feat(v3): Web 应用化基线（异步任务/SSE/llama-server 管理/Vue SPA 四页 + 设置页整页滚动修复）

config/config.yaml

@@ -88,8 +88,8 @@ runtime:
     cpu:   {ngl: 0,  ctx: 8192}
 
 architect:                              # 大模型（API）
-  model: deepseek-chat
+  model: deepseek-v4-flash
-  base_url: https://api.deepseek.com/v1
+  base_url: https://api.deepseek.com
   api_key_env: DEEPSEEK_API_KEY
   temperature: 0.2
   timeout_s: 60

gateway/jobs.py

@@ -0,0 +1,328 @@
+"""异步后台任务注册表（T1：后端异步化核心）。
+
+设计原则（对齐 v3 方案 D1-D6）：
+- asyncio 原生，无 Celery/Redis/外部队列
+- 每个 request_id -> TaskInfo（状态/开始时间/结果或错误）
+- 任务写 runs/{id}/workspace.json，SSE 生成器只读该文件（D3：不改 pipeline）
+- 定期清理已完成任务（防内存泄漏）
+"""
+from __future__ import annotations
+
+import asyncio
+import json
+import time
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any, AsyncGenerator, Dict, Optional
+
+# runs/ 目录（与 pipeline.py 默认一致）
+RUNS_DIR = Path("runs")
+
+# 任务状态
+STATE_PENDING = "pending"
+STATE_RUNNING = "running"
+STATE_DONE = "done"
+STATE_FAILED = "failed"
+
+
+@dataclass
+class TaskInfo:
+    """一个后台任务的状态快照。"""
+    request_id: str
+    state: str = STATE_PENDING           # pending | running | done | failed
+    started_at: float = 0.0             # time.time()
+    finished_at: float = 0.0            # time.time()（done/failed 时）
+    error: Optional[str] = None          # failed 时错误信息
+    # PipelineResult 字段（done 时填充）
+    response: Optional[str] = None
+    status: Optional[str] = None          # done | fast_path | escalated | failed
+    fast_path: bool = False
+    rounds_used: int = 0
+    api_input_tokens: int = 0
+    api_output_tokens: int = 0
+    cost_est: float = 0.0
+    model_used: Optional[str] = None
+    latency_ms: float = 0.0
+    workspace_path: Optional[str] = None
+    error_detail: Optional[str] = None
+    route: list = field(default_factory=list)
+
+    def to_dict(self) -> Dict[str, Any]:
+        return {
+            "request_id": self.request_id,
+            "state": self.state,
+            "started_at": self.started_at,
+            "finished_at": self.finished_at,
+            "error": self.error,
+            "response": self.response,
+            "status": self.status,
+            "fast_path": self.fast_path,
+            "rounds_used": self.rounds_used,
+            "api_input_tokens": self.api_input_tokens,
+            "api_output_tokens": self.api_output_tokens,
+            "cost_est": self.cost_est,
+            "model_used": self.model_used,
+            "latency_ms": round(self.latency_ms, 2),
+            "workspace_path": self.workspace_path,
+            "route": self.route,
+            "error_detail": self.error_detail,
+        }
+
+    def to_result_event(self) -> Dict[str, Any]:
+        """终态 SSE result 事件 payload。"""
+        return {
+            "type": "result",
+            "request_id": self.request_id,
+            "response": self.response or "",
+            "status": self.status,
+            "fast_path": self.fast_path,
+            "rounds_used": self.rounds_used,
+            "api_input_tokens": self.api_input_tokens,
+            "api_output_tokens": self.api_output_tokens,
+            "cost_est": self.cost_est,
+            "model_used": self.model_used,
+            "latency_ms": round(self.latency_ms, 2),
+            "route": self.route,
+            "error": self.error,
+        }
+
+
+class JobStore:
+    """asyncio 后台任务注册表。
+
+    线程安全（asyncio 事件循环单线程，不需要额外锁）。
+    任务由 register() 注册、由 _task_done() 填充结果。
+    SSE 生成器调用 watch_file() 轮询 workspace.json 变化。
+    """
+
+    def __init__(self, max_age_seconds: float = 3600.0, max_running: int = 50):
+        """
+        Args:
+            max_age_seconds: 已完成任务在内存中保留时间（秒），超时后自动清理
+            max_running: 最大同时运行任务数，超出后拒绝新任务
+        """
+        self._tasks: Dict[str, TaskInfo] = {}
+        self._asyncio_tasks: Dict[str, "asyncio.Task[None]"] = {}  # request_id -> asyncio.Task
+        self._cancel_events: Dict[str, asyncio.Event] = {}        # request_id -> cancel Event
+        self.max_age_seconds = max_age_seconds
+        self.max_running = max_running
+        self._poll_interval = 0.3  # workspace.json 轮询间隔（秒）
+
+    # ─────────────────────────────────────────────────────────────────
+    # 公共 API
+    # ─────────────────────────────────────────────────────────────────
+
+    def register(self, request_id: str) -> tuple[bool, str]:
+        """注册一个 pending 任务。返回 (True, "") 成功，(False, reason) 容量满。"""
+        running = [t for t in self._tasks.values() if t.state == STATE_RUNNING]
+        if len(running) >= self.max_running:
+            return False, f"同时运行任务已达上限 {self.max_running}，请稍后重试"
+        if request_id in self._tasks:
+            return False, f"任务 {request_id} 已存在"
+        self._tasks[request_id] = TaskInfo(
+            request_id=request_id,
+            state=STATE_PENDING,
+            started_at=time.time(),
+        )
+        return True, ""
+
+    def get(self, request_id: str) -> Optional[TaskInfo]:
+        """查询任务状态。"""
+        return self._tasks.get(request_id)
+
+    def list_all(self) -> Dict[str, TaskInfo]:
+        """列出所有任务（含已完成）。"""
+        return dict(self._tasks)
+
+    def list_running(self) -> Dict[str, TaskInfo]:
+        """只列出运行中任务。"""
+        return {k: v for k, v in self._tasks.items() if v.state == STATE_RUNNING}
+
+    def submit(
+        self,
+        request_id: str,
+        coro,  # type: asyncio.coroutine
+    ) -> "asyncio.Task[None]":
+        """提交协程到后台运行；内部注册 asyncio.Task。"""
+        task = asyncio.create_task(self._run_wrapper(request_id, coro))
+        self._asyncio_tasks[request_id] = task
+        return task
+
+    def new_cancel(self, request_id: str) -> asyncio.Event:
+        """为 SSE 连接创建一个取消事件。"""
+        evt = asyncio.Event()
+        self._cancel_events[request_id] = evt
+        return evt
+
+    def get_cancel(self, request_id: str) -> Optional[asyncio.Event]:
+        """获取已有取消事件。"""
+        return self._cancel_events.get(request_id)
+
+    # ─────────────────────────────────────────────────────────────────
+    # 内部：任务运行包装
+    # ─────────────────────────────────────────────────────────────────
+
+    async def _run_wrapper(self, request_id: str, coro):
+        """把用户协程包装成可追踪的任务：更新状态、捕获异常、清理。"""
+        info = self._tasks.get(request_id)
+        if info is None:
+            return
+        info.state = STATE_RUNNING
+        try:
+            await coro
+        except Exception as exc:  # pragma: no cover
+            if info:
+                info.state = STATE_FAILED
+                info.finished_at = time.time()
+                info.error = f"unhandled:{type(exc).__name__}:{exc}"
+        finally:
+            # 清理 asyncio task 引用
+            self._asyncio_tasks.pop(request_id, None)
+            # 定期 GC 已完成任务
+            self._cleanup_aged()
+
+    def _task_done(
+        self,
+        request_id: str,
+        result: "PipelineResult",  # from pipeline.PipelineResult
+    ) -> None:
+        """任务正常完成时由调用方调用，写入结果。"""
+        info = self._tasks.get(request_id)
+        if info is None:
+            return
+        info.state = STATE_DONE
+        info.finished_at = time.time()
+        info.response = result.response
+        info.status = result.status
+        info.fast_path = result.fast_path
+        info.rounds_used = result.rounds_used
+        info.api_input_tokens = result.api_input_tokens
+        info.api_output_tokens = result.api_output_tokens
+        info.cost_est = result.cost_est
+        info.model_used = result.model_used
+        info.latency_ms = result.latency_ms
+        info.workspace_path = result.workspace_path
+        info.route = list(result.route) if result.route else []
+        if result.error:
+            info.error_detail = result.error
+
+    def _task_failed(
+        self,
+        request_id: str,
+        error: str,
+        detail: Optional[str] = None,
+    ) -> None:
+        """任务异常结束时调用。"""
+        info = self._tasks.get(request_id)
+        if info is None:
+            return
+        info.state = STATE_FAILED
+        info.finished_at = time.time()
+        info.error = error
+        info.error_detail = detail
+
+    # ─────────────────────────────────────────────────────────────────
+    # SSE 专用：workspace.json 文件轮询
+    # ─────────────────────────────────────────────────────────────────
+
+    async def watch_workspace(
+        self,
+        request_id: str,
+        cancel_event: asyncio.Event,
+    ) -> AsyncGenerator[Dict[str, Any], None]:
+        """监视 runs/{id}/workspace.json，yield 单条事件。
+
+        事件类型：
+        - {"type": "status", "value": <state>, "request_id": <id>}
+        - {"type": "workspace", "version": <n>, "workspace": <dict>}
+        - {"type": "error", "detail": <str>}
+
+        当任务进入 done/failed 状态或 cancel_event 被 set 时停止。
+        """
+        ws_path = RUNS_DIR / request_id / "workspace.json"
+        seen_mtime: float = 0.0
+        seen_size: int = 0
+
+        while not cancel_event.is_set():
+            info = self.get(request_id)
+            # 检查终态
+            if info and info.state in (STATE_DONE, STATE_FAILED):
+                if info.state == STATE_FAILED:
+                    yield {"type": "error", "detail": info.error or "任务失败"}
+                # result 由 /runs/{id}/status 提供，此处只推送终态 status
+                yield {"type": "status", "value": info.state, "request_id": request_id}
+                break
+
+            # 读文件变化（mtime + size 双检）
+            if ws_path.exists():
+                try:
+                    stat = ws_path.stat()
+                    if stat.st_mtime != seen_mtime or stat.st_size != seen_size:
+                        raw = ws_path.read_text(encoding="utf-8")
+                        data = json.loads(raw)
+                        version = int(data.get("meta", {}).get("round", 0))
+                        seen_mtime = stat.st_mtime
+                        seen_size = stat.st_size
+                        yield {
+                            "type": "workspace",
+                            "version": version,
+                            "state": info.state if info else STATE_RUNNING,
+                            "request_id": request_id,
+                            "workspace": data,
+                        }
+                except (json.JSONDecodeError, OSError):
+                    pass  # 文件正在写入，忽略
+
+            await asyncio.sleep(self._poll_interval)
+
+        # 连接关闭前最后推一次终态
+        info = self.get(request_id)
+        if info:
+            yield {"type": "status", "value": info.state, "request_id": request_id}
+
+    # ─────────────────────────────────────────────────────────────────
+    # 内存清理
+    # ─────────────────────────────────────────────────────────────────
+
+    def _cleanup_aged(self) -> None:
+        """删除超过 max_age_seconds 的已完成任务引用。"""
+        now = time.time()
+        to_remove = [
+            rid for rid, info in self._tasks.items()
+            if info.state in (STATE_DONE, STATE_FAILED)
+            and (now - info.finished_at) > self.max_age_seconds
+        ]
+        for rid in to_remove:
+            self._tasks.pop(rid, None)
+
+    def cancel(self, request_id: str) -> bool:
+        """取消运行中的任务。返回 True 找到并取消，False 未找到。"""
+        task = self._asyncio_tasks.get(request_id)
+        if task is None:
+            return False
+        task.cancel()
+        info = self.get(request_id)
+        if info:
+            info.state = STATE_FAILED
+            info.finished_at = time.time()
+            info.error = "cancelled_by_user"
+        return True
+
+
+# ─────────────────────────────────────────────────────────────────────────────
+# 全局单例（gateway 进程内共享）
+# ─────────────────────────────────────────────────────────────────────────────
+_store: Optional[JobStore] = None
+
+
+def get_job_store() -> JobStore:
+    global _store
+    if _store is None:
+        _store = JobStore(max_age_seconds=3600.0, max_running=50)
+    return _store
+
+
+def reset_job_store() -> None:
+    """测试用：重置全局注册表。"""
+    global _store
+    _store = None

gateway/llama_manager.py

@@ -0,0 +1,453 @@
+"""llama-server 进程管理 & 模型下载。
+
+职责：
+- 启动/停止本地 llama-server 子进程（Windows 兼容）
+- 探测已有 .gguf 模型文件
+- 从 HuggingFace URL 下载模型（支持 huggingface.co 路径别名）
+- 下载进度可通过 SSE /llama/download/stream 订阅
+
+用法：
+  from gateway.llama_manager import get_llama_manager
+  lm = get_llama_manager()
+  await lm.start(model="models/qwen3.5-4b-q4_k_m.gguf")
+  await lm.stop()
+"""
+from __future__ import annotations
+
+import asyncio
+import json
+import os
+import re
+import subprocess
+import sys
+import threading
+import time
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Any, AsyncGenerator, Optional
+
+# ─────────────────────────────────────────────────────────────────────────────
+# 路径配置（与 config.yaml runtime.llama_server 段保持一致）
+# ─────────────────────────────────────────────────────────────────────────────
+_ROOT = Path(__file__).resolve().parent.parent          # E:\projectAIpopular
+BIN_DIR = _ROOT / "bin"
+MODELS_DIR = _ROOT / "models"
+PID_FILE = _ROOT / "data" / "llama-server.pid"
+LOG_FILE = _ROOT / "data" / "llama-server.log"
+
+# 确保目录存在
+BIN_DIR.mkdir(parents=True, exist_ok=True)
+MODELS_DIR.mkdir(parents=True, exist_ok=True)
+PID_FILE.parent.mkdir(parents=True, exist_ok=True)
+
+
+# ─────────────────────────────────────────────────────────────────────────────
+# 数据模型
+# ─────────────────────────────────────────────────────────────────────────────
+
+@dataclass
+class LlamaServerStatus:
+    running: bool
+    pid: Optional[int] = None
+    model: Optional[str] = None
+    port: Optional[int] = None
+    base_url: Optional[str] = None
+    started_at: Optional[float] = None
+    error: Optional[str] = None
+
+
+@dataclass
+class DownloadProgress:
+    url: str
+    dest: str
+    total_bytes: Optional[int] = None
+    downloaded_bytes: int = 0
+    progress_pct: float = 0.0
+    speed: str = ""
+    eta: str = ""
+    done: bool = False
+    error: Optional[str] = None
+
+
+# ─────────────────────────────────────────────────────────────────────────────
+# llama_manager 单例
+# ─────────────────────────────────────────────────────────────────────────────
+
+class LlamaManager:
+    _instance: Optional["LlamaManager"] = None
+
+    def __init__(self):
+        self._proc: Optional[subprocess.Popen] = None
+        self._pid: Optional[int] = None
+        self._model: Optional[str] = None
+        self._port: Optional[int] = None
+        self._started_at: Optional[float] = None
+        self._downloading: dict[str, DownloadProgress] = {}   # url -> progress
+        self._dl_lock = threading.Lock()
+        # 加载已有进程
+        self._load_pid()
+
+    # ── 进程持久化 ─────────────────────────────────────────────────────────
+
+    def _load_pid(self) -> None:
+        """从 pid 文件恢复进程引用（进程仍在运行时）。"""
+        if not PID_FILE.exists():
+            return
+        try:
+            pid = int(PID_FILE.read_text().strip())
+            os.kill(pid, 0)   # 检查进程是否存活
+            # 进程还在，尝试接管（通过 cmdline 判断是否是 llama-server）
+            self._pid = pid
+            self._proc = self._attach_to_process(pid)
+        except (ValueError, FileNotFoundError, OSError):
+            PID_FILE.unlink(missing_ok=True)
+
+    def _attach_to_process(self, pid: int) -> Optional[subprocess.Popen]:
+        """通过 pid 重新关联到 Popen（仅作状态恢复，不拥有 stdout）。"""
+        try:
+            return subprocess.Popen(
+                [sys.executable, "-c",
+                 f"import os; os.kill({pid}, 0)"],  # 存活检查
+                stdout=subprocess.DEVNULL,
+                stderr=subprocess.DEVNULL,
+            )
+        except Exception:
+            return None
+
+    def _write_pid(self, pid: int) -> None:
+        PID_FILE.write_text(str(pid), encoding="utf-8")
+
+    def _clear_pid(self) -> None:
+        PID_FILE.unlink(missing_ok=True)
+
+    # ── 进程管理 ─────────────────────────────────────────────────────────
+
+    def find_binary(self) -> Optional[Path]:
+        """查找 llama-server 可执行文件。"""
+        candidates = [
+            BIN_DIR / "llama-server.exe",
+            BIN_DIR / "llama-server",
+            _ROOT / "llama-server.exe",
+            _ROOT / "llama-server",
+        ]
+        for p in candidates:
+            if p.exists():
+                return p
+        # PATH 中查找
+        import shutil
+        found = shutil.which("llama-server") or shutil.which("llama-server.exe")
+        if found:
+            return Path(found)
+        return None
+
+    def status(self) -> LlamaServerStatus:
+        """返回当前服务状态。"""
+        if self._proc is None or self._pid is None:
+            return LlamaServerStatus(running=False)
+        try:
+            # 检查进程是否存活
+            os.kill(self._pid, 0)
+        except OSError:
+            # 进程已死
+            self._proc = None
+            self._pid = None
+            self._model = None
+            self._port = None
+            self._started_at = None
+            self._clear_pid()
+            return LlamaServerStatus(running=False)
+        return LlamaServerStatus(
+            running=True,
+            pid=self._pid,
+            model=self._model,
+            port=self._port,
+            base_url=f"http://127.0.0.1:{self._port}/v1",
+            started_at=self._started_at,
+        )
+
+    async def start(
+        self,
+        model: str,
+        port: int = 8901,
+        ngl: int = 99,
+        ctx: int = 4096,
+        extra_args: Optional[list] = None,
+    ) -> LlamaServerStatus:
+        """启动 llama-server，阻塞直到监听就绪或超时。"""
+        if self.status().running:
+            s = self.status()
+            if s.model == model and s.port == port:
+                return s   # 已是同一模型，无需重启
+            await self.stop()
+
+        binary = self.find_binary()
+        if binary is None:
+            return LlamaServerStatus(
+                running=False,
+                error="未找到 llama-server 可执行文件。"
+                        "请将 llama-server.exe 放入 bin/ 目录，"
+                        "或从 https://github.com/ggerganov/llama.cpp/releases 下载。",
+            )
+
+        model_path = Path(model)
+        if not model_path.is_absolute():
+            model_path = MODELS_DIR / model
+        if not model_path.exists():
+            return LlamaServerStatus(
+                running=False,
+                error=f"模型文件不存在：{model_path}。"
+                        "请先下载模型，或在设置页填写 HuggingFace URL 下载。",
+            )
+
+        args = [
+            str(binary),
+            "-m", str(model_path),
+            "-c", str(ctx),
+            "-ngl", str(ngl),
+            "--port", str(port),
+            "--host", "127.0.0.1",
+        ]
+        if extra_args:
+            args.extend(extra_args)
+
+        LOG_FILE.parent.mkdir(parents=True, exist_ok=True)
+        log_f = open(LOG_FILE, "w", encoding="utf-8", buffering=1)
+
+        try:
+            self._proc = subprocess.Popen(
+                args,
+                stdout=log_f,
+                stderr=subprocess.STDOUT,
+                cwd=str(_ROOT),
+                creationflags=subprocess.CREATE_NEW_PROCESS_GROUP if sys.platform == "win32" else 0,
+            )
+        except OSError as e:
+            log_f.close()
+            return LlamaServerStatus(running=False, error=f"启动失败：{e}")
+
+        self._pid = self._proc.pid
+        self._model = str(model_path)
+        self._port = port
+        self._started_at = time.time()
+        self._write_pid(self._pid)
+
+        # 等待服务就绪
+        ok = await self._wait_until_ready(port, timeout=30)
+        if not ok:
+            await self.stop()
+            return LlamaServerStatus(
+                running=False,
+                error=f"llama-server 启动后 {port} 端口在 30 秒内未响应",
+            )
+
+        return self.status()
+
+    async def stop(self) -> None:
+        """优雅停止 llama-server。"""
+        if self._pid is None:
+            self._proc = None
+            return
+
+        try:
+            if sys.platform == "win32":
+                # Windows: CTRL_BREAK_EVENT 或 taskkill
+                subprocess.run(
+                    ["taskkill", "/PID", str(self._pid), "/T", "/F"],
+                    capture_output=True,
+                    timeout=10,
+                )
+            else:
+                os.kill(self._pid, 15)   # SIGTERM
+                time.sleep(1)
+                try:
+                    os.kill(self._pid, 0)
+                    os.kill(self._pid, 9)
+                except OSError:
+                    pass
+        except Exception:
+            pass
+        finally:
+            self._proc = None
+            self._pid = None
+            self._model = None
+            self._port = None
+            self._started_at = None
+            self._clear_pid()
+
+    async def _wait_until_ready(self, port: int, timeout: float = 30) -> bool:
+        """轮询检查端口是否开始监听。"""
+        import httpx
+        url = f"http://127.0.0.1:{port}/v1/models"
+        deadline = time.time() + timeout
+        async with httpx.AsyncClient(timeout=5.0) as client:
+            while time.time() < deadline:
+                try:
+                    r = await client.get(url)
+                    if r.status_code < 500:
+                        return True
+                except (httpx.ConnectError, httpx.ReadTimeout, OSError):
+                    pass
+                await asyncio.sleep(0.5)
+        return False
+
+    # ── 模型列表 ─────────────────────────────────────────────────────────
+
+    def list_local_models(self) -> list[dict[str, str]]:
+        """列出 models/ 目录下所有 .gguf 文件。"""
+        models = []
+        for p in MODELS_DIR.glob("*.gguf"):
+            size_mb = p.stat().st_size // (1024 * 1024)
+            models.append({
+                "id": p.name,
+                "name": p.name,
+                "size_mb": size_mb,
+                "path": str(p),
+            })
+        return sorted(models, key=lambda m: m["name"])
+
+    # ── 模型下载 ─────────────────────────────────────────────────────────
+
+    async def download_model(
+        self,
+        url: str,
+        dest: Optional[str] = None,
+    ) -> DownloadProgress:
+        """从 HuggingFace 或直链下载 .gguf 模型文件。
+
+        HuggingFace 路径别名：用户输入 "Qwen/Qwen3-4B-GGUF/Qwen3-4B-Q4_K_M.gguf"
+        自动转换为 "https://huggingface.co/<repo>/resolve/main/<file>"
+
+        支持断点续传（Content-Range）。
+
+        返回 DownloadProgress 对象（含当前进度），进度通过 get_download_progress() 查询。
+        """
+        import httpx
+
+        # 路径别名转换
+        if not url.startswith("http"):
+            url = f"https://huggingface.co/{url}/resolve/main"
+
+        # 解析文件名
+        filename = url.rstrip("/").split("/")[-1]
+        if not filename.endswith(".gguf"):
+            filename += ".gguf"
+
+        if dest:
+            dest_path = Path(dest)
+        else:
+            dest_path = MODELS_DIR / filename
+
+        # 构造 HTTP 头
+        headers = {}
+        resume_bytes = 0
+        if dest_path.exists():
+            resume_bytes = dest_path.stat().st_size
+            headers["Range"] = f"bytes={resume_bytes}-"
+
+        # 获取文件大小
+        total_bytes: Optional[int] = None
+        try:
+            async with httpx.AsyncClient(timeout=httpx.Timeout(30.0, connect=10.0, read=60.0, write=30.0, pool=10.0), follow_redirects=True) as client:
+                head = await client.head(url, headers={"Range": "bytes=0-0"})
+                total_raw = head.headers.get("Content-Length")
+                if total_raw:
+                    total_bytes = int(total_raw)
+                # Content-Range 响应时 total_bytes 在 Content-Range 头里
+                cr = head.headers.get("Content-Range", "")
+                m = re.search(r"/(\d+)", cr)
+                if m:
+                    total_bytes = int(m.group(1))
+        except Exception:
+            pass
+
+        prog = DownloadProgress(
+            url=url,
+            dest=str(dest_path),
+            total_bytes=total_bytes,
+            downloaded_bytes=resume_bytes,
+        )
+        with self._dl_lock:
+            self._downloading[url] = prog
+
+        try:
+            mode = "ab" if resume_bytes > 0 else "wb"
+            async with httpx.AsyncClient(
+                timeout=httpx.Timeout(300.0, connect=10.0, read=300.0, write=30.0, pool=10.0),
+                follow_redirects=True,
+            ) as client:
+                t0 = time.time()
+                last_bytes = resume_bytes
+                async with client.stream("GET", url, headers=headers) as resp:
+                    if resp.status_code not in (200, 206):
+                        raise RuntimeError(f"HTTP {resp.status_code}")
+                    with open(dest_path, mode) as f:
+                        async for chunk in resp.aiter_bytes(chunk_size=8192):
+                            f.write(chunk)
+                            prog.downloaded_bytes += len(chunk)
+
+                            # 速度 & ETA
+                            elapsed = time.time() - t0
+                            if elapsed > 0.5:
+                                speed_bps = (prog.downloaded_bytes - last_bytes) / elapsed
+                                speed_str = _format_speed(speed_bps)
+                                if prog.total_bytes and speed_bps > 0:
+                                    remain = prog.total_bytes - prog.downloaded_bytes
+                                    eta_s = remain / speed_bps
+                                    prog.eta = _format_eta(eta_s)
+                                else:
+                                    prog.eta = ""
+                                prog.speed = speed_str
+                                last_bytes = prog.downloaded_bytes
+                                t0 = time.time()
+
+                            if prog.total_bytes:
+                                prog.progress_pct = min(prog.downloaded_bytes / prog.total_bytes * 100, 100)
+        except Exception as e:
+            prog.error = str(e)
+        finally:
+            prog.done = True
+            with self._dl_lock:
+                self._downloading[url] = prog
+
+        return prog
+
+    def get_download_progress(self, url: str) -> Optional[DownloadProgress]:
+        """查询下载进度。"""
+        with self._dl_lock:
+            return self._downloading.get(url)
+
+    def list_downloads(self) -> list[DownloadProgress]:
+        """列出所有活跃下载。"""
+        with self._dl_lock:
+            return list(self._downloading.values())
+
+
+def _format_speed(bps: float) -> str:
+    if bps >= 1e9:
+        return f"{bps/1e9:.1f} GB/s"
+    if bps >= 1e6:
+        return f"{bps/1e6:.1f} MB/s"
+    if bps >= 1e3:
+        return f"{bps/1e3:.1f} KB/s"
+    return f"{bps:.0f} B/s"
+
+
+def _format_eta(seconds: float) -> str:
+    if seconds < 60:
+        return f"{seconds:.0f}s"
+    if seconds < 3600:
+        return f"{seconds/60:.0f}m"
+    return f"{seconds/3600:.1f}h"
+
+
+# ─────────────────────────────────────────────────────────────────────────────
+# 全局单例
+# ─────────────────────────────────────────────────────────────────────────────
+_lm: Optional[LlamaManager] = None
+
+
+def get_llama_manager() -> LlamaManager:
+    global _lm
+    if _lm is None:
+        _lm = LlamaManager()
+    return _lm

gateway/static/assets/MetricsView-CE-HhrJg.css

@@ -0,0 +1 @@
+.metrics-view[data-v-51f92136]{height:100%;padding:20px 24px;overflow-y:auto}.metrics-header[data-v-51f92136]{justify-content:space-between;align-items:center;margin-bottom:20px;display:flex}.metrics-header h2[data-v-51f92136]{margin:0;font-size:20px}.refresh[data-v-51f92136]{cursor:pointer;background:#fff;border:1px solid #d1d5db;border-radius:6px;padding:6px 14px}.loading[data-v-51f92136],.error[data-v-51f92136]{text-align:center;color:#9ca3af;padding:40px}.error[data-v-51f92136]{color:#dc2626}.card-grid[data-v-51f92136]{grid-template-columns:repeat(auto-fill,minmax(280px,1fr));gap:16px;margin-bottom:24px;display:grid}.metric-card[data-v-51f92136]{background:#fff;border:1px solid #e5e7eb;border-radius:10px;padding:16px}.metric-card.highlight[data-v-51f92136]{background:#eff6ff;border-color:#2563eb}.metric-card h3[data-v-51f92136]{color:#374151;margin:0 0 12px;font-size:14px}.kv-list[data-v-51f92136]{grid-template-columns:1fr 1fr;gap:6px 12px;font-size:13px;display:grid}.kv-list span[data-v-51f92136]{color:#6b7280}.kv-list b[data-v-51f92136]{color:#111;text-align:right}.review-card[data-v-51f92136]{grid-column:span 2}.review-stats[data-v-51f92136]{gap:24px;margin-bottom:12px;display:flex}.stat-item[data-v-51f92136]{flex-direction:column;align-items:center;display:flex}.stat-num[data-v-51f92136]{color:#2563eb;font-size:28px;font-weight:700}.stat-label[data-v-51f92136]{color:#6b7280;font-size:12px}.progress-wrap[data-v-51f92136]{background:#e5e7eb;border-radius:99px;height:8px;margin-bottom:6px;overflow:hidden}.reviewed-bar[data-v-51f92136]{background:#16a34a;height:100%;transition:width .5s}.review-rate[data-v-51f92136]{color:#6b7280;margin:0;font-size:13px}.raw-json[data-v-51f92136]{background:#f9fafb;border:1px solid #e5e7eb;border-radius:8px}.raw-json summary[data-v-51f92136]{cursor:pointer;color:#6b7280;padding:10px 14px;font-size:13px}.raw-json pre[data-v-51f92136]{white-space:pre-wrap;border-top:1px solid #e5e7eb;margin:0;padding:10px 14px;font-size:12px}

gateway/static/assets/MetricsView-rQCk8WKw.js

@@ -0,0 +1 @@
+import{A as e,C as t,D as n,F as r,O as i,R as a,T as o,j as s,o as c,t as l,w as u,x as d,z as f}from"./index-BnlAnYlB.js";var p={class:`metrics-view`},m={key:0,class:`loading`},h={key:1,class:`error`},g={class:`card-grid`},_={class:`metric-card`},v={class:`kv-list`},y={class:`metric-card`},b={class:`kv-list`},x={key:0,class:`metric-card highlight`},S={class:`kv-list`},C={key:1,class:`metric-card review-card`},w={class:`review-stats`},T={class:`stat-item`},E={class:`stat-num`},D={class:`stat-item`},O={class:`stat-num`},k={key:0,class:`progress-wrap`},A={class:`review-rate`},j={class:`raw-json`},M=l(n({__name:`MetricsView`,setup(n){let l=r(null),M=r(!1),N=r(``);async function P(){M.value=!0,N.value=``;try{l.value=await c()}catch(e){N.value=e instanceof Error?e.message:`指标加载失败，请检查后端服务`}finally{M.value=!1}}return i(P),(n,r)=>(e(),o(`div`,p,[t(`header`,{class:`metrics-header`},[r[0]||=t(`h2`,null,`系统指标`,-1),t(`button`,{class:`refresh`,onClick:P},`🔄 刷新`)]),M.value?(e(),o(`div`,m,`加载中…`)):N.value?(e(),o(`div`,h,f(N.value),1)):l.value?(e(),o(d,{key:2},[t(`div`,g,[t(`div`,_,[r[1]||=t(`h3`,null,`路由器（v1）`,-1),t(`div`,v,[(e(!0),o(d,null,s(l.value.router,(n,r)=>(e(),o(d,{key:r},[t(`span`,null,f(r),1),t(`b`,null,f(n),1)],64))),128))])]),t(`div`,y,[r[2]||=t(`h3`,null,`缓存`,-1),t(`div`,b,[(e(!0),o(d,null,s(l.value.cache,(n,r)=>(e(),o(d,{key:r},[t(`span`,null,f(r),1),t(`b`,null,f(n),1)],64))),128))])]),l.value.v2?(e(),o(`div`,x,[r[3]||=t(`h3`,null,`协作管线（v2）`,-1),t(`div`,S,[(e(!0),o(d,null,s(l.value.v2,(n,r)=>(e(),o(d,{key:r},[t(`span`,null,f(r),1),t(`b`,null,f(n),1)],64))),128))])])):u(``,!0),l.value.review?(e(),o(`div`,C,[r[6]||=t(`h3`,null,`人工检验`,-1),t(`div`,w,[t(`div`,T,[t(`span`,E,f(l.value.review.pending),1),r[4]||=t(`span`,{class:`stat-label`},`待审核`,-1)]),t(`div`,D,[t(`span`,O,f(l.value.review.total),1),r[5]||=t(`span`,{class:`stat-label`},`总提交`,-1)])]),l.value.review.total>0?(e(),o(`div`,k,[t(`div`,{class:`reviewed-bar`,style:a({width:`${(l.value.review.total-l.value.review.pending)/l.value.review.total*100}%`})},null,4)])):u(``,!0),t(`p`,A,` 通过率： `+f(((l.value.review.total-l.value.review.pending)/l.value.review.total*100).toFixed(1))+`% `,1)])):u(``,!0)]),t(`details`,j,[r[7]||=t(`summary`,null,`原始 JSON`,-1),t(`pre`,null,f(JSON.stringify(l.value,null,2)),1)])],64)):u(``,!0)]))}}),[[`__scopeId`,`data-v-51f92136`]]);export{M as default};

gateway/static/assets/ReviewView-CaOmFJEu.css

@@ -0,0 +1 @@
+.review-view[data-v-19c16eff]{height:100%;padding:20px 24px;overflow-y:auto}.review-header[data-v-19c16eff]{justify-content:space-between;align-items:center;margin-bottom:20px;display:flex}.review-header h2[data-v-19c16eff]{margin:0;font-size:20px}.controls[data-v-19c16eff]{gap:8px;display:flex}button[data-v-19c16eff]{cursor:pointer;background:#fff;border:1px solid #d1d5db;border-radius:6px;padding:6px 14px;font-size:13px}button.active[data-v-19c16eff]{color:#fff;background:#2563eb;border-color:#2563eb}.refresh-btn[data-v-19c16eff]{margin-left:auto}.loading[data-v-19c16eff],.error[data-v-19c16eff],.empty[data-v-19c16eff]{text-align:center;color:#9ca3af;padding:40px}.error[data-v-19c16eff]{color:#dc2626}.queue-list[data-v-19c16eff]{flex-direction:column;gap:16px;display:flex}.review-card[data-v-19c16eff]{background:#fff;border:1px solid #e5e7eb;border-radius:10px;padding:16px}.card-header[data-v-19c16eff]{flex-wrap:wrap;align-items:center;gap:10px;margin-bottom:10px;display:flex}.card-id[data-v-19c16eff]{color:#6b7280;font-family:monospace;font-size:12px}.verdict-badge[data-v-19c16eff]{border-radius:99px;padding:2px 8px;font-size:12px;font-weight:600}.verdict-badge.pending[data-v-19c16eff]{color:#92400e;background:#fef3c7}.verdict-badge.approved[data-v-19c16eff]{color:#16a34a;background:#dcfce7}.verdict-badge.rejected[data-v-19c16eff]{color:#dc2626;background:#fee2e2}.tags[data-v-19c16eff]{gap:4px;display:flex}.tag[data-v-19c16eff]{color:#3730a3;background:#e0e7ff;border-radius:4px;padding:1px 6px;font-size:11px}.date[data-v-19c16eff]{color:#9ca3af;margin-left:auto;font-size:11px}.query-block[data-v-19c16eff],.response-block[data-v-19c16eff]{margin-bottom:8px;font-size:13px;line-height:1.6}.query-block pre[data-v-19c16eff],.response-block pre[data-v-19c16eff]{white-space:pre-wrap;background:#f9fafb;border:1px solid #e5e7eb;border-radius:4px;margin:4px 0 0;padding:6px 10px;font-size:13px}.actions[data-v-19c16eff]{flex-direction:column;gap:8px;margin-top:10px;display:flex}textarea[data-v-19c16eff]{resize:vertical;box-sizing:border-box;border:1px solid #d1d5db;border-radius:6px;width:100%;padding:8px 10px;font-family:inherit;font-size:13px}.btn-row[data-v-19c16eff]{gap:8px;display:flex}.approve[data-v-19c16eff]{color:#16a34a;background:#dcfce7;border-color:#86efac}.reject[data-v-19c16eff]{color:#dc2626;background:#fee2e2;border-color:#fca5a5}.correction[data-v-19c16eff]{background:#fffbeb;border:1px solid #fcd34d;border-radius:4px;margin-top:8px;padding:6px 10px;font-size:13px}

gateway/static/assets/ReviewView-nv7jUZLH.js

@@ -0,0 +1 @@
+import{A as e,C as t,D as n,E as r,F as i,L as a,N as o,O as s,S as c,T as l,b as u,j as d,l as f,m as p,t as m,w as h,x as g,z as _}from"./index-BnlAnYlB.js";var v={class:`review-view`},y={class:`review-header`},b={class:`controls`},x={key:0,class:`loading`},S={key:1,class:`error`},C={key:2,class:`queue-list`},w={key:0,class:`empty`},T={class:`card-header`},E={class:`card-id`},D={class:`tags`},O={class:`date`},k={class:`query-block`},A={class:`response-block`},j={key:0,class:`actions`},M=[`onUpdate:modelValue`],N={class:`btn-row`},P=[`onClick`],F=[`onClick`],I={key:1,class:`correction`},L=m(n({__name:`ReviewView`,setup(n){let m=i([]),L=i(!1),R=i(``),z=i(`pending`),B=i({}),V=c(()=>z.value===`all`?m.value:m.value.filter(e=>e.verdict===z.value));async function H(){L.value=!0,R.value=``;try{m.value=await f()}catch(e){R.value=e instanceof Error?e.message:String(e)}finally{L.value=!1}}async function U(e,t){try{await p(e,t,B.value[e]||void 0),await H()}catch(e){R.value=e instanceof Error?e.message:String(e)}}return s(H),(n,i)=>(e(),l(`div`,v,[t(`header`,y,[i[4]||=t(`h2`,null,`人工检验队列`,-1),t(`div`,b,[t(`button`,{class:a({active:z.value===`all`}),onClick:i[0]||=e=>z.value=`all`},`全部`,2),t(`button`,{class:a({active:z.value===`pending`}),onClick:i[1]||=e=>z.value=`pending`},`待审核`,2),t(`button`,{class:a({active:z.value===`approved`}),onClick:i[2]||=e=>z.value=`approved`},`已通过`,2),t(`button`,{class:a({active:z.value===`rejected`}),onClick:i[3]||=e=>z.value=`rejected`},`已拒绝`,2),t(`button`,{class:`refresh-btn`,onClick:H},`🔄 刷新`)])]),L.value?(e(),l(`div`,x,`加载中…`)):R.value?(e(),l(`div`,S,_(R.value),1)):(e(),l(`div`,C,[V.value.length?h(``,!0):(e(),l(`div`,w,`队列为空。`)),(e(!0),l(g,null,d(V.value,n=>(e(),l(`div`,{key:n.id,class:`review-card`},[t(`div`,T,[t(`span`,E,`#`+_(n.id),1),t(`span`,{class:a([`verdict-badge`,n.verdict])},_(n.verdict),3),t(`span`,D,[(e(!0),l(g,null,d(n.tags,t=>(e(),l(`span`,{key:t,class:`tag`},_(t),1))),128))]),t(`span`,O,_(n.created_at),1)]),t(`div`,k,[i[5]||=t(`strong`,null,`Query：`,-1),r(_(n.query),1)]),t(`div`,A,[i[6]||=t(`strong`,null,`Response：`,-1),t(`pre`,null,_(n.response),1)]),n.verdict===`pending`?(e(),l(`div`,j,[o(t(`textarea`,{"onUpdate:modelValue":e=>B.value[n.id]=e,placeholder:`修正意见（可选）`,rows:`2`},null,8,M),[[u,B.value[n.id]]]),t(`div`,N,[t(`button`,{class:`approve`,onClick:e=>U(n.id,`approved`)},`✅ 通过`,8,P),t(`button`,{class:`reject`,onClick:e=>U(n.id,`rejected`)},`❌ 拒绝`,8,F)])])):n.correction?(e(),l(`div`,I,[i[7]||=t(`strong`,null,`修正：`,-1),r(_(n.correction),1)])):h(``,!0)]))),128))]))]))}}),[[`__scopeId`,`data-v-19c16eff`]]);export{L as default};

gateway/static/assets/index-G8RWfJ2Y.css

@@ -0,0 +1 @@
+*{box-sizing:border-box;margin:0;padding:0}html,body,#app{color:#111;background:#fff;height:100%;font-family:-apple-system,BlinkMacSystemFont,Segoe UI,Roboto,sans-serif;font-size:14px}.app{flex-direction:column;height:100%;display:flex}.topbar{background:#fff;border-bottom:1px solid #e5e7eb;flex-shrink:0;align-items:center;gap:24px;height:52px;padding:0 24px;display:flex}.brand{color:#1e40af;white-space:nowrap;font-size:16px;font-weight:700}.nav-links{gap:4px;display:flex}.nav-links a{color:#374151;border-radius:6px;padding:6px 14px;font-size:14px;font-weight:500;text-decoration:none;transition:background .15s}.nav-links a:hover{background:#f3f4f6}.nav-links a.router-link-active{color:#2563eb;background:#eff6ff}.content{flex-direction:column;flex:1;min-height:0;display:flex;overflow:hidden}.chat-view[data-v-6692005c]{gap:0;height:100%;display:flex}.sidebar[data-v-6692005c]{background:#f9fafb;border-right:1px solid #e5e7eb;flex-direction:column;width:240px;display:flex;overflow:hidden}.sidebar h3[data-v-6692005c]{color:#6b7280;border-bottom:1px solid #e5e7eb;margin:0;padding:12px 16px;font-size:14px}.session-list[data-v-6692005c]{flex:1;margin:0;padding:8px;list-style:none;overflow-y:auto}.session-item[data-v-6692005c]{cursor:pointer;border-radius:6px;flex-direction:column;gap:2px;margin-bottom:4px;padding:8px 10px;font-size:13px;display:flex}.session-item[data-v-6692005c]:hover{background:#e5e7eb}.session-item.active[data-v-6692005c]{background:#dbeafe}.s-query[data-v-6692005c]{color:#111}.s-status[data-v-6692005c]{color:#9ca3af;font-size:11px}.s-status.done[data-v-6692005c]{color:#16a34a}.s-status.failed[data-v-6692005c]{color:#dc2626}.s-status.running[data-v-6692005c]{color:#2563eb}.main[data-v-6692005c]{flex-direction:column;flex:1;gap:12px;padding:20px 24px;display:flex;overflow:hidden}.empty[data-v-6692005c]{color:#9ca3af;flex-direction:column;flex:1;justify-content:center;align-items:center;gap:8px;display:flex}.hint[data-v-6692005c]{font-size:13px}.user-msg[data-v-6692005c],.assistant-msg[data-v-6692005c]{background:#f3f4f6;border-radius:8px;gap:12px;padding:12px 16px;font-size:14px;line-height:1.6;display:flex}.assistant-msg[data-v-6692005c]{background:#eff6ff}.role-label[data-v-6692005c]{color:#6b7280;min-width:32px;font-size:12px;font-weight:700}.user-msg pre[data-v-6692005c],.assistant-msg pre[data-v-6692005c]{white-space:pre-wrap;margin:0;font-family:inherit}.status-bar[data-v-6692005c]{flex-wrap:wrap;align-items:center;gap:12px;display:flex}.badge[data-v-6692005c]{border-radius:99px;padding:3px 10px;font-size:12px;font-weight:600}.badge.pending[data-v-6692005c]{color:#6b7280;background:#f3f4f6}.badge.running[data-v-6692005c]{color:#2563eb;background:#dbeafe}.badge.done[data-v-6692005c]{color:#16a34a;background:#dcfce7}.badge.failed[data-v-6692005c]{color:#dc2626;background:#fee2e2}.route-path[data-v-6692005c]{color:#9ca3af;font-size:12px}.ws-meta[data-v-6692005c]{color:#9ca3af;gap:16px;font-size:12px;display:flex}.ws-preview[data-v-6692005c]{background:#fafafa;border:1px solid #e5e7eb;border-radius:8px;padding:12px;font-size:13px}.brief-block[data-v-6692005c]{margin-bottom:8px}.tags[data-v-6692005c]{color:#6b7280;margin-left:8px;font-size:12px}.plan-list[data-v-6692005c],.progress-list[data-v-6692005c]{margin:4px 0 0;padding:0;list-style:none}.plan-list li[data-v-6692005c],.progress-list li[data-v-6692005c]{gap:6px;padding:2px 0;display:flex}.step-id[data-v-6692005c]{color:#6b7280;min-width:48px;font-family:monospace}.deps[data-v-6692005c]{color:#9ca3af;font-size:12px}.done[data-v-6692005c]{color:#16a34a}.pending[data-v-6692005c]{color:#9ca3af}.running[data-v-6692005c]{color:#2563eb}.error-msg[data-v-6692005c]{color:#dc2626;background:#fee2e2;border-radius:6px;padding:8px 12px;font-size:13px}.input-bar[data-v-6692005c]{border-top:1px solid #e5e7eb;gap:8px;margin-top:auto;padding-top:12px;display:flex}.input-bar input[data-v-6692005c]{border:1px solid #d1d5db;border-radius:8px;outline:none;flex:1;padding:10px 14px;font-size:14px}.input-bar input[data-v-6692005c]:focus{border-color:#2563eb}.input-bar button[data-v-6692005c]{color:#fff;cursor:pointer;background:#2563eb;border:none;border-radius:8px;padding:10px 20px;font-size:14px}.input-bar button[data-v-6692005c]:disabled{cursor:not-allowed;background:#9ca3af}

gateway/static/icons.svg

@@ -0,0 +1,24 @@
+<svg xmlns="http://www.w3.org/2000/svg">
+  <symbol id="bluesky-icon" viewBox="0 0 16 17">
+    <g clip-path="url(#bluesky-clip)"><path fill="#08060d" d="M7.75 7.735c-.693-1.348-2.58-3.86-4.334-5.097-1.68-1.187-2.32-.981-2.74-.79C.188 2.065.1 2.812.1 3.251s.241 3.602.398 4.13c.52 1.744 2.367 2.333 4.07 2.145-2.495.37-4.71 1.278-1.805 4.512 3.196 3.309 4.38-.71 4.987-2.746.608 2.036 1.307 5.91 4.93 2.746 2.72-2.746.747-4.143-1.747-4.512 1.702.189 3.55-.4 4.07-2.145.156-.528.397-3.691.397-4.13s-.088-1.186-.575-1.406c-.42-.19-1.06-.395-2.741.79-1.755 1.24-3.64 3.752-4.334 5.099"/></g>
+    <defs><clipPath id="bluesky-clip"><path fill="#fff" d="M.1.85h15.3v15.3H.1z"/></clipPath></defs>
+  </symbol>
+  <symbol id="discord-icon" viewBox="0 0 20 19">
+    <path fill="#08060d" d="M16.224 3.768a14.5 14.5 0 0 0-3.67-1.153c-.158.286-.343.67-.47.976a13.5 13.5 0 0 0-4.067 0c-.128-.306-.317-.69-.476-.976A14.4 14.4 0 0 0 3.868 3.77C1.546 7.28.916 10.703 1.231 14.077a14.7 14.7 0 0 0 4.5 2.306q.545-.748.965-1.587a9.5 9.5 0 0 1-1.518-.74q.191-.14.372-.293c2.927 1.369 6.107 1.369 8.999 0q.183.152.372.294-.723.437-1.52.74.418.838.963 1.588a14.6 14.6 0 0 0 4.504-2.308c.37-3.911-.63-7.302-2.644-10.309m-9.13 8.234c-.878 0-1.599-.82-1.599-1.82 0-.998.705-1.82 1.6-1.82.894 0 1.614.82 1.599 1.82.001 1-.705 1.82-1.6 1.82m5.91 0c-.878 0-1.599-.82-1.599-1.82 0-.998.705-1.82 1.6-1.82.893 0 1.614.82 1.599 1.82 0 1-.706 1.82-1.6 1.82"/>
+  </symbol>
+  <symbol id="documentation-icon" viewBox="0 0 21 20">
+    <path fill="none" stroke="#aa3bff" stroke-linecap="round" stroke-linejoin="round" stroke-width="1.35" d="m15.5 13.333 1.533 1.322c.645.555.967.833.967 1.178s-.322.623-.967 1.179L15.5 18.333m-3.333-5-1.534 1.322c-.644.555-.966.833-.966 1.178s.322.623.966 1.179l1.534 1.321"/>
+    <path fill="none" stroke="#aa3bff" stroke-linecap="round" stroke-linejoin="round" stroke-width="1.35" d="M17.167 10.836v-4.32c0-1.41 0-2.117-.224-2.68-.359-.906-1.118-1.621-2.08-1.96-.599-.21-1.349-.21-2.848-.21-2.623 0-3.935 0-4.983.369-1.684.591-3.013 1.842-3.641 3.428C3 6.449 3 7.684 3 10.154v2.122c0 2.558 0 3.838.706 4.726q.306.383.713.671c.76.536 1.79.64 3.581.66"/>
+    <path fill="none" stroke="#aa3bff" stroke-linecap="round" stroke-linejoin="round" stroke-width="1.35" d="M3 10a2.78 2.78 0 0 1 2.778-2.778c.555 0 1.209.097 1.748-.047.48-.129.854-.503.982-.982.145-.54.048-1.194.048-1.749a2.78 2.78 0 0 1 2.777-2.777"/>
+  </symbol>
+  <symbol id="github-icon" viewBox="0 0 19 19">
+    <path fill="#08060d" fill-rule="evenodd" d="M9.356 1.85C5.05 1.85 1.57 5.356 1.57 9.694a7.84 7.84 0 0 0 5.324 7.44c.387.079.528-.168.528-.376 0-.182-.013-.805-.013-1.454-2.165.467-2.616-.935-2.616-.935-.349-.91-.864-1.143-.864-1.143-.71-.48.051-.48.051-.48.787.051 1.2.805 1.2.805.695 1.194 1.817.857 2.268.649.064-.507.27-.857.49-1.052-1.728-.182-3.545-.857-3.545-3.87 0-.857.31-1.558.8-2.104-.078-.195-.349-1 .077-2.078 0 0 .657-.208 2.14.805a7.5 7.5 0 0 1 1.946-.26c.657 0 1.328.092 1.946.26 1.483-1.013 2.14-.805 2.14-.805.426 1.078.155 1.883.078 2.078.502.546.799 1.247.799 2.104 0 3.013-1.818 3.675-3.558 3.87.284.247.528.714.528 1.454 0 1.052-.012 1.896-.012 2.156 0 .208.142.455.528.377a7.84 7.84 0 0 0 5.324-7.441c.013-4.338-3.48-7.844-7.773-7.844" clip-rule="evenodd"/>
+  </symbol>
+  <symbol id="social-icon" viewBox="0 0 20 20">
+    <path fill="none" stroke="#aa3bff" stroke-linecap="round" stroke-linejoin="round" stroke-width="1.35" d="M12.5 6.667a4.167 4.167 0 1 0-8.334 0 4.167 4.167 0 0 0 8.334 0"/>
+    <path fill="none" stroke="#aa3bff" stroke-linecap="round" stroke-linejoin="round" stroke-width="1.35" d="M2.5 16.667a5.833 5.833 0 0 1 8.75-5.053m3.837.474.513 1.035c.07.144.257.282.414.309l.93.155c.596.1.736.536.307.965l-.723.73a.64.64 0 0 0-.152.531l.207.903c.164.715-.213.991-.84.618l-.872-.52a.63.63 0 0 0-.577 0l-.872.52c-.624.373-1.003.094-.84-.618l.207-.903a.64.64 0 0 0-.152-.532l-.723-.729c-.426-.43-.289-.864.306-.964l.93-.156a.64.64 0 0 0 .412-.31l.513-1.034c.28-.562.735-.562 1.012 0"/>
+  </symbol>
+  <symbol id="x-icon" viewBox="0 0 19 19">
+    <path fill="#08060d" fill-rule="evenodd" d="M1.893 1.98c.052.072 1.245 1.769 2.653 3.77l2.892 4.114c.183.261.333.48.333.486s-.068.089-.152.183l-.522.593-.765.867-3.597 4.087c-.375.426-.734.834-.798.905a1 1 0 0 0-.118.148c0 .01.236.017.664.017h.663l.729-.83c.4-.457.796-.906.879-.999a692 692 0 0 0 1.794-2.038c.034-.037.301-.34.594-.675l.551-.624.345-.392a7 7 0 0 1 .34-.374c.006 0 .93 1.306 2.052 2.903l2.084 2.965.045.063h2.275c1.87 0 2.273-.003 2.266-.021-.008-.02-1.098-1.572-3.894-5.547-2.013-2.862-2.28-3.246-2.273-3.266.008-.019.282-.332 2.085-2.38l2-2.274 1.567-1.782c.022-.028-.016-.03-.65-.03h-.674l-.3.342a871 871 0 0 1-1.782 2.025c-.067.075-.405.458-.75.852a100 100 0 0 1-.803.91c-.148.172-.299.344-.99 1.127-.304.343-.32.358-.345.327-.015-.019-.904-1.282-1.976-2.808L6.365 1.85H1.8zm1.782.91 8.078 11.294c.772 1.08 1.413 1.973 1.425 1.984.016.017.241.02 1.05.017l1.03-.004-2.694-3.766L7.796 5.75 5.722 2.852l-1.039-.004-1.039-.004z" clip-rule="evenodd"/>
+  </symbol>
+</svg>

gateway/static/index.html

@@ -1,224 +1,14 @@
-<!DOCTYPE html>
+<!doctype html>
-<html lang="zh-CN">
+<html lang="en">
-<head>
+  <head>
-<meta charset="UTF-8">
+    <meta charset="UTF-8" />
-<meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <link rel="icon" type="image/svg+xml" href="/static/favicon.svg" />
-<title>端云协同 LLM 协作系统</title>
+    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
-<style>
+    <title>webapp</title>
-:root{--bg:#0f1115;--panel:#171a21;--panel2:#1d212b;--line:#2a2f3a;--txt:#e6e9ef;--muted:#8b93a3;--acc:#4f8cff;--acc2:#23c46d;--warn:#f5a623;--bad:#ff5c5c;--mono:Consolas,"Courier New",monospace}
+    <script type="module" crossorigin src="/static/assets/index-BnlAnYlB.js"></script>
-*{box-sizing:border-box;margin:0;padding:0}
+    <link rel="stylesheet" crossorigin href="/static/assets/index-G8RWfJ2Y.css">
-body{background:var(--bg);color:var(--txt);font:14px/1.6 -apple-system,"Segoe UI","Microsoft YaHei",sans-serif;height:100vh;display:flex;flex-direction:column}
+  </head>
-header{display:flex;align-items:center;gap:16px;padding:12px 20px;background:var(--panel);border-bottom:1px solid var(--line)}
+  <body>
-header h1{font-size:16px;font-weight:600}
+    <div id="app"></div>
-header .sub{color:var(--muted);font-size:12px}
+  </body>
-.pill{font-size:11px;padding:2px 8px;border-radius:10px;border:1px solid var(--line);color:var(--muted)}
-nav{display:flex;gap:4px;padding:8px 20px;background:var(--panel);border-bottom:1px solid var(--line)}
-nav button{background:none;border:1px solid transparent;color:var(--muted);padding:6px 14px;border-radius:8px;cursor:pointer;font-size:13px}
-nav button.active{color:var(--txt);background:var(--panel2);border-color:var(--line)}
-nav button:hover{color:var(--txt)}
-main{flex:1;overflow:auto;padding:16px 20px}
-.panel{display:none;max-width:1200px;margin:0 auto}
-.panel.active{display:block}
-.chatrow{display:flex;gap:16px;align-items:flex-start}
-.chatbox{flex:1}
-.prompt{display:flex;gap:8px;margin-bottom:12px}
-textarea{flex:1;resize:vertical;min-height:70px;background:var(--panel2);border:1px solid var(--line);color:var(--txt);border-radius:10px;padding:10px;font-size:14px;font-family:inherit}
-textarea:focus{outline:none;border-color:var(--acc)}
-button.primary{background:var(--acc);color:#fff;border:none;border-radius:10px;padding:10px 20px;font-size:14px;cursor:pointer}
-button.primary:disabled{opacity:.5;cursor:not-allowed}
-.card{background:var(--panel);border:1px solid var(--line);border-radius:12px;padding:16px;margin-bottom:14px}
-.card h3{font-size:13px;color:var(--muted);margin-bottom:10px;font-weight:600;letter-spacing:.5px}
-.resp{white-space:pre-wrap;word-break:break-word;font-size:14px}
-.badges{display:flex;flex-wrap:wrap;gap:6px;margin:10px 0}
-.b{font-size:11px;padding:2px 8px;border-radius:10px}
-.b.ok{background:rgba(35,196,109,.15);color:var(--acc2);border:1px solid rgba(35,196,109,.4)}
-.b.fast{background:rgba(79,140,255,.15);color:var(--acc);border:1px solid rgba(79,140,255,.4)}
-.b.warn{background:rgba(245,166,35,.15);color:var(--warn);border:1px solid rgba(245,166,35,.4)}
-.b.err{background:rgba(255,92,92,.15);color:var(--bad);border:1px solid rgba(255,92,92,.4)}
-.b.plain{background:var(--panel2);color:var(--muted);border:1px solid var(--line)}
-.chips{display:flex;flex-wrap:wrap;gap:4px;margin-top:8px}
-.chip{font-size:11px;font-family:var(--mono);background:var(--panel2);color:var(--muted);padding:2px 7px;border-radius:6px;border:1px solid var(--line)}
-.grid{display:grid;grid-template-columns:repeat(auto-fill,minmax(200px,1fr));gap:12px}
-.stat{background:var(--panel);border:1px solid var(--line);border-radius:12px;padding:14px}
-.stat .v{font-size:22px;font-weight:700}
-.stat .k{font-size:12px;color:var(--muted);margin-top:2px}
-.step{display:flex;gap:8px;align-items:baseline;padding:6px 0;border-bottom:1px dashed var(--line)}
-.step:last-child{border:none}
-.step .id{font-family:var(--mono);color:var(--acc);font-weight:600}
-.step .st{font-size:11px;margin-left:auto}
-.done{color:var(--acc2)}.fail{color:var(--bad)}.block{color:var(--warn)}
-details{background:var(--panel);border:1px solid var(--line);border-radius:10px;padding:10px 12px;margin-bottom:8px}
-summary{cursor:pointer;font-weight:600;font-size:13px}
-pre.trace{background:var(--bg);border:1px solid var(--line);border-radius:8px;padding:10px;font-family:var(--mono);font-size:12px;overflow:auto;color:var(--muted)}
-.kw{font-family:var(--mono);color:var(--acc)}
-.row-btns{display:flex;gap:8px;margin-top:8px;flex-wrap:wrap}
-button.small{background:var(--panel2);border:1px solid var(--line);color:var(--txt);border-radius:8px;padding:6px 12px;font-size:12px;cursor:pointer}
-button.small:hover{border-color:var(--acc)}
-input,select{background:var(--panel2);border:1px solid var(--line);color:var(--txt);border-radius:8px;padding:6px 10px;font-size:13px}
-.muted{color:var(--muted)}
-.history{max-height:220px;overflow:auto;margin-top:8px}
-table{width:100%;border-collapse:collapse;font-size:13px}
-.form-grid{display:grid;grid-template-columns:repeat(auto-fill,minmax(220px,1fr));gap:12px}
-.form-grid label{display:flex;flex-direction:column;gap:5px;font-size:12px;color:var(--muted)}
-.form-grid input,.form-grid select{width:100%}
-th,td{text-align:left;padding:8px;border-bottom:1px solid var(--line)}
-th{color:var(--muted);font-weight:600}
-footer{padding:8px 20px;color:var(--muted);font-size:11px;border-top:1px solid var(--line);background:var(--panel)}
-</style>
-</head>
-<body>
-<header>
-  <h1>⚙️ 端云协同 LLM 协作系统</h1>
-  <span class="sub">Architect（大模型 API）＋ Worker（本地小模型）＋ 交流文本协议</span>
-  <span class="pill" id="healthPill">…</span>
-</header>
-<nav>
-  <button data-tab="chat" class="active">💬 对话</button>
-  <button data-tab="work">🔗 协作过程</button>
-  <button data-tab="review">🧑‍💻 人工检验</button>
-  <button data-tab="metrics">📊 指标</button>
-  <button data-tab="settings">⚙️ 模型设置</button>
-</nav>
-<main>
-  <!-- 对话 -->
-  <section class="panel active" id="panel-chat">
-    <div class="chatrow">
-      <div class="chatbox">
-        <div class="prompt">
-          <textarea id="query" placeholder="输入你的问题，例如：用 Python 写一个快速排序并分析复杂度…"></textarea>
-          <button class="primary" id="sendBtn">发送</button>
-        </div>
-        <div class="row-btns">
-          <button class="small" id="legacyBtn">🔁 与 v1 legacy 对照</button>
-          <button class="small" id="sampleBtn">🎲 示例</button>
-        </div>
-        <div id="chatResult"></div>
-        <div id="legacyResult" style="margin-top:12px"></div>
-      </div>
-      <div id="workspacePanel" style="flex:1;max-width:480px"></div>
-    </div>
-  </section>
-  <!-- 协作过程 -->
-  <section class="panel" id="panel-work">
-    <div class="row-btns" style="margin-bottom:10px">
-      <input id="workRid" placeholder="输入 request_id，留空用最近一次" style="width:300px">
-      <button class="small" id="loadWorkBtn">加载交流文本</button>
-    </div>
-    <div id="workResult"></div>
-  </section>
-  <!-- 人工检验 -->
-  <section class="panel" id="panel-review">
-    <div class="row-btns" style="margin-bottom:10px">
-      <select id="reviewStatus"><option value="">全部</option><option value="pending">待审</option><option value="reviewed">已审</option></select>
-      <button class="small" id="refreshReview">刷新</button>
-    </div>
-    <div id="reviewResult"></div>
-  </section>
-  <!-- 指标 -->
-  <section class="panel" id="panel-metrics">
-    <div class="row-btns" style="margin-bottom:10px"><button class="small" id="refreshMetrics">刷新</button></div>
-    <div id="metricsResult"></div>
-  </section>
-  <!-- 模型设置 -->
-  <section class="panel" id="panel-settings">
-    <div class="card">
-      <h3>⚙️ 内置小模型（Worker）</h3>
-      <div class="form-grid">
-        <label>后端 backend
-          <select id="set-worker-backend">
-            <option value="llama_server">llama_server（内置本地模型）</option>
-            <option value="openai">openai/api（Ollama / vLLM 等兼容端点）</option>
-            <option value="mock">mock（演示，零运行时）</option>
-          </select>
-        </label>
-        <label>模型路径/名称 <input id="set-worker-model" type="text" placeholder="models/xxx.gguf 或模型名"></label>
-        <label>端点 base_url（openai 用） <input id="set-worker-baseurl" type="text" placeholder="http://127.0.0.1:11434/v1"></label>
-        <label>端口（llama_server） <input id="set-worker-port" type="number"></label>
-        <label>temperature <input id="set-worker-temperature" type="number" step="0.1"></label>
-        <label>max_fix_attempts <input id="set-worker-fix" type="number"></label>
-      </div>
-    </div>
-    <div class="card">
-      <h3>🧠 大模型（Architect / API）</h3>
-      <div class="form-grid">
-        <label>model <input id="set-arch-model" type="text"></label>
-        <label>base_url <input id="set-arch-baseurl" type="text"></label>
-        <label>api_key_env <input id="set-arch-keyenv" type="text"></label>
-      </div>
-    </div>
-    <div class="card">
-      <h3>🔀 管线 pipeline</h3>
-      <div class="form-grid">
-        <label>fast_path（快路径直答） <input id="set-pipe-fast" type="checkbox"></label>
-        <label>rounds_cap <input id="set-pipe-rounds" type="number"></label>
-        <label>api_token_cap <input id="set-pipe-tokens" type="number"></label>
-        <label>breach_policy
-          <select id="set-pipe-breach">
-            <option value="architect_do">architect_do（兜底代做）</option>
-            <option value="local_only">local_only（本地降级）</option>
-          </select>
-        </label>
-      </div>
-    </div>
-    <div class="row-btns">
-      <button class="primary" id="saveSettingsBtn">保存并生效</button>
-      <button class="small" id="resetSettingsBtn">恢复默认</button>
-      <span class="muted" id="settingsMsg"></span>
-    </div>
-  </section>
-</main>
-<footer>端云协同 LLM 协作系统 · v2 · 交流文本协议 · 北极星：token 下降 ≥80%</footer>
-<script>
-const $=id=>document.getElementById(id);
-let lastRequestId=null;
-const esc=s=>{const d=document.createElement("div");d.textContent=s==null?"":String(s);return d.innerHTML;};
-function el(tag,cls,txt){const n=document.createElement(tag);if(cls)n.className=cls;if(txt!=null)n.textContent=txt;return n;}
-function setTab(t){document.querySelectorAll("nav button").forEach(b=>b.classList.toggle("active",b.dataset.tab===t));document.querySelectorAll(".panel").forEach(p=>p.classList.toggle("active",p.id==="panel-"+t));}
-document.querySelectorAll("nav button").forEach(b=>b.onclick=()=>{setTab(b.dataset.tab);if(b.dataset.tab==="review")loadReview();if(b.dataset.tab==="metrics")loadMetrics();if(b.dataset.tab==="settings")loadSettings();});
-
-// ---- 健康状态 ----
-async function health(){try{const r=await fetch("/health");const d=await r.json();$("healthPill").textContent="健康 · "+d.domains.length+" 领域";}catch(e){$("healthPill").textContent="离线";}}
-
-// ---- 对话 ----
-async function sendChat(){const q=$("query").value.trim();if(!q)return;const btn=$("sendBtn");btn.disabled=true;btn.textContent="运行中…";$("chatResult").innerHTML="";$("legacyResult").innerHTML="";$("workspacePanel").innerHTML="";try{const r=await fetch("/chat",{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({query:q})});const d=await r.json();lastRequestId=d.request_id||null;renderChatResult(d);if(d.request_id)loadWorkspace(d.request_id);}catch(e){$("chatResult").appendChild(el("div","resp","请求失败："+e.message));}finally{btn.disabled=false;btn.textContent="发送";}}
-function renderChatResult(d){const c=$("chatResult");c.innerHTML="";const card=el("div","card");card.appendChild(el("h3","","响应"));const b=$("badges0")||el("div","badges");b.innerHTML="";const st=(d.status||"done");b.appendChild(chip(st==="fast_path"?"⚡ 快路径":"🤝 协作",st==="fast_path"?"b fast":(st==="done"?"b ok":"b warn")));b.appendChild(chip("status: "+esc(st),"b plain"));b.appendChild(chip("model: "+esc(d.model_used),"b plain"));b.appendChild(chip("输入 "+d.api_input_tokens+" / 输出 "+d.api_output_tokens+" token","b plain"));b.appendChild(chip("耗时 "+d.latency_ms+" ms","b plain"));b.appendChild(chip("rounds "+d.rounds_used,"b plain"));b.id="badges0";card.appendChild(b);const resp=el("div","resp",d.response||"(空)");card.appendChild(resp);if(d.error)card.appendChild(el("div","resp muted","⚠️ "+d.error));const route=el("div","chips");(d.route||[]).forEach(r=>route.appendChild(el("span","chip",r)));card.appendChild(el("h3","","路由轨迹"));card.appendChild(route);card.appendChild(el("div","muted","request_id: "+(d.request_id||"-")+"（见右侧协作过程）"));c.appendChild(card);}
-function chip(txt,cls){const s=el("span","b "+cls,txt);return s;}
-
-// ---- 协作过程 / 交流文本 ----
-async function loadWorkspace(rid){try{const r=await fetch("/runs/"+rid+"/workspace");if(!r.ok)throw new Error("无 workspace");const ws=await r.json();renderWorkspace($("workspacePanel"),ws);if($("workRid"))$("workRid").value=rid;renderWorkTab(ws);}catch(e){$("workspacePanel").appendChild(el("div","muted","（该请求无交流文本："+e.message+"）"));}}
-function renderWorkspace(host,ws){host.innerHTML="";const card=el("div","card");card.appendChild(el("h3","","🤝 交流文本 · "+(ws.request_id||"")));const b=ws.brief||{};card.appendChild(el("div","resp","🎯 "+(b.goal||"")));const plan=el("div","");(b.plan||[]).forEach(p=>{const row=el("div","step");row.appendChild(el("span","id",p.id));row.appendChild(el("span","",p.task||""));plan.appendChild(row);});card.appendChild(el("h3","","计划 (plan)"));card.appendChild(plan);card.appendChild(el("h3","","进度 (progress)"));const pg=el("div","");(ws.progress||[]).forEach(p=>{const row=el("div","step");row.appendChild(el("span","id",p.step));row.appendChild(el("span","",p.summary||""));const st=el("span","st "+(p.status==="done"?"done":p.status==="failed"?"fail":"block"),p.status);row.appendChild(st);pg.appendChild(row);});card.appendChild(pg);if((ws.issues||[]).length){card.appendChild(el("h3","","问题 (issues)"));(ws.issues||[]).forEach(i=>{const dt=el("details","");dt.appendChild(el("summary","","#"+(i.id||"")+" · step "+(i.step||"")));dt.appendChild(el("div","resp","observed: "+(i.observed||"")+"\nask: "+(i.ask||"")));card.appendChild(dt);});}if((ws.decisions||[]).length){card.appendChild(el("h3","","裁决 (decisions)"));(ws.decisions||[]).forEach(dc=>{const dt=el("details","");dt.appendChild(el("summary","","#"+esc(dc.ref)+" · 裁决"));dt.appendChild(el("div","resp",dc.reply||""));card.appendChild(dt);});}if((ws.archive||[]).length){card.appendChild(el("h3","","归档 (archive)"));const ar=el("div","resp");(ws.archive||[]).forEach(a=>ar.appendChild(el("div","muted","▸ "+a)));card.appendChild(ar);}host.appendChild(card);}
-function renderWorkTab(ws){const host=$("workResult");if(!host)return;host.innerHTML="";renderWorkspace(host,ws);}
-async function loadWorkByRid(){const rid=$("workRid").value.trim()||lastRequestId;if(!rid){$("workResult").innerHTML="<div class=muted>没有 request_id</div>";return;}await loadWorkspace(rid);}
-
-// ---- 人工检验 ----
-async function loadReview(){const st=$("reviewStatus").value;const r=await fetch("/review/queue?limit=50&status="+st);const rows=await r.json();const host=$("reviewResult");host.innerHTML="";if(!rows.length){host.appendChild(el("div","muted","暂无记录"));return;}rows.forEach(x=>{const card=el("div","card");card.appendChild(el("h3","","#"+(x.id||"")+" · "+(x.request_id||"")+" · "+(x.status||"")));card.appendChild(el("div","resp",x.query||""));card.appendChild(el("div","muted","tags: "+(x.tags||[]).join(", ")+" · reason: "+(x.reason||"")));if(x.verdict)card.appendChild(el("div","muted","verdict: "+x.verdict+(x.correction?" · correction: "+x.correction:"")));if(x.status==="pending"){const btns=el("div","row-btns");["approve","edit","reject"].forEach(v=>{const b=el("button","small",v);b.onclick=()=>submitReview(x.id,v);btns.appendChild(b);});card.appendChild(btns);}host.appendChild(card);});}
-async function submitReview(id,verdict){let correction=null;if(verdict==="edit"){correction=prompt("修正内容：");if(correction==null)return;}const r=await fetch("/review/"+id+"?verdict="+verdict+(correction?"&correction="+encodeURIComponent(correction):""),{method:"POST"});const d=await r.json();if(d.ok)loadReview();else alert(d.detail||"提交失败");}
-
-// ---- 指标 ----
-async function loadMetrics(){const r=await fetch("/metrics");const d=await r.json();const host=$("metricsResult");host.innerHTML="";const v2=d.v2||{};const grid=el("div","grid");stat(grid,"总请求",v2.total_requests||0);stat(grid,"快路径命中率",((v2.fast_path_rate||0)*100).toFixed(1)+"%");stat(grid,"熔断次数",v2.breach_count||0);stat(grid,"API 输入 token",v2.api_tokens?v2.api_tokens.input:0);stat(grid,"API 输出 token",v2.api_tokens?v2.api_tokens.output:0);stat(grid,"回合数均值",v2.rounds_used?v2.rounds_used.avg:0);host.appendChild(grid);const card=el("div","card");card.appendChild(el("h3","","状态分布"));const tb=el("table","");const tr=el("tr","");["status","count"].forEach(h=>tr.appendChild(el("th","",h)));tb.appendChild(tr);Object.entries(v2.status_distribution||{}).forEach(([k,v])=>{const r=el("tr","");r.appendChild(el("td","",k));r.appendChild(el("td","",v));tb.appendChild(r);});card.appendChild(tb);host.appendChild(card);const rc=el("div","card");rc.appendChild(el("h3","","回合数分布"));const tb2=el("table","");const tr2=el("tr","");["rounds","count"].forEach(h=>tr2.appendChild(el("th","",h)));tb2.appendChild(tr2);Object.entries((v2.rounds_used||{}).distribution||{}).forEach(([k,v])=>{const r=el("tr","");r.appendChild(el("td","",k));r.appendChild(el("td","",v));tb2.appendChild(r);});rc.appendChild(tb2);host.appendChild(rc);}
-function stat(grid,v,k){const s=el("div","stat");s.appendChild(el("div","v",v));s.appendChild(el("div","k",k));grid.appendChild(s);}
-
-// ---- v1 legacy 对照 ----
-async function sendLegacy(){const q=$("query").value.trim();if(!q)return;const host=$("legacyResult");host.innerHTML="";try{const r=await fetch("/chat/legacy",{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({query:q})});const d=await r.json();const card=el("div","card");card.appendChild(el("h3","","🔁 v1 legacy（规则路由）"));card.appendChild(el("div","resp",d.response||""));const chips=el("div","chips");(d.route||[]).forEach(x=>chips.appendChild(el("span","chip",x)));card.appendChild(chips);card.appendChild(el("div","muted","domain: "+(d.domain||"")+" · conf "+(d.confidence||"")+" · latency "+(d.latency_ms||0)+" ms"));host.appendChild(card);}catch(e){host.appendChild(el("div","muted","legacy 调用失败："+e.message));}}
-
-// ---- 示例 ----
-const SAMPLES=["用 Python 实现快速排序，并分析时间与空间复杂度","求解方程 x^2 - 5x + 6 = 0","高血压患者日常饮食需要注意什么","解释一下深度学习中的注意力机制"];
-function sample(){$("query").value=SAMPLES[Math.floor(Math.random()*SAMPLES.length)];}
-
-// ---- 模型设置 ----
-async function loadSettings(){try{const r=await fetch("/config");const s=await r.json();const w=s.worker||{};$("set-worker-backend").value=w.backend||"llama_server";$("set-worker-model").value=w.model||"";$("set-worker-baseurl").value=w.base_url||"";$("set-worker-port").value=w.port||8901;$("set-worker-temperature").value=w.temperature||0.3;$("set-worker-fix").value=w.max_fix_attempts||2;const a=s.architect||{};$("set-arch-model").value=a.model||"";$("set-arch-baseurl").value=a.base_url||"";$("set-arch-keyenv").value=a.api_key_env||"";const p=s.pipeline||{};$("set-pipe-fast").checked=!!p.fast_path;$("set-pipe-rounds").value=p.rounds_cap||6;$("set-pipe-tokens").value=p.api_token_cap||8000;$("set-pipe-breach").value=p.breach_policy||"architect_do";$("settingsMsg").textContent="";}catch(e){$("settingsMsg").textContent="读取失败："+e.message;}}
-async function saveSettings(){const patch={worker:{backend:$("set-worker-backend").value,model:$("set-worker-model").value,base_url:$("set-worker-baseurl").value,port:parseInt($("set-worker-port").value||8901),temperature:parseFloat($("set-worker-temperature").value||0.3),max_fix_attempts:parseInt($("set-worker-fix").value||2)},architect:{model:$("set-arch-model").value,base_url:$("set-arch-baseurl").value,api_key_env:$("set-arch-keyenv").value},pipeline:{fast_path:$("set-pipe-fast").checked,rounds_cap:parseInt($("set-pipe-rounds").value||6),api_token_cap:parseInt($("set-pipe-tokens").value||8000),breach_policy:$("set-pipe-breach").value}};try{const r=await fetch("/config",{method:"PUT",headers:{"Content-Type":"application/json"},body:JSON.stringify(patch)});const d=await r.json();$("settingsMsg").textContent=r.ok?"✅ 已保存并重建管线":"❌ "+(d.detail||"保存失败");}catch(e){$("settingsMsg").textContent="保存失败："+e.message;}}
-async function resetSettings(){try{await fetch("/config/reset",{method:"POST"});$("settingsMsg").textContent="已恢复默认设置";loadSettings();}catch(e){$("settingsMsg").textContent="重置失败："+e.message;}}
-$("saveSettingsBtn").onclick=saveSettings;
-$("resetSettingsBtn").onclick=resetSettings;
-$("sendBtn").onclick=sendChat;
-$("legacyBtn").onclick=sendLegacy;
-$("sampleBtn").onclick=sample;
-$("loadWorkBtn").onclick=loadWorkByRid;
-$("refreshReview").onclick=loadReview;
-$("refreshMetrics").onclick=loadMetrics;
-$("query").addEventListener("keydown",e=>{if(e.key==="Enter"&&(e.ctrlKey||e.metaKey))sendChat();});
-health();loadMetrics();
-</script>
-</body>
 </html>

router_system/architect.py

@@ -59,7 +59,7 @@ class ArchitectClient:
     def __init__(
         self,
         model: str,
-        base_url: str = "https://api.deepseek.com/v1",
+        base_url: str = "https://api.deepseek.com",
         api_key: Optional[str] = None,
         temperature: float = 0.2,
         timeout_s: float = 60.0,
@@ -220,8 +220,8 @@ def build_architect(cfg: Dict[str, Any],
     _env = get_env or os.environ.get
     key = cfg.get("api_key") or _env(cfg.get("api_key_env", "DEEPSEEK_API_KEY"))
     return ArchitectClient(
-        model=cfg.get("model", "deepseek-chat"),
+        model=cfg.get("model", "deepseek-v4-flash"),
-        base_url=cfg.get("base_url", "https://api.deepseek.com/v1"),
+        base_url=cfg.get("base_url", "https://api.deepseek.com"),
         api_key=key,
         temperature=float(cfg.get("temperature", 0.2)),
         timeout_s=float(cfg.get("timeout_s", 60)),

router_system/classifier.py

@@ -10,7 +10,7 @@ confidence = 1 - exp(-s)，保证 s=1 -> 0.63，s=2 -> 0.86，s=3 -> 0.95。
 from __future__ import annotations
 
 import math
-from typing import Dict, List, Tuple
+from typing import Dict, List, Optional, Tuple
 
 from .difficulty import estimate_difficulty
 from .models import Classification
@@ -51,6 +51,14 @@ DOMAIN_RULES: Dict[str, List[Tuple[str, float]]] = {
         ("知识产权", 1.1), ("版权", 0.9), ("专利", 0.9), ("违约", 0.9), ("赔偿", 0.8),
         ("仲裁", 0.9), ("劳动法", 1.0), ("刑法", 1.0), ("民法典", 1.0),
         ("法规", 0.8), ("条款", 0.7), ("律师", 0.8), ("起诉", 0.9), ("判决", 0.9),
+        # 劳动法
+        ("加班", 0.9), ("加班费", 1.0), ("工资", 0.8), ("辞退", 0.9), ("裁员", 0.9),
+        ("试用期", 0.9), ("社保", 0.8), ("公积金", 0.8), ("年假", 0.9), ("离职", 0.8),
+        ("解除劳动合同", 1.1), ("经济补偿", 1.0), ("竞业", 1.0),
+        # 房产/婚姻/消费者
+        ("租房", 0.9), ("买房", 0.9), ("购房", 0.9), ("押金", 0.8), ("房贷", 0.9),
+        ("离婚", 1.0), ("继承", 0.9), ("遗产", 0.9), ("抚养权", 0.9), ("遗嘱", 0.9),
+        ("退款", 0.9), ("退货", 0.8), ("消费者", 0.8), ("七天无理由", 1.0), ("维权", 0.8),
         ("law", 1.0), ("legal", 1.1), ("contract", 1.0), ("compliance", 1.0),
         ("litigation", 1.0), ("copyright", 0.9), ("patent", 0.9), ("trademark", 0.9),
         ("liability", 0.9), ("regulatory", 0.8), ("jurisdiction", 0.9),
@@ -61,11 +69,17 @@ DOMAIN_RULES: Dict[str, List[Tuple[str, float]]] = {
         ("医生", 0.9), ("血压", 0.9), ("高血压", 1.0), ("糖尿病", 1.0), ("感冒", 0.9),
         ("剂量", 0.9), ("副作用", 0.9), ("手术", 0.9), ("患者", 0.9),
         ("吃药", 0.9), ("发烧", 1.0), ("疫苗", 0.9), ("感染", 0.9), ("体检", 0.7),
+        # 急救/消化/心理/营养/儿科
+        ("烫伤", 1.0), ("烧伤", 1.0), ("止血", 0.9), ("扭伤", 0.9), ("中暑", 1.0),
+        ("急救", 0.9), ("腹泻", 0.9), ("拉肚子", 0.9), ("便秘", 0.9), ("胃", 0.7),
+        ("失眠", 0.9), ("焦虑", 0.9), ("抑郁", 0.9), ("压力", 0.6), ("睡眠", 0.7),
+        ("减肥", 0.8), ("营养", 0.7), ("卡路里", 0.9), ("儿童", 0.8), ("婴儿", 0.9),
+        ("宝宝", 0.8), ("抗生素", 0.9), ("止咳", 0.9),
         ("medical", 1.0), ("patient", 0.9), ("symptom", 1.0), ("disease", 0.9),
         ("diagnosis", 1.0), ("treatment", 0.8), ("prescription", 1.0),
         ("dosage", 0.9), ("side effect", 0.9), ("hypertension", 1.0),
         ("diabetes", 1.0), ("surgery", 0.8), ("clinic", 0.7), ("vaccine", 0.9),
-        ("infection", 0.9),
+        ("infection", 0.9), ("first aid", 0.9), ("insomnia", 0.9),
     ],
     "general": [
         ("总结", 0.4), ("翻译", 0.4), ("介绍", 0.4), ("解释", 0.3),
@@ -74,6 +88,40 @@ DOMAIN_RULES: Dict[str, List[Tuple[str, float]]] = {
         ("write an essay", 0.4), ("邮件", 0.4), ("email", 0.3),
         ("推荐", 0.3), ("评价", 0.3),
     ],
+    "finance": [
+        ("理财", 1.0), ("投资", 1.0), ("基金", 1.0), ("股票", 1.0), ("债券", 0.9),
+        ("存款", 0.9), ("储蓄", 0.8), ("利率", 0.8), ("利息", 0.8), ("贷款", 1.0),
+        ("房贷", 1.0), ("月供", 0.9), ("保险", 0.9), ("理赔", 0.9), ("保费", 0.8),
+        ("信用卡", 1.0), ("征信", 0.9), ("逾期", 0.9), ("分期", 0.8), ("记账", 0.7),
+        ("预算", 0.7), ("理财规划", 1.0), ("收益率", 0.9), ("定投", 0.9),
+        ("invest", 0.8), ("fund", 0.8), ("stock", 0.9), ("loan", 0.9),
+        ("mortgage", 0.9), ("insurance", 0.9), ("credit card", 0.9),
+        ("finance", 0.8), ("money", 0.6), ("lpr", 0.9), ("投资理财", 1.1),
+    ],
+    "life": [
+        ("菜谱", 0.9), ("做饭", 0.8), ("烹饪", 0.9), ("美食", 0.8), ("做法", 0.7),
+        ("旅行", 0.9), ("旅游", 0.9), ("攻略", 0.8), ("机票", 0.8), ("酒店", 0.7),
+        ("签证", 0.9), ("景点", 0.8), ("自驾", 0.8),
+        ("装修", 0.9), ("收纳", 0.8), ("家居", 0.7), ("清洁", 0.7), ("打扫", 0.7),
+        ("宠物", 0.9), ("猫", 0.7), ("狗", 0.7), ("猫粮", 0.9), ("驱虫", 0.9),
+        ("健身", 0.9), ("锻炼", 0.8), ("跑步", 0.8), ("增肌", 0.9), ("减脂", 0.9),
+        ("瑜伽", 0.8), ("天气", 0.7), ("气温", 0.7),
+        ("recipe", 0.8), ("travel", 0.9), ("trip", 0.8), ("pet", 0.8),
+        ("workout", 0.9), ("gym", 0.8), ("weather", 0.7), ("cook", 0.8),
+    ],
+    "education": [
+        ("学习方法", 1.0), ("怎么学", 0.7), ("高效学习", 1.0), ("记忆", 0.6), ("复习", 0.7),
+        ("预习", 0.7), ("笔记", 0.6), ("专注", 0.6), ("拖延", 0.7), ("学习效率", 0.9),
+        ("考试", 0.9), ("备考", 1.0), ("刷题", 0.9), ("模拟考", 0.9), ("中考", 0.9),
+        ("高考", 0.9), ("考研", 0.9), ("考前", 0.7),
+        ("英语", 0.8), ("单词", 0.7), ("口语", 0.8), ("听力", 0.7), ("雅思", 1.0),
+        ("托福", 1.0), ("四级", 0.9), ("六级", 0.9), ("背单词", 0.9),
+        ("选课", 0.9), ("课程", 0.6), ("专业选择", 0.9), ("报班", 0.8), ("网课", 0.7),
+        ("自学", 0.7), ("职业规划", 1.0), ("求职", 0.9), ("面试", 0.8), ("简历", 0.8),
+        ("实习", 0.7), ("跳槽", 0.8), ("转行", 0.9),
+        ("study", 0.8), ("exam", 0.9), ("language", 0.7), ("career", 0.8),
+        ("interview", 0.8), ("education", 0.7), ("learn", 0.6),
+    ],
 }
 
 _STOPWORDS = {
@@ -92,11 +140,21 @@ class BaseClassifier:
 
 
 class RuleClassifier(BaseClassifier):
-    """基于关键词规则的分类器（零依赖）。"""
+    """基于关键词规则的分类器（零依赖）。
 
-    def __init__(self, confidence_floor: float = 0.55):
+    domains 参数（可选）：限定只对部分领域打分 —— 两级路由中，
+    每个大领域的组内路由模型用 RuleClassifier(domains=组内领域)，
+    只认识本组领域，体积与匹配开销约为统一分类器的 1/4。
+    """
+
+    def __init__(self, confidence_floor: float = 0.55,
+                 domains: Optional[List[str]] = None):
         self.confidence_floor = confidence_floor
-        self.rules = DOMAIN_RULES
+        if domains is None:
+            self.rules = DOMAIN_RULES
+        else:
+            self.rules = {d: DOMAIN_RULES[d] for d in domains if d in DOMAIN_RULES}
+        self.domains = list(self.rules.keys())
 
     def _score(self, query: str) -> Tuple[Dict[str, float], Dict[str, List[str]]]:
         q = query.lower()
@@ -159,7 +217,7 @@ class HuggingFaceClassifier(BaseClassifier):
     仅当安装 torch+transformers 且模型可加载时可用；否则抛错提示。
     """
 
-    def __init__(self, model_name: str, num_labels: int = 5, confidence_floor: float = 0.55):
+    def __init__(self, model_name: str, num_labels: int = 8, confidence_floor: float = 0.55):
         try:
             from transformers import AutoModelForSequenceClassification, AutoTokenizer
         except ImportError as e:
@@ -170,7 +228,8 @@ class HuggingFaceClassifier(BaseClassifier):
         self.model = AutoModelForSequenceClassification.from_pretrained(
             model_name, num_labels=num_labels
         )
-        self.labels = ["code", "math", "legal", "medical", "general"]
+        self.labels = ["code", "math", "legal", "medical", "general",
+                       "finance", "life", "education"]
         self.confidence_floor = confidence_floor
 
     def classify(self, query: str) -> Classification:

router_system/config.py

@@ -15,23 +15,41 @@ DEFAULT_CONFIG_PATH = Path(__file__).resolve().parent.parent / "config" / "confi
 _DEFAULTS: Dict[str, Any] = {
     "system": {"name": "multi-expert-router", "version": "0.1.0"},
     "router": {
-        "low_confidence_threshold": 0.60,   # 分类置信度低于此值 -> 直接走大模型
+        "low_confidence_threshold": 0.60,   # 分类置信度低于此值 -> 直接走最后处理者
-        "judge_fallback_threshold": 0.70,   # Judge 质量分低于此值 -> 升级大模型
+        "judge_fallback_threshold": 0.70,   # Judge 质量分低于此值 -> 升级最后处理者
         "default_temperature": 0.2,
     },
+    "execution": {
+        "mode": "rule",              # rule（L0 零参数）| hybrid
+        "planner": "rule",           # rule | hf
+        "expert_backend": "rule",    # rule（规则执行器）| hf | api
+        "model_level": "L0",         # L0 | L1 | L2
+        "max_plan_depth": 3,
+    },
     "classifier": {"type": "rule", "model": "Qwen/Qwen3-0.6B", "confidence_floor": 0.55},
-    "domains": ["code", "math", "legal", "medical", "general"],
+    "domains": ["code", "math", "legal", "medical", "finance", "life", "education", "general"],
+    # 两级路由：大领域分组（用户接口指定 group → 组内路由模型 → 组内专业小模型）
+    # 组内路由模型只识别本组领域，体积约为统一路由模型的 1/4
+    "domain_groups": {
+        "tech": ["code", "math"],
+        "professional": ["legal", "medical", "finance"],
+        "lifestyle": ["life", "education"],
+        "general": ["general"],
+    },
     "experts": {
         "code": {"type": "mock", "model": "Qwen/Qwen2.5-Coder-7B-Instruct"},
         "math": {"type": "mock", "model": "Qwen/Qwen3-4B-Instruct"},
         "legal": {"type": "mock", "model": "Qwen/Qwen3-4B-Instruct"},
         "medical": {"type": "mock", "model": "Qwen/Qwen3-4B-Instruct"},
+        "finance": {"type": "mock", "model": "Qwen/Qwen3-4B-Instruct"},
+        "life": {"type": "mock", "model": "Qwen/Qwen3-1.7B-Instruct"},
+        "education": {"type": "mock", "model": "Qwen/Qwen3-1.7B-Instruct"},
         "general": {"type": "mock", "model": "Qwen/Qwen3-1.7B-Instruct"},
     },
     "fallback": {
         "type": "mock",
-        "model": "deepseek-chat",
+        "model": "deepseek-v4-flash",
-        "base_url": "https://api.deepseek.com/v1",
+        "base_url": "https://api.deepseek.com",
         "api_key_env": "DEEPSEEK_API_KEY",
     },
     "judge": {"type": "rule", "model": "Qwen/Qwen3-1.7B-Instruct"},

router_system/fallback.py

@@ -1,4 +1,10 @@
-"""大模型回退层：Mock 与 OpenAI 兼容 API 两种后端。"""
+"""大模型回退层（最后处理者）：Mock / 降级模板 / 本地模型 / OpenAI 兼容 API。
+
+- NoneFallback   ：降级模板（零参数，明确告知超出知识库范围）—— L0 最小可用
+- MockFallback   ：确定性 mock 大模型（零参数，测试升级路径用）
+- LocalFallback  ：本地小模型（≤8B，Q4 量化，OpenAI 兼容端点如 Ollama/vLLM）—— 按需加载
+- APIFallback    ：远程 OpenAI 兼容 API（可选，默认关闭）
+"""
 from __future__ import annotations
 
 import asyncio
@@ -14,6 +20,30 @@ class FallbackProvider:
         raise NotImplementedError
 
 
+class NoneFallback(FallbackProvider):
+    """降级模板：零参数兜底，明确告知查询超出知识库范围。"""
+
+    def __init__(self, model: str = "none"):
+        self.model = model
+        self.name = "fallback-none"
+
+    async def generate(self, query: str) -> ExpertResponse:
+        body = (
+            f"（降级响应）「{query}」\n\n"
+            "当前查询超出本地知识库可处理范围（低置信度或质量校验未通过）。\n"
+            "可选处理：\n"
+            "1. 换个更明确的问法重试\n"
+            "2. 启用 L2 本地小模型或配置最后处理者（fallback.type: local）\n"
+        )
+        return ExpertResponse(
+            text=body,
+            model_used=self.model,
+            latency_ms=0.0,
+            tokens=80,
+            cost_est=0.0,
+        )
+
+
 class MockFallback(FallbackProvider):
     """确定性 mock 大模型：标识为 fallback，便于测试升级路径。"""
 
@@ -40,6 +70,54 @@ class MockFallback(FallbackProvider):
         )
 
 
+class LocalFallback(FallbackProvider):
+    """本地小模型最后处理者（≤8B，如 DeepSeek-R1-Distill-Qwen-7B Q4）。
+
+    通过 OpenAI 兼容端点调用（Ollama 默认 11434/v1，vLLM 默认 8001/v1），
+    模型按需加载、用完即卸载（由本地推理服务管理），不常驻显存。
+    """
+
+    def __init__(self, model: str, base_url: str = "http://127.0.0.1:11434/v1",
+                 api_key: str = ""):
+        self.model = model
+        self.base_url = base_url.rstrip("/")
+        self.api_key = api_key
+        self.name = f"fallback-local-{model}"
+        self._client = None
+
+    def _get_client(self):
+        if self._client is None:
+            import httpx
+            self._client = httpx.AsyncClient(timeout=120.0)
+        return self._client
+
+    async def generate(self, query: str) -> ExpertResponse:
+        client = self._get_client()
+        headers = {"Authorization": f"Bearer {self.api_key}"} if self.api_key else {}
+        resp = await client.post(
+            f"{self.base_url}/chat/completions",
+            headers=headers,
+            json={
+                "model": self.model,
+                "messages": [{"role": "user", "content": query}],
+                "temperature": 0.3,
+                "max_tokens": 2048,
+            },
+        )
+        resp.raise_for_status()
+        data = resp.json()
+        body = data["choices"][0]["message"]["content"]
+        usage = data.get("usage", {})
+        tokens = usage.get("completion_tokens", int(len(body) / 2.2))
+        return ExpertResponse(
+            text=body,
+            model_used=self.model,
+            latency_ms=0.0,
+            tokens=tokens,
+            cost_est=0.0,  # 本地推理成本按电费计，模型层成本记为 0（相对 API）
+        )
+
+
 class APIFallback(FallbackProvider):
     """OpenAI 兼容大模型 API（如 DeepSeek / OpenAI / 本地 vLLM）。"""
 
@@ -85,9 +163,17 @@ class APIFallback(FallbackProvider):
 def build_fallback(cfg: Dict) -> FallbackProvider:
     """cfg 为 fallback 段配置。"""
     ftype = cfg.get("type", "mock")
-    model = cfg.get("model", "deepseek-chat")
+    model = cfg.get("model", "deepseek-v4-flash")
+    if ftype == "none":
+        return NoneFallback(model=model)
     if ftype == "mock":
         return MockFallback(model=model)
+    if ftype == "local":
+        return LocalFallback(
+            model,
+            cfg.get("base_url", "http://127.0.0.1:11434/v1"),
+            cfg.get("api_key", ""),
+        )
     if ftype == "api":
         import os
         api_key = cfg.get("api_key") or os.environ.get(cfg.get("api_key_env", "API_KEY"))
@@ -95,5 +181,5 @@ def build_fallback(cfg: Dict) -> FallbackProvider:
             raise RuntimeError(
                 f"APIFallback 缺少 API Key：请设置环境变量 {cfg.get('api_key_env')} 或配置 api_key"
             )
-        return APIFallback(model, cfg.get("base_url", "https://api.deepseek.com/v1"), api_key)
+        return APIFallback(model, cfg.get("base_url", "https://api.deepseek.com"), api_key)
-    raise ValueError(f"未知 fallback 类型: {ftype}（支持 mock | api）")␍
+    raise ValueError(f"未知 fallback 类型: {ftype}（支持 none | mock | local | api）")

router_system/judge.py

@@ -56,8 +56,9 @@ class BaseJudge:
 class RuleJudge(BaseJudge):
     """启发式质量评估（零依赖）。"""
 
-    def __init__(self, fallback_threshold: float = 0.70):
+    def __init__(self, fallback_threshold: float = 0.70, kb=None):
         self.fallback_threshold = fallback_threshold
+        self.kb = kb  # 可选知识库：facts 维度校验用
 
     async def evaluate(self, query: str, response: str, domain: str) -> QualityEvaluation:
         scores: Dict[str, float] = {}
@@ -103,7 +104,25 @@ class RuleJudge(BaseJudge):
         else:
             scores["safety"] = 1.0
 
-        weights = {"coverage": 0.4, "length": 0.2, "format": 0.2, "safety": 0.2}
+        # 5) 领域知识引用（facts 维度）：法律/医疗响应应覆盖查询命中的知识条目
+        if domain in ("legal", "medical") and self.kb is not None:
+            facts = self.kb.facts(domain)
+            hit_facts = [f for f in facts if any(k in query for k in f.get("keywords", []))]
+            if hit_facts:
+                ok = sum(
+                    1 for f in hit_facts
+                    if any(k in response for k in f.get("keywords", []))
+                )
+                scores["facts"] = ok / len(hit_facts)
+                if scores["facts"] < 0.6:
+                    reasons.append(f"领域知识引用不足 ({scores['facts']:.0%})")
+            else:
+                scores["facts"] = 1.0
+        else:
+            scores["facts"] = 1.0
+
+        weights = {"coverage": 0.35, "length": 0.15, "format": 0.15,
+                   "safety": 0.15, "facts": 0.2}
         overall = sum(scores.get(k, 0.0) * w for k, w in weights.items())
         needs = overall < self.fallback_threshold
         if needs:
@@ -157,17 +176,17 @@ class LLMJudge(BaseJudge):
         )
 
 
-def build_judge(cfg: Dict, fallback_threshold: float = 0.70) -> BaseJudge:
+def build_judge(cfg: Dict, fallback_threshold: float = 0.70, kb=None) -> BaseJudge:
     """cfg 为 judge 段配置。"""
     jtype = cfg.get("type", "rule")
     if jtype == "rule":
-        return RuleJudge(fallback_threshold=fallback_threshold)
+        return RuleJudge(fallback_threshold=fallback_threshold, kb=kb)
     if jtype == "llm":
         import os
         api_key = cfg.get("api_key") or os.environ.get(cfg.get("api_key_env", "API_KEY"))
         return LLMJudge(
-            cfg.get("model", "deepseek-chat"),
+            cfg.get("model", "deepseek-v4-flash"),
-            cfg.get("base_url", "https://api.deepseek.com/v1"),
+            cfg.get("base_url", "https://api.deepseek.com"),
             api_key or "",
             fallback_threshold=fallback_threshold,
         )

router_system/models.py

@@ -44,6 +44,10 @@ class RouterResult:
     cache_level: Optional[str] = None    # exact | semantic
     cost_est: float = 0.0
     error: Optional[str] = None
+    subdomain: Optional[str] = None      # 二级子领域（如 investing/labor）
+    subdomain2: Optional[str] = None     # 三级子领域（如 fund/overtime）
+    domain_group: Optional[str] = None   # 大领域组（两级路由第一级：tech/professional/...）
+    request_id: Optional[str] = None     # 请求 ID（配合 /traces/{id} 查询完整推理链）
 
     def to_dict(self) -> Dict[str, Any]:
         return {
@@ -61,6 +65,10 @@ class RouterResult:
             "cache_level": self.cache_level,
             "cost_est": round(self.cost_est, 6),
             "error": self.error,
+            "subdomain": self.subdomain,
+            "subdomain2": self.subdomain2,
+            "domain_group": self.domain_group,
+            "request_id": self.request_id,
         }
 
 

router_system/pipeline.py

@@ -68,8 +68,9 @@ class CollaborativePipeline:
     # ---------------------------------------------------------------
     # 入口
     # ---------------------------------------------------------------
-    async def run(self, query: str) -> PipelineResult:
+    async def run(self, query: str, request_id: Optional[str] = None) -> PipelineResult:
-        request_id = uuid.uuid4().hex[:12]
+        if request_id is None:
+            request_id = uuid.uuid4().hex[:12]
         ws = Workspace.new(request_id, query, self.api_token_cap, self.rounds_cap)
         route: List[str] = ["v2"]
         t0 = time.perf_counter() * 1000.0
@@ -100,11 +101,15 @@ class CollaborativePipeline:
         route.append("loop")
         plan = brief.get("plan") or []
         pending = [p.get("id") for p in plan]
+        plan_ids = {p.get("id") for p in plan}          # 用于 _deps_done 过滤
         while pending and not ws.exhausted():
             progressed = False
             for sid in list(pending):
                 step = next((p for p in plan if p.get("id") == sid), {})
-                if not self._deps_done(ws, step.get("deps") or []):
+                deps = step.get("deps") or []
+                # 只检查在 plan 中的依赖；不在 plan 的 ID 视为"不存在"→自动满足
+                relevant = [d for d in deps if d in plan_ids]
+                if not self._deps_done(ws, relevant):
                     continue
                 existing = self._read_artifact(request_id, self._artifact_name(sid, ws))
                 outcome = await self.worker.run_step(ws, sid, existing_artifact=existing,
@@ -115,6 +120,8 @@ class CollaborativePipeline:
                     ws.rollup()
                     route.append(f"step:{sid}:done")
                     progressed = True
+                    if not pending:                              # 所有步骤完成，退出循环
+                        break
                 else:  # issue -> Architect 裁决
                     route.append(f"step:{sid}:issue")
                     try:
@@ -227,7 +234,17 @@ class CollaborativePipeline:
         return artifact_name_for(sid, domain)
 
     def _deps_done(self, ws: Workspace, deps: List[str]) -> bool:
+        # progress 中 done 的条目；done 条目 rollup 后移到 archive（字符串格式如 "s1: ..."）
         done = {p["step"] for p in ws.get("progress", []) if p.get("status") == "done"}
+        for entry in ws.get("archive", []):
+            if isinstance(entry, dict):
+                sid = entry.get("step", "")
+            elif isinstance(entry, str):
+                sid = entry.split(":")[0].strip() if ":" in entry else ""
+            else:
+                sid = ""
+            if sid in deps:
+                done.add(sid)
         return all(d in done for d in deps)
 
     def _last_decision_for(self, ws: Workspace, sid: str) -> str:

router_system/router.py

@@ -1,26 +1,39 @@
-"""主路由器：协调 缓存 -> 分类 -> 专家 -> Judge -> 大模型回退 的完整链路。
+"""主路由器：两级路由（大领域组 → 组内路由模型 → 专业执行器）专家系统编排。
 
-流程（对齐实现方案）：
+两级体系（对齐用户架构决策）：
-1. 检查缓存（L1 精确 / L2 语义）
+  第一级：用户通过接口指定大领域组（domain_group: tech/professional/lifestyle/general），
-2. 低置信度查询直接走大模型（should_fallback）
+          或系统自动检测（8 领域分类 → 映射到组）
-3. 分类器输出领域 + 难度
+  第二级：组内路由模型（RuleClassifier(domains=组内领域) + 组内知识/模板）识别具体
-4. 选择专家模型生成
+          领域、子领域、拆解子任务 → 组内专业小模型/规则执行器
-5. Judge 评估质量
+  组内路由模型只认识本组领域：体积与匹配开销约为统一路由模型的 1/4，
-6. 质量不达标 -> 升级大模型
+  且未来 L2 模型层可每组一个更小的路由模型，按需加载不常驻。
-7. 记录指标、写缓存、返回结果
+
+链路：缓存 → 组路由(分类/子领域/拆解) → 黑板+前向链 → DAG 执行 → 合并
+      → Judge 校验 → （不达标）最后处理者升级 → 缓存/指标
+
+L0 模式（默认）：规则分类 + 规则拆解 + 规则执行器 + 规则 Judge —— 零模型参数、零 API。
+L2 模式（可选）：execution.expert_backend = hf/api 时，子任务改由专家池小模型执行
+（≤8B，按需加载），其余流程不变。
 """
 from __future__ import annotations
 
-from typing import Any, Dict, Optional
+import uuid
+from typing import Any, Dict, List, Optional
 
 from .cache import RouterCache
-from .classifier import BaseClassifier, build_classifier
+from .classifier import BaseClassifier, RuleClassifier, build_classifier
 from .config import load_config
+from .executors import NodeExecutor, build_node_executor
 from .experts import Expert, build_expert_pool
 from .fallback import FallbackProvider, build_fallback
+from .inference import InferenceEngine
 from .judge import BaseJudge, build_judge
+from .knowledge import KnowledgeBase
+from .memory import TaskGraph, TaskNode, WorkingMemory
 from .models import Classification, ExpertResponse, RouterResult, now_ms
+from .planner import Planner
 from .stats import Stats
+from .trace import TraceStore
 
 
 class Router:
@@ -33,6 +46,8 @@ class Router:
         cache: Optional[RouterCache] = None,
         stats: Optional[Stats] = None,
         config: Optional[Dict[str, Any]] = None,
+        kb: Optional[KnowledgeBase] = None,
+        planner: Optional[Planner] = None,
     ):
         self.classifier = classifier
         self.experts = experts
@@ -45,11 +60,43 @@ class Router:
         self.low_confidence_threshold = rcfg.get("low_confidence_threshold", 0.60)
         self.judge_fallback_threshold = rcfg.get("judge_fallback_threshold", 0.70)
         self.cache_enabled = cfg.get("cache", {}).get("enabled", True)
+        # ---- 专家系统内核 ----
+        self.kb = kb or KnowledgeBase()
+        self.planner = planner or Planner(self.kb)
+        self.inference = InferenceEngine(self.kb)
+        ecfg = cfg.get("execution", {})
+        self.expert_backend = ecfg.get("expert_backend", "rule")  # rule | hf | api
+        # 子任务执行后端（T1 抽象：NodeExecutor 工厂，新增后端无需改 Router）
+        self.node_executor: NodeExecutor = build_node_executor(
+            self.expert_backend, kb=self.kb, experts=experts)
+        # 推理链轨迹存储（T3：可解释性产品化）
+        self.trace_store = TraceStore()
+        # ---- 两级路由：大领域分组 + 组内路由模型（更小更专） ----
+        self.domain_groups: Dict[str, List[str]] = cfg.get("domain_groups", {}) or {}
+        if not self.domain_groups:
+            # 兜底：未配置时按单组（全部领域）处理，行为退化为一级路由
+            self.domain_groups = {"all": list(self.experts.keys())}
+        self._group_of_domain: Dict[str, str] = {}
+        for g, domains in self.domain_groups.items():
+            for d in domains:
+                self._group_of_domain[d] = g
+        # 组内路由模型：每组一个轻量分类器（只认识组内领域）
+        self._group_classifiers: Dict[str, RuleClassifier] = {
+            g: RuleClassifier(domains=domains)
+            for g, domains in self.domain_groups.items()
+        }
 
     # ---------------------------------------------------------------
-    async def route(self, query: str) -> RouterResult:
+    async def route(self, query: str, domain_group: Optional[str] = None) -> RouterResult:
+        """两级路由入口。
+
+        domain_group 指定时：跳过 8 领域统一分类器，直接用组内路由模型
+        （RuleClassifier(domains=组内领域)）识别组内领域 —— 更小更专。
+        未指定时：统一分类器识别领域 → 自动映射到大领域组（向后兼容）。
+        """
         start = now_ms()
         route: list = []
+        request_id = uuid.uuid4().hex[:12]
 
         # ---- Step 1: 缓存 ----
         if self.cache_enabled:
@@ -71,71 +118,132 @@ class Router:
                     cache_hit=True,
                     cache_level=level,
                     cost_est=0.0,
+                    subdomain=cached.get("subdomain"),
+                    subdomain2=cached.get("subdomain2"),
+                    domain_group=cached.get("domain_group"),
+                    request_id=request_id,
+                )
+                self._store_trace(
+                    request_id=request_id, query=query, group=cached.get("domain_group"),
+                    domain=result.domain, difficulty=result.difficulty,
+                    confidence=result.confidence, subdomain=result.subdomain,
+                    subdomain2=result.subdomain2, route=route, quality=result.quality_score,
+                    upgraded=False, model=result.model_used, latency=latency,
+                    cache_hit=True, cache_level=level,
                 )
                 self.stats.record(latency, result.domain, result.difficulty, False, True, level, 0.0, result.model_used)
                 return result
             route.append("cache:miss")
 
-        # ---- Step 2: 分类 ----
+        # ---- Step 2: 组路由（两级第一级）→ 组内分类（两级第二级） ----
-        classification = self.classifier.classify(query)
+        classifier = self.classifier
+        group = domain_group
+        if group is not None:
+            # 用户指定大领域：校验 + 使用组内路由模型
+            if group not in self.domain_groups:
+                raise ValueError(
+                    f"未知大领域组: {group}（可用: {sorted(self.domain_groups)}）"
+                )
+            classifier = self._group_classifiers[group]
+            route.append(f"group:{group}@explicit")
+        classification = classifier.classify(query)
+        if group is None:
+            # 自动检测：8 领域分类 → 映射大领域组
+            group = self._group_of_domain.get(classification.domain, "general")
+            route.append(f"group:{group}@auto")
         route.append(f"classify:{classification.domain}@{classification.confidence:.2f}/{classification.difficulty}")
+        subdomain, subdomain2 = self._detect_subdomain(query, classification.domain)
+        if subdomain:
+            route.append(f"subdomain:{subdomain}")
+        if subdomain2:
+            route.append(f"subdomain2:{subdomain2}")
 
-        # 低置信度 -> 直接走大模型
+        # ---- Step 3: 低置信度 -> 直接走最后处理者 ----
-        if self.classifier.should_fallback(classification, self.low_confidence_threshold):
+        if classifier.should_fallback(classification, self.low_confidence_threshold):
             route.append("direct_fallback")
             fb = await self._call_fallback(query)
             latency = now_ms() - start
-            result = self._finalize(query, classification, fb, quality_score=0.0,
-                                    upgraded=True, route=route, latency_ms=latency,
-                                    model_used=fb.model_used, cost_est=fb.cost_est)
-            self._record(result, latency)
-            return result
-
-        # ---- Step 3: 选择专家 ----
-        domain = classification.domain
-        expert = self.experts.get(domain)
-        if expert is None:
-            expert = self.experts.get("general")
-            route.append("expert:fallback-to-general")
-        else:
-            route.append(f"expert:{expert.name}")
-
-        # ---- Step 4: 生成 ----
-        try:
-            expert_resp = await expert.generate(query, classification.difficulty)
-        except Exception as e:
-            self.stats.record_error()
-            route.append(f"expert_error:{type(e).__name__}")
-            fb = await self._call_fallback(query)
-            latency = now_ms() - start
             result = self._finalize(query, classification, fb, quality_score=0.0,
                                     upgraded=True, route=route, latency_ms=latency,
                                     model_used=fb.model_used, cost_est=fb.cost_est,
-                                    error=str(e))
+                                    subdomain=subdomain, subdomain2=subdomain2,
+                                    domain_group=group)
+            result.request_id = request_id
+            self._store_trace(
+                request_id=request_id, query=query, group=group,
+                domain=result.domain, difficulty=result.difficulty,
+                confidence=result.confidence, subdomain=subdomain,
+                subdomain2=subdomain2, route=route, quality=0.0,
+                upgraded=True, model=fb.model_used, latency=latency,
+            )
             self._record(result, latency)
             return result
 
-        # ---- Step 5: Judge 评估 ----
+        # ---- Step 4: Planner 任务拆解（DAG） ----
+        graph = self.planner.plan(query, classification)
+        route.extend(self.planner.explain_plan(graph))
+
+        # ---- Step 5: 黑板初始化 + 前向链（规则轨迹） ----
+        memory = WorkingMemory()
+        self.inference.initialize(
+            query, classification.domain, classification.difficulty,
+            classification.confidence, memory,
+        )
+        fired = self.inference.run(query, classification.domain, memory)
+        if fired:
+            route.append(f"rules:{','.join(fired[:5])}")
+
+        # ---- Step 6: DAG 顺序执行（拓扑序） ----
+        order = graph.topo_order()
+        last_model = f"rule:{classification.domain}"
+        for node in order:
+            last_model = await self._execute_node(graph, node, classification, memory, route) or last_model
+
+        # ---- Step 7: 黑板合并（节点输出 + 推理机规则产出） ----
+        response = memory.merge([n.id for n in order])
+        # 追加推理机规则产出的部分解（带 output 的知识规则，如 git/docker/常识条目）
+        node_ids = {n.id for n in order}
+        extra_sections = [sid for sid in memory.sections if sid not in node_ids]
+        extras = [memory.section(s) for s in extra_sections if memory.section(s)]
+        if extras:
+            extra_text = "\n\n".join(extras)
+            response = (response + "\n\n" + extra_text) if response.strip() else extra_text
+        if not response.strip():
+            response = "（规则执行器）未能生成有效回答：任务均未产出内容。"
+            route.append("merge:empty")
+
+        # ---- Step 8: Judge 校验 ----
         try:
-            evaluation = await self.judge.evaluate(query, expert_resp.text, domain)
+            evaluation = await self.judge.evaluate(query, response, classification.domain)
         except Exception:
             evaluation = None
             route.append("judge_error")
-
         quality_score = evaluation.overall_score if evaluation else 0.0
         route.append(f"judge:{quality_score:.2f}")
 
         upgraded = False
-        final_resp = expert_resp
         if evaluation is not None and evaluation.needs_fallback:
             route.append("upgrade")
-            final_resp = await self._call_fallback(query)
+            fb = await self._call_fallback(query)
+            response = fb.text
+            last_model = fb.model_used
             upgraded = True
 
         latency = now_ms() - start
-        result = self._finalize(query, classification, final_resp, quality_score=quality_score,
+        result = self._finalize(query, classification, ExpertResponse(
-                                upgraded=upgraded, route=route, latency_ms=latency,
+            text=response, model_used=last_model, latency_ms=latency,
-                                model_used=final_resp.model_used, cost_est=final_resp.cost_est)
+            tokens=max(8, int(len(response) / 2.2)), cost_est=0.0,
+        ), quality_score=quality_score, upgraded=upgraded, route=route,
+            latency_ms=latency, model_used=last_model, cost_est=0.0,
+            subdomain=subdomain, subdomain2=subdomain2, domain_group=group)
+        result.request_id = request_id
+        self._store_trace(
+            request_id=request_id, query=query, group=group,
+            domain=result.domain, difficulty=result.difficulty,
+            confidence=result.confidence, subdomain=subdomain,
+            subdomain2=subdomain2, route=route, quality=quality_score,
+            upgraded=upgraded, model=last_model, latency=latency,
+        )
         self._record(result, latency)
 
         # 未升级的结果写缓存
@@ -144,6 +252,80 @@ class Router:
 
         return result
 
+    # ---------------------------------------------------------------
+    def _store_trace(self, request_id: str, query: str, group: Optional[str],
+                     domain: str, difficulty: str, confidence: float,
+                     subdomain: Optional[str], subdomain2: Optional[str],
+                     route: list, quality: float, upgraded: bool,
+                     model: str, latency: float,
+                     cache_hit: bool = False, cache_level: Optional[str] = None) -> None:
+        """记录完整推理链到轨迹存储（T3：可解释性产品化）。"""
+        self.trace_store.put(request_id, {
+            "request_id": request_id,
+            "query": query,
+            "domain_group": group,
+            "domain": domain,
+            "difficulty": difficulty,
+            "confidence": round(confidence, 4),
+            "subdomain": subdomain,
+            "subdomain2": subdomain2,
+            "route": list(route),
+            "quality_score": round(quality, 4),
+            "upgraded": upgraded,
+            "model_used": model,
+            "latency_ms": round(latency, 2),
+            "cache_hit": cache_hit,
+            "cache_level": cache_level,
+        })
+
+    # ---------------------------------------------------------------
+    def _detect_subdomain(self, query: str, domain: str) -> tuple:
+        """子领域识别：返回 (二级 subdomain, 三级 subdomain2)。
+
+        二级取领域内最高优先级带 subdomain 的命中规则；
+        三级取最高优先级带 subdomain2 的命中规则（可与二级来自不同规则）。
+        """
+        hits = self.kb.match(query, domain=domain)
+        sub = None
+        sub2 = None
+        for h in hits:
+            if sub is None and h.subdomain:
+                sub = h.subdomain
+            if sub2 is None and h.subdomain2:
+                sub2 = h.subdomain2
+            if sub is not None and sub2 is not None:
+                break
+        return sub, sub2
+
+    # ---------------------------------------------------------------
+    async def _execute_node(self, graph: TaskGraph, node: TaskNode,
+                            classification: Classification, memory: WorkingMemory,
+                            route: list) -> Optional[str]:
+        """执行一个子任务节点；返回使用的 model_used（失败返回 None）。"""
+        # 依赖检查：依赖失败/跳过 → 本节点跳过
+        for dep_id in node.deps:
+            dep = graph.get(dep_id)
+            if dep is not None and dep.status in ("failed", "skipped"):
+                node.status = "skipped"
+                route.append(f"{node.id}:{node.kind}:skip")
+                return None
+        node.status = "running"
+        try:
+            # NodeExecutor 后端执行（rule 零参数 / model 专家池 ≤8B）
+            resp = await self.node_executor.execute(
+                node, classification.domain, classification.difficulty, memory)
+            node.output = resp.text
+            node.status = "done"
+            memory.write_section(node.id, resp.text)
+            route.append(f"{node.id}:{node.kind}")
+            return resp.model_used
+        except Exception as e:
+            node.status = "failed"
+            node.error = str(e)
+            self.stats.record_error()
+            route.append(f"{node.id}:{node.kind}:error:{type(e).__name__}")
+            return None
+
     # ---------------------------------------------------------------
     async def _call_fallback(self, query: str) -> ExpertResponse:
         try:
@@ -151,7 +333,7 @@ class Router:
         except Exception as e:
             # 回退也失败：返回错误占位响应
             return ExpertResponse(
-                text=f"[系统错误] 专家与大模型回退均失败：{type(e).__name__}: {e}",
+                text=f"[系统错误] 专家与最后处理者均失败：{type(e).__name__}: {e}",
                 model_used=f"error:{self.fallback.name}",
                 cost_est=0.0,
             )
@@ -160,7 +342,10 @@ class Router:
     def _finalize(query: str, classification: Classification, resp: ExpertResponse,
                   quality_score: float, upgraded: bool, route: list,
                   latency_ms: float, model_used: str, cost_est: float,
-                  error: Optional[str] = None) -> RouterResult:
+                  error: Optional[str] = None,
+                  subdomain: Optional[str] = None,
+                  subdomain2: Optional[str] = None,
+                  domain_group: Optional[str] = None) -> RouterResult:
         return RouterResult(
             query=query,
             response=resp.text,
@@ -175,6 +360,9 @@ class Router:
             cache_hit=False,
             cost_est=cost_est,
             error=error,
+            subdomain=subdomain,
+            subdomain2=subdomain2,
+            domain_group=domain_group,
         )
 
     def _record(self, result: RouterResult, latency_ms: float):
@@ -194,18 +382,25 @@ class Router:
         return {
             "status": "ok",
             "domains": list(self.experts.keys()),
+            "domain_groups": self.domain_groups,
             "classifier": type(self.classifier).__name__,
             "judge": type(self.judge).__name__,
             "fallback": type(self.fallback).__name__,
+            "planner": type(self.planner).__name__,
+            "execution_mode": self.expert_backend,
+            "rules": self.kb.rules_count(),
         }
 
 
 def build_router(config_path: Optional[str] = None) -> Router:
-    """从配置构建完整 Router（默认 mock 全链路，零依赖可跑）。"""
+    """从配置构建完整 Router（默认 L0 专家系统模式：零参数可跑）。"""
     config = load_config(config_path)
+    kb = KnowledgeBase()
     classifier = build_classifier(config.get("classifier", {}))
     experts = build_expert_pool(config.get("experts", {}), config.get("domains", []))
-    judge = build_judge(config.get("judge", {}), config.get("router", {}).get("judge_fallback_threshold", 0.70))
+    judge = build_judge(config.get("judge", {}),
+                        config.get("router", {}).get("judge_fallback_threshold", 0.70),
+                        kb=kb)
     fallback = build_fallback(config.get("fallback", {}))
     cache_cfg = config.get("cache", {})
     cache = RouterCache(
@@ -213,5 +408,8 @@ def build_router(config_path: Optional[str] = None) -> Router:
         similarity_threshold=cache_cfg.get("similarity_threshold", 0.88),
         promote_frequency=cache_cfg.get("promote_frequency", 5),
     )
+    ecfg = config.get("execution", {})
+    planner = Planner(kb, max_depth=ecfg.get("max_plan_depth", 3))
     stats = Stats()
-    return Router(classifier, experts, judge, fallback, cache, stats, config)␍
+    return Router(classifier, experts, judge, fallback, cache, stats, config,
+                  kb=kb, planner=planner)

router_system/workspace.py

@@ -39,7 +39,7 @@ LIMITS = {
 
 # 允许的领域标签（4.2 brief.tags；仅用于安全标记与验证接地，不做路由 D3）
 ALLOWED_TAGS = {"code", "math", "legal", "medical", "finance",
-                "life", "education", "general", "safety"}
+                "life", "education", "general", "safety", "science"}
 
 STATUS_FLOW = {
     "draft": {"in_progress"},

scripts/demo.py

@@ -1,7 +1,8 @@
-"""CLI 演示：构建 mock 全链路路由系统，跑一组样例查询并打印结果。
+"""CLI 演示：构建专家系统内核路由（默认 L0 零参数模式），跑样例查询并打印推理链。
 
 用法：
-    python scripts/demo.py [--query "自定义查询"] [--batch]
+    python scripts/demo.py [--query "自定义查询"] [--batch] [--trace] [--verbose]
+--trace  打印完整推理链（分类 → 拆解 DAG → 规则轨迹 → 子任务执行 → Judge）
 """
 from __future__ import annotations
 
@@ -10,6 +11,10 @@ import asyncio
 import sys
 from pathlib import Path
 
+if hasattr(sys.stdout, "reconfigure"):
+    sys.stdout.reconfigure(encoding="utf-8")
+    sys.stderr.reconfigure(encoding="utf-8")
+
 sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
 
 from router_system.router import build_router
@@ -26,7 +31,7 @@ SAMPLE_QUERIES = [
 ]
 
 
-async def run_demo(router, queries, verbose: bool = False):
+async def run_demo(router, queries, verbose: bool = False, trace: bool = False):
     for q in queries:
         r = await router.route(q)
         print("=" * 72)
@@ -35,15 +40,20 @@ async def run_demo(router, queries, verbose: bool = False):
               f"upgraded={r.upgraded} quality={r.quality_score:.2f} model={r.model_used} "
               f"latency={r.latency_ms:.1f}ms cache={r.cache_hit}({r.cache_level}) cost=${r.cost_est:.6f}")
         print(f"  route: {' -> '.join(r.route)}")
+        if trace:
+            # 拆解轨迹从 route 中展开为更可读的形式
+            plan_steps = [s for s in r.route if s.startswith("plan:") or ":" in s]
+            print("  推理链: " + " -> ".join(r.route))
         if verbose:
-            print(f"  --- response ---\n{r.response[:400]}")
+            print(f"  --- response ---\n{r.response[:600]}")
 
 
 async def main():
-    parser = argparse.ArgumentParser(description="多专家路由系统 CLI 演示")
+    parser = argparse.ArgumentParser(description="多专家路由系统（专家系统内核）CLI 演示")
     parser.add_argument("--query", type=str, default=None, help="单条查询（覆盖默认样例）")
     parser.add_argument("--batch", action="store_true", help="批量模式（打印全部响应）")
     parser.add_argument("--verbose", action="store_true", help="打印响应正文")
+    parser.add_argument("--trace", action="store_true", help="打印完整推理链")
     args = parser.parse_args()
 
     router = build_router()
@@ -51,9 +61,9 @@ async def main():
     print()
 
     if args.query:
-        await run_demo(router, [args.query], verbose=True)
+        await run_demo(router, [args.query], verbose=True, trace=True)
     else:
-        await run_demo(router, SAMPLE_QUERIES, verbose=args.verbose)
+        await run_demo(router, SAMPLE_QUERIES, verbose=args.verbose, trace=args.trace)
 
     print()
     print("=" * 72)

scripts/eval.py

@@ -12,27 +12,46 @@ import sys
 from collections import Counter
 from pathlib import Path
 
+# 兼容 GBK 控制台（中文 Windows 默认编码），避免打印 ✅/⚠️ 时 UnicodeEncodeError
+if hasattr(sys.stdout, "reconfigure"):
+    sys.stdout.reconfigure(encoding="utf-8")
+    sys.stderr.reconfigure(encoding="utf-8")
+
 sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
 
 from router_system.router import build_router
 
-# (query, 期望领域)
+# (query, 期望领域, 期望大领域组, 期望子领域)
+# 覆盖 8 领域 × 3 条 = 24 条 + 两级路由/三级子领域指标
 BENCH = [
-    ("用 Python 实现二分查找", "code"),
+    # ---- tech：code / math ----
-    ("这段 JavaScript 为什么报错：undefined is not a function", "code"),
+    ("用 Python 实现二分查找", "code", "tech", "algorithm"),
-    ("帮我优化这个 SQL 查询的索引", "code"),
+    ("这段 JavaScript 为什么报错：undefined is not a function", "code", "tech", "debugging"),
-    ("求解一元二次方程 ax^2+bx+c=0 的求根公式", "math"),
+    ("帮我优化这个 SQL 查询的索引", "code", "tech", "database"),
-    ("证明勾股定理", "math"),
+    ("求解一元二次方程 ax^2+bx+c=0 的求根公式", "math", "tech", "algebra"),
-    ("计算 3x + 5 = 20，x 等于多少", "math"),
+    ("证明勾股定理", "math", "tech", "geometry"),
-    ("劳动合同到期不续签，公司需要支付经济补偿吗", "legal"),
+    ("计算 3x + 5 = 20，x 等于多少", "math", "tech", "algebra"),
-    ("在合同中约定违约金上限 30%，是否合规", "legal"),
+    # ---- professional：legal / medical / finance ----
-    ("专利申请的流程和费用大概是多少", "legal"),
+    ("劳动合同到期不续签，公司需要支付经济补偿吗", "legal", "professional", "labor"),
-    ("高血压患者可以吃哪些降压药，副作用是什么", "medical"),
+    ("在合同中约定违约金上限 30%，是否合规", "legal", "professional", "contract"),
-    ("感冒发烧 38.5 度，需要吃退烧药吗", "medical"),
+    ("加班费怎么计算", "legal", "professional", "labor"),
-    ("糖尿病患者的日常饮食建议", "medical"),
+    ("高血压患者可以吃哪些降压药，副作用是什么", "medical", "professional", "medication"),
-    ("介绍一下 Transformer 架构", "general"),
+    ("感冒发烧 38.5 度，需要吃退烧药吗", "medical", "professional", "medication"),
-    ("写一封请假邮件", "general"),
+    ("烫伤后怎么处理", "medical", "professional", "firstaid"),
-    ("为什么天空是蓝色的", "general"),
+    ("基金定投的收益率怎么计算", "finance", "professional", "investing"),
+    ("信用卡逾期了怎么办", "finance", "professional", "credit"),
+    ("房贷利率是 LPR 加多少", "finance", "professional", "loan"),
+    # ---- lifestyle：life / education ----
+    ("日本旅行攻略", "life", "lifestyle", "travel"),
+    ("健身增肌计划怎么安排", "life", "lifestyle", "fitness"),
+    ("家常菜谱推荐", "life", "lifestyle", "food"),
+    ("考研英语怎么备考", "education", "lifestyle", "exam"),
+    ("高效学习方法", "education", "lifestyle", "study"),
+    ("面试技巧有哪些", "education", "lifestyle", "career"),
+    # ---- general ----
+    ("介绍一下 Transformer 架构", "general", "general", "explain"),
+    ("写一封请假邮件", "general", "general", "writing"),
+    ("为什么天空是蓝色的", "general", "general", "explain"),
 ]
 
 
@@ -40,32 +59,54 @@ async def main():
     parser = argparse.ArgumentParser()
     parser.add_argument("--repeat", type=int, default=2, help="重复轮数（验证缓存）")
     parser.add_argument("--config", type=str, default=None)
+    parser.add_argument("--group", type=str, default=None,
+                        help="指定大领域组测试两级路由（如 tech）；默认自动检测")
     args = parser.parse_args()
 
     router = build_router(args.config)
+    # 指定组时只评测组内样例（组路由只认识本组领域）
+    bench = BENCH
+    if args.group is not None:
+        bench = [b for b in BENCH if b[2] == args.group]
+        if not bench:
+            print(f"组 {args.group} 无评测样例，可用组: {sorted({b[2] for b in BENCH})}")
+            return
     correct = Counter()
+    group_correct = 0
+    subdomain_correct = 0
     total = 0
     upgraded = 0
     cache_hits = 0
+    decomposed = 0  # 被 Planner 拆解为多子任务的请求数
 
     for round_i in range(args.repeat):
-        for q, expected in BENCH:
+        for q, expected, expected_group, expected_sub in bench:
-            r = await router.route(q)
+            r = await router.route(q, domain_group=args.group)
             total += 1
             if r.domain == expected:
                 correct["total"] += 1
             else:
                 correct[f"misclass->{r.domain}"] += 1
+            if args.group is None and r.domain_group == expected_group:
+                group_correct += 1
+            if r.subdomain == expected_sub:
+                subdomain_correct += 1
             if r.upgraded:
                 upgraded += 1
             if r.cache_hit:
                 cache_hits += 1
+            if any("plan:multi" in s for s in r.route):
+                decomposed += 1
 
     acc = correct["total"] / total
-    print(f"样例数: {len(BENCH)} x {args.repeat} 轮 = {total} 次请求")
+    print(f"样例数: {len(bench)} x {args.repeat} 轮 = {total} 次请求")
     print(f"分类准确率: {acc:.1%}  ({correct['total']}/{total})")
+    if args.group is None:
+        print(f"大领域组识别准确率: {group_correct/total:.1%} ({group_correct}/{total})")
+    print(f"子领域识别准确率: {subdomain_correct/total:.1%} ({subdomain_correct}/{total})")
     print(f"升级率: {upgraded/total:.1%} ({upgraded}/{total})")
     print(f"缓存命中率: {cache_hits/total:.1%} ({cache_hits}/{total})")
+    print(f"任务拆解率: {decomposed/total:.1%} ({decomposed}/{total})")
     print()
     print("运行指标:", router.stats.summary())
     print("缓存统计:", router.cache.stats())

scripts/serve.py

@@ -1,4 +1,4 @@
-"""启动路由网关服务（后台、无窗口）。
+"""启动路由网关服务（后台、无窗口）。
 
 用法：
     python scripts/serve.py [--port 8000] [--stop]
@@ -41,11 +41,16 @@ def stop():
         return
     pid = int(PID_FILE.read_text().strip())
     try:
-        import signal
+        # Windows 下 SIGTERM 对 detached 进程不可靠，改用 taskkill 强制结束进程树
-        os.kill(pid, signal.SIGTERM)
+        subprocess.run(
-        print(f"已发送终止信号 pid={pid}")
+            ["taskkill", "/PID", str(pid), "/T", "/F"],
-    except ProcessLookupError:
+            capture_output=True, text=True, timeout=15,
-        print(f"进程 {pid} 不存在，清理 pid 文件。")
+        )
+        print(f"已终止服务 pid={pid}")
+    except subprocess.TimeoutExpired:
+        print(f"终止超时 pid={pid}，请手动结束进程。")
+    except Exception as e:
+        print(f"终止失败（进程可能不存在）: {e}")
     PID_FILE.unlink(missing_ok=True)
 
 

tests/test_cache.py

@@ -13,9 +13,9 @@ def test_exact_hit():
 
 def test_semantic_hit():
     c = RouterCache(semantic_enabled=True, similarity_threshold=0.5)
-    c.put("?python?????", {"response": "code", "domain": "code"})
+    c.put("用python写一个快速排序", {"response": "code", "domain": "code"})
-    # ?????????? L2
+    # 相似改写查询命中 L2 语义缓存
-    hit = c.get("?python????????")
+    hit = c.get("用python写一个快速排序算法")
     assert hit is not None
     assert hit[0] == "semantic"
 
@@ -24,7 +24,7 @@ def test_promote_to_exact():
     c = RouterCache(promote_frequency=3)
     result = {"response": "x", "domain": "general"}
     c.put("query", result)
-    # ?????? 3 ? ? ???????
+    # 语义命中 3 次后提升为精确缓存
     for _ in range(3):
         hit = c.get("query")
         assert hit is not None

tests/test_classifier.py

@@ -30,8 +30,8 @@ def test_medical_classification():
 
 def test_general_low_confidence():
     clf = RuleClassifier()
-    r = clf.classify("今天天气怎么样")
+    r = clf.classify("你好呀")
-    # 未命中任何领域 -> 低置信度，触发 should_fallback
+    # 未命中任何领域 -> general，低置信度，触发 should_fallback
     assert r.domain == "general"
     assert clf.should_fallback(r, 0.6) is True
 

tests/test_gateway.py

@@ -41,14 +41,30 @@ def test_chat_legacy(client):
 
 
 def test_chat_v2(v2_client):
+    # POST /chat 立即返回 request_id（异步协议）
     resp = v2_client.post("/chat", json={"query": "请介绍快速排序算法"})
     assert resp.status_code == 200
     data = resp.json()
-    assert data["response"]
+    assert "request_id" in data
-    assert data["request_id"]
+    assert data["status"] == "pending"
-    assert "fast_path" in data
+
-    assert "route" in data
+    # 轮询 /runs/{id}/status 直到完成
-    assert data["status"] in ("fast_path", "done", "escalated", "failed")
+    import time
+    for _ in range(50):          # 最多 5s
+        time.sleep(0.1)
+        status_resp = v2_client.get(f"/runs/{data['request_id']}/status")
+        assert status_resp.status_code == 200
+        s = status_resp.json()
+        if s["status"] in ("done", "failed"):
+            break
+
+    assert s["status"] == "done", f"期望 done，实际 {s['status']}，error={s.get('error')}"
+    assert s["response"]
+    assert "pipeline_status" in s
+    assert s["pipeline_status"] in ("fast_path", "done", "escalated")
+    # fast_path 不写 workspace.json，所以 workspace_path 可能为 None
+    if s["pipeline_status"] != "fast_path":
+        assert s["workspace_path"] is not None
 
 
 def test_chat_empty_query(client):
@@ -93,8 +109,9 @@ def test_web_ui_served(client):
     resp = client.get("/")
     assert resp.status_code == 200
     assert resp.headers["content-type"].startswith("text/html")
-    assert "端云协同" in resp.text
+    # Vue SPA：由 Vite 生成，特征是 <div id="app"> 和 /static/assets/ 引用
-    assert "发送" in resp.text
+    assert '<div id="app">' in resp.text
+    assert '/static/assets/' in resp.text
 
 
 def test_review_flow(client):

tests/test_router.py

@@ -11,7 +11,7 @@ async def test_normal_flow_code(router):
     assert r.response
     assert r.model_used
     assert r.latency_ms >= 0
-    assert "expert" in r.route[1] or "classify" in r.route[1]
+    assert any("classify:" in s for s in r.route)
 
 
 @pytest.mark.asyncio

webapp/.gitignore

@@ -0,0 +1,24 @@
+# Logs
+logs
+*.log
+npm-debug.log*
+yarn-debug.log*
+yarn-error.log*
+pnpm-debug.log*
+lerna-debug.log*
+
+node_modules
+dist
+dist-ssr
+*.local
+
+# Editor directories and files
+.vscode/*
+!.vscode/extensions.json
+.idea
+.DS_Store
+*.suo
+*.ntvs*
+*.njsproj
+*.sln
+*.sw?

webapp/README.md

@@ -0,0 +1,5 @@
+# Vue 3 + TypeScript + Vite
+
+This template should help get you started developing with Vue 3 and TypeScript in Vite. The template uses Vue 3 `<script setup>` SFCs, check out the [script setup docs](https://v3.vuejs.org/api/sfc-script-setup.html#sfc-script-setup) to learn more.
+
+Learn more about the recommended Project Setup and IDE Support in the [Vue Docs TypeScript Guide](https://vuejs.org/guide/typescript/overview.html#project-setup).

webapp/index.html

@@ -0,0 +1,13 @@
+<!doctype html>
+<html lang="en">
+  <head>
+    <meta charset="UTF-8" />
+    <link rel="icon" type="image/svg+xml" href="/favicon.svg" />
+    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
+    <title>webapp</title>
+  </head>
+  <body>
+    <div id="app"></div>
+    <script type="module" src="/src/main.ts"></script>
+  </body>
+</html>

webapp/package.json

@@ -0,0 +1,26 @@
+{
+  "name": "webapp",
+  "private": true,
+  "version": "0.0.0",
+  "type": "module",
+  "scripts": {
+    "dev": "vite",
+    "build": "vue-tsc -b && vite build",
+    "preview": "vite preview"
+  },
+  "dependencies": {
+    "@vueuse/core": "^14.4.0",
+    "axios": "^1.20.0",
+    "pinia": "^4.0.3",
+    "vue": "^3.5.41",
+    "vue-router": "^4.6.4"
+  },
+  "devDependencies": {
+    "@types/node": "^24.13.3",
+    "@vitejs/plugin-vue": "^6.0.8",
+    "@vue/tsconfig": "^0.9.1",
+    "typescript": "~6.0.2",
+    "vite": "^8.2.2",
+    "vue-tsc": "^3.3.11"
+  }
+}

webapp/public/icons.svg

@@ -0,0 +1,24 @@
+<svg xmlns="http://www.w3.org/2000/svg">
+  <symbol id="bluesky-icon" viewBox="0 0 16 17">
+    <g clip-path="url(#bluesky-clip)"><path fill="#08060d" d="M7.75 7.735c-.693-1.348-2.58-3.86-4.334-5.097-1.68-1.187-2.32-.981-2.74-.79C.188 2.065.1 2.812.1 3.251s.241 3.602.398 4.13c.52 1.744 2.367 2.333 4.07 2.145-2.495.37-4.71 1.278-1.805 4.512 3.196 3.309 4.38-.71 4.987-2.746.608 2.036 1.307 5.91 4.93 2.746 2.72-2.746.747-4.143-1.747-4.512 1.702.189 3.55-.4 4.07-2.145.156-.528.397-3.691.397-4.13s-.088-1.186-.575-1.406c-.42-.19-1.06-.395-2.741.79-1.755 1.24-3.64 3.752-4.334 5.099"/></g>
+    <defs><clipPath id="bluesky-clip"><path fill="#fff" d="M.1.85h15.3v15.3H.1z"/></clipPath></defs>
+  </symbol>
+  <symbol id="discord-icon" viewBox="0 0 20 19">
+    <path fill="#08060d" d="M16.224 3.768a14.5 14.5 0 0 0-3.67-1.153c-.158.286-.343.67-.47.976a13.5 13.5 0 0 0-4.067 0c-.128-.306-.317-.69-.476-.976A14.4 14.4 0 0 0 3.868 3.77C1.546 7.28.916 10.703 1.231 14.077a14.7 14.7 0 0 0 4.5 2.306q.545-.748.965-1.587a9.5 9.5 0 0 1-1.518-.74q.191-.14.372-.293c2.927 1.369 6.107 1.369 8.999 0q.183.152.372.294-.723.437-1.52.74.418.838.963 1.588a14.6 14.6 0 0 0 4.504-2.308c.37-3.911-.63-7.302-2.644-10.309m-9.13 8.234c-.878 0-1.599-.82-1.599-1.82 0-.998.705-1.82 1.6-1.82.894 0 1.614.82 1.599 1.82.001 1-.705 1.82-1.6 1.82m5.91 0c-.878 0-1.599-.82-1.599-1.82 0-.998.705-1.82 1.6-1.82.893 0 1.614.82 1.599 1.82 0 1-.706 1.82-1.6 1.82"/>
+  </symbol>
+  <symbol id="documentation-icon" viewBox="0 0 21 20">
+    <path fill="none" stroke="#aa3bff" stroke-linecap="round" stroke-linejoin="round" stroke-width="1.35" d="m15.5 13.333 1.533 1.322c.645.555.967.833.967 1.178s-.322.623-.967 1.179L15.5 18.333m-3.333-5-1.534 1.322c-.644.555-.966.833-.966 1.178s.322.623.966 1.179l1.534 1.321"/>
+    <path fill="none" stroke="#aa3bff" stroke-linecap="round" stroke-linejoin="round" stroke-width="1.35" d="M17.167 10.836v-4.32c0-1.41 0-2.117-.224-2.68-.359-.906-1.118-1.621-2.08-1.96-.599-.21-1.349-.21-2.848-.21-2.623 0-3.935 0-4.983.369-1.684.591-3.013 1.842-3.641 3.428C3 6.449 3 7.684 3 10.154v2.122c0 2.558 0 3.838.706 4.726q.306.383.713.671c.76.536 1.79.64 3.581.66"/>
+    <path fill="none" stroke="#aa3bff" stroke-linecap="round" stroke-linejoin="round" stroke-width="1.35" d="M3 10a2.78 2.78 0 0 1 2.778-2.778c.555 0 1.209.097 1.748-.047.48-.129.854-.503.982-.982.145-.54.048-1.194.048-1.749a2.78 2.78 0 0 1 2.777-2.777"/>
+  </symbol>
+  <symbol id="github-icon" viewBox="0 0 19 19">
+    <path fill="#08060d" fill-rule="evenodd" d="M9.356 1.85C5.05 1.85 1.57 5.356 1.57 9.694a7.84 7.84 0 0 0 5.324 7.44c.387.079.528-.168.528-.376 0-.182-.013-.805-.013-1.454-2.165.467-2.616-.935-2.616-.935-.349-.91-.864-1.143-.864-1.143-.71-.48.051-.48.051-.48.787.051 1.2.805 1.2.805.695 1.194 1.817.857 2.268.649.064-.507.27-.857.49-1.052-1.728-.182-3.545-.857-3.545-3.87 0-.857.31-1.558.8-2.104-.078-.195-.349-1 .077-2.078 0 0 .657-.208 2.14.805a7.5 7.5 0 0 1 1.946-.26c.657 0 1.328.092 1.946.26 1.483-1.013 2.14-.805 2.14-.805.426 1.078.155 1.883.078 2.078.502.546.799 1.247.799 2.104 0 3.013-1.818 3.675-3.558 3.87.284.247.528.714.528 1.454 0 1.052-.012 1.896-.012 2.156 0 .208.142.455.528.377a7.84 7.84 0 0 0 5.324-7.441c.013-4.338-3.48-7.844-7.773-7.844" clip-rule="evenodd"/>
+  </symbol>
+  <symbol id="social-icon" viewBox="0 0 20 20">
+    <path fill="none" stroke="#aa3bff" stroke-linecap="round" stroke-linejoin="round" stroke-width="1.35" d="M12.5 6.667a4.167 4.167 0 1 0-8.334 0 4.167 4.167 0 0 0 8.334 0"/>
+    <path fill="none" stroke="#aa3bff" stroke-linecap="round" stroke-linejoin="round" stroke-width="1.35" d="M2.5 16.667a5.833 5.833 0 0 1 8.75-5.053m3.837.474.513 1.035c.07.144.257.282.414.309l.93.155c.596.1.736.536.307.965l-.723.73a.64.64 0 0 0-.152.531l.207.903c.164.715-.213.991-.84.618l-.872-.52a.63.63 0 0 0-.577 0l-.872.52c-.624.373-1.003.094-.84-.618l.207-.903a.64.64 0 0 0-.152-.532l-.723-.729c-.426-.43-.289-.864.306-.964l.93-.156a.64.64 0 0 0 .412-.31l.513-1.034c.28-.562.735-.562 1.012 0"/>
+  </symbol>
+  <symbol id="x-icon" viewBox="0 0 19 19">
+    <path fill="#08060d" fill-rule="evenodd" d="M1.893 1.98c.052.072 1.245 1.769 2.653 3.77l2.892 4.114c.183.261.333.48.333.486s-.068.089-.152.183l-.522.593-.765.867-3.597 4.087c-.375.426-.734.834-.798.905a1 1 0 0 0-.118.148c0 .01.236.017.664.017h.663l.729-.83c.4-.457.796-.906.879-.999a692 692 0 0 0 1.794-2.038c.034-.037.301-.34.594-.675l.551-.624.345-.392a7 7 0 0 1 .34-.374c.006 0 .93 1.306 2.052 2.903l2.084 2.965.045.063h2.275c1.87 0 2.273-.003 2.266-.021-.008-.02-1.098-1.572-3.894-5.547-2.013-2.862-2.28-3.246-2.273-3.266.008-.019.282-.332 2.085-2.38l2-2.274 1.567-1.782c.022-.028-.016-.03-.65-.03h-.674l-.3.342a871 871 0 0 1-1.782 2.025c-.067.075-.405.458-.75.852a100 100 0 0 1-.803.91c-.148.172-.299.344-.99 1.127-.304.343-.32.358-.345.327-.015-.019-.904-1.282-1.976-2.808L6.365 1.85H1.8zm1.782.91 8.078 11.294c.772 1.08 1.413 1.973 1.425 1.984.016.017.241.02 1.05.017l1.03-.004-2.694-3.766L7.796 5.75 5.722 2.852l-1.039-.004-1.039-.004z" clip-rule="evenodd"/>
+  </symbol>
+</svg>

webapp/src/App.vue

@@ -0,0 +1,86 @@
+<template>
+  <div class="app">
+    <!-- 顶部导航栏 -->
+    <nav class="topbar">
+      <span class="brand">🤖 端云协同 LLM 系统</span>
+      <div class="nav-links">
+        <router-link to="/chat">💬 对话</router-link>
+        <router-link to="/collaboration">🔄 协作</router-link>
+        <router-link to="/review">🔍 检验</router-link>
+        <router-link to="/metrics">📊 指标</router-link>
+        <router-link to="/settings">⚙️ 设置</router-link>
+      </div>
+    </nav>
+
+    <!-- 路由视图 -->
+    <router-view class="content" />
+  </div>
+</template>
+
+<script setup lang="ts">
+// App.vue — 根布局，导航栏 + 路由出口
+</script>
+
+<style>
+* { box-sizing: border-box; margin: 0; padding: 0; }
+
+html, body, #app {
+  height: 100%;
+  font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif;
+  font-size: 14px;
+  color: #111;
+  background: #fff;
+}
+
+.app {
+  display: flex;
+  flex-direction: column;
+  height: 100%;
+}
+
+.topbar {
+  display: flex;
+  align-items: center;
+  gap: 24px;
+  padding: 0 24px;
+  height: 52px;
+  border-bottom: 1px solid #e5e7eb;
+  background: #fff;
+  flex-shrink: 0;
+}
+
+.brand {
+  font-size: 16px;
+  font-weight: 700;
+  color: #1e40af;
+  white-space: nowrap;
+}
+
+.nav-links {
+  display: flex;
+  gap: 4px;
+}
+
+.nav-links a {
+  padding: 6px 14px;
+  border-radius: 6px;
+  text-decoration: none;
+  color: #374151;
+  font-size: 14px;
+  font-weight: 500;
+  transition: background 0.15s;
+}
+.nav-links a:hover { background: #f3f4f6; }
+.nav-links a.router-link-active {
+  background: #eff6ff;
+  color: #2563eb;
+}
+
+.content {
+  flex: 1;
+  overflow: hidden;   /* 限制自己高度，不被子内容撑开 */
+  display: flex;
+  flex-direction: column;
+  min-height: 0;
+}
+</style>

webapp/src/api/index.ts

@@ -0,0 +1,262 @@
+import axios from 'axios'
+import type {
+  ChatInitResponse,
+  RunStatus,
+  SSEEvent,
+  ReviewItem,
+  Metrics,
+} from '@/types'
+
+const http = axios.create({ timeout: 10_000 })
+
+// ── Chat ─────────────────────────────────────────────────────────────────────
+
+/** POST /chat：异步提交，立即返回 request_id */
+export async function chat(query: string, domain_group?: string) {
+  const { data } = await http.post<ChatInitResponse>('/chat', { query, domain_group })
+  return data
+}
+
+/** GET /runs/{id}/status：查询任务状态 */
+export async function getRunStatus(requestId: string) {
+  const { data } = await http.get<RunStatus>(`/runs/${requestId}/status`)
+  return data
+}
+
+/** GET /runs/{id}/workspace：获取交流文本全文 */
+export async function getWorkspace(requestId: string) {
+  const { data } = await http.get(`/runs/${requestId}/workspace`)
+  return data
+}
+
+/** SSE /runs/{id}/stream：订阅实时协作事件 */
+export function watchRun(requestId: string) {
+  const es = new EventSource(`/runs/${requestId}/stream`)
+  return {
+    /** 监听器会在组件卸载时自动断开，调用者只需提供 onXxx 回调 */
+    subscribe(opts: {
+      onWorkspace: (ws: import('@/types').Workspace) => void
+      onStatus: (s: string) => void
+      onError: (detail: string) => void
+      onDone: () => void
+    }) {
+      es.addEventListener('message', (ev) => {
+        const d: SSEEvent = JSON.parse(ev.data)
+        if (d.type === 'workspace') opts.onWorkspace(d.workspace)
+        else if (d.type === 'status') {
+          opts.onStatus(d.value)
+          if (d.value === 'done' || d.value === 'failed') opts.onDone()
+        }
+        else if (d.type === 'error') opts.onError(d.detail)
+      })
+      es.addEventListener('error', () => {
+        // EventSource 自己会重连，这里只记录
+      })
+    },
+    close() { es.close() },
+  }
+}
+
+// ── 人工检验 ─────────────────────────────────────────────────────────────────
+
+export async function listReviews(status?: string) {
+  const params = status ? { status } : {}
+  const { data } = await http.get<ReviewItem[]>('/review/queue', { params })
+  return data
+}
+
+export async function submitReview(reviewId: number, verdict: string, correction?: string) {
+  const { data } = await http.post(`/review/${reviewId}`, null, {
+    params: { verdict, correction },
+  })
+  return data
+}
+
+// ── 指标 ─────────────────────────────────────────────────────────────────────
+
+export async function getMetrics() {
+  const { data } = await http.get<Metrics>('/api/metrics')
+  return data
+}
+
+// ── 模型设置 ─────────────────────────────────────────────────────────────────
+
+export interface ModelSettings {
+  worker: {
+    backend: string
+    model: string
+    base_url: string
+    port: number
+    temperature: number
+    max_fix_attempts: number
+    code_timeout_s: number
+  }
+  architect: {
+    model: string
+    base_url: string
+    api_key: string
+  }
+  pipeline: {
+    fast_path: boolean
+    rounds_cap: number
+    api_token_cap: number
+    breach_policy: string
+  }
+}
+
+/** GET /config：读取当前模型设置 */
+export async function getConfig() {
+  const { data } = await http.get<ModelSettings>('/config')
+  return data
+}
+
+/** PUT /config：更新模型设置（部分更新） */
+export async function updateConfig(patch: Partial<ModelSettings>) {
+  const { data } = await http.put<ModelSettings>('/config', patch)
+  return data
+}
+
+/** POST /config/reset：恢复默认设置 */
+export async function resetConfig() {
+  const { data } = await http.post<ModelSettings>('/config/reset')
+  return data
+}
+
+// ── 模型发现 & 连接验证 ───────────────────────────────────────────────────────
+
+export interface ModelInfo {
+  id: string
+  name: string
+}
+
+/**
+ * GET /config/models?backend=llama_server&base_url=...&api_key=...
+ * 返回 { models: ModelInfo[] } 或 { error: string }
+ */
+export async function listModels(
+  backend: string,
+  base_url: string,
+  api_key: string,
+) {
+  const params: Record<string, string> = { backend }
+  if (base_url) params.base_url = base_url
+  if (api_key) params.api_key = api_key
+  const { data } = await http.get<{ models?: ModelInfo[]; error?: string }>(
+    '/config/models', { params },
+  )
+  return data
+}
+
+/**
+ * GET /config/ping?backend=...&base_url=...&api_key=...
+ * 返回 { ok: boolean, detail?: string }
+ */
+export async function pingBackend(
+  backend: string,
+  base_url: string,
+  api_key: string,
+) {
+  const params: Record<string, string> = { backend }
+  if (base_url) params.base_url = base_url
+  if (api_key) params.api_key = api_key
+  const { data } = await http.get<{ ok: boolean; detail?: string; status_code?: number }>(
+    '/config/ping', { params },
+  )
+  return data
+}
+
+// ── llama-server 内置管理 ─────────────────────────────────────────────────────
+
+export interface LlamaStatus {
+  running: boolean
+  pid?: number | null
+  model?: string | null
+  port?: number | null
+  base_url?: string | null
+  started_at?: number | null
+  error?: string | null
+}
+
+export interface LocalModel {
+  id: string
+  name: string
+  size_mb: number
+  path: string
+}
+
+export interface DownloadProgress {
+  url: string
+  dest: string
+  downloaded_bytes: number
+  total_bytes?: number | null
+  progress_pct: number
+  speed: string
+  eta: string
+  done: boolean
+  error?: string | null
+}
+
+/** GET /llama/status */
+export async function getLlamaStatus() {
+  const { data } = await http.get<LlamaStatus>('/llama/status')
+  return data
+}
+
+/** GET /llama/models */
+export async function listLocalModels() {
+  const { data } = await http.get<{ models: LocalModel[] }>('/llama/models')
+  return data.models
+}
+
+/** POST /llama/start */
+export async function startLlama(model: string, port = 8901, ngl = 99, ctx = 4096) {
+  const { data } = await http.post<LlamaStatus>('/llama/start', null, {
+    params: { model, port, ngl, ctx },
+  })
+  return data
+}
+
+/** POST /llama/stop */
+export async function stopLlama() {
+  const { data } = await http.post<{ running: boolean }>('/llama/stop')
+  return data
+}
+
+/** POST /llama/download */
+export async function downloadModel(url: string, dest?: string) {
+  const params: Record<string, string> = { url }
+  if (dest) params.dest = dest
+  const { data } = await http.post<DownloadProgress>('/llama/download', null, { params })
+  return data
+}
+
+/** SSE /llama/download/stream?url=... */
+export function watchDownload(url: string) {
+  const es = new EventSource(`/llama/download/stream?url=${encodeURIComponent(url)}`)
+  return {
+    subscribe(opts: { onProgress: (p: DownloadProgress) => void; onDone: (p: DownloadProgress) => void }) {
+      es.addEventListener('message', (ev) => {
+        const p: DownloadProgress = JSON.parse(ev.data)
+        opts.onProgress(p)
+        if (p.done) opts.onDone(p)
+      })
+      es.addEventListener('error', () => { es.close() })
+    },
+    close() { es.close() },
+  }
+}
+
+// ── 辅助：轮询直到完成（用于不需要 SSE 的场景）──────────────────────────────────
+
+export async function pollUntilDone(
+  requestId: string,
+  { timeout = 60_000, interval = 500 }: { timeout?: number; interval?: number } = {},
+) {
+  const t0 = Date.now()
+  while (Date.now() - t0 < timeout) {
+    const s = await getRunStatus(requestId)
+    if (s.status === 'done' || s.status === 'failed') return s
+    await new Promise((r) => setTimeout(r, interval))
+  }
+  throw new Error('poll timeout')
+}

webapp/src/assets/vue.svg

@@ -0,0 +1 @@
+<svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" class="iconify iconify--logos" width="37.07" height="36" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 198"><path fill="#41B883" d="M204.8 0H256L128 220.8L0 0h97.92L128 51.2L157.44 0h47.36Z"></path><path fill="#41B883" d="m0 0l128 220.8L256 0h-51.2L128 132.48L50.56 0H0Z"></path><path fill="#35495E" d="M50.56 0L128 133.12L204.8 0h-47.36L128 51.2L97.92 0H50.56Z"></path></svg>

webapp/src/components/HelloWorld.vue

@@ -0,0 +1,95 @@
+<script setup lang="ts">
+import { ref } from 'vue'
+import heroImg from '../assets/hero.png'
+import viteLogo from '../assets/vite.svg'
+import vueLogo from '../assets/vue.svg'
+
+const count = ref(0)
+</script>
+
+<template>
+  <section id="center">
+    <div class="hero">
+      <img :src="heroImg" class="base" width="170" height="179" alt="" />
+      <img :src="vueLogo" class="framework" alt="Vue logo" />
+      <img :src="viteLogo" class="vite" alt="Vite logo" />
+    </div>
+    <div>
+      <h1>Get started</h1>
+      <p>Edit <code>src/App.vue</code> and save to test <code>HMR</code></p>
+    </div>
+    <button type="button" class="counter" @click="count++">
+      Count is {{ count }}
+    </button>
+  </section>
+
+  <div class="ticks"></div>
+
+  <section id="next-steps">
+    <div id="docs">
+      <svg class="icon" role="presentation" aria-hidden="true">
+        <use href="/icons.svg#documentation-icon"></use>
+      </svg>
+      <h2>Documentation</h2>
+      <p>Your questions, answered</p>
+      <ul>
+        <li>
+          <a href="https://vite.dev/" target="_blank">
+            <img class="logo" :src="viteLogo" alt="" />
+            Explore Vite
+          </a>
+        </li>
+        <li>
+          <a href="https://vuejs.org/" target="_blank">
+            <img class="button-icon" :src="vueLogo" alt="" />
+            Learn more
+          </a>
+        </li>
+      </ul>
+    </div>
+    <div id="social">
+      <svg class="icon" role="presentation" aria-hidden="true">
+        <use href="/icons.svg#social-icon"></use>
+      </svg>
+      <h2>Connect with us</h2>
+      <p>Join the Vite community</p>
+      <ul>
+        <li>
+          <a href="https://github.com/vitejs/vite" target="_blank">
+            <svg class="button-icon" role="presentation" aria-hidden="true">
+              <use href="/icons.svg#github-icon"></use>
+            </svg>
+            GitHub
+          </a>
+        </li>
+        <li>
+          <a href="https://chat.vite.dev/" target="_blank">
+            <svg class="button-icon" role="presentation" aria-hidden="true">
+              <use href="/icons.svg#discord-icon"></use>
+            </svg>
+            Discord
+          </a>
+        </li>
+        <li>
+          <a href="https://x.com/vite_js" target="_blank">
+            <svg class="button-icon" role="presentation" aria-hidden="true">
+              <use href="/icons.svg#x-icon"></use>
+            </svg>
+            X.com
+          </a>
+        </li>
+        <li>
+          <a href="https://bsky.app/profile/vite.dev" target="_blank">
+            <svg class="button-icon" role="presentation" aria-hidden="true">
+              <use href="/icons.svg#bluesky-icon"></use>
+            </svg>
+            Bluesky
+          </a>
+        </li>
+      </ul>
+    </div>
+  </section>
+
+  <div class="ticks"></div>
+  <section id="spacer"></section>
+</template>

webapp/src/main.ts

@@ -0,0 +1,9 @@
+import { createApp } from 'vue'
+import { createPinia } from 'pinia'
+import App from './App.vue'
+import router from './router'
+
+const app = createApp(App)
+app.use(createPinia())
+app.use(router)
+app.mount('#app')

webapp/src/router/index.ts

@@ -0,0 +1,41 @@
+import { createRouter, createWebHistory } from 'vue-router'
+import ChatView from '@/views/ChatView.vue'
+
+// Vite 构建时 base=/static/，但 FastAPI 在 /metrics 等路径提供 SPA（不在 /static/ 下），
+// 所以 history 固定用 '/'，避免 Vue Router 把 /metrics 当作 /static/metrics 解析导致路由不匹配。
+const router = createRouter({
+  history: createWebHistory('/'),
+  routes: [
+    {
+      path: '/',
+      redirect: '/chat',
+    },
+    {
+      path: '/chat',
+      name: 'chat',
+      component: ChatView,
+    },
+    {
+      path: '/collaboration',
+      name: 'collaboration',
+      component: () => import('@/views/CollaborationView.vue'),
+    },
+    {
+      path: '/review',
+      name: 'review',
+      component: () => import('@/views/ReviewView.vue'),
+    },
+    {
+      path: '/metrics',
+      name: 'metrics',
+      component: () => import('@/views/MetricsView.vue'),
+    },
+    {
+      path: '/settings',
+      name: 'settings',
+      component: () => import('@/views/SettingsView.vue'),
+    },
+  ],
+})
+
+export default router

webapp/src/stores/chat.ts

@@ -0,0 +1,61 @@
+import { defineStore } from 'pinia'
+import { ref } from 'vue'
+import { chat, getRunStatus, pollUntilDone } from '@/api'
+import type { ChatInitResponse, RunStatus, Workspace } from '@/types'
+
+// 单次会话记录
+export interface ChatSession {
+  requestId: string
+  query: string
+  init: ChatInitResponse
+  status: RunStatus | null
+  workspace: Workspace | null
+  error: string | null
+}
+
+export const useChatStore = defineStore('chat', () => {
+  const sessions = ref<ChatSession[]>([])
+  const currentId = ref<string | null>(null)
+
+  const current = () => sessions.value.find((s) => s.requestId === currentId.value) ?? null
+
+  async function sendQuery(query: string) {
+    const init = await chat(query)
+    const session: ChatSession = {
+      requestId: init.request_id,
+      query,
+      init,
+      status: null,
+      workspace: null,
+      error: null,
+    }
+    sessions.value.unshift(session)
+    currentId.value = init.request_id
+    return session
+  }
+
+  async function pollStatus(requestId: string) {
+    const s = await getRunStatus(requestId)
+    const session = sessions.value.find((x) => x.requestId === requestId)
+    if (session) session.status = s
+    return s
+  }
+
+  async function waitForDone(requestId: string) {
+    const s = await pollUntilDone(requestId)
+    const session = sessions.value.find((x) => x.requestId === requestId)
+    if (session) session.status = s
+    return s
+  }
+
+  function updateWorkspace(requestId: string, ws: Workspace) {
+    const session = sessions.value.find((x) => x.requestId === requestId)
+    if (session) session.workspace = ws
+  }
+
+  function setCurrent(id: string | null) {
+    currentId.value = id
+  }
+
+  return { sessions, currentId, current, sendQuery, pollStatus, waitForDone, updateWorkspace, setCurrent }
+})

webapp/src/style.css

@@ -0,0 +1,296 @@
+:root {
+  --text: #6b6375;
+  --text-h: #08060d;
+  --bg: #fff;
+  --border: #e5e4e7;
+  --code-bg: #f4f3ec;
+  --accent: #aa3bff;
+  --accent-bg: rgba(170, 59, 255, 0.1);
+  --accent-border: rgba(170, 59, 255, 0.5);
+  --social-bg: rgba(244, 243, 236, 0.5);
+  --shadow:
+    rgba(0, 0, 0, 0.1) 0 10px 15px -3px, rgba(0, 0, 0, 0.05) 0 4px 6px -2px;
+
+  --sans: system-ui, 'Segoe UI', Roboto, sans-serif;
+  --heading: system-ui, 'Segoe UI', Roboto, sans-serif;
+  --mono: ui-monospace, Consolas, monospace;
+
+  font: 18px/145% var(--sans);
+  letter-spacing: 0.18px;
+  color-scheme: light dark;
+  color: var(--text);
+  background: var(--bg);
+  font-synthesis: none;
+  text-rendering: optimizeLegibility;
+  -webkit-font-smoothing: antialiased;
+  -moz-osx-font-smoothing: grayscale;
+
+  @media (max-width: 1024px) {
+    font-size: 16px;
+  }
+}
+
+@media (prefers-color-scheme: dark) {
+  :root {
+    --text: #9ca3af;
+    --text-h: #f3f4f6;
+    --bg: #16171d;
+    --border: #2e303a;
+    --code-bg: #1f2028;
+    --accent: #c084fc;
+    --accent-bg: rgba(192, 132, 252, 0.15);
+    --accent-border: rgba(192, 132, 252, 0.5);
+    --social-bg: rgba(47, 48, 58, 0.5);
+    --shadow:
+      rgba(0, 0, 0, 0.4) 0 10px 15px -3px, rgba(0, 0, 0, 0.25) 0 4px 6px -2px;
+  }
+
+  #social .button-icon {
+    filter: invert(1) brightness(2);
+  }
+}
+
+body {
+  margin: 0;
+}
+
+h1,
+h2 {
+  font-family: var(--heading);
+  font-weight: 500;
+  color: var(--text-h);
+}
+
+h1 {
+  font-size: 56px;
+  letter-spacing: -1.68px;
+  margin: 32px 0;
+  @media (max-width: 1024px) {
+    font-size: 36px;
+    margin: 20px 0;
+  }
+}
+h2 {
+  font-size: 24px;
+  line-height: 118%;
+  letter-spacing: -0.24px;
+  margin: 0 0 8px;
+  @media (max-width: 1024px) {
+    font-size: 20px;
+  }
+}
+p {
+  margin: 0;
+}
+
+code,
+.counter {
+  font-family: var(--mono);
+  display: inline-flex;
+  border-radius: 4px;
+  color: var(--text-h);
+}
+
+code {
+  font-size: 15px;
+  line-height: 135%;
+  padding: 4px 8px;
+  background: var(--code-bg);
+}
+
+.counter {
+  font-size: 16px;
+  padding: 5px 10px;
+  border-radius: 5px;
+  color: var(--accent);
+  background: var(--accent-bg);
+  border: 2px solid transparent;
+  transition: border-color 0.3s;
+  margin-bottom: 24px;
+
+  &:hover {
+    border-color: var(--accent-border);
+  }
+  &:focus-visible {
+    outline: 2px solid var(--accent);
+    outline-offset: 2px;
+  }
+}
+
+.hero {
+  position: relative;
+
+  .base,
+  .framework,
+  .vite {
+    inset-inline: 0;
+    margin: 0 auto;
+  }
+
+  .base {
+    width: 170px;
+    position: relative;
+    z-index: 0;
+  }
+
+  .framework,
+  .vite {
+    position: absolute;
+  }
+
+  .framework {
+    z-index: 1;
+    top: 34px;
+    height: 28px;
+    transform: perspective(2000px) rotateZ(300deg) rotateX(44deg) rotateY(39deg)
+      scale(1.4);
+  }
+
+  .vite {
+    z-index: 0;
+    top: 107px;
+    height: 26px;
+    width: auto;
+    transform: perspective(2000px) rotateZ(300deg) rotateX(40deg) rotateY(39deg)
+      scale(0.8);
+  }
+}
+
+#app {
+  width: 1126px;
+  max-width: 100%;
+  margin: 0 auto;
+  text-align: center;
+  border-inline: 1px solid var(--border);
+  min-height: 100svh;
+  display: flex;
+  flex-direction: column;
+  box-sizing: border-box;
+}
+
+#center {
+  display: flex;
+  flex-direction: column;
+  gap: 25px;
+  place-content: center;
+  place-items: center;
+  flex-grow: 1;
+
+  @media (max-width: 1024px) {
+    padding: 32px 20px 24px;
+    gap: 18px;
+  }
+}
+
+#next-steps {
+  display: flex;
+  border-top: 1px solid var(--border);
+  text-align: left;
+
+  & > div {
+    flex: 1 1 0;
+    padding: 32px;
+    @media (max-width: 1024px) {
+      padding: 24px 20px;
+    }
+  }
+
+  .icon {
+    margin-bottom: 16px;
+    width: 22px;
+    height: 22px;
+  }
+
+  @media (max-width: 1024px) {
+    flex-direction: column;
+    text-align: center;
+  }
+}
+
+#docs {
+  border-right: 1px solid var(--border);
+
+  @media (max-width: 1024px) {
+    border-right: none;
+    border-bottom: 1px solid var(--border);
+  }
+}
+
+#next-steps ul {
+  list-style: none;
+  padding: 0;
+  display: flex;
+  gap: 8px;
+  margin: 32px 0 0;
+
+  .logo {
+    height: 18px;
+  }
+
+  a {
+    color: var(--text-h);
+    font-size: 16px;
+    border-radius: 6px;
+    background: var(--social-bg);
+    display: flex;
+    padding: 6px 12px;
+    align-items: center;
+    gap: 8px;
+    text-decoration: none;
+    transition: box-shadow 0.3s;
+
+    &:hover {
+      box-shadow: var(--shadow);
+    }
+    .button-icon {
+      height: 18px;
+      width: 18px;
+    }
+  }
+
+  @media (max-width: 1024px) {
+    margin-top: 20px;
+    flex-wrap: wrap;
+    justify-content: center;
+
+    li {
+      flex: 1 1 calc(50% - 8px);
+    }
+
+    a {
+      width: 100%;
+      justify-content: center;
+      box-sizing: border-box;
+    }
+  }
+}
+
+#spacer {
+  height: 88px;
+  border-top: 1px solid var(--border);
+  @media (max-width: 1024px) {
+    height: 48px;
+  }
+}
+
+.ticks {
+  position: relative;
+  width: 100%;
+
+  &::before,
+  &::after {
+    content: '';
+    position: absolute;
+    top: -4.5px;
+    border: 5px solid transparent;
+  }
+
+  &::before {
+    left: 0;
+    border-left-color: var(--border);
+  }
+  &::after {
+    right: 0;
+    border-right-color: var(--border);
+  }
+}

webapp/src/types/index.ts

@@ -0,0 +1,117 @@
+// ── 协作任务状态 ──────────────────────────────────────────────────────────────
+
+/** POST /chat 立即返回 */
+export interface ChatInitResponse {
+  request_id: string
+  status: 'pending'
+}
+
+/** GET /runs/{id}/status 返回 */
+export interface RunStatus {
+  request_id: string
+  status: 'pending' | 'running' | 'done' | 'failed'   // JobStore 任务状态
+  ws_state: string | null                              // workspace.json 内 state
+  started_at: number
+  finished_at: number
+  error: string | null
+  // PipelineResult 等效字段（扁平时 TaskInfo）
+  response: string | null
+  pipeline_status: string | null   // done | fast_path | escalated | failed
+  fast_path: boolean
+  rounds_used: number
+  api_input_tokens: number
+  api_output_tokens: number
+  cost_est: number
+  model_used: string | null
+  latency_ms: number | null
+  route: string[]
+  workspace_path: string | null
+}
+
+// ── SSE 事件 ─────────────────────────────────────────────────────────────────
+
+export type SSEEvent =
+  | { type: 'workspace'; version: number; state: string; request_id: string; workspace: Workspace }
+  | { type: 'status';    value: string;  request_id: string }
+  | { type: 'error';     detail: string }
+
+// ── 交流文本（workspace.json）────────────────────────────────────────────────
+
+export interface Workspace {
+  request_id: string
+  query: string
+  brief: Brief | null
+  plan: PlanStep[]
+  progress: ProgressStep[]
+  issues: Issue[]
+  decisions: Decision[]
+  archive: string[]
+  meta: WsMeta
+  status?: string
+  error?: string
+}
+
+export interface Brief {
+  goal: string
+  tags: string[]
+  domain: string
+  constraints: string[]
+}
+
+export interface PlanStep {
+  id: string
+  task: string
+  deps: string[]
+  status?: string
+}
+
+export interface ProgressStep {
+  step: string
+  status: 'pending' | 'running' | 'done'
+  artifact?: string
+  note?: string
+}
+
+export interface Issue {
+  id: string
+  step: string
+  description: string
+  suggestion?: string
+}
+
+export interface Decision {
+  ref: string
+  reply: string
+  patch_plan?: PlanStep[]
+}
+
+export interface WsMeta {
+  state: 'draft' | 'in_progress' | 'reviewing' | 'done' | 'failed'
+  round: number
+  rounds_cap: number
+  version?: number
+  api_input_tokens: number
+  api_output_tokens: number
+}
+
+// ── 人工检验 ─────────────────────────────────────────────────────────────────
+
+export interface ReviewItem {
+  id: number
+  request_id: string
+  query: string
+  response: string
+  verdict: 'pending' | 'approved' | 'rejected'
+  tags: string[]
+  created_at: string
+  correction?: string
+}
+
+// ── 指标 ─────────────────────────────────────────────────────────────────────
+
+export interface Metrics {
+  router: Record<string, unknown>
+  cache: Record<string, unknown>
+  v2?: Record<string, unknown>
+  review?: { pending: number; total: number }
+}

webapp/src/views/ChatView.vue

@@ -0,0 +1,338 @@
+<template>
+  <div class="chat-view">
+    <!-- 历史会话侧边栏 -->
+    <aside class="sidebar">
+      <h3>会话记录</h3>
+      <ul class="session-list">
+        <li
+          v-for="s in chatStore.sessions"
+          :key="s.requestId"
+          :class="['session-item', { active: s.requestId === chatStore.currentId }]"
+          @click="chatStore.setCurrent(s.requestId)"
+        >
+          <span class="s-query">{{ s.query.slice(0, 28) }}{{ s.query.length > 28 ? '…' : '' }}</span>
+          <span class="s-status" :class="s.status?.status ?? 'pending'">
+            {{ s.status?.status ?? 'pending' }}
+          </span>
+        </li>
+      </ul>
+    </aside>
+
+    <!-- 主聊天区 -->
+    <main class="main">
+      <div v-if="!chatStore.current()" class="empty">
+        <p>输入问题，开启端云协同协作之旅。</p>
+        <p class="hint">结果通过 SSE 实时推送，可切换「协作」页面查看交流文本可视化。</p>
+      </div>
+
+      <template v-else>
+        <!-- 用户问题 -->
+        <div class="user-msg">
+          <span class="role-label">你</span>
+          <p>{{ chatStore.current()!.query }}</p>
+        </div>
+
+        <!-- 状态指示器 -->
+        <div class="status-bar">
+          <span v-if="runStatus === 'pending'" class="badge pending">⏳ 排队中…</span>
+          <span v-else-if="runStatus === 'running'" class="badge running">
+            🔄 协作中 ({{ ws?.meta.round ?? 0 }}/{{ ws?.meta.rounds_cap ?? 6 }})
+          </span>
+          <span v-else-if="runStatus === 'done'" class="badge done">✅ 完成</span>
+          <span v-else-if="runStatus === 'failed'" class="badge failed">❌ 失败</span>
+
+          <span v-if="ws" class="route-path">
+            路由：{{ chatStore.current()!.status?.route?.join(' → ') ?? '—' }}
+          </span>
+        </div>
+
+        <!-- 协作元信息 -->
+        <div v-if="ws" class="ws-meta">
+          <span>tokens: {{ ws.meta.api_input_tokens }} in / {{ ws.meta.api_output_tokens }} out</span>
+          <span>延迟: {{ chatStore.current()!.status?.latency_ms?.toFixed(0) }} ms</span>
+          <span>模型: {{ chatStore.current()!.status?.model_used ?? '—' }}</span>
+        </div>
+
+        <!-- 响应正文 -->
+        <div v-if="response" class="assistant-msg">
+          <span class="role-label">系统</span>
+          <pre>{{ response }}</pre>
+        </div>
+
+        <!-- 交流文本预览（紧凑折叠） -->
+        <details v-if="ws" class="ws-preview">
+          <summary>📄 交流文本预览</summary>
+          <div v-if="ws.brief" class="brief-block">
+            <strong>Brief:</strong> {{ ws.brief.goal }}
+            <span class="tags">{{ ws.brief.tags.join(', ') }}</span>
+          </div>
+          <ul v-if="ws.plan?.length" class="plan-list">
+            <li v-for="p in ws.plan" :key="p.id" :class="p.status">
+              <span class="step-id">{{ p.id }}</span>
+              <span>{{ p.task }}</span>
+              <span class="deps" v-if="p.deps.length">←{{ p.deps.join(',') }}</span>
+            </li>
+          </ul>
+          <ul v-if="ws.progress?.length" class="progress-list">
+            <li v-for="pg in ws.progress" :key="pg.step" :class="pg.status">
+              {{ pg.step }}: {{ pg.status }}
+            </li>
+          </ul>
+        </details>
+
+        <!-- 错误 -->
+        <div v-if="chatStore.current()!.error" class="error-msg">
+          {{ chatStore.current()!.error }}
+        </div>
+      </template>
+
+      <!-- 输入框 -->
+      <form class="input-bar" @submit.prevent="handleSend">
+        <input
+          v-model="input"
+          placeholder="输入你的问题…"
+          :disabled="sending"
+          autofocus
+        />
+        <button type="submit" :disabled="sending || !input.trim()">
+          {{ sending ? '发送中…' : '发送' }}
+        </button>
+      </form>
+    </main>
+  </div>
+</template>
+
+<script setup lang="ts">
+import { ref, computed, watch } from 'vue'
+import { useChatStore } from '@/stores/chat'
+import { watchRun } from '@/api'
+import type { Workspace } from '@/types'
+
+const chatStore = useChatStore()
+const input = ref('')
+const sending = ref(false)
+
+const current = computed(() => chatStore.current())
+const runStatus = computed(() => current.value?.status?.status ?? 'pending')
+const ws = computed(() => current.value?.workspace)
+const response = computed(() => current.value?.status?.response ?? null)
+
+// SSE 订阅
+let sseHandle: ReturnType<typeof watchRun> | null = null
+
+function subscribeSSE(requestId: string) {
+  sseHandle?.close()
+  sseHandle = watchRun(requestId)
+  sseHandle.subscribe({
+    onWorkspace(workspace: Workspace) {
+      chatStore.updateWorkspace(requestId, workspace)
+    },
+    onStatus(_state: string) {
+      chatStore.pollStatus(requestId)
+    },
+    onError(detail: string) {
+      const s = chatStore.sessions.find((x) => x.requestId === requestId)
+      if (s) s.error = detail
+    },
+    onDone() {
+      chatStore.pollStatus(requestId)
+    },
+  })
+}
+
+watch(
+  () => chatStore.currentId,
+  (id) => {
+    if (id) {
+      subscribeSSE(id)
+      // 若已完成立即拉一次状态
+      if (current.value?.status?.status === 'done') {
+        chatStore.pollStatus(id)
+      }
+    }
+  },
+  { immediate: true },
+)
+
+async function handleSend() {
+  if (!input.value.trim() || sending.value) return
+  const query = input.value.trim()
+  input.value = ''
+  sending.value = true
+  try {
+    const session = await chatStore.sendQuery(query)
+    subscribeSSE(session.requestId)
+  } finally {
+    sending.value = false
+  }
+}
+</script>
+
+<style scoped>
+.chat-view {
+  display: flex;
+  height: 100%;
+  gap: 0;
+}
+
+.sidebar {
+  width: 240px;
+  border-right: 1px solid #e5e7eb;
+  background: #f9fafb;
+  display: flex;
+  flex-direction: column;
+  overflow: hidden;
+}
+.sidebar h3 {
+  padding: 12px 16px;
+  margin: 0;
+  font-size: 14px;
+  color: #6b7280;
+  border-bottom: 1px solid #e5e7eb;
+}
+.session-list {
+  list-style: none;
+  margin: 0;
+  padding: 8px;
+  overflow-y: auto;
+  flex: 1;
+}
+.session-item {
+  padding: 8px 10px;
+  border-radius: 6px;
+  cursor: pointer;
+  margin-bottom: 4px;
+  display: flex;
+  flex-direction: column;
+  gap: 2px;
+  font-size: 13px;
+}
+.session-item:hover { background: #e5e7eb; }
+.session-item.active { background: #dbeafe; }
+.s-query { color: #111; }
+.s-status { font-size: 11px; color: #9ca3af; }
+.s-status.done { color: #16a34a; }
+.s-status.failed { color: #dc2626; }
+.s-status.running { color: #2563eb; }
+
+.main {
+  flex: 1;
+  display: flex;
+  flex-direction: column;
+  overflow: hidden;
+  padding: 20px 24px;
+  gap: 12px;
+}
+
+.empty {
+  flex: 1;
+  display: flex;
+  flex-direction: column;
+  align-items: center;
+  justify-content: center;
+  color: #9ca3af;
+  gap: 8px;
+}
+.hint { font-size: 13px; }
+
+.user-msg, .assistant-msg {
+  background: #f3f4f6;
+  border-radius: 8px;
+  padding: 12px 16px;
+  display: flex;
+  gap: 12px;
+  font-size: 14px;
+  line-height: 1.6;
+}
+.assistant-msg { background: #eff6ff; }
+.role-label {
+  font-weight: 700;
+  font-size: 12px;
+  color: #6b7280;
+  min-width: 32px;
+}
+.user-msg pre, .assistant-msg pre {
+  margin: 0;
+  white-space: pre-wrap;
+  font-family: inherit;
+}
+
+.status-bar {
+  display: flex;
+  align-items: center;
+  gap: 12px;
+  flex-wrap: wrap;
+}
+.badge {
+  padding: 3px 10px;
+  border-radius: 99px;
+  font-size: 12px;
+  font-weight: 600;
+}
+.badge.pending { background: #f3f4f6; color: #6b7280; }
+.badge.running { background: #dbeafe; color: #2563eb; }
+.badge.done { background: #dcfce7; color: #16a34a; }
+.badge.failed { background: #fee2e2; color: #dc2626; }
+.route-path { font-size: 12px; color: #9ca3af; }
+
+.ws-meta {
+  display: flex;
+  gap: 16px;
+  font-size: 12px;
+  color: #9ca3af;
+}
+
+.ws-preview {
+  background: #fafafa;
+  border: 1px solid #e5e7eb;
+  border-radius: 8px;
+  padding: 12px;
+  font-size: 13px;
+}
+.brief-block { margin-bottom: 8px; }
+.tags { margin-left: 8px; color: #6b7280; font-size: 12px; }
+.plan-list, .progress-list {
+  list-style: none;
+  padding: 0;
+  margin: 4px 0 0;
+}
+.plan-list li, .progress-list li {
+  padding: 2px 0;
+  display: flex;
+  gap: 6px;
+}
+.step-id { font-family: monospace; color: #6b7280; min-width: 48px; }
+.deps { color: #9ca3af; font-size: 12px; }
+.done { color: #16a34a; }
+.pending { color: #9ca3af; }
+.running { color: #2563eb; }
+
+.error-msg { color: #dc2626; font-size: 13px; background: #fee2e2; padding: 8px 12px; border-radius: 6px; }
+
+.input-bar {
+  display: flex;
+  gap: 8px;
+  padding-top: 12px;
+  border-top: 1px solid #e5e7eb;
+  margin-top: auto;
+}
+.input-bar input {
+  flex: 1;
+  padding: 10px 14px;
+  border: 1px solid #d1d5db;
+  border-radius: 8px;
+  font-size: 14px;
+  outline: none;
+}
+.input-bar input:focus { border-color: #2563eb; }
+.input-bar button {
+  padding: 10px 20px;
+  background: #2563eb;
+  color: #fff;
+  border: none;
+  border-radius: 8px;
+  font-size: 14px;
+  cursor: pointer;
+}
+.input-bar button:disabled { background: #9ca3af; cursor: not-allowed; }
+</style>

webapp/src/views/CollaborationView.vue

@@ -0,0 +1,450 @@
+<template>
+  <div class="collab-view">
+    <!-- 左侧：会话选择 + 协作流程图 -->
+    <aside class="collab-sidebar">
+      <h3>协作会话</h3>
+      <ul class="session-list">
+        <li
+          v-for="s in chatStore.sessions"
+          :key="s.requestId"
+          :class="['session-item', { active: s.requestId === activeId }]"
+          @click="selectSession(s.requestId)"
+        >
+          <span>{{ s.query.slice(0, 24) }}{{ s.query.length > 24 ? '…' : '' }}</span>
+          <span class="meta">
+            {{ s.status?.rounds_used ?? '?' }} 轮
+            · {{ s.status?.model_used ?? '—' }}
+          </span>
+        </li>
+      </ul>
+
+      <!-- 当前会话统计 -->
+      <div v-if="currentSession" class="stats">
+        <h4>运行统计</h4>
+        <div class="stat-grid">
+          <span>输入 Token</span><b>{{ currentSession.status?.api_input_tokens ?? 0 }}</b>
+          <span>输出 Token</span><b>{{ currentSession.status?.api_output_tokens ?? 0 }}</b>
+          <span>延迟</span><b>{{ currentSession.status?.latency_ms?.toFixed(0) ?? '—' }} ms</b>
+          <span>快路径</span><b>{{ currentSession.status?.fast_path ? '是' : '否' }}</b>
+        </div>
+
+        <h4>路由路径</h4>
+        <div class="route-flow">
+          <span
+            v-for="(r, i) in currentSession.status?.route ?? []"
+            :key="i"
+            class="route-node"
+          >{{ r }}</span>
+        </div>
+      </div>
+    </aside>
+
+    <!-- 右侧：交流文本实时可视化 -->
+    <main class="collab-main">
+      <div v-if="!ws" class="empty">
+        <p>从左侧选择一个会话，或在「对话」页发起新提问。</p>
+        <p>协作过程通过 SSE 实时推送，无需刷新。</p>
+      </div>
+
+      <template v-else>
+        <!-- Brief 阶段 -->
+        <section class="section brief-section">
+          <h2 class="section-title">📋 Brief（任务简报）</h2>
+          <div class="brief-card">
+            <div class="goal">{{ ws.brief?.goal ?? '(未生成)' }}</div>
+            <div class="tags">
+              <span v-for="t in ws.brief?.tags ?? []" :key="t" class="tag">{{ t }}</span>
+              <span v-if="ws.brief?.domain" class="domain">{{ ws.brief.domain }}</span>
+            </div>
+            <ul v-if="ws.brief?.constraints?.length" class="constraints">
+              <li v-for="(c, i) in ws.brief.constraints" :key="i">{{ c }}</li>
+            </ul>
+          </div>
+        </section>
+
+        <!-- Plan 阶段 -->
+        <section class="section plan-section">
+          <h2 class="section-title">📌 Plan（执行计划）</h2>
+          <div class="plan-timeline">
+            <div
+              v-for="step in ws.plan ?? []"
+              :key="step.id"
+              :class="['plan-step', getStepStatus(step.id)]"
+            >
+              <div class="step-dot" />
+              <div class="step-content">
+                <div class="step-header">
+                  <span class="step-id">{{ step.id }}</span>
+                  <span class="step-status">{{ getStepStatus(step.id) }}</span>
+                </div>
+                <p class="step-task">{{ step.task }}</p>
+                <div v-if="step.deps.length" class="step-deps">
+                  依赖：<span v-for="d in step.deps" :key="d" class="dep">{{ d }}</span>
+                </div>
+                <!-- 步骤产出工件 -->
+                <div v-if="getStepArtifact(step.id)" class="step-artifact">
+                  <details>
+                    <summary>📄 工件内容</summary>
+                    <pre>{{ getStepArtifact(step.id) }}</pre>
+                  </details>
+                </div>
+              </div>
+            </div>
+          </div>
+        </section>
+
+        <!-- Progress 实时滚动 -->
+        <section class="section progress-section">
+          <h2 class="section-title">🔄 进度（实时）</h2>
+          <div class="progress-bar-wrap">
+            <div class="progress-label">
+              第 {{ ws.meta.round }} / {{ ws.meta.rounds_cap }} 轮
+            </div>
+            <div class="progress-bar">
+              <div
+                class="progress-fill"
+                :style="{ width: `${(ws.meta.round / ws.meta.rounds_cap) * 100}%` }"
+              />
+            </div>
+          </div>
+
+          <div class="progress-steps">
+            <div
+              v-for="pg in ws.progress ?? []"
+              :key="pg.step"
+              :class="['pg-step', pg.status]"
+            >
+              <span class="pg-icon">
+                {{ pg.status === 'done' ? '✅' : pg.status === 'running' ? '⏳' : '⭕' }}
+              </span>
+              <span>{{ pg.step }}</span>
+              <span v-if="pg.note" class="pg-note">{{ pg.note }}</span>
+            </div>
+          </div>
+        </section>
+
+        <!-- Issues + Decisions -->
+        <section v-if="ws.issues?.length" class="section issues-section">
+          <h2 class="section-title">⚠️ Issues &amp; 裁决</h2>
+          <div v-for="issue in ws.issues" :key="issue.id" class="issue-card">
+            <div class="issue-header">
+              <span class="issue-id">{{ issue.id }}</span>
+              <span class="issue-step">Step: {{ issue.step }}</span>
+            </div>
+            <p>{{ issue.description }}</p>
+            <div v-if="getDecision(issue.id)" class="decision">
+              <strong>Architect 裁决：</strong>{{ getDecision(issue.id)?.reply }}
+            </div>
+          </div>
+        </section>
+
+        <!-- Archive -->
+        <section v-if="ws.archive?.length" class="section archive-section">
+          <h2 class="section-title">📦 Archive（摘要归档）</h2>
+          <ul class="archive-list">
+            <li v-for="(item, i) in ws.archive" :key="i">{{ item }}</li>
+          </ul>
+        </section>
+
+        <!-- 最终交付 -->
+        <section v-if="ws.meta.state === 'done'" class="section deliver-section">
+          <h2 class="section-title">🎉 交付</h2>
+          <div class="response-block">
+            <pre>{{ currentSession?.status?.response ?? '(无内容)' }}</pre>
+          </div>
+        </section>
+      </template>
+    </main>
+  </div>
+</template>
+
+<script setup lang="ts">
+import { ref, computed, watch } from 'vue'
+import { useChatStore } from '@/stores/chat'
+import { watchRun } from '@/api'
+import type { Workspace } from '@/types'
+
+const chatStore = useChatStore()
+const activeId = ref<string | null>(chatStore.currentId)
+const liveWs = ref<Workspace | null>(null)
+
+const currentSession = computed(() =>
+  chatStore.sessions.find((s) => s.requestId === activeId.value) ?? null,
+)
+const ws = computed(() => liveWs.value ?? currentSession.value?.workspace ?? null)
+
+function selectSession(id: string) {
+  activeId.value = id
+  liveWs.value = null
+}
+
+// SSE 实时推送
+let sse: ReturnType<typeof watchRun> | null = null
+
+function startSSE(requestId: string) {
+  sse?.close()
+  sse = watchRun(requestId)
+  sse.subscribe({
+    onWorkspace(workspace: Workspace) {
+      if (workspace.request_id === activeId.value) {
+        liveWs.value = workspace
+        chatStore.updateWorkspace(requestId, workspace)
+      }
+    },
+    onStatus() {
+      chatStore.pollStatus(requestId)
+    },
+    onDone() {
+      chatStore.pollStatus(requestId)
+    },
+    onError() {},
+  })
+}
+
+watch(
+  () => chatStore.sessions,
+  (sessions) => {
+    if (activeId.value) {
+      const found = sessions.find((s) => s.requestId === activeId.value)
+      if (!found) activeId.value = sessions[0]?.requestId ?? null
+    }
+  },
+  { deep: true },
+)
+
+watch(
+  activeId,
+  (id) => {
+    if (!id) return
+    const s = chatStore.sessions.find((x) => x.requestId === id)
+    if (s?.workspace) liveWs.value = s.workspace
+    startSSE(id)
+  },
+  { immediate: true },
+)
+
+function getStepStatus(stepId: string) {
+  return ws.value?.progress?.find((p: import('@/types').ProgressStep) => p.step === stepId)?.status ?? 'pending'
+}
+
+function getStepArtifact(stepId: string) {
+  return ws.value?.progress?.find((p: import('@/types').ProgressStep) => p.step === stepId)?.artifact ?? null
+}
+
+function getDecision(issueId: string) {
+  return ws.value?.decisions?.find((d: import('@/types').Decision) => d.ref === issueId) ?? null
+}
+</script>
+
+<style scoped>
+.collab-view {
+  display: flex;
+  height: 100%;
+}
+
+.collab-sidebar {
+  width: 260px;
+  border-right: 1px solid #e5e7eb;
+  background: #f9fafb;
+  display: flex;
+  flex-direction: column;
+  overflow: hidden;
+}
+.collab-sidebar h3 {
+  padding: 12px 16px;
+  margin: 0;
+  font-size: 13px;
+  color: #6b7280;
+  border-bottom: 1px solid #e5e7eb;
+  font-weight: 600;
+}
+.session-list {
+  list-style: none;
+  padding: 8px;
+  overflow-y: auto;
+  flex: 0 0 auto;
+  max-height: 35%;
+}
+.session-item {
+  padding: 8px 10px;
+  border-radius: 6px;
+  cursor: pointer;
+  margin-bottom: 4px;
+  display: flex;
+  flex-direction: column;
+  gap: 2px;
+  font-size: 13px;
+}
+.session-item:hover { background: #e5e7eb; }
+.session-item.active { background: #dbeafe; }
+.meta { font-size: 11px; color: #9ca3af; }
+
+.stats {
+  padding: 12px 16px;
+  border-top: 1px solid #e5e7eb;
+  font-size: 12px;
+  overflow-y: auto;
+}
+.stats h4 { margin: 0 0 6px; color: #374151; font-size: 12px; }
+.stat-grid {
+  display: grid;
+  grid-template-columns: 1fr 1fr;
+  gap: 4px 8px;
+  margin-bottom: 12px;
+}
+.stat-grid span { color: #6b7280; }
+.stat-grid b { color: #111; text-align: right; }
+
+.route-flow {
+  display: flex;
+  flex-wrap: wrap;
+  gap: 4px;
+}
+.route-node {
+  background: #e0e7ff;
+  color: #3730a3;
+  padding: 2px 6px;
+  border-radius: 4px;
+  font-size: 11px;
+  font-family: monospace;
+}
+
+.collab-main {
+  flex: 1;
+  overflow-y: auto;
+  padding: 20px 28px;
+}
+
+.empty {
+  height: 100%;
+  display: flex;
+  flex-direction: column;
+  align-items: center;
+  justify-content: center;
+  color: #9ca3af;
+  gap: 8px;
+}
+
+.section { margin-bottom: 28px; }
+.section-title {
+  font-size: 16px;
+  color: #111;
+  margin: 0 0 12px;
+  padding-bottom: 6px;
+  border-bottom: 2px solid #2563eb;
+}
+
+.brief-card {
+  background: #eff6ff;
+  border: 1px solid #bfdbfe;
+  border-radius: 8px;
+  padding: 14px 16px;
+}
+.goal { font-size: 14px; margin-bottom: 8px; color: #1e40af; font-weight: 600; }
+.tags { display: flex; gap: 6px; flex-wrap: wrap; }
+.tag { background: #dbeafe; color: #1d4ed8; padding: 2px 8px; border-radius: 99px; font-size: 12px; }
+.domain { background: #fce7f3; color: #9d174d; padding: 2px 8px; border-radius: 99px; font-size: 12px; }
+.constraints { list-style: disc inside; font-size: 13px; color: #374151; margin-top: 8px; }
+
+/* Plan timeline */
+.plan-timeline { display: flex; flex-direction: column; gap: 0; }
+.plan-step {
+  display: flex;
+  gap: 12px;
+  padding-bottom: 16px;
+  position: relative;
+}
+.plan-step::before {
+  content: '';
+  position: absolute;
+  left: 6px;
+  top: 16px;
+  bottom: 0;
+  width: 2px;
+  background: #e5e7eb;
+}
+.plan-step:last-child::before { display: none; }
+.step-dot {
+  width: 14px;
+  height: 14px;
+  border-radius: 50%;
+  border: 2px solid #d1d5db;
+  background: #fff;
+  flex-shrink: 0;
+  margin-top: 2px;
+  z-index: 1;
+}
+.plan-step.done .step-dot { background: #16a34a; border-color: #16a34a; }
+.plan-step.running .step-dot { background: #2563eb; border-color: #2563eb; }
+.plan-step.pending .step-dot { background: #fff; }
+.step-content { flex: 1; }
+.step-header { display: flex; align-items: center; gap: 8px; margin-bottom: 4px; }
+.step-id { font-family: monospace; font-size: 12px; color: #6b7280; font-weight: 700; }
+.step-status { font-size: 11px; padding: 1px 6px; border-radius: 99px; }
+.plan-step.done .step-status { background: #dcfce7; color: #16a34a; }
+.plan-step.running .step-status { background: #dbeafe; color: #2563eb; }
+.plan-step.pending .step-status { background: #f3f4f6; color: #9ca3af; }
+.step-task { margin: 0; font-size: 13px; color: #374151; }
+.step-deps { font-size: 11px; color: #9ca3af; margin-top: 2px; }
+.dep { background: #f3f4f6; padding: 0 4px; border-radius: 3px; font-family: monospace; margin-right: 4px; }
+.step-artifact { margin-top: 6px; }
+.step-artifact details { background: #fafafa; border: 1px solid #e5e7eb; border-radius: 4px; }
+.step-artifact summary { padding: 4px 8px; cursor: pointer; font-size: 12px; color: #6b7280; }
+.step-artifact pre { padding: 6px 10px; font-size: 12px; margin: 0; white-space: pre-wrap; max-height: 120px; overflow-y: auto; }
+
+/* Progress */
+.progress-bar-wrap { margin-bottom: 10px; }
+.progress-label { font-size: 12px; color: #6b7280; margin-bottom: 4px; }
+.progress-bar { height: 6px; background: #e5e7eb; border-radius: 99px; overflow: hidden; }
+.progress-fill { height: 100%; background: #2563eb; border-radius: 99px; transition: width 0.4s ease; }
+.progress-steps { display: flex; flex-wrap: wrap; gap: 8px; }
+.pg-step {
+  display: flex;
+  align-items: center;
+  gap: 4px;
+  padding: 4px 10px;
+  border-radius: 6px;
+  font-size: 12px;
+  border: 1px solid #e5e7eb;
+}
+.pg-step.done { background: #dcfce7; border-color: #bbf7d0; }
+.pg-step.running { background: #dbeafe; border-color: #bfdbfe; }
+.pg-note { color: #9ca3af; font-size: 11px; }
+
+/* Issues */
+.issue-card {
+  border: 1px solid #fca5a5;
+  background: #fff5f5;
+  border-radius: 8px;
+  padding: 12px 14px;
+  margin-bottom: 10px;
+}
+.issue-header { display: flex; gap: 8px; margin-bottom: 6px; }
+.issue-id { font-family: monospace; font-size: 12px; color: #dc2626; font-weight: 700; }
+.issue-step { font-size: 11px; color: #6b7280; }
+.decision {
+  margin-top: 8px;
+  background: #eff6ff;
+  border: 1px solid #bfdbfe;
+  border-radius: 4px;
+  padding: 6px 10px;
+  font-size: 13px;
+}
+
+/* Archive */
+.archive-list { list-style: disc inside; font-size: 13px; color: #374151; }
+
+/* Deliver */
+.response-block {
+  background: #f0fdf4;
+  border: 1px solid #86efac;
+  border-radius: 8px;
+  padding: 14px 16px;
+}
+.response-block pre {
+  margin: 0;
+  white-space: pre-wrap;
+  font-size: 14px;
+  line-height: 1.7;
+}
+</style>

webapp/src/views/MetricsView.vue

@@ -0,0 +1,171 @@
+<template>
+  <div class="metrics-view">
+    <header class="metrics-header">
+      <h2>系统指标</h2>
+      <button class="refresh" @click="load">🔄 刷新</button>
+    </header>
+
+    <div v-if="loading" class="loading">加载中…</div>
+    <div v-else-if="error" class="error">{{ error }}</div>
+
+    <template v-else-if="data">
+      <!-- 卡片网格 -->
+      <div class="card-grid">
+        <div class="metric-card">
+          <h3>路由器（v1）</h3>
+          <div class="kv-list">
+            <template v-for="(v, k) in data.router" :key="k">
+              <span>{{ k }}</span><b>{{ v }}</b>
+            </template>
+          </div>
+        </div>
+
+        <div class="metric-card">
+          <h3>缓存</h3>
+          <div class="kv-list">
+            <template v-for="(v, k) in data.cache" :key="k">
+              <span>{{ k }}</span><b>{{ v }}</b>
+            </template>
+          </div>
+        </div>
+
+        <div v-if="data.v2" class="metric-card highlight">
+          <h3>协作管线（v2）</h3>
+          <div class="kv-list">
+            <template v-for="(v, k) in data.v2" :key="k">
+              <span>{{ k }}</span><b>{{ v }}</b>
+            </template>
+          </div>
+        </div>
+
+        <div v-if="data.review" class="metric-card review-card">
+          <h3>人工检验</h3>
+          <div class="review-stats">
+            <div class="stat-item">
+              <span class="stat-num">{{ data.review.pending }}</span>
+              <span class="stat-label">待审核</span>
+            </div>
+            <div class="stat-item">
+              <span class="stat-num">{{ data.review.total }}</span>
+              <span class="stat-label">总提交</span>
+            </div>
+          </div>
+          <div v-if="data.review.total > 0" class="progress-wrap">
+            <div
+              class="reviewed-bar"
+              :style="{
+                width: `${((data.review.total - data.review.pending) / data.review.total) * 100}%`,
+              }"
+            />
+          </div>
+          <p class="review-rate">
+            通过率：
+            {{ (((data.review.total - data.review.pending) / data.review.total) * 100).toFixed(1) }}%
+          </p>
+        </div>
+      </div>
+
+      <!-- 原始 JSON -->
+      <details class="raw-json">
+        <summary>原始 JSON</summary>
+        <pre>{{ JSON.stringify(data, null, 2) }}</pre>
+      </details>
+    </template>
+  </div>
+</template>
+
+<script setup lang="ts">
+import { ref, onMounted } from 'vue'
+import { getMetrics } from '@/api'
+import type { Metrics } from '@/types'
+
+const data = ref<Metrics | null>(null)
+const loading = ref(false)
+const error = ref('')
+
+async function load() {
+  loading.value = true
+  error.value = ''
+  try {
+    data.value = await getMetrics()
+  } catch (e: unknown) {
+    // 网络超时或服务端错误：显示友好错误而不是无限 loading
+    error.value = e instanceof Error ? e.message : '指标加载失败，请检查后端服务'
+  } finally {
+    loading.value = false
+  }
+}
+
+onMounted(load)
+</script>
+
+<style scoped>
+.metrics-view { padding: 20px 24px; height: 100%; overflow-y: auto; }
+
+.metrics-header {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  margin-bottom: 20px;
+}
+.metrics-header h2 { margin: 0; font-size: 20px; }
+.refresh { padding: 6px 14px; border: 1px solid #d1d5db; border-radius: 6px; cursor: pointer; background: #fff; }
+
+.loading, .error { text-align: center; padding: 40px; color: #9ca3af; }
+.error { color: #dc2626; }
+
+.card-grid {
+  display: grid;
+  grid-template-columns: repeat(auto-fill, minmax(280px, 1fr));
+  gap: 16px;
+  margin-bottom: 24px;
+}
+
+.metric-card {
+  background: #fff;
+  border: 1px solid #e5e7eb;
+  border-radius: 10px;
+  padding: 16px;
+}
+.metric-card.highlight { border-color: #2563eb; background: #eff6ff; }
+.metric-card h3 { margin: 0 0 12px; font-size: 14px; color: #374151; }
+
+.kv-list {
+  display: grid;
+  grid-template-columns: 1fr 1fr;
+  gap: 6px 12px;
+  font-size: 13px;
+}
+.kv-list span { color: #6b7280; }
+.kv-list b { color: #111; text-align: right; }
+
+.review-card { grid-column: span 2; }
+.review-stats { display: flex; gap: 24px; margin-bottom: 12px; }
+.stat-item { display: flex; flex-direction: column; align-items: center; }
+.stat-num { font-size: 28px; font-weight: 700; color: #2563eb; }
+.stat-label { font-size: 12px; color: #6b7280; }
+
+.progress-wrap {
+  height: 8px;
+  background: #e5e7eb;
+  border-radius: 99px;
+  overflow: hidden;
+  margin-bottom: 6px;
+}
+.reviewed-bar { height: 100%; background: #16a34a; transition: width 0.5s ease; }
+.review-rate { font-size: 13px; color: #6b7280; margin: 0; }
+
+.raw-json {
+  background: #f9fafb;
+  border: 1px solid #e5e7eb;
+  border-radius: 8px;
+}
+.raw-json summary { padding: 10px 14px; cursor: pointer; font-size: 13px; color: #6b7280; }
+.raw-json pre {
+  padding: 10px 14px;
+  margin: 0;
+  font-size: 12px;
+  white-space: pre-wrap;
+  border-top: 1px solid #e5e7eb;
+}
+</style>

webapp/src/views/ReviewView.vue

@@ -0,0 +1,196 @@
+<template>
+  <div class="review-view">
+    <header class="review-header">
+      <h2>人工检验队列</h2>
+      <div class="controls">
+        <button :class="{ active: filter === 'all' }" @click="filter = 'all'">全部</button>
+        <button :class="{ active: filter === 'pending' }" @click="filter = 'pending'">待审核</button>
+        <button :class="{ active: filter === 'approved' }" @click="filter = 'approved'">已通过</button>
+        <button :class="{ active: filter === 'rejected' }" @click="filter = 'rejected'">已拒绝</button>
+        <button class="refresh-btn" @click="load">🔄 刷新</button>
+      </div>
+    </header>
+
+    <div v-if="loading" class="loading">加载中…</div>
+    <div v-else-if="error" class="error">{{ error }}</div>
+
+    <div v-else class="queue-list">
+      <div v-if="!filtered.length" class="empty">队列为空。</div>
+
+      <div v-for="item in filtered" :key="item.id" class="review-card">
+        <div class="card-header">
+          <span class="card-id">#{{ item.id }}</span>
+          <span class="verdict-badge" :class="item.verdict">{{ item.verdict }}</span>
+          <span class="tags">
+            <span v-for="t in item.tags" :key="t" class="tag">{{ t }}</span>
+          </span>
+          <span class="date">{{ item.created_at }}</span>
+        </div>
+
+        <div class="query-block">
+          <strong>Query：</strong>{{ item.query }}
+        </div>
+
+        <div class="response-block">
+          <strong>Response：</strong>
+          <pre>{{ item.response }}</pre>
+        </div>
+
+        <div v-if="item.verdict === 'pending'" class="actions">
+          <textarea
+            v-model="correctionInputs[item.id]"
+            placeholder="修正意见（可选）"
+            rows="2"
+          />
+          <div class="btn-row">
+            <button class="approve" @click="submit(item.id, 'approved')">✅ 通过</button>
+            <button class="reject" @click="submit(item.id, 'rejected')">❌ 拒绝</button>
+          </div>
+        </div>
+        <div v-else-if="item.correction" class="correction">
+          <strong>修正：</strong>{{ item.correction }}
+        </div>
+      </div>
+    </div>
+  </div>
+</template>
+
+<script setup lang="ts">
+import { ref, computed, onMounted } from 'vue'
+import { listReviews, submitReview } from '@/api'
+import type { ReviewItem } from '@/types'
+
+const items = ref<ReviewItem[]>([])
+const loading = ref(false)
+const error = ref('')
+const filter = ref<'all' | 'pending' | 'approved' | 'rejected'>('pending')
+const correctionInputs = ref<Record<number, string>>({})
+
+const filtered = computed(() =>
+  filter.value === 'all'
+    ? items.value
+    : items.value.filter((i) => i.verdict === filter.value),
+)
+
+async function load() {
+  loading.value = true
+  error.value = ''
+  try {
+    items.value = await listReviews()
+  } catch (e: unknown) {
+    error.value = e instanceof Error ? e.message : String(e)
+  } finally {
+    loading.value = false
+  }
+}
+
+async function submit(id: number, verdict: 'approved' | 'rejected') {
+  try {
+    await submitReview(id, verdict, correctionInputs.value[id] || undefined)
+    await load()
+  } catch (e: unknown) {
+    error.value = e instanceof Error ? e.message : String(e)
+  }
+}
+
+onMounted(load)
+</script>
+
+<style scoped>
+.review-view { padding: 20px 24px; height: 100%; overflow-y: auto; }
+
+.review-header {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  margin-bottom: 20px;
+}
+.review-header h2 { margin: 0; font-size: 20px; }
+
+.controls { display: flex; gap: 8px; }
+button {
+  padding: 6px 14px;
+  border: 1px solid #d1d5db;
+  background: #fff;
+  border-radius: 6px;
+  cursor: pointer;
+  font-size: 13px;
+}
+button.active { background: #2563eb; color: #fff; border-color: #2563eb; }
+.refresh-btn { margin-left: auto; }
+
+.loading, .error, .empty {
+  text-align: center;
+  padding: 40px;
+  color: #9ca3af;
+}
+.error { color: #dc2626; }
+
+.queue-list { display: flex; flex-direction: column; gap: 16px; }
+
+.review-card {
+  border: 1px solid #e5e7eb;
+  border-radius: 10px;
+  padding: 16px;
+  background: #fff;
+}
+.card-header {
+  display: flex;
+  align-items: center;
+  gap: 10px;
+  margin-bottom: 10px;
+  flex-wrap: wrap;
+}
+.card-id { font-family: monospace; font-size: 12px; color: #6b7280; }
+.verdict-badge {
+  padding: 2px 8px;
+  border-radius: 99px;
+  font-size: 12px;
+  font-weight: 600;
+}
+.verdict-badge.pending { background: #fef3c7; color: #92400e; }
+.verdict-badge.approved { background: #dcfce7; color: #16a34a; }
+.verdict-badge.rejected { background: #fee2e2; color: #dc2626; }
+.tags { display: flex; gap: 4px; }
+.tag { background: #e0e7ff; color: #3730a3; padding: 1px 6px; border-radius: 4px; font-size: 11px; }
+.date { margin-left: auto; font-size: 11px; color: #9ca3af; }
+
+.query-block, .response-block {
+  font-size: 13px;
+  margin-bottom: 8px;
+  line-height: 1.6;
+}
+.query-block pre, .response-block pre {
+  margin: 4px 0 0;
+  background: #f9fafb;
+  border: 1px solid #e5e7eb;
+  border-radius: 4px;
+  padding: 6px 10px;
+  white-space: pre-wrap;
+  font-size: 13px;
+}
+
+.actions { display: flex; flex-direction: column; gap: 8px; margin-top: 10px; }
+textarea {
+  width: 100%;
+  padding: 8px 10px;
+  border: 1px solid #d1d5db;
+  border-radius: 6px;
+  font-size: 13px;
+  resize: vertical;
+  box-sizing: border-box;
+  font-family: inherit;
+}
+.btn-row { display: flex; gap: 8px; }
+.approve { background: #dcfce7; border-color: #86efac; color: #16a34a; }
+.reject { background: #fee2e2; border-color: #fca5a5; color: #dc2626; }
+
+.correction {
+  margin-top: 8px;
+  background: #fffbeb;
+  border: 1px solid #fcd34d;
+  border-radius: 4px;
+  padding: 6px 10px;
+  font-size: 13px;
+}
+</style>

webapp/tsconfig.app.json

@@ -0,0 +1,20 @@
+{
+  "extends": "@vue/tsconfig/tsconfig.dom.json",
+  "compilerOptions": {
+    "tsBuildInfoFile": "./node_modules/.tmp/tsconfig.app.tsbuildinfo",
+    "types": ["vite/client"],
+    "allowArbitraryExtensions": true,
+    "ignoreDeprecations": "6.0",
+    "baseUrl": ".",
+    "paths": {
+      "@/*": ["src/*"]
+    },
+
+    /* Linting */
+    "noUnusedLocals": true,
+    "noUnusedParameters": true,
+    "erasableSyntaxOnly": true,
+    "noFallthroughCasesInSwitch": true
+  },
+  "include": ["src/**/*.ts", "src/**/*.tsx", "src/**/*.vue"]
+}

webapp/tsconfig.json

@@ -0,0 +1,7 @@
+{
+  "files": [],
+  "references": [
+    { "path": "./tsconfig.app.json" },
+    { "path": "./tsconfig.node.json" }
+  ]
+}

webapp/tsconfig.node.json

@@ -0,0 +1,23 @@
+{
+  "compilerOptions": {
+    "tsBuildInfoFile": "./node_modules/.tmp/tsconfig.node.tsbuildinfo",
+    "target": "es2023",
+    "lib": ["ES2023"],
+    "types": ["node"],
+    "skipLibCheck": true,
+
+    /* Bundler mode */
+    "module": "nodenext",
+    "allowImportingTsExtensions": true,
+    "verbatimModuleSyntax": true,
+    "moduleDetection": "force",
+    "noEmit": true,
+
+    /* Linting */
+    "noUnusedLocals": true,
+    "noUnusedParameters": true,
+    "erasableSyntaxOnly": true,
+    "noFallthroughCasesInSwitch": true
+  },
+  "include": ["vite.config.ts"]
+}

webapp/vite.config.ts

@@ -0,0 +1,27 @@
+import vue from '@vitejs/plugin-vue'
+import { defineConfig } from 'vite'
+import { resolve } from 'path'
+
+export default defineConfig({
+  plugins: [vue()],
+  base: '/static/',
+  resolve: {
+    alias: { '@': resolve(__dirname, 'src') },
+  },
+  server: {
+    proxy: {
+      // 开发时：前端同源请求（无 /api 前缀）代理到 FastAPI
+      '/chat':   { target: 'http://localhost:8000', changeOrigin: true },
+      '/runs':   { target: 'http://localhost:8000', changeOrigin: true },
+      '/review': { target: 'http://localhost:8000', changeOrigin: true },
+      '/api/metrics':{ target: 'http://localhost:8000', changeOrigin: true },
+      '/config': { target: 'http://localhost:8000', changeOrigin: true },
+      '/health': { target: 'http://localhost:8000', changeOrigin: true },
+      '/traces':{ target: 'http://localhost:8000', changeOrigin: true },
+    },
+  },
+  build: {
+    outDir: '../gateway/static',
+    emptyOutDir: true,
+  },
+})

任务拆解与执行计划.md

@@ -81,3 +81,4 @@ P0 完成后的能力：干净的后端抽象 + 可量化的评测 + 可追溯
 | T12 | 实验脚本 bench_tokens.py + 数据集 | ✅ 完成 | T12 |
 | T13 | E1 本地跑数完成（E2–E5 待 live 接入） | ✅ 完成 | T13 |
 | T14 | 文档收口（README v2 改写） | ✅ 完成 | T14 |
+| T15 | Pipeline 死锁修复（_deps_done 依赖过滤 + pending-empty break）+ /metrics SPA 路由冲突修复 | ✅ 完成 | T15 |