agentworkshop 0.2.1

package/server/services/workshop/agents/tool-approvals.ts

@@ -0,0 +1,174 @@
+/**
+ * ToolApprovals —— 手动确认模式的工具执行审批服务(进程内)。
+ *
+ * manual 模式绑定的数控下发:工具调用 → request() 挂起等待 →
+ * 孪生侧栏审批面板 批准/拒绝(可附备注)→ 挂起 Promise 落定,
+ * 结果(含用户备注)回给 Agent 作为 tool result。超时自动按拒绝收敛。
+ */
+
+import { randomUUID } from 'node:crypto'
+import { getOps } from '../ops/ops'
+
+export interface ToolApproval {
+  id: string
+  agentId: string
+  nodeId: string
+  kind: 'dcw' | 'daq'
+  /** 人读摘要(节点名/物理量/目标值/窗口) */
+  detail: string
+  createdAt: string
+  /** 到期时刻(createdAt + 超时窗);UI 据此显示自动拒绝倒计时 */
+  expiresAt: string
+  status: 'pending' | 'approved' | 'denied' | 'expired'
+  comment: string
+  decidedAt: string | null
+  /** S4:裁决人留痕(空 = 超时/系统收敛) */
+  decidedBy: string
+  decidedName: string
+}
+
+/** 审批超时窗(超时默认拒绝,指令不执行);可用环境变量 HITL_TIMEOUT_MS 调整(测试/产线策略) */
+const TIMEOUT_MS = Number(process.env.HITL_TIMEOUT_MS) > 0 ? Number(process.env.HITL_TIMEOUT_MS) : 180_000
+const HISTORY_CAP = 50
+
+class ToolApprovalService {
+  private pending = new Map<string, {
+    approval: ToolApproval
+    resolve: (r: { approved: boolean, comment: string, id: string }) => void
+    timer: NodeJS.Timeout
+  }>()
+
+  private history: ToolApproval[] = []
+
+  /** 挂起一次执行审批(工具侧 await;批准/拒绝/超时三向落定) */
+  request(agentId: string, nodeId: string, kind: 'dcw' | 'daq', detail: string): Promise<{ approved: boolean, comment: string, id: string }> {
+    const approval: ToolApproval = {
+      id: `ap-${randomUUID().slice(0, 8)}`,
+      agentId,
+      nodeId,
+      kind,
+      detail,
+      createdAt: new Date().toISOString(),
+      expiresAt: new Date(Date.now() + TIMEOUT_MS).toISOString(),
+      status: 'pending',
+      comment: '',
+      decidedAt: null,
+      decidedBy: '',
+      decidedName: '',
+    }
+    return new Promise((resolve) => {
+      const timer = setTimeout(() => {
+        if (!this.pending.has(approval.id)) return
+        this.pending.delete(approval.id)
+        approval.status = 'expired'
+        approval.decidedAt = new Date().toISOString()
+        approval.comment = '审批超时未处理,默认拒绝,指令未执行'
+        this.remember(approval)
+        resolve({ approved: false, comment: approval.comment, id: approval.id })
+      }, TIMEOUT_MS)
+      timer.unref?.()
+      this.pending.set(approval.id, { approval, resolve, timer })
+    })
+  }
+
+  decide(id: string, approved: boolean, comment: string, decidedBy = '', decidedName = ''): ToolApproval {
+    const entry = this.pending.get(id)
+    if (!entry) throw new Error(`审批不存在或已处理: ${id}`)
+    clearTimeout(entry.timer)
+    this.pending.delete(id)
+    entry.approval.status = approved ? 'approved' : 'denied'
+    entry.approval.comment = String(comment ?? '').trim()
+    entry.approval.decidedAt = new Date().toISOString()
+    entry.approval.decidedBy = decidedBy
+    entry.approval.decidedName = decidedName
+    this.remember(entry.approval)
+    entry.resolve({ approved, comment: entry.approval.comment, id })
+    return entry.approval
+  }
+
+  /** 审批面板拉取(指定 Agent 的待处理;agentId 空 = 全部) */
+  listPending(agentId = ''): ToolApproval[] {
+    return [...this.pending.values()]
+      .map(e => e.approval)
+      .filter(a => !agentId || a.agentId === agentId)
+  }
+
+  /** 解绑/换线时取消挂起审批:该 Agent 对某节点的全部 pending 按拒绝收敛(备注说明原因) */
+  cancelPendingFor(agentId: string, nodeId: string): number {
+    let n = 0
+    for (const [id, entry] of [...this.pending.entries()]) {
+      if (entry.approval.agentId !== agentId || entry.approval.nodeId !== nodeId) continue
+      clearTimeout(entry.timer)
+      this.pending.delete(id)
+      entry.approval.status = 'denied'
+      entry.approval.comment = '绑定已解除,审批失效'
+      entry.approval.decidedAt = new Date().toISOString()
+      this.remember(entry.approval)
+      entry.resolve({ approved: false, comment: entry.approval.comment, id })
+      n++
+    }
+    return n
+  }
+
+  /** 同一 Agent 同一节点的挂起审批去重:已有 pending 时拒绝新挂起(防审批面板堆积) */
+  hasPendingFor(agentId: string, nodeId: string): boolean {
+    for (const entry of this.pending.values()) {
+      if (entry.approval.agentId === agentId && entry.approval.nodeId === nodeId) return true
+    }
+    return false
+  }
+
+  historyList(): ToolApproval[] {
+    // S4:优先读持久化表(重启后仍可查);未接线(测试/降级)回退内存窗口
+    const repo = getOps()?.approvalHistory
+    if (repo) {
+      return repo.list(HISTORY_CAP * 4).map(r => ({
+        id: r.id,
+        agentId: r.agentId,
+        nodeId: r.nodeId,
+        kind: r.kind as ToolApproval['kind'],
+        detail: r.detail,
+        status: r.status as ToolApproval['status'],
+        comment: r.comment,
+        decidedAt: r.decidedAt,
+        decidedBy: r.decidedBy,
+        decidedName: r.decidedName,
+        createdAt: r.createdAt,
+        // 持久化表未存到期时刻:按超时窗从 createdAt 派生(仅展示用)
+        expiresAt: new Date(Date.parse(r.createdAt) + TIMEOUT_MS).toISOString(),
+      }))
+    }
+    return this.history
+  }
+
+  private remember(a: ToolApproval): void {
+    this.history.unshift(a)
+    if (this.history.length > HISTORY_CAP) this.history.splice(HISTORY_CAP)
+    // S4:同步持久化(失败不影响审批主流程——工具侧已拿到裁决结果)
+    try {
+      getOps()?.approvalHistory.upsert({
+        id: a.id,
+        agentId: a.agentId,
+        nodeId: a.nodeId,
+        kind: a.kind,
+        detail: a.detail,
+        status: a.status,
+        comment: a.comment,
+        decidedBy: a.decidedBy,
+        decidedName: a.decidedName,
+        createdAt: a.createdAt,
+        decidedAt: a.decidedAt,
+      })
+    }
+    catch {
+      // 持久化失败降级为仅内存历史
+    }
+  }
+}
+
+const g = globalThis as typeof globalThis & { __toolApprovals?: ToolApprovalService }
+
+export function getToolApprovals(): ToolApprovalService {
+  g.__toolApprovals ??= new ToolApprovalService()
+  return g.__toolApprovals
+}

package/server/services/workshop/assets/character-asset.repo.ts

@@ -0,0 +1,120 @@
+/**
+ * CharacterAsset 注册表 —— JSON 文件持久化(轻量,无需迁移 DB)。
+ *
+ * 记录用户上传/内置的角色模型元信息(id/file/name/kind/sheet/anims/anchor/scale/author/createdAt/appliedTo)。
+ * 应用级单例,写入 server/data/character-assets.json,进程内缓存,启动读盘。
+ * 引删保护:appliedTo 里仍有 agent 绑定时,删除改为"停用"而非硬删。
+ */
+
+import fs from 'node:fs'
+import { join } from 'node:path'
+import { ensureDataDir } from '@/shared/config/home.mjs'
+
+export interface CharacterAsset {
+  id: string
+  /** 作用域:空字符串 = 全局共享库;否则为该 workspace 私有 */
+  workspaceId: string
+  name: string
+  /** 相对 public 的路径,供 Phaser 加载 */
+  file: string
+  kind: 'single' | 'sheet' | 'zip' | 'glb'
+  /** 精灵表帧布局(与 town-anim 的 ModelAnimSpec 对齐) */
+  sheet?: { frameWidth: number, frameHeight: number, frames: number }
+  anims?: Partial<Record<'idle' | 'walk' | 'work', { start: number, count: number, frameRate?: number }>>
+  anchor?: { x: number, y: number }
+  scale?: number
+  /** 已绑定该模型的 agentId 反向索引(引删保护用) */
+  appliedTo: string[]
+  author: string
+  createdAt: string
+}
+
+// 配置根 .AgentWorkShop/data（ensureDataDir 自动迁移旧 cwd/server/data 位置）
+const DB_PATH = join(ensureDataDir(), 'character-assets.json')
+
+function load(): CharacterAsset[] {
+  try {
+    const raw = fs.readFileSync(DB_PATH, 'utf-8')
+    const parsed = JSON.parse(raw)
+    return Array.isArray(parsed) ? parsed : []
+  }
+  catch {
+    return []
+  }
+}
+
+function save(list: CharacterAsset[]): void {
+  fs.mkdirSync(join(DB_PATH, '..'), { recursive: true })
+  fs.writeFileSync(DB_PATH, JSON.stringify(list, null, 2), 'utf-8')
+}
+
+class CharacterAssetRepo {
+  private list: CharacterAsset[] = load()
+  private readonly dbPath = DB_PATH
+
+  listAll(workspaceId?: string): CharacterAsset[] {
+    if (!workspaceId) return this.list
+    return this.list.filter(a => a.workspaceId === workspaceId || a.workspaceId === '')
+  }
+
+  findById(id: string): CharacterAsset | undefined {
+    return this.list.find(a => a.id === id)
+  }
+
+  create(asset: Omit<CharacterAsset, 'appliedTo' | 'createdAt'>): CharacterAsset {
+    if (this.list.some(a => a.id === asset.id)) {
+      const existing = this.list.find(a => a.id === asset.id)!
+      // 幂等覆盖元信息,保留 appliedTo
+      const merged: CharacterAsset = { ...existing, ...asset, appliedTo: existing.appliedTo ?? [] }
+      this.list = this.list.map(a => a.id === asset.id ? merged : a)
+      save(this.list)
+      return merged
+    }
+    const full: CharacterAsset = { ...asset, appliedTo: [], createdAt: new Date().toISOString() }
+    this.list.push(full)
+    save(this.list)
+    return full
+  }
+
+  bind(assetId: string, agentId: string): void {
+    const asset = this.findById(assetId)
+    if (!asset) return
+    if (!asset.appliedTo.includes(agentId)) {
+      asset.appliedTo.push(agentId)
+      save(this.list)
+    }
+  }
+
+  unbind(assetId: string, agentId: string): void {
+    const asset = this.findById(assetId)
+    if (!asset) return
+    asset.appliedTo = asset.appliedTo.filter(id => id !== agentId)
+    save(this.list)
+  }
+
+  /**
+   * 删除(引删保护):仍被绑定 → 改为停用标记并返回 {ok, used};未绑定 → 硬删。
+   * 返回是否成功删除文件(由调用方决定是否移除磁盘文件)。
+   */
+  remove(id: string): { ok: boolean, used: number, kept: boolean } {
+    const asset = this.findById(id)
+    if (!asset) return { ok: false, used: 0, kept: false }
+    if (asset.appliedTo.length > 0) {
+      // 仍被使用 → 保留记录(调用方据 used>0 提示"仍被 N 个 Agent 使用",不删文件)
+      return { ok: true, used: asset.appliedTo.length, kept: true }
+    }
+    this.list = this.list.filter(a => a.id !== id)
+    save(this.list)
+    return { ok: true, used: 0, kept: false }
+  }
+
+  /** 是否存在空 workspaceId 的全局库(供前端展示区分) */
+  readonly dbFile = this.dbPath
+}
+
+// 应用级单例
+let singleton: CharacterAssetRepo | null = null
+export function getCharacterAssetRepo(): CharacterAssetRepo {
+  if (!singleton) singleton = new CharacterAssetRepo()
+  return singleton
+}

package/server/services/workshop/assets/device-twin.repo.ts

@@ -0,0 +1,236 @@
+/**
+ * DeviceTwin 注册表 —— 数字孪生设备/实体模型(JSON 文件持久化)。
+ *
+ * 记录数字孪生设备:id/name/modelRef(绑定的实体模型 assetId)/boundAgentId(控制该设备的 Agent)/
+ * telemetry(实时数据采集)/desired(期望状态)/state(实际运行状态)/controls(可下发指令)。
+ * 模拟域:无真实设备时用 telemetry 模拟驱动(采样函数);接入真实设备时由 MCP 或采集器写入。
+ *
+ * 应用级单例,写入 server/data/device-twins.json,进程内缓存,启动读盘。
+ */
+
+import fs from 'node:fs'
+import { join } from 'node:path'
+import { ensureDataDir } from '@/shared/config/home.mjs'
+import { AppError } from '../../../utils/errors'
+
+export interface DeviceTwin {
+  id: string
+  /** 作用域:空字符串 = 全局;否则该 workspace 私有 */
+  workspaceId: string
+  name: string
+  /** 绑定的实体模型 assetId(拖入场景的 prop/设备模型) */
+  modelRef: string
+  /** 控制该设备的 Agent(数字人);经 MCP device.* 控制 */
+  boundAgentId: string | null
+  /** 实体类别:device 设备 / environment 环境 / asset 资产 / daq 数采节点(程序化传感网格) */
+  kind: 'device' | 'environment' | 'asset' | 'daq'
+  /** 实时数据采集(数字孪生字段集合) */
+  telemetry: Record<string, number | string | boolean>
+  /** 期望状态(Agent 下发) */
+  desired: Record<string, number | string | boolean>
+  /** 实际运行状态 */
+  state: 'idle' | 'running' | 'offline' | 'alarm'
+  /** 可下发指令集(供 Agent 经 MCP 调用) */
+  controls: string[]
+  /** 场景落点(3D 小镇 x/z 世界坐标;undefined = 未放入场景,仅注册表记录) */
+  posX?: number
+  posZ?: number
+  /** 场景朝向(绕 y 轴角度,度;缺省 0) */
+  rotationY?: number
+  /** 场景内缩放倍率(缺省 1 = 默认归一化尺寸) */
+  scale?: number
+  /** 最近更新时间(数据采集心跳) */
+  updatedAt: string
+  createdAt: string
+}
+
+// 配置根 .AgentWorkShop/data（ensureDataDir 自动迁移旧 cwd/server/data 位置）
+const DB_PATH = join(ensureDataDir(), 'device-twins.json')
+
+function load(): DeviceTwin[] {
+  try {
+    const raw = fs.readFileSync(DB_PATH, 'utf-8')
+    const parsed = JSON.parse(raw)
+    return Array.isArray(parsed) ? parsed : []
+  }
+  catch {
+    return []
+  }
+}
+function save(list: DeviceTwin[]): void {
+  fs.mkdirSync(join(DB_PATH, '..'), { recursive: true })
+  fs.writeFileSync(DB_PATH, JSON.stringify(list, null, 2), 'utf-8')
+}
+
+export class DeviceTwinRepo {
+  private list: DeviceTwin[] = load()
+  private flushTimer: NodeJS.Timeout | null = null
+
+  /** 遥测写盘防抖(采样回写每帧触发,逐次全文件序列化会 fsync 抖动;2s 合并) */
+  private flushDebounced(ms = 2000): void {
+    if (this.flushTimer) return
+    this.flushTimer = setTimeout(() => {
+      this.flushTimer = null
+      this.flushNow()
+    }, ms)
+    this.flushTimer.unref?.()
+  }
+
+  private flushNow(): void {
+    if (this.flushTimer) {
+      clearTimeout(this.flushTimer)
+      this.flushTimer = null
+    }
+    save(this.list)
+  }
+
+  listAll(workspaceId?: string): DeviceTwin[] {
+    if (!workspaceId) return this.list
+    return this.list.filter(d => d.workspaceId === workspaceId || d.workspaceId === '')
+  }
+
+  findById(id: string): DeviceTwin | undefined {
+    return this.list.find(d => d.id === id)
+  }
+
+  create(input: Omit<DeviceTwin, 'id' | 'createdAt' | 'updatedAt' | 'state' | 'boundAgentId' | 'desired'> & { state?: DeviceTwin['state'], boundAgentId?: string | null, desired?: DeviceTwin['desired'] }): DeviceTwin {
+    const full: DeviceTwin = {
+      id: `dev-${Date.now().toString(36)}-${Math.random().toString(36).slice(2, 7)}`,
+      workspaceId: input.workspaceId,
+      name: input.name,
+      modelRef: input.modelRef,
+      boundAgentId: input.boundAgentId ?? null,
+      kind: input.kind,
+      telemetry: input.telemetry ?? {},
+      desired: input.desired ?? {},
+      state: input.state ?? 'idle',
+      controls: input.controls ?? [],
+      posX: input.posX,
+      posZ: input.posZ,
+      rotationY: input.rotationY,
+      scale: input.scale,
+      updatedAt: new Date().toISOString(),
+      createdAt: new Date().toISOString(),
+    }
+    this.list.push(full)
+    save(this.list)
+    return full
+  }
+
+  update(id: string, patch: Partial<Pick<DeviceTwin, 'name' | 'modelRef' | 'boundAgentId' | 'controls' | 'desired' | 'posX' | 'posZ' | 'rotationY' | 'scale'>>): DeviceTwin | undefined {
+    const d = this.findById(id)
+    if (!d) return undefined
+    if (patch.name !== undefined) d.name = patch.name
+    if (patch.modelRef !== undefined) d.modelRef = patch.modelRef
+    if (patch.boundAgentId !== undefined) d.boundAgentId = patch.boundAgentId
+    if (patch.controls !== undefined) d.controls = patch.controls
+    if (patch.desired !== undefined) d.desired = { ...d.desired, ...patch.desired }
+    if (patch.posX !== undefined) d.posX = patch.posX
+    if (patch.posZ !== undefined) d.posZ = patch.posZ
+    if (patch.rotationY !== undefined) d.rotationY = patch.rotationY
+    if (patch.scale !== undefined) d.scale = patch.scale
+    d.updatedAt = new Date().toISOString()
+    save(this.list)
+    return d
+  }
+
+  /**
+   * 采集:更新 telemetry + 派生 state;写盘走防抖(热路径)。
+   * state 派生两级:
+   *  - nodeState 透传(DAQ 回写路径):节点态由用户设置的量程/预警带数据派生,
+   *    设备信任量测侧结论(数据驱动,零硬编码);多节点绑定已在 DAQ 侧聚合最严重态;
+   *  - 无 nodeState(REST/MCP 直写遥测):保留既有阈值规则兜底(向后兼容)。
+   */
+  applyTelemetry(id: string, patch: Record<string, number | string | boolean>, nodeState?: 'alarm' | 'warn' | 'ok'): DeviceTwin | undefined {
+    const d = this.findById(id)
+    if (!d) return undefined
+    d.telemetry = { ...d.telemetry, ...patch }
+    if (nodeState) {
+      if (nodeState === 'alarm') d.state = 'alarm'
+      else if (d.state === 'alarm') d.state = d.desired.on === false ? 'idle' : 'running'
+    }
+    else {
+      // legacy 派生:温度/压力越限 → alarm;否则 running/idle
+      const temp = Number(d.telemetry.temperature ?? 0)
+      const pressure = Number(d.telemetry.pressure ?? 0)
+      if ((temp > 0 && temp > 85) || (pressure > 0 && pressure > 2.0)) d.state = 'alarm'
+      else if (d.state === 'alarm') d.state = 'running'
+      else if (d.state !== 'running' && d.desired.on !== false) d.state = 'running'
+    }
+    d.updatedAt = new Date().toISOString()
+    this.flushDebounced()
+    return d
+  }
+
+  /** 下发指令:白名单校验(twin.controls 非空时)→ 写 desired + state 变化,拒绝静默成功 */
+  applyControl(id: string, cmd: string, args: Record<string, unknown>): DeviceTwin | undefined {
+    const d = this.findById(id)
+    if (!d) return undefined
+    if (d.controls.length > 0 && !d.controls.includes(cmd)) {
+      throw new AppError(400, 'BAD_REQUEST', `未知指令 "${cmd}"(设备支持:${d.controls.join(', ') || '无'})`)
+    }
+    switch (cmd) {
+      case 'power_on':
+        d.desired.on = true
+        break
+      case 'power_off':
+        d.desired.on = false
+        break
+      case 'stop':
+        d.desired.on = false
+        break
+      case 'set_speed':
+        if (typeof args.value !== 'number') throw new AppError(400, 'BAD_REQUEST', 'set_speed 需要 number 型 args.value')
+        d.desired.speed = args.value
+        break
+      case 'set_temperature':
+        if (typeof args.value !== 'number') throw new AppError(400, 'BAD_REQUEST', 'set_temperature 需要 number 型 args.value')
+        d.desired.temperature = args.value
+        break
+      default:
+        throw new AppError(400, 'BAD_REQUEST', `未知指令 "${cmd}"`)
+    }
+    d.state = d.desired.on === false ? 'idle' : 'running'
+    d.updatedAt = new Date().toISOString()
+    this.flushNow()
+    return d
+  }
+
+  remove(id: string): boolean {
+    const before = this.list.length
+    this.list = this.list.filter(d => d.id !== id)
+    if (this.list.length !== before) {
+      save(this.list)
+      return true
+    }
+    return false
+  }
+}
+
+let singleton: DeviceTwinRepo | null = null
+export function getDeviceTwinRepo(): DeviceTwinRepo {
+  if (!singleton) singleton = new DeviceTwinRepo()
+  return singleton
+}
+
+/**
+ * 场景事件载荷(设备孪生 → device.created/updated/deleted 广播)。
+ * 与前端 DeviceTwinView 同构,供其他小镇客户端即时同步场景节点。
+ */
+export function deviceScenePayload(d: DeviceTwin): Record<string, unknown> {
+  return {
+    id: d.id,
+    name: d.name,
+    modelRef: d.modelRef,
+    kind: d.kind,
+    state: d.state,
+    posX: d.posX,
+    posZ: d.posZ,
+    rotationY: d.rotationY,
+    scale: d.scale,
+    workspaceId: d.workspaceId,
+    boundAgentId: d.boundAgentId,
+    telemetry: d.telemetry,
+    updatedAt: d.updatedAt,
+  }
+}

package/server/services/workshop/daq/alarm-notify.ts

@@ -0,0 +1,100 @@
+/**
+ * AlarmNotifier —— 报警外送 + 升级扫描(S5,production-readiness-plan)。
+ *
+ * - notifyAlarm():webhook 外送(env ALARM_WEBHOOK_URL;POST JSON,天然兼容
+ *   钉钉/企微/飞书机器人),失败重试 3 次退避;fire-and-forget,绝不阻塞采集主链路;
+ * - startAlarmEscalator():60s 扫描未确认且超 ALARM_ESCALATE_MINUTES(默认 15)的
+ *   报警 → escalation+1 并重发外送(升级通知);globalThis 守卫防 HMR 重复。
+ */
+import { randomUUID } from 'node:crypto'
+import { getOps } from '../ops/ops'
+
+const RETRIES = 3
+const ESCALATE_SWEEP_MS = 60_000
+
+const g = globalThis as typeof globalThis & { __awAlarmEscalator?: NodeJS.Timeout }
+
+export interface AlarmNotifyPayload {
+  id: string
+  nodeId: string
+  nodeName: string
+  metric: string
+  value: number | null
+  rule: string
+  threshold: number | null
+  escalation: number
+  createdAt: string
+}
+
+/** webhook 外送(单次;成功返回 true) */
+async function postWebhook(payload: AlarmNotifyPayload, url: string): Promise<boolean> {
+  try {
+    const res = await fetch(url, {
+      method: 'POST',
+      headers: { 'content-type': 'application/json' },
+      body: JSON.stringify({ kind: 'alarm', text: `[报警] ${payload.nodeName} ${payload.metric} ${payload.rule} 阈值 ${payload.threshold ?? '-'} 当前 ${payload.value ?? '-'}(升级 x${payload.escalation})`, ...payload }),
+      signal: AbortSignal.timeout(5000),
+    })
+    return res.ok
+  }
+  catch {
+    return false
+  }
+}
+
+/** 外送 + 重试退避(1s/2s/4s);结果记录进 alarm_events.notified_json */
+export function notifyAlarm(payload: AlarmNotifyPayload): void {
+  const url = process.env.ALARM_WEBHOOK_URL
+  if (!url) return // 未配置外送 = 仅 WS/面板可见(合法降级)
+  const repo = getOps()?.alarmEvents
+  void (async () => {
+    let ok = false
+    for (let attempt = 1; attempt <= RETRIES && !ok; attempt++) {
+      ok = await postWebhook(payload, url)
+      if (!ok && attempt < RETRIES) await new Promise(r => setTimeout(r, 2 ** (attempt - 1) * 1000))
+    }
+    if (!repo) return
+    try {
+      const row = repo.list(500).find(a => a.id === payload.id)
+      if (!row) return
+      const log = JSON.parse(row.notifiedJson || '[]') as Array<{ url: string, ok: boolean, at: string, attempt: number }>
+      log.push({ url, ok, at: new Date().toISOString(), attempt: RETRIES })
+      repo.recordNotify(payload.id, JSON.stringify(log.slice(-10)))
+    }
+    catch {
+      // 外送记录失败不影响业务
+    }
+  })()
+}
+
+/** 升级扫描器(懒启动;进程内单例) */
+export function startAlarmEscalator(): void {
+  if (g.__awAlarmEscalator) return
+  const escalateMin = Math.max(1, Number(process.env.ALARM_ESCALATE_MINUTES) || 15)
+  const timer = setInterval(() => {
+    const repo = getOps()?.alarmEvents
+    if (!repo) return
+    try {
+      const cutoff = Date.now() - escalateMin * 60_000
+      for (const a of repo.listOpen(200)) {
+        if (Date.parse(a.createdAt) > cutoff) continue
+        repo.escalate(a.id)
+        notifyAlarm({
+          id: a.id, nodeId: a.nodeId, nodeName: a.nodeName, metric: a.metric,
+          value: a.value, rule: a.rule, threshold: a.threshold,
+          escalation: a.escalation + 1, createdAt: a.createdAt,
+        })
+      }
+    }
+    catch {
+      // 升级扫描失败不致命(下一轮再扫)
+    }
+  }, ESCALATE_SWEEP_MS)
+  timer.unref?.()
+  g.__awAlarmEscalator = timer
+}
+
+/** 生成报警事件 id(raise 侧用) */
+export function newAlarmId(): string {
+  return `al-${randomUUID().slice(0, 8)}`
+}

package/server/services/workshop/daq/bus/index.ts

@@ -0,0 +1,46 @@
+/**
+ * DaqQueuePort 工厂 —— 配置驱动(DAQ_MQTT_URL 由 infra 提供),缺省进程内 mock 队列。
+ * 单例 globalThis;rebuildDaqQueue() 支持重连时切回真实 MQTT。
+ */
+import { createLogger } from '../../logger'
+import { InProcQueueAdapter } from './inproc.adapter'
+import { MqttQueueAdapter } from './mqtt.adapter'
+import type { DaqQueuePort } from './queue-port'
+
+const log = createLogger('daq.queue')
+
+const g = globalThis as typeof globalThis & { __daqQueue?: DaqQueuePort }
+
+export async function getDaqQueue(): Promise<DaqQueuePort> {
+  if (!g.__daqQueue) {
+    g.__daqQueue = new InProcQueueAdapter()
+    await g.__daqQueue.init()
+    log.info('[daq-queue] 进程内 mock 队列就绪(启动插件将按 infra 状态装配 MQTT)')
+  }
+  return g.__daqQueue
+}
+
+/** 重连/初始装配:按 infra 判定结果(重新)构建 MQTT 或进程内队列 */
+export async function rebuildDaqQueue(online: boolean, url?: string | null): Promise<DaqQueuePort> {
+  if (!online || !url) {
+    if (g.__daqQueue?.backend !== 'inproc') {
+      g.__daqQueue = new InProcQueueAdapter()
+      await g.__daqQueue.init()
+      log.warn('[daq-queue] 降级进程内队列')
+    }
+    return g.__daqQueue
+  }
+  try {
+    const adapter = new MqttQueueAdapter(url)
+    await adapter.init()
+    g.__daqQueue = adapter
+    log.info('[daq-queue] MQTT 队列就绪:', url)
+    return adapter
+  }
+  catch (err) {
+    log.error('[daq-queue] MQTT 初始化失败,回退进程内队列:', err instanceof Error ? err.message : err)
+    g.__daqQueue = new InProcQueueAdapter()
+    await g.__daqQueue.init()
+    return g.__daqQueue
+  }
+}