_admin_page_progress_status_badge_reuse(admin_html: str) -> None:
+    """进度表格状态徽章复用 INGEST_STATUS_TEXT 中文映射与 status-* 配色"""
+    # statusBadgeEl 共享构造函数（无 id 可重复使用）
+    assert "function statusBadgeEl" in admin_html
+    # makeStatusBadge 复用 statusBadgeEl + 加 id
+    assert "statusBadgeEl(status)" in admin_html
+    # 状态徽章 CSS 类复用
+    for cls in ("status-running", "status-done", "status-failed"):
+        assert cls in admin_html
+
+
+def test_admin_page_progress_actions(admin_html: str) -> None:
+    """进度表格操作列：done 查看文档+重新入库、failed 重试、进行中无操作；reingest 端点"""
+    # done 状态操作按钮
+    assert "查看文档" in admin_html
+    assert "重新入库" in admin_html
+    # failed 状态重试按钮
+    assert "重试" in admin_html
+    # reingestDocument 函数与 POST /documents/{doc_id}/reingest 端点
+    assert "function reingestDocument" in admin_html
+    assert "/reingest" in admin_html
+    # 查看文档：切到 section-docs 并调用 loadDocDetail
+    assert 'activateSection("section-docs")' in admin_html
+    assert "loadDocDetail" in admin_html
+    # 状态统计文案
+    assert "进行中" in admin_html
+
+
+def test_admin_page_batch_upload(admin_html: str) -> None:
+    """批量上传：input multiple 属性、upload-batch 端点、批量结果展示、自动切进度区块"""
+    # file input multiple 属性
+    assert 'id="upload-file"' in admin_html
+    assert "multiple" in admin_html
+    # 批量上传分支：files > 1 时走 upload-batch 端点
+    assert "/api/v1/documents/upload-batch" in admin_html
+    assert 'fileInput.files.length > 1' in admin_html
+    # FormData 用 files 字段逐文件 append
+    assert 'batchForm.append("files"' in admin_html
+    # 批量结果展示函数
+    assert "function renderBatchUploadResult" in admin_html
+    assert "批量上传完成" in admin_html
+    # 成功后自动切到入库进度区块
+    assert 'activateSection("section-progress")' in admin_html
+    # 单文件上传仍走原 upload 端点（保持既有逻辑）
+    assert "/api/v1/documents/upload\"" in admin_html or "/api/v1/documents/upload'," in admin_html
+
+
+def test_admin_page_no_prompt_confirm_calls(admin_html: str) -> None:
+    """清理：弹窗组件已替换原生 prompt/confirm，全页面无 prompt( 与 confirm( 调用残留"""
+    assert "prompt(" not in admin_html
+    assert "confirm(" not in admin_html
 
 
 def test_admin_page_login_card(admin_html: str) -> None:
@@ -193,8 +309,52 @@ def test_admin_page_password_form(admin_html: str) -> None:
     assert "passwordForced" in admin_html
 
 
+def test_admin_page_profile_section(admin_html: str) -> None:
+    """个人中心区块：导航按钮、section、账号信息卡片、改密表单、退出按钮、
+    user 角色默认进入、GET /auth/me 与 POST /auth/password 路径"""
+    # section 与导航按钮（不限 admin，所有登录用户可见）
+    assert 'id="section-profile"' in admin_html
+    assert 'id="nav-profile"' in admin_html
+    assert 'data-target="section-profile"' in admin_html
+    assert "个人中心" in admin_html
+    # 账号信息卡片字段：用户名 / 角色 / 创建时间 / 须改密 / 启用状态
+    assert 'id="profile-cards"' in admin_html
+    for label in ("用户名", "角色", "创建时间", "须改密", "启用状态"):
+        assert label in admin_html
+    # 改密表单：旧密码 / 新密码 / 确认 / 强度提示 / 提交按钮
+    assert 'id="profile-password-form"' in admin_html
+    assert 'id="profile-old-password"' in admin_html
+    assert 'id="profile-new-password"' in admin_html
+    assert 'id="profile-confirm-password"' in admin_html
+    assert 'id="profile-strength-hint"' in admin_html
+    assert 'id="btn-profile-password-submit"' in admin_html
+    # 强度提示复用 Task 3 文案（<8 弱 / ≥8 中 / ≥12 强）
+    assert "密码强度：" in admin_html
+    for level in ("弱", "中", "强"):
+        assert level in admin_html
+    # 两次不一致本地拦截
+    assert "两次输入的新密码不一致" in admin_html
+    # 提交按钮 loading 态
+    assert "提交中…" in admin_html
+    # 退出登录按钮（复用现有 logout 逻辑）
+    assert 'id="btn-profile-logout"' in admin_html
+    assert "退出登录" in admin_html
+    assert "doLogout" in admin_html
+    # 数据来源：GET /auth/me 加载账号信息；POST /auth/password 改密
+    assert "/api/v1/auth/me" in admin_html
+    assert "/api/v1/auth/password" in admin_html
+    # 个人中心加载与渲染函数
+    assert "loadProfile" in admin_html
+    assert "renderProfile" in admin_html
+    # user 角色默认进个人中心（admin 仍默认进概览）
+    assert 'activateSection("section-profile")' in admin_html
+    assert 'activateSection("section-overview")' in admin_html
+    # activateSection 触发 loadProfile
+    assert 'targetId === "section-profile"' in admin_html
+
+
 def test_admin_page_users_section(admin_html: str) -> None:
-    """用户管理区块：用户列表/创建表单/角色下拉/重置密码/删除 confirm/角色门禁"""
+    """用户管理区块：用户列表/创建表单/角色下拉/重置密码弹窗/删除确认弹窗/角色门禁"""
     assert 'id="section-users"' in admin_html
     assert 'id="users-tbody"' in admin_html
     # 表头列：用户名/角色/须改密/创建时间/操作
@@ -208,10 +368,10 @@ def test_admin_page_users_section(admin_html: str) -> None:
     assert '<option value="admin"' in admin_html
     assert '<option value="user"' in admin_html
     assert "创建用户" in admin_html
-    # 操作列：重置密码（弹输入）与删除（confirm 二次确认）
+    # 操作列：重置密码 + 删除均走通用弹窗组件（无 prompt/confirm 调用）
     assert "重置密码" in admin_html
-    assert "prompt(" in admin_html
+    assert "resetUserPassword" in admin_html
-    assert "确定删除用户" in admin_html
+    assert "deleteUser" in admin_html
     # 用户管理端点
     assert "/api/v1/auth/users" in admin_html
     # 角色门禁：仅 admin 挂载该区块进 DOM，非 admin 完全不渲染
@@ -220,6 +380,115 @@ def test_admin_page_users_section(admin_html: str) -> None:
     assert "unmountUsersSection" in admin_html
 
 
+def test_admin_page_users_filter_and_inline_edit(admin_html: str) -> None:
+    """用户管理：列表搜索筛选 + 行内角色编辑 + 启用开关 + 当前 admin 自身行防降级"""
+    # 列表搜索筛选：用户名搜索框 + 角色筛选下拉
+    assert 'id="user-search"' in admin_html
+    assert 'id="user-role-filter"' in admin_html
+    assert '<option value="all"' in admin_html
+    # 输入/变更事件触发本地过滤（不重新请求后端）
+    assert 'addEventListener("input", applyUsersFilter)' in admin_html
+    assert 'addEventListener("change", applyUsersFilter)' in admin_html
+    assert "applyUsersFilter" in admin_html
+    # 表头加「启用」列
+    assert "启用" in admin_html
+    # 行内角色编辑：select + 保存按钮（仅当值改变时启用）
+    assert "user-role-select" in admin_html
+    assert "user-role-save" in admin_html
+    assert "updateUserRole" in admin_html
+    # 行内启用/禁用开关：按钮文字 + 样式区分
+    assert "user-enabled-toggle" in admin_html
+    assert "toggleUserEnabled" in admin_html
+    # PATCH /auth/users/{username} 端点调用（role 与 enabled 两条分支）
+    assert 'method: "PATCH"' in admin_html
+    assert "encodeURIComponent(username)" in admin_html
+    # 当前登录 admin 自身行防自锁降级：role select disabled + 启用开关 disabled
+    assert "isSelf" in admin_html
+    assert "不能禁用当前登录账号" in admin_html
+
+
+def test_admin_page_modal_component(admin_html: str) -> None:
+    """通用 overlay 弹窗组件：modal-overlay/modal-card 结构 + openModal/closeModal API"""
+    # CSS 类
+    assert "modal-overlay" in admin_html
+    assert "modal-card" in admin_html
+    assert "modal-title" in admin_html
+    assert "modal-content" in admin_html
+    assert "modal-actions" in admin_html
+    # 挂载点容器
+    assert 'id="modal-root"' in admin_html
+    # 通用 JS API
+    assert "function openModal(" in admin_html
+    assert "function closeModal(" in admin_html
+    assert "function isModalSubmitting(" in admin_html
+    assert "function setModalSubmitting(" in admin_html
+    assert "function showModelError(" in admin_html
+    assert "function hideModelError(" in admin_html
+    # 提交中标记属性
+    assert "data-submitting" in admin_html
+
+
+def test_admin_page_modal_cancel_and_loading(admin_html: str) -> None:
+    """弹窗可取消（点遮罩/取消按钮/ESC）+ 提交中不可取消 + loading 态（提交中禁用）"""
+    # 点遮罩关闭：e.target === overlay
+    assert "e.target === overlay" in admin_html
+    # 取消按钮存在
+    assert "取消" in admin_html
+    # ESC 键关闭最顶层弹窗
+    assert "Escape" in admin_html
+    # 提交中不可关闭：遮罩点击与 ESC 均检查 data-submitting !== "true"
+    assert 'data-submitting") !== "true"' in admin_html
+    assert 'data-submitting") === "true"' in admin_html
+    # 提交 loading 文案 + 按钮禁用
+    assert "提交中…" in admin_html
+    assert 'submitBtn.disabled = true' in admin_html
+    # isModalSubmitting 守卫：取消按钮在提交中不响应
+    assert "isModalSubmitting(" in admin_html
+
+
+def test_admin_page_reset_password_modal(admin_html: str) -> None:
+    """重置密码弹窗：新密码/确认/强度提示/两次不一致本地拦截/loading 态"""
+    # 弹窗标题含用户名占位
+    assert "重置用户" in admin_html
+    assert "密码" in admin_html
+    # 新密码 + 确认密码输入框（动态创建，id 经 JS 属性赋值）
+    assert 'newInput.id = "modal-reset-new"' in admin_html
+    assert 'confirmInput.id = "modal-reset-confirm"' in admin_html
+    # 密码强度提示节点
+    assert 'hint.id = "modal-reset-strength"' in admin_html
+    assert "modal-strength-hint" in admin_html
+    # 强度三档纯文案：<8 弱 / ≥8 中 / ≥12 强
+    assert "密码强度：" in admin_html
+    for level in ("弱", "中", "强"):
+        assert level in admin_html
+    # 两次不一致本地拦截
+    assert "两次输入的新密码不一致" in admin_html
+    # 重置密码端点
+    assert "/api/v1/auth/users/" in admin_html
+    assert "/password" in admin_html
+    # 提交按钮 loading 态
+    assert "setModalSubmitting" in admin_html
+
+
+def test_admin_page_delete_user_modal(admin_html: str) -> None:
+    """删除确认弹窗：警告文案 + 输入用户名匹配才可提交 + loading 态"""
+    # 弹窗标题
+    assert "删除用户" in admin_html
+    # 警告文案
+    assert "此操作不可恢复，将清除该用户及其全部会话" in admin_html
+    # 「请输入用户名 {username} 以确认」+ 文本输入框（动态创建，id 经 JS 属性赋值）
+    assert "请输入用户名" in admin_html
+    assert "以确认" in admin_html
+    assert 'input.id = "modal-delete-input"' in admin_html
+    # 输入 !== username 时删除按钮禁用；匹配后启用
+    assert "submitBtn.disabled = true" in admin_html
+    assert "input.value !== username" in admin_html
+    # 删除端点
+    assert 'method: "DELETE"' in admin_html
+    # 提交按钮 loading 态
+    assert "setModalSubmitting" in admin_html
+
+
 def test_admin_page_auth_interceptor(admin_html: str) -> None:
     """请求拦截：统一注入 Authorization Bearer；1005 回登录；1006 错误条提示"""
     assert 'options.headers["Authorization"] = "Bearer " + token' in admin_html
@@ -231,6 +500,22 @@ def test_admin_page_auth_interceptor(admin_html: str) -> None:
     assert "1006" in admin_html
 
 
+def test_admin_page_doc_detail_file_link(admin_html: str) -> None:
+    """文档详情区：有关联文件时展示原始文件行与下载链接（textContent 防 XSS）"""
+    # 条件判断与字段访问
+    assert "data.file" in admin_html
+    assert "file.filename" in admin_html
+    assert "file.size_bytes" in admin_html
+    # 大小格式化辅助
+    assert "formatFileSize" in admin_html
+    # 下载链接用 el("a") 创建，设 href 与 download 属性，不 innerHTML
+    assert 'el("a"' in admin_html
+    assert "download" in admin_html
+    # 行标签文案
+    assert "原始文件" in admin_html
+    assert "下载" in admin_html
+
+
 def test_admin_page_api_guide_section(admin_html: str) -> None:
     """API 指南区块：导航按钮、section、静态清单与鉴权标注（user 角色也可见）"""
     assert 'data-target="section-api-guide"' in admin_html
@@ -320,15 +605,19 @@ def test_admin_page_api_guide_paths_in_real_routes(admin_html: str) -> None:
         assert (method, path) in route_methods, f"API_GUIDE 端点 {method} {path} 不在后端路由集合内"
 
     # 全部真实端点覆盖：health/search/documents*/knowledge*/auth*
+    # Task 2 新增端点：upload-batch / tasks 列表 / reingest
     expected = {
         ("GET", "/api/v1/health"),
         ("POST", "/api/v1/search"),
         ("POST", "/api/v1/documents"),
         ("POST", "/api/v1/documents/upload"),
+        ("POST", "/api/v1/documents/upload-batch"),
+        ("GET", "/api/v1/documents/tasks"),
         ("GET", "/api/v1/documents/tasks/{task_id}"),
         ("GET", "/api/v1/documents"),
         ("GET", "/api/v1/documents/{doc_id}"),
         ("DELETE", "/api/v1/documents/{doc_id}"),
+        ("POST", "/api/v1/documents/{doc_id}/reingest"),
         ("GET", "/api/v1/knowledge/categories"),
         ("GET", "/api/v1/knowledge/stats"),
         ("POST", "/api/v1/auth/login"),

tests/test_auth_api.py

@@ -245,8 +245,9 @@ class TestUsersCrud:
         users = body["data"]
         assert len(users) == 1
         admin = users[0]
-        assert set(admin.keys()) == {"username", "role", "must_change_password", "created_at"}
+        assert set(admin.keys()) == {"username", "role", "must_change_password", "enabled", "created_at"}
         assert admin["username"] == "admin"
+        assert admin["enabled"] is True
         assert "password_hash" not in admin
         assert "salt" not in admin
 
@@ -271,10 +272,11 @@ class TestUsersCrud:
 
         assert body["code"] == 0
         data = body["data"]
-        assert set(data.keys()) == {"username", "role", "must_change_password", "created_at"}
+        assert set(data.keys()) == {"username", "role", "must_change_password", "enabled", "created_at"}
         assert data["username"] == "carol"
         assert data["role"] == "user"
         assert data["must_change_password"] is False
+        assert data["enabled"] is True
         # 新用户可登录
         assert _login(client, "carol", "carol-pass-123")["code"] == 0
 
@@ -386,6 +388,170 @@ class TestUsersCrud:
         assert body["code"] == 1001
 
 
+class TestUpdateUser:
+    """PATCH /api/v1/auth/users/{username}：更新角色/启用状态"""
+
+    def test_update_role_success(self, client: TestClient, admin_headers: dict[str, str]) -> None:
+        client.post(
+            "/api/v1/auth/users",
+            json={"username": "carol", "password": "carol-pass-123"},
+            headers=admin_headers,
+        )
+        body = client.patch(
+            "/api/v1/auth/users/carol", json={"role": "admin"}, headers=admin_headers
+        ).json()
+
+        assert body["code"] == 0
+        assert body["data"]["role"] == "admin"
+        assert body["data"]["enabled"] is True
+
+    def test_update_enabled_success(
+        self, client: TestClient, auth_stores, admin_headers: dict[str, str]
+    ) -> None:
+        user_store, _ = auth_stores
+        _create_user(user_store, "dave", "dave-pass-123", role="user")
+        body = client.patch(
+            "/api/v1/auth/users/dave", json={"enabled": False}, headers=admin_headers
+        ).json()
+
+        assert body["code"] == 0
+        assert body["data"]["enabled"] is False
+        assert body["data"]["role"] == "user"
+
+    def test_update_both_role_and_enabled(
+        self, client: TestClient, auth_stores, admin_headers: dict[str, str]
+    ) -> None:
+        user_store, _ = auth_stores
+        _create_user(user_store, "erin", "erin-pass-123", role="user")
+        body = client.patch(
+            "/api/v1/auth/users/erin",
+            json={"role": "admin", "enabled": False},
+            headers=admin_headers,
+        ).json()
+
+        assert body["code"] == 0
+        assert body["data"]["role"] == "admin"
+        assert body["data"]["enabled"] is False
+
+    def test_at_least_one_field_required(self, client: TestClient, admin_headers: dict[str, str]) -> None:
+        body = client.patch(
+            "/api/v1/auth/users/admin", json={}, headers=admin_headers
+        ).json()
+
+        assert body["code"] == 1001
+        assert body["data"] is None
+
+    def test_user_not_found(self, client: TestClient, admin_headers: dict[str, str]) -> None:
+        body = client.patch(
+            "/api/v1/auth/users/ghost", json={"role": "user"}, headers=admin_headers
+        ).json()
+
+        assert body["code"] == 1004
+        assert body["data"] is None
+
+    def test_invalid_role(self, client: TestClient, admin_headers: dict[str, str]) -> None:
+        body = client.patch(
+            "/api/v1/auth/users/admin", json={"role": "superuser"}, headers=admin_headers
+        ).json()
+
+        assert body["code"] == 1001
+
+    def test_last_admin_demote_forbidden(self, client: TestClient, admin_headers: dict[str, str]) -> None:
+        # 唯一 admin 降级为 user → 1001
+        body = client.patch(
+            "/api/v1/auth/users/admin", json={"role": "user"}, headers=admin_headers
+        ).json()
+
+        assert body["code"] == 1001
+        # 未生效
+        assert client.get("/api/v1/auth/users", headers=admin_headers).json()["code"] == 0
+
+    def test_last_admin_disable_forbidden(self, client: TestClient, admin_headers: dict[str, str]) -> None:
+        # 唯一 admin 禁用 → 1001
+        body = client.patch(
+            "/api/v1/auth/users/admin", json={"enabled": False}, headers=admin_headers
+        ).json()
+
+        assert body["code"] == 1001
+
+    def test_non_admin_forbidden(self, client: TestClient, auth_stores) -> None:
+        user_store, session_store = auth_stores
+        _create_user(user_store, "bob", "bob-pass-123", role="user")
+        headers = _headers(session_store, "bob", "user")
+
+        body = client.patch(
+            "/api/v1/auth/users/bob", json={"role": "admin"}, headers=headers
+        ).json()
+
+        assert body["code"] == 1006
+
+    def test_demote_when_multiple_admins_ok(
+        self, client: TestClient, auth_stores, admin_headers: dict[str, str]
+    ) -> None:
+        # 存在第二个 admin 时，可降级其中一个
+        user_store, _ = auth_stores
+        _create_user(user_store, "admin2", "admin2-pass-123", role="admin")
+        body = client.patch(
+            "/api/v1/auth/users/admin2", json={"role": "user"}, headers=admin_headers
+        ).json()
+
+        assert body["code"] == 0
+        assert body["data"]["role"] == "user"
+
+
+class TestDisabledUser:
+    """禁用用户：登录拦截 + 旧 token 失效 + 业务端点拦截"""
+
+    def test_disabled_user_login_blocked(
+        self, client: TestClient, auth_stores, admin_headers: dict[str, str]
+    ) -> None:
+        user_store, _ = auth_stores
+        _create_user(user_store, "bob", "bob-pass-123", role="user")
+        # 禁用 bob
+        client.patch(
+            "/api/v1/auth/users/bob", json={"enabled": False}, headers=admin_headers
+        )
+        # 登录 → 1005 账号已禁用
+        body = _login(client, "bob", "bob-pass-123")
+
+        assert body["code"] == 1005
+        assert body["data"] is None
+
+    def test_disabled_user_old_token_invalidated(
+        self, client: TestClient, auth_stores, admin_headers: dict[str, str]
+    ) -> None:
+        user_store, _ = auth_stores
+        _create_user(user_store, "bob", "bob-pass-123", role="user")
+        # bob 登录拿到 token
+        login_body = _login(client, "bob", "bob-pass-123")
+        assert login_body["code"] == 0
+        bob_headers = {"Authorization": f"Bearer {login_body['data']['token']}"}
+        # 禁用 bob → session 清除
+        client.patch(
+            "/api/v1/auth/users/bob", json={"enabled": False}, headers=admin_headers
+        )
+        # 旧 token 调业务端点 → 1005（session 已清）
+        body = client.get("/api/v1/auth/me", headers=bob_headers).json()
+
+        assert body["code"] == 1005
+
+    def test_disabled_user_blocked_even_with_valid_session(
+        self, client: TestClient, auth_stores, admin_headers: dict[str, str]
+    ) -> None:
+        """禁用用户即使持有效 session（直接签发绕过登录），业务端点仍拦截 1005"""
+        user_store, session_store = auth_stores
+        _create_user(user_store, "bob", "bob-pass-123", role="user")
+        # 禁用 bob → session 清除
+        client.patch(
+            "/api/v1/auth/users/bob", json={"enabled": False}, headers=admin_headers
+        )
+        # 直接为 bob 签发新 session 绕过登录与清理，验证 get_current_user 的 enabled 拦截
+        bob_headers = _headers(session_store, "bob", "user")
+        body = client.get("/api/v1/auth/me", headers=bob_headers).json()
+
+        assert body["code"] == 1005
+
+
 class TestDocumentAuth:
     """文档端点鉴权：变更类需登录，GET 系列免登录"""
 

tests/test_batch_upload_api.py

@@ -0,0 +1,335 @@
+"""POST /documents/upload-batch、POST /documents/{id}/reingest、GET /documents/tasks 端点测试
+
+TestClient + FakeManager + FakeQdrant，不真实联网。
+"""
+
+from collections.abc import Iterator
+from pathlib import Path
+from typing import Any
+
+import pytest
+from fastapi.testclient import TestClient
+
+from app.api.v1 import document as document_module
+from app.main import app
+from app.models.document import DocumentInput
+from app.services.qdrant import QdrantService
+
+
+class FakeManager:
+    """假入库任务管理器：支持 submit/get/list_tasks"""
+
+    def __init__(self, task_id: str = "task-1") -> None:
+        self.task_id = task_id
+        self._counter = 0
+        self.submitted: list[DocumentInput] = []
+        self._tasks: list[dict[str, Any]] = []
+
+    async def submit(self, doc: DocumentInput) -> str:
+        self.submitted.append(doc)
+        self._counter += 1
+        tid = f"{self.task_id}-{self._counter}"
+        self._tasks.append(
+            {
+                "task_id": tid,
+                "status": "pending",
+                "filename": doc.metadata.get("original_filename") or doc.title,
+                "created_at": f"2026-01-0{self._counter}T00:00:00+00:00",
+                "updated_at": f"2026-01-0{self._counter}T00:00:00+00:00",
+                "doc_id": None,
+            }
+        )
+        return tid
+
+    async def get(self, task_id: str) -> dict[str, Any] | None:
+        return None
+
+    async def list_tasks(self, limit: int = 20) -> list[dict[str, Any]]:
+        return list(self._tasks[:limit])
+
+
+class FakeQdrant:
+    """假 Qdrant 服务：可控的 get_l1_metadata/delete_by_doc_id"""
+
+    def __init__(self, meta: dict[str, str] | None = None) -> None:
+        self.meta = meta
+        self.deleted: list[str] = []
+
+    async def get_l1_metadata(self, doc_id: str) -> dict[str, str] | None:
+        return self.meta
+
+    async def delete_by_doc_id(self, doc_id: str) -> dict[str, int]:
+        self.deleted.append(doc_id)
+        return {"doc_l1": 1}
+
+
+@pytest.fixture
+def client(
+    monkeypatch: pytest.MonkeyPatch, tmp_path: Path, admin_headers: dict[str, str]
+) -> Iterator[TestClient]:
+    async def _noop_ensure_collections(self: QdrantService) -> None:
+        return None
+
+    monkeypatch.setattr(QdrantService, "ensure_collections", _noop_ensure_collections)
+    monkeypatch.setattr(document_module.settings, "upload_dir", str(tmp_path / "uploads"))
+    with TestClient(app) as test_client:
+        test_client.headers.update(admin_headers)
+        yield test_client
+
+
+def _inject_manager(monkeypatch: pytest.MonkeyPatch, manager: FakeManager) -> None:
+    monkeypatch.setattr(document_module, "_get_task_manager", lambda: manager)
+
+
+def _inject_qdrant(monkeypatch: pytest.MonkeyPatch, qdrant: FakeQdrant) -> None:
+    monkeypatch.setattr(document_module, "_get_qdrant", lambda: qdrant)
+
+
+# ----------------------- upload-batch -----------------------
+
+
+def test_batch_upload_multiple_files_success(
+    client: TestClient, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """多文件全部成功：3 文件 → 3 tasks，failed 为空"""
+    manager = FakeManager()
+    _inject_manager(monkeypatch, manager)
+
+    resp = client.post(
+        "/api/v1/documents/upload-batch",
+        files=[
+            ("files", ("a.txt", b"content a", "text/plain")),
+            ("files", ("b.md", b"# B", "text/markdown")),
+            ("files", ("c.txt", b"content c", "text/plain")),
+        ],
+    )
+
+    assert resp.status_code == 202
+    data = resp.json()["data"]
+    assert len(data["tasks"]) == 3
+    assert data["failed"] == []
+    assert len(manager.submitted) == 3
+    filenames = [t["filename"] for t in data["tasks"]]
+    assert filenames == ["a.txt", "b.md", "c.txt"]
+
+
+def test_batch_upload_partial_failure(
+    client: TestClient, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """部分失败：1 个不支持扩展名进 failed，其他成功"""
+    manager = FakeManager()
+    _inject_manager(monkeypatch, manager)
+
+    resp = client.post(
+        "/api/v1/documents/upload-batch",
+        files=[
+            ("files", ("ok.txt", b"good", "text/plain")),
+            ("files", ("bad.xlsx", b"binary", "application/octet-stream")),
+        ],
+    )
+
+    assert resp.status_code == 202
+    data = resp.json()["data"]
+    assert len(data["tasks"]) == 1
+    assert data["tasks"][0]["filename"] == "ok.txt"
+    assert len(data["failed"]) == 1
+    assert data["failed"][0]["filename"] == "bad.xlsx"
+    assert "不支持" in data["failed"][0]["error"]
+    assert len(manager.submitted) == 1
+
+
+def test_batch_upload_all_fail(
+    client: TestClient, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """全部失败：tasks 空，failed 2 条，未提交任何任务"""
+    manager = FakeManager()
+    _inject_manager(monkeypatch, manager)
+
+    resp = client.post(
+        "/api/v1/documents/upload-batch",
+        files=[
+            ("files", ("a.exe", b"x", "application/octet-stream")),
+            ("files", ("b.bin", b"y", "application/octet-stream")),
+        ],
+    )
+
+    assert resp.status_code == 202
+    data = resp.json()["data"]
+    assert data["tasks"] == []
+    assert len(data["failed"]) == 2
+    assert len(manager.submitted) == 0
+
+
+def test_batch_upload_empty_list_returns_1001(
+    client: TestClient, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """空文件列表：1001"""
+    manager = FakeManager()
+    _inject_manager(monkeypatch, manager)
+
+    resp = client.post("/api/v1/documents/upload-batch", files=[])
+
+    body = resp.json()
+    assert body["code"] == 1001
+    assert manager.submitted == []
+
+
+# ----------------------- reingest -----------------------
+
+
+def test_reingest_success(
+    client: TestClient, monkeypatch: pytest.MonkeyPatch, tmp_path: Path
+) -> None:
+    """有原始文件：202 + task_id，删旧数据后提交新任务"""
+    upload_dir = tmp_path / "uploads"
+    upload_dir.mkdir(parents=True, exist_ok=True)
+    raw = upload_dir / "raw_report.txt"
+    raw.write_bytes(b"original content")
+    meta = {
+        "raw_file_path": str(raw),
+        "original_filename": "raw_report.txt",
+    }
+    qdrant = FakeQdrant(meta=meta)
+    _inject_qdrant(monkeypatch, qdrant)
+    manager = FakeManager(task_id="reingest-task")
+    _inject_manager(monkeypatch, manager)
+
+    resp = client.post("/api/v1/documents/some-doc/reingest")
+
+    assert resp.status_code == 202
+    data = resp.json()["data"]
+    assert data["task_id"] == "reingest-task-1"
+    assert data["status"] == "pending"
+    # 旧数据已删
+    assert qdrant.deleted == ["some-doc"]
+    # 新任务已提交，文本来自原文件
+    assert len(manager.submitted) == 1
+    assert manager.submitted[0].text == "original content"
+    assert manager.submitted[0].title == "raw_report"  # 文件名 stem
+
+
+def test_reingest_no_raw_file_returns_1001(
+    client: TestClient, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """无原始文件（meta 无 raw_file_path）：1001"""
+    qdrant = FakeQdrant(meta={"some": "metadata"})
+    _inject_qdrant(monkeypatch, qdrant)
+    manager = FakeManager()
+    _inject_manager(monkeypatch, manager)
+
+    resp = client.post("/api/v1/documents/doc1/reingest")
+
+    body = resp.json()
+    assert body["code"] == 1001
+    assert "无原始文件" in body["message"]
+    assert len(manager.submitted) == 0
+
+
+def test_reingest_missing_file_returns_1004(
+    client: TestClient, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """文件不存在：1004"""
+    qdrant = FakeQdrant(
+        meta={"raw_file_path": "/nonexistent/path.txt", "original_filename": "path.txt"}
+    )
+    _inject_qdrant(monkeypatch, qdrant)
+    manager = FakeManager()
+    _inject_manager(monkeypatch, manager)
+
+    resp = client.post("/api/v1/documents/doc1/reingest")
+
+    body = resp.json()
+    assert body["code"] == 1004
+    assert len(manager.submitted) == 0
+
+
+def test_reingest_doc_not_found_returns_1004(
+    client: TestClient, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """文档不存在（meta 为 None）：1004"""
+    qdrant = FakeQdrant(meta=None)
+    _inject_qdrant(monkeypatch, qdrant)
+    manager = FakeManager()
+    _inject_manager(monkeypatch, manager)
+
+    resp = client.post("/api/v1/documents/nope/reingest")
+
+    body = resp.json()
+    assert body["code"] == 1004
+    assert len(manager.submitted) == 0
+
+
+# ----------------------- tasks 列表 -----------------------
+
+
+def test_list_ingest_tasks_returns_items(
+    client: TestClient, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """任务列表：返回近期任务，含 items 与 total"""
+    manager = FakeManager()
+    _inject_manager(monkeypatch, manager)
+    manager._tasks = [
+        {
+            "task_id": "t1",
+            "status": "done",
+            "filename": "a.txt",
+            "created_at": "2026-01-01T00:00:00+00:00",
+            "updated_at": "2026-01-01T00:00:00+00:00",
+            "doc_id": "d1",
+        },
+        {
+            "task_id": "t2",
+            "status": "failed",
+            "filename": "b.txt",
+            "created_at": "2026-01-02T00:00:00+00:00",
+            "updated_at": "2026-01-02T00:00:00+00:00",
+            "doc_id": None,
+        },
+    ]
+
+    resp = client.get("/api/v1/documents/tasks")
+
+    assert resp.status_code == 200
+    data = resp.json()["data"]
+    assert data["total"] == 2
+    assert len(data["items"]) == 2
+    assert data["items"][0]["task_id"] == "t1"
+    assert data["items"][0]["doc_id"] == "d1"
+
+
+def test_list_ingest_tasks_limit(
+    client: TestClient, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """limit 截断生效"""
+    manager = FakeManager()
+    _inject_manager(monkeypatch, manager)
+    manager._tasks = [
+        {
+            "task_id": f"t{i}",
+            "status": "done",
+            "filename": "f",
+            "created_at": "",
+            "updated_at": "",
+            "doc_id": None,
+        }
+        for i in range(10)
+    ]
+
+    resp = client.get("/api/v1/documents/tasks?limit=5")
+
+    data = resp.json()["data"]
+    assert data["total"] == 5
+    assert len(data["items"]) == 5
+
+
+def test_list_ingest_tasks_requires_auth(
+    client: TestClient, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """无 token：1005 未认证"""
+    manager = FakeManager()
+    _inject_manager(monkeypatch, manager)
+
+    resp = client.get("/api/v1/documents/tasks", headers={"Authorization": ""})
+
+    body = resp.json()
+    assert body["code"] == 1005

tests/test_batch_upload_integration.py

@@ -0,0 +1,339 @@
+"""批量上传与入库进度集成验证
+
+在真实内存 Qdrant + FakeOllama + 确定性向量环境下，全链路验证：
+- POST /documents/upload-batch 批量上传闭环（全部成功 / 部分失败）
+- GET /documents/tasks 任务列表（按 updated_at 降序、每项含 filename/doc_id）
+- POST /documents/{doc_id}/reingest 重新入库（成功 / 边界 1001/1004）
+
+复用 tests/test_e2e_integration.py 的 FakeOllama / DeterministicEmbedding / FakeCache；
+upload_dir 用 tmp_path 真实落盘，monkeypatch 模块级 _task_manager / _qdrant 单例。
+"""
+
+from pathlib import Path
+
+import pytest
+from fastapi.testclient import TestClient
+from qdrant_client import AsyncQdrantClient
+
+from app.api.v1 import document as document_module
+from app.config import Settings
+from app.core.chunker import Chunker
+from app.core.classifier import Classifier
+from app.core.ingest_tasks import IngestTaskManager, IngestTaskStatus
+from app.core.ingestion import Ingester
+from app.core.query_parser import QueryParser
+from app.core.retriever import Retriever
+from app.core.sparse import SparseEncoder
+from app.core.summarizer import Summarizer
+from app.main import app
+from app.models.knowledge import load_taxonomy
+from app.services.qdrant import COLLECTION_CHUNKS, QdrantService
+from tests.test_e2e_integration import (
+    FAKE_L1_SUMMARY,
+    DeterministicEmbedding,
+    FakeCache,
+    FakeOllama,
+)
+
+
+async def _make_env() -> tuple[QdrantService, Ingester, Retriever]:
+    """构建集成环境：真实组件 + 内存 Qdrant + FakeOllama + 确定性向量"""
+    qdrant = QdrantService(client=AsyncQdrantClient(location=":memory:"))
+    await qdrant.ensure_collections()
+    ollama = FakeOllama()
+    taxonomy = load_taxonomy()
+    embedding = DeterministicEmbedding()
+    ingester = Ingester(
+        summarizer=Summarizer(ollama=ollama),  # type: ignore[arg-type]
+        classifier=Classifier(ollama=ollama, taxonomy=taxonomy),  # type: ignore[arg-type]
+        chunker=Chunker(),
+        embedding=embedding,  # type: ignore[arg-type]
+        sparse=SparseEncoder(),
+        qdrant=qdrant,
+    )
+    retriever = Retriever(
+        qdrant=qdrant,
+        query_parser=QueryParser(ollama=ollama, taxonomy=taxonomy, cache=FakeCache()),  # type: ignore[arg-type]
+        embedding=embedding,  # type: ignore[arg-type]
+        sparse_encoder=SparseEncoder(),
+    )
+    return qdrant, ingester, retriever
+
+
+def _patch_app(
+    monkeypatch: pytest.MonkeyPatch,
+    manager: IngestTaskManager,
+    qdrant: QdrantService,
+    retriever: Retriever,
+    upload_dir: Path,
+) -> None:
+    """替换模块级单例：任务管理器 / Qdrant / Retriever / upload_dir，lifespan 建集合改空操作"""
+
+    async def _noop_ensure_collections(self: QdrantService) -> None:
+        return None
+
+    monkeypatch.setattr(QdrantService, "ensure_collections", _noop_ensure_collections)
+    monkeypatch.setattr(document_module, "_task_manager", manager)
+    monkeypatch.setattr(document_module, "_qdrant", qdrant)
+    monkeypatch.setattr(document_module.settings, "upload_dir", str(upload_dir))
+
+
+def _md_content(name: str) -> bytes:
+    """构造可触发完整三级总结的 .md 文件内容（不同 name 保证文本不同，避免去重）"""
+    paragraph = f"这是 {name} 章节的正文内容，包含足够信息量用于测试切分与向量化流程。" * 20
+    text = f"# {name} 安装指南\n{paragraph}\n\n## 环境准备\n{paragraph}\n\n## 安装步骤\n{paragraph}"
+    return text.encode("utf-8")
+
+
+class TestBatchUploadIntegration:
+    """批量上传闭环：3 文件全成功 → 任务列表 → 文档详情含原始文件信息"""
+
+    async def test_batch_upload_full_loop(
+        self,
+        monkeypatch: pytest.MonkeyPatch,
+        tmp_path: Path,
+        admin_headers: dict[str, str],
+    ) -> None:
+        qdrant, ingester, retriever = await _make_env()
+        manager = IngestTaskManager(ingester, None, Settings())
+        _patch_app(monkeypatch, manager, qdrant, retriever, tmp_path / "uploads")
+
+        with TestClient(app) as client:
+            # 1. 批量上传 3 个 .md：202 + tasks 含 3 个 + failed 空
+            resp = client.post(
+                "/api/v1/documents/upload-batch",
+                files=[
+                    ("files", ("a.md", _md_content("Alpha"), "text/markdown")),
+                    ("files", ("b.md", _md_content("Beta"), "text/markdown")),
+                    ("files", ("c.md", _md_content("Gamma"), "text/markdown")),
+                ],
+                headers=admin_headers,
+            )
+            assert resp.status_code == 202
+            data = resp.json()["data"]
+            assert len(data["tasks"]) == 3
+            assert data["failed"] == []
+            task_ids = [t["task_id"] for t in data["tasks"]]
+            filenames = [t["filename"] for t in data["tasks"]]
+            assert filenames == ["a.md", "b.md", "c.md"]
+
+            # 2. 逐个等待终态：全部 done
+            doc_ids: list[str] = []
+            for tid in task_ids:
+                final = await manager.wait_done(tid)
+                assert final["status"] == IngestTaskStatus.DONE
+                doc_ids.append(final["result"]["document_id"])
+
+            # 3. GET /documents/tasks：items 含 3 项，按 updated_at 降序，每项含 filename/doc_id
+            resp_list = client.get(
+                "/api/v1/documents/tasks?limit=10", headers=admin_headers
+            )
+            body_list = resp_list.json()
+            assert body_list["code"] == 0
+            items = body_list["data"]["items"]
+            assert body_list["data"]["total"] == 3
+            assert len(items) == 3
+            # 按 updated_at 降序
+            updated = [it["updated_at"] for it in items]
+            assert updated == sorted(updated, reverse=True)
+            # 每项含 filename 与 doc_id
+            for it in items:
+                assert it["filename"] in {"a.md", "b.md", "c.md"}
+                assert it["doc_id"] in doc_ids
+
+            # 4. GET /documents/{doc_id}：file 字段非 null（有原始文件信息）
+            resp_detail = client.get(f"/api/v1/documents/{doc_ids[0]}")
+            body_detail = resp_detail.json()
+            assert body_detail["code"] == 0
+            file_info = body_detail["data"]["file"]
+            assert file_info is not None
+            assert file_info["filename"] == filenames[0]
+            assert file_info["size_bytes"] > 0
+            assert file_info["url"] == f"/api/v1/documents/{doc_ids[0]}/file"
+
+
+class TestBatchUploadPartialFailure:
+    """部分失败：1 个 .unsupported 扩展名进 failed，其余 2 个成功"""
+
+    async def test_batch_upload_partial_failure(
+        self,
+        monkeypatch: pytest.MonkeyPatch,
+        tmp_path: Path,
+        admin_headers: dict[str, str],
+    ) -> None:
+        qdrant, ingester, retriever = await _make_env()
+        manager = IngestTaskManager(ingester, None, Settings())
+        _patch_app(monkeypatch, manager, qdrant, retriever, tmp_path / "uploads")
+
+        with TestClient(app) as client:
+            resp = client.post(
+                "/api/v1/documents/upload-batch",
+                files=[
+                    ("files", ("ok1.md", _md_content("Ok1"), "text/markdown")),
+                    ("files", ("ok2.md", _md_content("Ok2"), "text/markdown")),
+                    ("files", ("bad.unsupported", b"whatever", "application/octet-stream")),
+                ],
+                headers=admin_headers,
+            )
+            assert resp.status_code == 202
+            data = resp.json()["data"]
+            assert len(data["tasks"]) == 2
+            assert len(data["failed"]) == 1
+            assert data["failed"][0]["filename"] == "bad.unsupported"
+            assert "不支持" in data["failed"][0]["error"]
+            # 2 个成功任务均可进入终态 done
+            for t in data["tasks"]:
+                final = await manager.wait_done(t["task_id"])
+                assert final["status"] == IngestTaskStatus.DONE
+
+
+class TestReingestIntegration:
+    """重新入库：成功路径 + 边界（无原始文件 / 文档不存在 / 文件已删）"""
+
+    async def test_reingest_success(
+        self,
+        monkeypatch: pytest.MonkeyPatch,
+        tmp_path: Path,
+        admin_headers: dict[str, str],
+    ) -> None:
+        qdrant, ingester, retriever = await _make_env()
+        manager = IngestTaskManager(ingester, None, Settings())
+        _patch_app(monkeypatch, manager, qdrant, retriever, tmp_path / "uploads")
+
+        with TestClient(app) as client:
+            # 先上传 1 个文件，拿到 doc_id
+            resp_up = client.post(
+                "/api/v1/documents/upload-batch",
+                files=[("files", ("re.md", _md_content("Reingest"), "text/markdown"))],
+                headers=admin_headers,
+            )
+            assert resp_up.status_code == 202
+            up_data = resp_up.json()["data"]
+            assert len(up_data["tasks"]) == 1
+            old_task_id = up_data["tasks"][0]["task_id"]
+            old_final = await manager.wait_done(old_task_id)
+            old_doc_id = old_final["result"]["document_id"]
+
+            # 重新入库：202 + task_id
+            resp_re = client.post(
+                f"/api/v1/documents/{old_doc_id}/reingest", headers=admin_headers
+            )
+            assert resp_re.status_code == 202
+            re_body = resp_re.json()
+            assert re_body["code"] == 0
+            new_task_id = re_body["data"]["task_id"]
+            assert new_task_id
+
+            # 等待终态：done
+            new_final = await manager.wait_done(new_task_id)
+            assert new_final["status"] == IngestTaskStatus.DONE
+            new_doc_id = new_final["result"]["document_id"]
+
+            # 旧 doc_id 数据已删：GET /documents/{old_doc_id} → 1004
+            resp_old = client.get(f"/api/v1/documents/{old_doc_id}")
+            assert resp_old.json()["code"] == 1004
+
+            # 新 doc_id 详情：L1 summary 存在，file 非 null（保留原文件信息）
+            resp_new = client.get(f"/api/v1/documents/{new_doc_id}")
+            body_new = resp_new.json()
+            assert body_new["code"] == 0
+            assert body_new["data"]["l1"]["text"] == FAKE_L1_SUMMARY
+            assert body_new["data"]["file"] is not None
+            assert body_new["data"]["file"]["filename"] == "re.md"
+
+            # 旧 chunks 已删：chunks 集合中 old_doc_id 点数为 0
+            old_count = await qdrant.client.count(
+                collection_name=COLLECTION_CHUNKS,
+                count_filter=qdrant.build_filter(doc_ids=[old_doc_id]),
+                exact=True,
+            )
+            assert old_count.count == 0
+
+    async def test_reingest_text_only_doc_returns_1001(
+        self,
+        monkeypatch: pytest.MonkeyPatch,
+        tmp_path: Path,
+        admin_headers: dict[str, str],
+    ) -> None:
+        """纯文本入库的 doc_id → reingest → 1001（无原始文件）"""
+        qdrant, ingester, retriever = await _make_env()
+        manager = IngestTaskManager(ingester, None, Settings())
+        _patch_app(monkeypatch, manager, qdrant, retriever, tmp_path / "uploads")
+
+        with TestClient(app) as client:
+            # 纯文本入库（无原始文件落盘记录）
+            resp_doc = client.post(
+                "/api/v1/documents",
+                json={
+                    "text": "# 纯文本\n这是一段纯文本入库的内容，用于测试 reingest 边界。",
+                    "title": "纯文本",
+                },
+                headers=admin_headers,
+            )
+            assert resp_doc.status_code == 202
+            task_id = resp_doc.json()["data"]["task_id"]
+            final = await manager.wait_done(task_id)
+            doc_id = final["result"]["document_id"]
+
+            # reingest → 1001
+            resp_re = client.post(
+                f"/api/v1/documents/{doc_id}/reingest", headers=admin_headers
+            )
+            body = resp_re.json()
+            assert body["code"] == 1001
+            assert "无原始文件" in body["message"]
+
+    async def test_reingest_doc_not_found_returns_1004(
+        self,
+        monkeypatch: pytest.MonkeyPatch,
+        tmp_path: Path,
+        admin_headers: dict[str, str],
+    ) -> None:
+        """不存在的 doc_id → reingest → 1004"""
+        qdrant, ingester, retriever = await _make_env()
+        manager = IngestTaskManager(ingester, None, Settings())
+        _patch_app(monkeypatch, manager, qdrant, retriever, tmp_path / "uploads")
+
+        with TestClient(app) as client:
+            resp = client.post(
+                "/api/v1/documents/nonexistent-doc/reingest", headers=admin_headers
+            )
+            body = resp.json()
+            assert body["code"] == 1004
+
+    async def test_reingest_file_deleted_returns_1004(
+        self,
+        monkeypatch: pytest.MonkeyPatch,
+        tmp_path: Path,
+        admin_headers: dict[str, str],
+    ) -> None:
+        """文件被删（删 upload_dir 下落盘文件后）→ reingest → 1004"""
+        qdrant, ingester, retriever = await _make_env()
+        manager = IngestTaskManager(ingester, None, Settings())
+        _patch_app(monkeypatch, manager, qdrant, retriever, tmp_path / "uploads")
+
+        with TestClient(app) as client:
+            # 上传 1 个文件
+            resp_up = client.post(
+                "/api/v1/documents/upload-batch",
+                files=[("files", ("del.md", _md_content("Deleted"), "text/markdown"))],
+                headers=admin_headers,
+            )
+            assert resp_up.status_code == 202
+            tid = resp_up.json()["data"]["tasks"][0]["task_id"]
+            final = await manager.wait_done(tid)
+            doc_id = final["result"]["document_id"]
+
+            # 删除 upload_dir 下落盘的原始文件
+            meta = await qdrant.get_l1_metadata(doc_id)
+            assert meta is not None
+            raw_path = Path(meta["raw_file_path"])
+            assert raw_path.is_file()
+            raw_path.unlink()
+
+            # reingest → 1004（文件不存在）
+            resp_re = client.post(
+                f"/api/v1/documents/{doc_id}/reingest", headers=admin_headers
+            )
+            body = resp_re.json()
+            assert body["code"] == 1004

tests/test_document_file_api.py

@@ -0,0 +1,169 @@
+"""GET /api/v1/documents/{doc_id}/file 文件下载端点测试（TestClient + FakeQdrant，不真实联网）
+
+覆盖分支：成功下载、文档不存在、无关联文件、文件已从磁盘删除、路径越界、免登录访问。
+"""
+
+from collections.abc import Iterator
+from pathlib import Path
+
+import pytest
+from fastapi.testclient import TestClient
+
+from app.api.v1 import document as document_module
+from app.main import app
+from app.services.qdrant import QdrantService
+
+
+class FakeQdrant:
+    """假 QdrantService：仅实现 get_l1_metadata，按 doc_id 返回固定 metadata 或 None"""
+
+    def __init__(self, meta: dict[str, str] | None = None) -> None:
+        self.meta = meta
+
+    async def get_l1_metadata(self, doc_id: str) -> dict[str, str] | None:
+        return self.meta
+
+
+@pytest.fixture
+def client(monkeypatch: pytest.MonkeyPatch, tmp_path: Path) -> Iterator[TestClient]:
+    """TestClient，Qdrant 集合初始化空操作；upload_dir 指向临时目录
+
+    不挂 admin_headers：下载端点免登录，验证无 token 可访问。
+    """
+
+    async def _noop_ensure_collections(self: QdrantService) -> None:
+        return None
+
+    monkeypatch.setattr(QdrantService, "ensure_collections", _noop_ensure_collections)
+    monkeypatch.setattr(document_module.settings, "upload_dir", str(tmp_path / "uploads"))
+    with TestClient(app) as test_client:
+        yield test_client
+
+
+def _install_fake(monkeypatch: pytest.MonkeyPatch, fake: FakeQdrant) -> None:
+    """将 _get_qdrant 单例替换为假服务"""
+    monkeypatch.setattr(document_module, "_get_qdrant", lambda: fake)
+
+
+def test_download_file_success(
+    client: TestClient, monkeypatch: pytest.MonkeyPatch, tmp_path: Path
+) -> None:
+    """文档存在且有关联文件 → 200 + 文件内容正确 + content-disposition 含文件名"""
+    upload_dir = tmp_path / "uploads"
+    upload_dir.mkdir(parents=True, exist_ok=True)
+    target = upload_dir / "doc-1_notes.txt"
+    content = b"hello world file content"
+    target.write_bytes(content)
+
+    fake = FakeQdrant(
+        meta={
+            "raw_file_path": str(target),
+            "original_filename": "notes.txt",
+            "original_size_bytes": str(len(content)),
+        }
+    )
+    _install_fake(monkeypatch, fake)
+
+    resp = client.get("/api/v1/documents/doc-1/file")
+
+    assert resp.status_code == 200
+    assert resp.content == content
+    disposition = resp.headers.get("content-disposition", "")
+    assert "attachment" in disposition
+    assert "notes.txt" in disposition
+
+
+def test_download_file_doc_not_found(
+    client: TestClient, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """文档不存在（metadata 为 None）→ 1004"""
+    fake = FakeQdrant(meta=None)
+    _install_fake(monkeypatch, fake)
+
+    resp = client.get("/api/v1/documents/missing/file")
+
+    body = resp.json()
+    assert body["code"] == 1004
+    assert "文件" in body["message"]
+
+
+def test_download_file_no_associated_file(
+    client: TestClient, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """文档存在但 raw_file_path 为空 → 1004"""
+    fake = FakeQdrant(meta={"raw_file_path": "", "original_filename": "x.txt"})
+    _install_fake(monkeypatch, fake)
+
+    resp = client.get("/api/v1/documents/doc-1/file")
+
+    body = resp.json()
+    assert body["code"] == 1004
+    assert "未关联文件" in body["message"]
+
+
+def test_download_file_missing_on_disk(
+    client: TestClient, monkeypatch: pytest.MonkeyPatch, tmp_path: Path
+) -> None:
+    """文件已从磁盘删除（path.is_file() False）→ 1004"""
+    missing = tmp_path / "uploads" / "gone.txt"
+    fake = FakeQdrant(meta={"raw_file_path": str(missing), "original_filename": "gone.txt"})
+    _install_fake(monkeypatch, fake)
+
+    resp = client.get("/api/v1/documents/doc-1/file")
+
+    body = resp.json()
+    assert body["code"] == 1004
+    assert "文件不存在" in body["message"]
+
+
+def test_download_file_path_traversal_rejected(
+    client: TestClient, monkeypatch: pytest.MonkeyPatch, tmp_path: Path
+) -> None:
+    """raw_file_path 指向 upload_dir 之外 → 1004（路径越界校验）"""
+    outside = tmp_path / "secret.txt"
+    outside.write_bytes(b"secret")
+
+    fake = FakeQdrant(meta={"raw_file_path": str(outside), "original_filename": "secret.txt"})
+    _install_fake(monkeypatch, fake)
+
+    resp = client.get("/api/v1/documents/doc-1/file")
+
+    body = resp.json()
+    assert body["code"] == 1004
+    assert "文件不存在" in body["message"]
+
+
+def test_download_file_no_auth_required(
+    client: TestClient, monkeypatch: pytest.MonkeyPatch, tmp_path: Path
+) -> None:
+    """免登录：不携带 Authorization 头可正常下载（端点不挂鉴权依赖）"""
+    upload_dir = tmp_path / "uploads"
+    upload_dir.mkdir(parents=True, exist_ok=True)
+    target = upload_dir / "doc-1_open.txt"
+    target.write_bytes(b"open content")
+
+    fake = FakeQdrant(meta={"raw_file_path": str(target), "original_filename": "open.txt"})
+    _install_fake(monkeypatch, fake)
+
+    resp = client.get("/api/v1/documents/doc-1/file")
+
+    assert resp.status_code == 200
+    assert resp.content == b"open content"
+
+
+def test_download_file_uses_original_filename_when_missing(
+    client: TestClient, monkeypatch: pytest.MonkeyPatch, tmp_path: Path
+) -> None:
+    """metadata 缺少 original_filename 时回退到 path.name 作为下载文件名"""
+    upload_dir = tmp_path / "uploads"
+    upload_dir.mkdir(parents=True, exist_ok=True)
+    target = upload_dir / "fallback_name.txt"
+    target.write_bytes(b"fb")
+
+    fake = FakeQdrant(meta={"raw_file_path": str(target)})
+    _install_fake(monkeypatch, fake)
+
+    resp = client.get("/api/v1/documents/doc-1/file")
+
+    assert resp.status_code == 200
+    assert "fallback_name.txt" in resp.headers.get("content-disposition", "")

tests/test_document_file_integration.py

@@ -0,0 +1,161 @@
+"""文件链接集成验证
+
+在真实内存 Qdrant + FakeOllama + 真实临时 upload_dir 环境下，验证文件上传到下载的全链路闭环：
+POST /documents/upload → 入库 wait_done → GET /documents/{id}（file 字段）→ GET /documents/{id}/file；
+并覆盖纯文本入库无文件关联的对照路径（详情 file=None、下载端点 1004）。
+"""
+
+from pathlib import Path
+from urllib.parse import unquote
+
+import pytest
+from fastapi.testclient import TestClient
+from qdrant_client import AsyncQdrantClient
+
+from app.api.v1 import document as document_module
+from app.config import Settings
+from app.core.chunker import Chunker
+from app.core.classifier import Classifier
+from app.core.ingest_tasks import IngestTaskManager, IngestTaskStatus
+from app.core.ingestion import Ingester
+from app.core.sparse import SparseEncoder
+from app.core.summarizer import Summarizer
+from app.main import app
+from app.models.knowledge import load_taxonomy
+from app.services.qdrant import QdrantService
+from tests.test_e2e_integration import (
+    FAKE_CATEGORY,
+    DeterministicEmbedding,
+    FakeOllama,
+)
+
+
+async def _make_env() -> tuple[QdrantService, Ingester]:
+    """构建集成环境：真实组件 + 内存 Qdrant + FakeOllama + 确定性向量"""
+    qdrant = QdrantService(client=AsyncQdrantClient(location=":memory:"))
+    await qdrant.ensure_collections()
+    ollama = FakeOllama()
+    taxonomy = load_taxonomy()
+    embedding = DeterministicEmbedding()
+    ingester = Ingester(
+        summarizer=Summarizer(ollama=ollama),  # type: ignore[arg-type]
+        classifier=Classifier(ollama=ollama, taxonomy=taxonomy),  # type: ignore[arg-type]
+        chunker=Chunker(),
+        embedding=embedding,
+        sparse=SparseEncoder(),
+        qdrant=qdrant,
+    )
+    return qdrant, ingester
+
+
+def _patch_app(
+    monkeypatch: pytest.MonkeyPatch,
+    manager: IngestTaskManager,
+    qdrant: QdrantService,
+    upload_dir: str,
+) -> None:
+    """替换模块级单例：任务管理器 / Qdrant / upload_dir，lifespan 建集合改空操作"""
+
+    async def _noop_ensure_collections(self: QdrantService) -> None:
+        return None
+
+    monkeypatch.setattr(QdrantService, "ensure_collections", _noop_ensure_collections)
+    monkeypatch.setattr(document_module, "_task_manager", manager)
+    monkeypatch.setattr(document_module, "_qdrant", qdrant)
+    monkeypatch.setattr(document_module.settings, "upload_dir", upload_dir)
+
+
+class TestDocumentFileIntegration:
+    """文件链接集成：上传闭环 + 文本入库无文件对照"""
+
+    async def test_upload_full_loop(
+        self,
+        monkeypatch: pytest.MonkeyPatch,
+        tmp_path: Path,
+        admin_headers: dict[str, str],
+    ) -> None:
+        """文件上传闭环：upload → wait_done done → 详情含 file → 下载返回原文"""
+        qdrant, ingester = await _make_env()
+        manager = IngestTaskManager(ingester, None, Settings())
+        _patch_app(monkeypatch, manager, qdrant, str(tmp_path / "uploads"))
+
+        filename = "上传测试.md"
+        original_content = "上传原文测试内容"
+        with TestClient(app) as client:
+            # 1. multipart 上传 .md 文件（变更类端点需 admin 认证头）
+            resp_upload = client.post(
+                "/api/v1/documents/upload",
+                files={"file": (filename, original_content.encode("utf-8"), "text/markdown")},
+                headers=admin_headers,
+            )
+            assert resp_upload.status_code == 202
+            body_upload = resp_upload.json()
+            assert body_upload["code"] == 0
+            assert body_upload["data"]["status"] == "pending"
+            task_id = body_upload["data"]["task_id"]
+            assert task_id
+
+            # 2. 等待入库终态：done + 结果完整
+            final = await manager.wait_done(task_id)
+            assert final["status"] == IngestTaskStatus.DONE
+            result = final["result"]
+            document_id = result["document_id"]
+            assert document_id
+            assert result["category"] == FAKE_CATEGORY
+
+            # 3. 文档详情：file 字段非空，含原文件名 / url / size
+            resp_detail = client.get(f"/api/v1/documents/{document_id}")
+            body_detail = resp_detail.json()
+            assert body_detail["code"] == 0
+            file_info = body_detail["data"]["file"]
+            assert file_info is not None
+            assert filename in file_info["filename"]
+            assert file_info["url"] == f"/api/v1/documents/{document_id}/file"
+            assert file_info["size_bytes"] > 0
+
+            # 4. 下载：200 + 响应体含上传原文 + Content-Disposition 含原文件名（中文按 RFC 5987 百分号编码，需 unquote）
+            resp_file = client.get(f"/api/v1/documents/{document_id}/file")
+            assert resp_file.status_code == 200
+            assert original_content in resp_file.content.decode("utf-8")
+            disposition = resp_file.headers.get("content-disposition", "")
+            assert "attachment" in disposition
+            assert filename in unquote(disposition)
+
+    async def test_text_ingest_no_file(
+        self,
+        monkeypatch: pytest.MonkeyPatch,
+        tmp_path: Path,
+        admin_headers: dict[str, str],
+    ) -> None:
+        """纯文本入库（无 metadata）：详情 file=None，下载端点返回 1004"""
+        qdrant, ingester = await _make_env()
+        manager = IngestTaskManager(ingester, None, Settings())
+        _patch_app(monkeypatch, manager, qdrant, str(tmp_path / "uploads"))
+
+        with TestClient(app) as client:
+            # 1. 纯文本入库（无 metadata）
+            resp_post = client.post(
+                "/api/v1/documents",
+                json={"text": "纯文本入库无附件的对照内容", "title": "无文件文档"},
+                headers=admin_headers,
+            )
+            assert resp_post.status_code == 202
+            body_post = resp_post.json()
+            assert body_post["code"] == 0
+            task_id = body_post["data"]["task_id"]
+
+            # 2. 等待终态 done
+            final = await manager.wait_done(task_id)
+            assert final["status"] == IngestTaskStatus.DONE
+            document_id = final["result"]["document_id"]
+
+            # 3. 详情：file 为 None
+            resp_detail = client.get(f"/api/v1/documents/{document_id}")
+            body_detail = resp_detail.json()
+            assert body_detail["code"] == 0
+            assert body_detail["data"]["file"] is None
+
+            # 4. 下载端点：文档未关联文件 → 1004
+            resp_file = client.get(f"/api/v1/documents/{document_id}/file")
+            body_file = resp_file.json()
+            assert body_file["code"] == 1004

tests/test_ingest_task_api.py

@@ -19,6 +19,8 @@ class FakeManager:
         self.tasks = tasks or {}
         self.task_id = task_id
         self.submitted: list[DocumentInput] = []
+        self.retry_result: str | None = None
+        self.delete_result: bool = False
 
     async def submit(self, doc: DocumentInput) -> str:
         self.submitted.append(doc)
@@ -27,6 +29,12 @@ class FakeManager:
     async def get(self, task_id: str) -> dict[str, Any] | None:
         return self.tasks.get(task_id)
 
+    async def retry(self, task_id: str) -> str | None:
+        return self.retry_result
+
+    async def delete(self, task_id: str) -> bool:
+        return self.delete_result
+
 
 def _task(task_id: str, status: str, **extra: Any) -> dict[str, Any]:
     """构造一条任务记录（时间字段为固定 ISO8601 字符串）"""
@@ -159,3 +167,44 @@ def test_post_empty_text_creates_no_task(client: TestClient, monkeypatch: pytest
     body = resp.json()
     assert body["code"] == 1001
     assert manager.submitted == []
+
+
+def test_retry_task_returns_202(client: TestClient, monkeypatch: pytest.MonkeyPatch) -> None:
+    """POST tasks/{id}/retry：HTTP 202，返回新 task_id 与 pending 状态"""
+    manager = FakeManager()
+    manager.retry_result = "new-task-7"
+    _inject_manager(monkeypatch, manager)
+
+    resp = client.post("/api/v1/documents/tasks/t-failed/retry")
+
+    assert resp.status_code == 202
+    body = resp.json()
+    assert body["code"] == 0
+    assert body["data"]["task_id"] == "new-task-7"
+    assert body["data"]["status"] == "pending"
+
+
+def test_retry_task_not_found(client: TestClient, monkeypatch: pytest.MonkeyPatch) -> None:
+    """POST tasks/{id}/retry：manager 返回 None → code=1004"""
+    manager = FakeManager()
+    manager.retry_result = None
+    _inject_manager(monkeypatch, manager)
+
+    resp = client.post("/api/v1/documents/tasks/nope/retry")
+
+    body = resp.json()
+    assert body["code"] == 1004
+
+
+def test_delete_task(client: TestClient, monkeypatch: pytest.MonkeyPatch) -> None:
+    """DELETE tasks/{id}：删除成功 → deleted=true"""
+    manager = FakeManager()
+    manager.delete_result = True
+    _inject_manager(monkeypatch, manager)
+
+    resp = client.delete("/api/v1/documents/tasks/t-x")
+
+    body = resp.json()
+    assert body["code"] == 0
+    assert body["data"]["task_id"] == "t-x"
+    assert body["data"]["deleted"] is True

tests/test_ingest_tasks.py

@@ -7,8 +7,8 @@ from datetime import datetime
 from typing import Any
 
 from app.config import Settings
+from app.core.dedup import DEDUP_KEY_PREFIX
 from app.core.ingest_tasks import (
-    DEDUP_KEY_PREFIX,
     REDIS_KEY_PREFIX,
     IngestTaskManager,
     IngestTaskStatus,
@@ -91,6 +91,19 @@ class FakeRedis:
     async def get_json(self, key: str) -> dict[str, Any] | None:
         return self.store.get(key)
 
+    def _get_client(self) -> Any:
+        """返回支持 scan_iter 的假客户端（供 list_tasks 扫描键）"""
+        store = self.store
+
+        class _FakeClient:
+            async def scan_iter(self, match: str = "*") -> Any:
+                prefix = match[:-1] if match.endswith("*") else match
+                for key in list(store.keys()):
+                    if key.startswith(prefix):
+                        yield key
+
+        return _FakeClient()
+
 
 async def test_submit_returns_immediately_and_completes() -> None:
     """submit 立即返回；任务后台跑完为 done，结果完整，阶段序列齐全，Redis 镜像同步"""
@@ -317,3 +330,132 @@ async def test_dedup_lookup_failure_falls_back_to_normal_pipeline() -> None:
     final = await manager.wait_done(task, timeout=5)
     assert final["status"] == IngestTaskStatus.DONE
     assert len(ingester.calls) == 1
+
+
+async def test_submit_records_filename_from_metadata_or_title() -> None:
+    """submit：filename 优先 metadata.original_filename，其次 title"""
+    ingester = FakeIngester()
+    manager = IngestTaskManager(ingester, None, Settings())
+
+    # 有 original_filename
+    t1 = await manager.submit(
+        DocumentInput(text="x", title="t1", metadata={"original_filename": "report.pdf"})
+    )
+    await manager.wait_done(t1, timeout=5)
+    # 无 original_filename，回退 title
+    t2 = await manager.submit(DocumentInput(text="y", title="my-title"))
+    await manager.wait_done(t2, timeout=5)
+
+    items = await manager.list_tasks(limit=20)
+    by_id = {it["task_id"]: it for it in items}
+    assert by_id[t1]["filename"] == "report.pdf"
+    assert by_id[t2]["filename"] == "my-title"
+
+
+async def test_list_tasks_merges_memory_and_redis_and_sorts() -> None:
+    """list_tasks：合并内存与 Redis 镜像，去重，按 updated_at 降序"""
+    ingester = FakeIngester()
+    redis = FakeRedis()
+    manager = IngestTaskManager(ingester, redis, Settings())
+
+    # 内存任务 1
+    t1 = await manager.submit(DocumentInput(text="内容A", title="t1"))
+    await manager.wait_done(t1, timeout=5)
+    # 内存任务 2
+    t2 = await manager.submit(DocumentInput(text="内容B", title="t2"))
+    await manager.wait_done(t2, timeout=5)
+
+    # Redis 镜像中独有任务（不在内存里，模拟重启后只存在 Redis 的历史记录）
+    redis.store[f"{REDIS_KEY_PREFIX}redis-only-1"] = {
+        "task_id": "redis-only-1",
+        "status": "done",
+        "filename": "legacy.md",
+        "created_at": "2020-01-01T00:00:00+00:00",
+        "updated_at": "2020-01-01T00:00:00+00:00",
+        "result": {"document_id": "doc-legacy"},
+        "error": None,
+    }
+
+    items = await manager.list_tasks(limit=20)
+    # 共 3 条（内存 2 + Redis 独有 1，内存的 2 也已镜像到 Redis 但按 task_id 去重）
+    assert len(items) == 3
+    # 按 updated_at 降序：内存任务（当前时间）排在 Redis 旧任务前
+    ids = [it["task_id"] for it in items]
+    assert "redis-only-1" in ids
+    assert ids[-1] == "redis-only-1"  # 最旧排最后
+    # Redis 独有任务提取 doc_id 与 filename
+    legacy = next(it for it in items if it["task_id"] == "redis-only-1")
+    assert legacy["filename"] == "legacy.md"
+    assert legacy["doc_id"] == "doc-legacy"
+    # 内存 done 任务也提取 doc_id
+    mem_item = next(it for it in items if it["task_id"] == t1)
+    assert mem_item["doc_id"] == "doc-1"
+
+
+async def test_list_tasks_dedups_memory_and_redis_by_task_id() -> None:
+    """list_tasks：内存与 Redis 都有的同一 task_id 仅保留内存版本（去重）"""
+    ingester = FakeIngester()
+    redis = FakeRedis()
+    manager = IngestTaskManager(ingester, redis, Settings())
+
+    task_id = await manager.submit(DocumentInput(text="唯一", title="t"))
+    await manager.wait_done(task_id, timeout=5)
+
+    # Redis 镜像中给同一 task_id 篡改一个旧 status，验证内存版本胜出
+    redis.store[f"{REDIS_KEY_PREFIX}{task_id}"]["status"] = "pending"
+
+    items = await manager.list_tasks(limit=20)
+    assert len(items) == 1
+    assert items[0]["task_id"] == task_id
+    assert items[0]["status"] == IngestTaskStatus.DONE  # 内存版本（done）
+
+
+async def test_list_tasks_respects_limit() -> None:
+    """list_tasks：limit 截断返回条数"""
+    ingester = FakeIngester()
+    manager = IngestTaskManager(ingester, None, Settings())
+    for i in range(5):
+        t = await manager.submit(DocumentInput(text=f"c{i}", title=f"t{i}"))
+        await manager.wait_done(t, timeout=5)
+
+    items = await manager.list_tasks(limit=3)
+    assert len(items) == 3
+
+
+async def test_list_tasks_without_redis_returns_memory_only() -> None:
+    """Redis 不可用：list_tasks 仅返回内存任务"""
+    ingester = FakeIngester()
+    manager = IngestTaskManager(ingester, None, Settings())
+    t1 = await manager.submit(DocumentInput(text="x", title="t1"))
+    await manager.wait_done(t1, timeout=5)
+
+    items = await manager.list_tasks(limit=20)
+    assert len(items) == 1
+    assert items[0]["task_id"] == t1
+    assert items[0]["filename"] == "t1"
+    assert items[0]["doc_id"] == "doc-1"
+
+
+async def test_list_tasks_extract_fields_for_in_progress_task() -> None:
+    """进行中任务：doc_id 为 None（done 时才从 result.document_id 提取）"""
+    ingester = FakeIngester()
+    ingester.gate = asyncio.Event()  # 阻塞任务使其停留在最后一个阶段
+    manager = IngestTaskManager(ingester, None, Settings(ingest_max_concurrency=1))
+
+    t1 = await manager.submit(
+        DocumentInput(text="x", title="t1", metadata={"original_filename": "a.txt"})
+    )
+    await asyncio.wait_for(ingester.started.wait(), timeout=1)
+    await asyncio.sleep(0.05)
+
+    items = await manager.list_tasks(limit=20)
+    assert len(items) == 1
+    item = items[0]
+    assert item["task_id"] == t1
+    # 任务已推进到 writing 阶段（gate 阻塞前最后一个 progress_cb）
+    assert item["status"] == IngestTaskStatus.WRITING
+    assert item["doc_id"] is None  # 未完成，doc_id 为 None
+    assert item["filename"] == "a.txt"
+
+    ingester.gate.set()
+    await manager.wait_done(t1, timeout=5)

tests/test_ingestion.py

@@ -3,13 +3,27 @@
 from typing import Any
 
 import pytest
+from qdrant_client import AsyncQdrantClient
 
+from app.config import settings
 from app.core.chunker import Chunker
 from app.core.ingestion import Ingester, IngestionError
 from app.core.sparse import SparseEncoder
 from app.models.document import DocumentInput, DocumentSummary, SummaryLevel
 from app.models.knowledge import CategoryResult
-from app.services.qdrant import COLLECTION_L2, COLLECTION_L3
+from app.services.qdrant import COLLECTION_L2, COLLECTION_L3, QdrantService
+
+_DIM = settings.embedding_dimension
+
+
+def _fake_vec(index: int, dim: int) -> list[float]:
+    """构造确定性伪向量：第 0 维放 index 标识，其余补 0，长度对齐 dim"""
+    vec = [0.0] * dim
+    if dim > 0:
+        vec[0] = float(index)
+    if dim > 1:
+        vec[1] = 1.0
+    return vec
 
 
 class FakeSummarizer:
@@ -37,14 +51,18 @@ class FakeClassifier:
 
 
 class FakeEmbedding:
-    """按输入数量返回伪向量的假 EmbeddingService，记录每次调用的文本"""
+    """按输入数量返回伪向量的假 EmbeddingService，记录每次调用的文本
 
-    def __init__(self) -> None:
+    dim 默认 2（FakeQdrant 不校验维度）；接真实 Qdrant 时需传 settings.embedding_dimension。
+    """
+
+    def __init__(self, dim: int = 2) -> None:
+        self.dim = dim
         self.calls: list[list[str]] = []
 
     async def embed(self, texts: list[str]) -> list[list[float]]:
         self.calls.append(list(texts))
-        return [[float(i), 1.0] for i in range(len(texts))]
+        return [_fake_vec(i, self.dim) for i in range(len(texts))]
 
 
 class FakeQdrant:
@@ -65,6 +83,7 @@ class FakeQdrant:
         tags: list[str],
         dense_vector: list[float],
         sparse_vector: Any = None,
+        metadata: dict[str, str] | None = None,
     ) -> None:
         if self.fail_on == "l1":
             raise RuntimeError("qdrant down")
@@ -77,6 +96,7 @@ class FakeQdrant:
                 "tags": tags,
                 "dense_vector": dense_vector,
                 "sparse_vector": sparse_vector,
+                "metadata": metadata,
             }
         )
 
@@ -271,3 +291,59 @@ class TestQdrantFailure:
         assert len(qdrant.l1_calls) == 1
         assert {c for c, _ in qdrant.nodes_calls} == {COLLECTION_L2, COLLECTION_L3}
         assert qdrant.chunks_calls == []
+
+
+class TestMetadataIntegration:
+    """L1 metadata 端到端：真实内存 Qdrant + 假总结/分类/向量化，验证 metadata 透传与 get_doc_detail.file"""
+
+    @pytest.fixture
+    async def real_service(self) -> QdrantService:
+        svc = QdrantService(client=AsyncQdrantClient(location=":memory:"))
+        await svc.ensure_collections()
+        return svc
+
+    def _make_real_ingester(self, service: QdrantService, summary: DocumentSummary) -> Ingester:
+        return Ingester(
+            summarizer=FakeSummarizer(summary),  # type: ignore[arg-type]
+            classifier=FakeClassifier(_category()),  # type: ignore[arg-type]
+            chunker=Chunker(),
+            embedding=FakeEmbedding(dim=_DIM),  # type: ignore[arg-type]
+            sparse=SparseEncoder(),
+            qdrant=service,
+        )
+
+    async def test_ingest_with_metadata_populates_file_field(self, real_service: QdrantService) -> None:
+        """DocumentInput 带 metadata 入库后，get_doc_detail.file 含正确信息"""
+        doc = DocumentInput(
+            text="这是一段用于测试 metadata 透传的正文内容。" * 5,
+            title="带文件元数据的文档",
+            metadata={
+                "raw_file_path": "/data/uploads/spec.md",
+                "original_filename": "spec.md",
+                "original_size_bytes": "5120",
+            },
+        )
+        ingester = self._make_real_ingester(real_service, _structured_summary())
+        result = await ingester.ingest(doc)
+
+        # FakeQdrant 已被真实 service 取代：l1_calls 不再可用，直接查 Qdrant
+        detail = await real_service.get_doc_detail(result.document_id)
+        assert detail is not None
+        assert detail["file"] == {
+            "filename": "spec.md",
+            "size_bytes": 5120,
+            "url": f"/api/v1/documents/{result.document_id}/file",
+        }
+        # L1 payload 也应带 metadata 字段
+        assert detail["l1"]["metadata"] == doc.metadata
+
+    async def test_ingest_without_metadata_file_is_none(self, real_service: QdrantService) -> None:
+        """文本入库（metadata 为空 dict）→ get_doc_detail.file=None"""
+        doc = DocumentInput(text="纯文本入库，无文件元数据。" * 5, title="纯文本")
+        ingester = self._make_real_ingester(real_service, _structured_summary())
+        result = await ingester.ingest(doc)
+
+        detail = await real_service.get_doc_detail(result.document_id)
+        assert detail is not None
+        assert detail["file"] is None
+        assert detail["l1"]["metadata"] == {}

tests/test_qdrant.py

@@ -234,3 +234,152 @@ async def test_search_hybrid_rrf(service: QdrantService) -> None:
     assert len(filtered) == 1
     assert filtered[0].payload is not None
     assert filtered[0].payload["doc_id"] == "doc-h2"
+
+
+# ---------- L1 metadata 存储与读取 ----------
+
+
+async def test_upsert_l1_metadata_roundtrip(service: QdrantService) -> None:
+    """upsert_l1 写入 metadata 后，get_l1_metadata 返回写入的 dict"""
+    meta = {
+        "raw_file_path": "/data/uploads/report.pdf",
+        "original_filename": "年报.pdf",
+        "original_size_bytes": "2048",
+    }
+    await service.upsert_l1(
+        doc_id="doc-meta",
+        title="带元数据文档",
+        summary="总结",
+        category="tech",
+        tags=["x"],
+        dense_vector=_dense(0.9),
+        metadata=meta,
+    )
+    assert await service.get_l1_metadata("doc-meta") == meta
+
+
+async def test_upsert_l1_without_metadata_stores_empty_dict(service: QdrantService) -> None:
+    """upsert_l1 不传 metadata（旧签名）→ payload 存空 dict，get_l1_metadata 返回 {}"""
+    await service.upsert_l1(
+        doc_id="doc-no-meta",
+        title="无元数据文档",
+        summary="总结",
+        category="tech",
+        tags=["x"],
+        dense_vector=_dense(0.9),
+    )
+    assert await service.get_l1_metadata("doc-no-meta") == {}
+
+
+async def test_get_l1_metadata_doc_not_found(service: QdrantService) -> None:
+    """文档不存在 → get_l1_metadata 返回 None"""
+    assert await service.get_l1_metadata("doc-missing") is None
+
+
+async def test_get_l1_metadata_legacy_payload_without_field(service: QdrantService) -> None:
+    """存量旧文档：payload 完全没有 metadata 字段（绕过 upsert_l1 直接写点）→ 返回 None 不报错"""
+    import uuid
+
+    from qdrant_client import models
+
+    point = models.PointStruct(
+        id=str(uuid.uuid5(uuid.NAMESPACE_URL, "legacy-doc:l1")),
+        vector={"dense": _dense(0.4)},
+        payload={"doc_id": "legacy-doc", "title": "旧文档", "category": "tech", "tags": [], "text": "旧总结"},
+    )
+    await service.client.upsert(collection_name=COLLECTION_L1, points=[point])
+    assert await service.get_l1_metadata("legacy-doc") is None
+
+
+# ---------- get_doc_detail file 字段 ----------
+
+
+async def test_get_doc_detail_file_field_with_raw_path(service: QdrantService) -> None:
+    """L1 metadata 含 raw_file_path → file 字段含 filename/size_bytes/url"""
+    await service.upsert_l1(
+        doc_id="doc-file",
+        title="文件文档",
+        summary="总结",
+        category="tech",
+        tags=["x"],
+        dense_vector=_dense(0.9),
+        metadata={
+            "raw_file_path": "/data/uploads/report.pdf",
+            "original_filename": "年报.pdf",
+            "original_size_bytes": "4096",
+        },
+    )
+    detail = await service.get_doc_detail("doc-file")
+    assert detail is not None
+    file_info = detail["file"]
+    assert file_info == {
+        "filename": "年报.pdf",
+        "size_bytes": 4096,
+        "url": "/api/v1/documents/doc-file/file",
+    }
+
+
+async def test_get_doc_detail_file_field_filename_fallback_to_path_name(service: QdrantService) -> None:
+    """metadata 缺 original_filename 时，filename 回退为 raw_file_path 的 Path.name"""
+    await service.upsert_l1(
+        doc_id="doc-file-fb",
+        title="文件文档",
+        summary="总结",
+        category="tech",
+        tags=["x"],
+        dense_vector=_dense(0.9),
+        metadata={"raw_file_path": "/data/uploads/notes.md", "original_size_bytes": "100"},
+    )
+    detail = await service.get_doc_detail("doc-file-fb")
+    assert detail is not None
+    assert detail["file"]["filename"] == "notes.md"
+    assert detail["file"]["size_bytes"] == 100
+
+
+async def test_get_doc_detail_file_none_when_no_metadata(service: QdrantService) -> None:
+    """无 metadata（文本入库，存空 dict）→ file=None"""
+    await service.upsert_l1(
+        doc_id="doc-text",
+        title="纯文本文档",
+        summary="总结",
+        category="tech",
+        tags=["x"],
+        dense_vector=_dense(0.9),
+    )
+    detail = await service.get_doc_detail("doc-text")
+    assert detail is not None
+    assert detail["file"] is None
+
+
+async def test_get_doc_detail_file_none_when_no_raw_path(service: QdrantService) -> None:
+    """metadata 无 raw_file_path → file=None"""
+    await service.upsert_l1(
+        doc_id="doc-meta-no-path",
+        title="文档",
+        summary="总结",
+        category="tech",
+        tags=["x"],
+        dense_vector=_dense(0.9),
+        metadata={"original_filename": "x.pdf", "original_size_bytes": "10"},
+    )
+    detail = await service.get_doc_detail("doc-meta-no-path")
+    assert detail is not None
+    assert detail["file"] is None
+
+
+async def test_get_doc_detail_file_none_for_legacy_doc(service: QdrantService) -> None:
+    """存量旧文档（payload 无 metadata 字段，绕过 upsert_l1 直接写点）→ file=None 不报错"""
+    import uuid
+
+    from qdrant_client import models
+
+    point = models.PointStruct(
+        id=str(uuid.uuid5(uuid.NAMESPACE_URL, "legacy-detail:l1")),
+        vector={"dense": _dense(0.3)},
+        payload={"doc_id": "legacy-detail", "title": "旧文档", "category": "tech", "tags": [], "text": "旧总结"},
+    )
+    await service.client.upsert(collection_name=COLLECTION_L1, points=[point])
+    detail = await service.get_doc_detail("legacy-detail")
+    assert detail is not None
+    assert detail["file"] is None
+    assert detail["l1"]["doc_id"] == "legacy-detail"

tests/test_reranker.py

@@ -0,0 +1,128 @@
+"""RerankerService 单元测试（mock httpx，不发起真实网络请求）"""
+
+from typing import Any
+
+import httpx
+import pytest
+
+from app.config import settings
+from app.services.reranker import OllamaRerankerService, create_reranker_service
+
+
+def _make_response(status_code: int, data: dict | None = None) -> httpx.Response:
+    """构造带 request 上下文的 httpx.Response（raise_for_status 依赖 request）"""
+    request = httpx.Request("POST", "http://localhost:11434/api/rerank")
+    if data is None:
+        return httpx.Response(status_code, request=request)
+    return httpx.Response(status_code, json=data, request=request)
+
+
+class _FakeAsyncClient:
+    """httpx.AsyncClient 替代品：记录请求参数，返回预设响应或抛出预设异常"""
+
+    def __init__(
+        self,
+        captured: dict,
+        response: httpx.Response | None = None,
+        error: Exception | None = None,
+        **kwargs: Any,
+    ) -> None:
+        self._captured = captured
+        self._response = response
+        self._error = error
+        captured["timeout"] = kwargs.get("timeout")
+
+    async def __aenter__(self) -> "_FakeAsyncClient":
+        return self
+
+    async def __aexit__(self, *args: object) -> bool:
+        return False
+
+    async def post(self, url: str, json: dict | None = None) -> httpx.Response:
+        self._captured["url"] = url
+        self._captured["json"] = json
+        if self._error is not None:
+            raise self._error
+        assert self._response is not None
+        return self._response
+
+
+def _client(monkeypatch: pytest.MonkeyPatch, captured: dict, **kw: Any) -> None:
+    monkeypatch.setattr(httpx, "AsyncClient", lambda **k: _FakeAsyncClient(captured, **k, **kw))
+
+
+class TestRerank:
+    async def test_returns_scores_aligned_to_documents(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        captured: dict = {}
+        _client(
+            monkeypatch,
+            captured,
+            response=_make_response(
+                200,
+                {
+                    "results": [
+                        {"index": 0, "relevance_score": 0.2},
+                        {"index": 1, "relevance_score": 0.9},
+                        {"index": 2, "relevance_score": 0.5},
+                    ]
+                },
+            ),
+        )
+        service = OllamaRerankerService(base_url="http://localhost:11434/", model="qwen3-reranker:0.6b")
+
+        scores = await service.rerank("查询", ["doc-a", "doc-b", "doc-c"])
+
+        assert scores == [0.2, 0.9, 0.5]
+        assert captured["url"] == "http://localhost:11434/api/rerank"
+        assert captured["json"] == {
+            "model": "qwen3-reranker:0.6b",
+            "query": "查询",
+            "documents": ["doc-a", "doc-b", "doc-c"],
+        }
+
+    async def test_empty_documents_returns_empty(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        captured: dict = {}
+        _client(monkeypatch, captured, response=_make_response(200, {"results": []}))
+        service = OllamaRerankerService(base_url="http://localhost:11434", model="qwen3-reranker:0.6b")
+
+        scores = await service.rerank("查询", [])
+
+        assert scores == []
+        # 空文档不发起请求
+        assert "url" not in captured
+
+    async def test_partial_results_zero_fill_missing(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        captured: dict = {}
+        # 仅返回 index 1，index 0/2 缺失 → 补 0
+        _client(
+            monkeypatch,
+            captured,
+            response=_make_response(200, {"results": [{"index": 1, "relevance_score": 0.7}]}),
+        )
+        service = OllamaRerankerService(base_url="http://localhost:11434", model="qwen3-reranker:0.6b")
+
+        scores = await service.rerank("查询", ["a", "b", "c"])
+
+        assert scores == [0.0, 0.7, 0.0]
+
+    async def test_http_error_raises(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        _client(monkeypatch, {}, response=_make_response(500))
+        service = OllamaRerankerService(base_url="http://localhost:11434", model="qwen3-reranker:0.6b")
+
+        with pytest.raises(httpx.HTTPStatusError):
+            await service.rerank("查询", ["a", "b"])
+
+
+class TestCreateRerankerService:
+    def test_disabled_returns_none(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        monkeypatch.setattr(settings, "reranker_enabled", False)
+        assert create_reranker_service() is None
+
+    def test_enabled_returns_service(self, monkeypatch: pytest.MonkeyPatch) -> None:
+        monkeypatch.setattr(settings, "reranker_enabled", True)
+        monkeypatch.setattr(settings, "reranker_model", "qwen3-reranker:0.6b")
+        monkeypatch.setattr(settings, "ollama_base_url", "http://ollama:11434")
+        monkeypatch.setattr(settings, "reranker_timeout", 30.0)
+        service = create_reranker_service()
+        assert isinstance(service, OllamaRerankerService)
+        assert service.model == "qwen3-reranker:0.6b"

tests/test_retriever.py

@@ -82,15 +82,29 @@ def _route(
     )
 
 
-def _make_retriever(qdrant: FakeQdrant, route: RouteDecision) -> Retriever:
+def _make_retriever(qdrant: FakeQdrant, route: RouteDecision, reranker: Any = None) -> Retriever:
     return Retriever(
         qdrant=qdrant,  # type: ignore[arg-type]
         query_parser=FakeParser(route),  # type: ignore[arg-type]
         embedding=FakeEmbedding(),
         sparse_encoder=SparseEncoder(),
+        reranker=reranker,
     )
 
 
+class FakeReranker:
+    """假 RerankerService：返回预设分数或抛出预设异常"""
+
+    def __init__(self, scores: list[float] | None = None, exc: Exception | None = None) -> None:
+        self.scores = scores if scores is not None else []
+        self.exc = exc
+
+    async def rerank(self, query: str, documents: list[str]) -> list[float]:
+        if self.exc is not None:
+            raise self.exc
+        return self.scores
+
+
 def _calls(qdrant: FakeQdrant, collection: str) -> list[dict[str, Any]]:
     return [c for c in qdrant.calls if c["collection"] == collection]
 
@@ -348,3 +362,52 @@ class TestSearchApi:
         resp = client.get("/api/v1/health")
         assert resp.status_code == 200
         assert resp.json() == {"status": "ok"}
+
+
+class TestReranker:
+    """重排启用时对 chunk 候选精排；关闭或失败则退化为 RRF 顺序"""
+
+    @staticmethod
+    def _full_qdrant() -> FakeQdrant:
+        return FakeQdrant(
+            {
+                COLLECTION_L1: [[_point("l1a", "d1")]],
+                COLLECTION_L2: [[_point("l2a", "d1", "章节A")]],
+                COLLECTION_L3: [[_point("l3a", "d1", "章节A")]],
+                COLLECTION_CHUNKS: [
+                    [
+                        _point("c1", "d1", "章节A"),
+                        _point("c2", "d1", "章节A"),
+                        _point("c3", "d1", "章节A"),
+                    ]
+                ],
+            }
+        )
+
+    async def test_enabled_reorders_by_relevance(self):
+        """重排按相关性分数降序重排，并将 score 写回相关性分数"""
+        reranker = FakeReranker(scores=[0.1, 0.9, 0.5])  # c1<c3<c2
+        retriever = _make_retriever(self._full_qdrant(), _route(), reranker=reranker)
+
+        resp = await retriever.search(SearchRequest(query="测试查询"))
+
+        # 重排后顺序为 c2, c3, c1
+        assert [h.text for h in resp.hits] == ["text-c2", "text-c3", "text-c1"]
+        assert resp.hits[0].score == 0.9
+
+    async def test_disabled_keeps_rrf_order(self):
+        """重排未注入（关闭）时保持 RRF 融合原始顺序"""
+        retriever = _make_retriever(self._full_qdrant(), _route(), reranker=None)
+
+        resp = await retriever.search(SearchRequest(query="测试查询"))
+
+        assert [h.text for h in resp.hits] == ["text-c1", "text-c2", "text-c3"]
+
+    async def test_failure_falls_back_to_rrf(self):
+        """重排调用异常时退化为 RRF 顺序，检索不中断"""
+        reranker = FakeReranker(exc=RuntimeError("ollama 不可用"))
+        retriever = _make_retriever(self._full_qdrant(), _route(), reranker=reranker)
+
+        resp = await retriever.search(SearchRequest(query="测试查询"))
+
+        assert [h.text for h in resp.hits] == ["text-c1", "text-c2", "text-c3"]

tests/test_user_management_integration.py

@@ -0,0 +1,247 @@
+"""Spec Task 5：用户管理增强（PATCH 端点 + enabled 字段）全链路集成验证
+
+内存 UserStore/SessionStore 经 conftest.auth_stores 夹具注入 app.api.deps 单例，
+通过 TestClient 走真实 HTTP 链路（不跑 lifespan，无需真实 Redis/Qdrant），覆盖：
+- 改角色生效：admin 创建 user → PATCH 改 admin → GET /auth/me 与重新登录均反映新角色
+- 禁用用户：旧 token 立即失效（session 已清，1005）；重新登录 1005（"账号已禁用"）
+- 重新启用：PATCH enabled=True → login 恢复成功
+- 最后 admin 保护：唯一 admin 降级/禁用自己 → 1001（且自身状态未变）
+- PATCH 校验：空 body / 非法 role → 1001；不存在用户 → 1004；非 admin 调用 → 1006
+"""
+
+from typing import Any
+
+from fastapi.testclient import TestClient
+
+from app.main import app
+
+
+def _login(client: TestClient, username: str, password: str) -> dict[str, Any]:
+    """调登录接口并返回响应体"""
+    return client.post("/api/v1/auth/login", json={"username": username, "password": password}).json()
+
+
+def _bearer(token: str) -> dict[str, str]:
+    """构造 Authorization Bearer 请求头"""
+    return {"Authorization": f"Bearer {token}"}
+
+
+def _admin_login(client: TestClient) -> dict[str, str]:
+    """以 auth_stores 预置的 admin（admin/admin-pass-123）登录，返回 Bearer 请求头"""
+    resp = _login(client, "admin", "admin-pass-123")
+    assert resp["code"] == 0, f"admin 登录失败: {resp}"
+    return _bearer(resp["data"]["token"])
+
+
+def _create_user(
+    client: TestClient,
+    admin_headers: dict[str, str],
+    username: str,
+    password: str,
+    role: str = "user",
+) -> None:
+    """admin 创建用户并断言成功"""
+    resp = client.post(
+        "/api/v1/auth/users",
+        json={"username": username, "password": password, "role": role},
+        headers=admin_headers,
+    ).json()
+    assert resp["code"] == 0, f"创建用户 {username} 失败: {resp}"
+
+
+class TestRoleChange:
+    """改角色生效全链路"""
+
+    def test_role_change_takes_effect(self, auth_stores) -> None:
+        client = TestClient(app)
+        admin_headers = _admin_login(client)
+
+        # admin 创建 user 角色账号 alice
+        _create_user(client, admin_headers, "alice", "alice-pass-123", role="user")
+
+        # alice 登录，初始角色 user
+        alice_login = _login(client, "alice", "alice-pass-123")
+        assert alice_login["code"] == 0
+        assert alice_login["data"]["role"] == "user"
+        alice_headers = _bearer(alice_login["data"]["token"])
+
+        # 改角色前 GET /auth/me 反映 user 角色
+        me_before = client.get("/api/v1/auth/me", headers=alice_headers).json()
+        assert me_before["code"] == 0
+        assert me_before["data"]["role"] == "user"
+
+        # admin PATCH 改 alice 角色为 admin
+        patched = client.patch(
+            "/api/v1/auth/users/alice",
+            json={"role": "admin"},
+            headers=admin_headers,
+        ).json()
+        assert patched["code"] == 0
+        assert patched["data"]["role"] == "admin"
+
+        # 同一 token 立即反映新角色（角色变更不清 session，用户记录实时读取）
+        me_after = client.get("/api/v1/auth/me", headers=alice_headers).json()
+        assert me_after["code"] == 0
+        assert me_after["data"]["role"] == "admin"
+
+        # 重新登录也反映新角色
+        relogin = _login(client, "alice", "alice-pass-123")
+        assert relogin["code"] == 0
+        assert relogin["data"]["role"] == "admin"
+
+
+class TestDisableUser:
+    """禁用用户：旧 session 清除 + 登录拒绝"""
+
+    def test_disable_user_clears_session_and_blocks_login(self, auth_stores) -> None:
+        client = TestClient(app)
+        admin_headers = _admin_login(client)
+
+        _create_user(client, admin_headers, "bob", "bob-pass-123")
+
+        # bob 登录拿 token
+        bob_login = _login(client, "bob", "bob-pass-123")
+        assert bob_login["code"] == 0
+        bob_headers = _bearer(bob_login["data"]["token"])
+
+        # admin 禁用 bob（清空其全部 session）
+        disabled = client.patch(
+            "/api/v1/auth/users/bob",
+            json={"enabled": False},
+            headers=admin_headers,
+        ).json()
+        assert disabled["code"] == 0
+        assert disabled["data"]["enabled"] is False
+
+        # bob 旧 token 调鉴权端点 → 1005（session 已清，凭证无效）
+        me = client.get("/api/v1/auth/me", headers=bob_headers).json()
+        assert me["code"] == 1005
+
+        # bob 重新登录 → 1005（账号已禁用）
+        relogin = _login(client, "bob", "bob-pass-123")
+        assert relogin["code"] == 1005
+        assert "禁用" in relogin["message"]
+
+
+class TestReenableUser:
+    """重新启用：login 恢复成功"""
+
+    def test_reenable_user_allows_login(self, auth_stores) -> None:
+        client = TestClient(app)
+        admin_headers = _admin_login(client)
+
+        _create_user(client, admin_headers, "carol", "carol-pass-123")
+
+        # 先禁用 carol，确认登录被拒
+        disabled = client.patch(
+            "/api/v1/auth/users/carol",
+            json={"enabled": False},
+            headers=admin_headers,
+        ).json()
+        assert disabled["code"] == 0
+        assert _login(client, "carol", "carol-pass-123")["code"] == 1005
+
+        # 重新启用
+        enabled = client.patch(
+            "/api/v1/auth/users/carol",
+            json={"enabled": True},
+            headers=admin_headers,
+        ).json()
+        assert enabled["code"] == 0
+        assert enabled["data"]["enabled"] is True
+
+        # login 恢复成功
+        relogin = _login(client, "carol", "carol-pass-123")
+        assert relogin["code"] == 0
+        assert relogin["data"]["username"] == "carol"
+
+
+class TestLastAdminProtection:
+    """最后 admin 保护：唯一 admin 不可降级/禁用"""
+
+    def test_cannot_demote_last_admin(self, auth_stores) -> None:
+        client = TestClient(app)
+        admin_headers = _admin_login(client)
+
+        # 唯一 admin（admin）尝试降级自己 role=user → 1001
+        resp = client.patch(
+            "/api/v1/auth/users/admin",
+            json={"role": "user"},
+            headers=admin_headers,
+        ).json()
+        assert resp["code"] == 1001
+        # admin 未被降级，仍可访问 admin 专属端点
+        assert client.get("/api/v1/auth/users", headers=admin_headers).json()["code"] == 0
+
+    def test_cannot_disable_last_admin(self, auth_stores) -> None:
+        client = TestClient(app)
+        admin_headers = _admin_login(client)
+
+        # 唯一 admin 尝试禁用自己 enabled=False → 1001
+        resp = client.patch(
+            "/api/v1/auth/users/admin",
+            json={"enabled": False},
+            headers=admin_headers,
+        ).json()
+        assert resp["code"] == 1001
+        # admin 未被禁用，旧 token 仍可用
+        assert client.get("/api/v1/auth/users", headers=admin_headers).json()["code"] == 0
+
+
+class TestPatchValidation:
+    """PATCH 端点校验：空 body / 非法 role / 不存在用户 / 非 admin"""
+
+    def test_empty_body_rejected(self, auth_stores) -> None:
+        client = TestClient(app)
+        admin_headers = _admin_login(client)
+        _create_user(client, admin_headers, "dave", "dave-pass-123")
+
+        # 空 body（role 与 enabled 均缺）→ 模型校验 1001
+        resp = client.patch(
+            "/api/v1/auth/users/dave",
+            json={},
+            headers=admin_headers,
+        ).json()
+        assert resp["code"] == 1001
+
+    def test_invalid_role_rejected(self, auth_stores) -> None:
+        client = TestClient(app)
+        admin_headers = _admin_login(client)
+        _create_user(client, admin_headers, "dave", "dave-pass-123")
+
+        # role 非法（非 admin/user）→ 模型校验 1001
+        resp = client.patch(
+            "/api/v1/auth/users/dave",
+            json={"role": "superuser"},
+            headers=admin_headers,
+        ).json()
+        assert resp["code"] == 1001
+
+    def test_nonexistent_user_rejected(self, auth_stores) -> None:
+        client = TestClient(app)
+        admin_headers = _admin_login(client)
+
+        # 不存在用户 → 1004
+        resp = client.patch(
+            "/api/v1/auth/users/ghost",
+            json={"role": "admin"},
+            headers=admin_headers,
+        ).json()
+        assert resp["code"] == 1004
+
+    def test_non_admin_forbidden(self, auth_stores) -> None:
+        client = TestClient(app)
+        admin_headers = _admin_login(client)
+        _create_user(client, admin_headers, "eve", "eve-pass-123", role="user")
+
+        # eve（user 角色）登录后调 PATCH → 1006
+        eve_login = _login(client, "eve", "eve-pass-123")
+        assert eve_login["code"] == 0
+        eve_headers = _bearer(eve_login["data"]["token"])
+
+        resp = client.patch(
+            "/api/v1/auth/users/eve",
+            json={"role": "admin"},
+            headers=eve_headers,
+        ).json()
+        assert resp["code"] == 1006

tests/test_users.py

@@ -17,6 +17,7 @@ import pytest
 
 from app.core.users import (
     UserExistsError,
+    UserNotFoundError,
     UserStore,
     UserStoreError,
     bootstrap_admin,
@@ -225,6 +226,101 @@ class TestCountAdmins:
         assert await store.count_admins() == 1
 
 
+class TestEnabledField:
+    """enabled 字段：默认 True、create 传参、get/list 回读、存量兼容"""
+
+    async def test_default_enabled_is_true(self):
+        record = await UserStore(FakeRedis()).create("alice", "password123")
+        assert record.enabled is True
+
+    async def test_create_with_enabled_false(self):
+        redis = FakeRedis()
+        store = UserStore(redis)
+        record = await store.create("alice", "password123", enabled=False)
+        assert record.enabled is False
+        # 持久化后回读仍为 False
+        assert (await store.get("alice")).enabled is False
+        # 落库 JSON 含 enabled 字段
+        assert json.loads(redis.store["user:alice"])["enabled"] is False
+
+    async def test_get_list_roundtrip_enabled(self):
+        store = UserStore(FakeRedis())
+        await store.create("alice", "password123", enabled=False)
+        await store.create("bob", "password123", role="admin")
+        assert (await store.get("alice")).enabled is False
+        assert (await store.get("bob")).enabled is True
+        records = {r.username: r for r in await store.list()}
+        assert records["alice"].enabled is False
+        assert records["bob"].enabled is True
+
+    async def test_legacy_record_without_enabled_defaults_true(self):
+        """存量记录无 enabled 字段时按 True 兼容（get/list）"""
+        redis = FakeRedis()
+        store = UserStore(redis)
+        # 直接写入无 enabled 字段的存量记录
+        redis.store["user:legacy"] = json.dumps(
+            {
+                "username": "legacy",
+                "role": "user",
+                "password_hash": "hash",
+                "salt": "00" * 16,
+                "must_change_password": False,
+                "created_at": "2024-01-01T00:00:00+00:00",
+            }
+        )
+        record = await store.get("legacy")
+        assert record is not None
+        assert record.enabled is True
+        records = await store.list()
+        assert records[0].enabled is True
+
+
+class TestUpdateUser:
+    """update_user：角色/启用状态更新与校验"""
+
+    async def test_update_role(self):
+        store = UserStore(FakeRedis())
+        await store.create("alice", "password123")
+        updated = await store.update_user("alice", role="admin")
+        assert updated.role == "admin"
+        assert updated.enabled is True  # 未改动
+        # 持久化
+        record = await store.get("alice")
+        assert record is not None
+        assert record.role == "admin"
+
+    async def test_update_enabled(self):
+        store = UserStore(FakeRedis())
+        await store.create("alice", "password123", role="admin")
+        updated = await store.update_user("alice", enabled=False)
+        assert updated.enabled is False
+        assert updated.role == "admin"  # 未改动
+
+    async def test_update_both(self):
+        store = UserStore(FakeRedis())
+        await store.create("alice", "password123")
+        updated = await store.update_user("alice", role="admin", enabled=False)
+        assert updated.role == "admin"
+        assert updated.enabled is False
+
+    async def test_user_not_found_raises(self):
+        with pytest.raises(UserNotFoundError):
+            await UserStore(FakeRedis()).update_user("nobody", role="admin")
+
+    async def test_invalid_role_raises_value_error(self):
+        store = UserStore(FakeRedis())
+        await store.create("alice", "password123")
+        with pytest.raises(ValueError):
+            await store.update_user("alice", role="superuser")
+
+    async def test_no_fields_noop(self):
+        store = UserStore(FakeRedis())
+        await store.create("alice", "password123", role="admin")
+        updated = await store.update_user("alice")
+        assert updated.role == "admin"
+        assert updated.enabled is True
+
+
 class TestBootstrapAdmin:
     """空库引导创建默认管理员"""