agentworkshop 0.7.5 → 0.7.7

package/scripts/_dbg-real-scenario-timeline.log

@@ -0,0 +1,123 @@
+[03:22:04.847] 测试用户已注册(token=ut-b725d…)BASE=http://127.0.0.1:3000
+[03:22:04.968]   PASS  服务可达且为多引擎版本(注册表含 codex/dsh/opencode) — harnesses=mock,omp,opencode,codex,dsh,claude
+[03:22:05.003]   PASS  新建产线 — line=多引擎联调线
+[03:22:05.042]   PASS  新建 DCW 数控节点(mock 温控) — id=dw-c5cc2e42
+[03:22:05.050]   PASS  新建 DAQ 数采节点(mock 温度) — id=dn-960ca8da driver=mock
+[03:22:05.056]   PASS  创建 AgentChannel(lead=omp) — channelId=7decac99-c3f6-45d7-820d-a571976c0381
+[03:22:05.065]   PASS  创建 worker-codex 并入组(harness=codex) — 051d24dc-2b2f-4f2e-94a5-0711b187866e
+[03:22:05.078]   PASS  创建 worker-dsh 并入组(harness=dsh) — 4eb98cb2-fcc9-4781-88c9-ca7cbc135d3d
+[03:22:05.096] WS 监控已接入(AEP 实时帧)
+[03:22:07.640]   PASS  绑定 dsh↔DCW(manual,HITL 模式) — mode=manual
+[03:22:07.646]   PASS  绑定 dsh↔DAQ(auto,数采)
+[03:22:07.652]   PASS  绑定 codex↔DCW(manual)
+[03:22:07.659]   PASS  dsh 工具面可见绑定节点(my_industrial_nodes) — #### ◆ 联调温度设定器 [id=dw-c5cc2e42] - 物理量: 烘箱温度设定,单位 ℃,精度 1 位小数 - 工艺语义: 烘箱/熔体温度设定:升高使热塑温度上升(成膜更均匀但能耗高、过热降解风险),降低则偏冷易厚度不均。调整后
+[03:22:07.711]   PASS  人类→codex 消息已投递 — status=200
+[03:22:08.767]   [事件] #3 a2a.message ?→?: 兼容性检查:请用一句话汇报你正在运行的引擎类型(如 codex),然后调用 report_progress 工具上报 10% 进度(无任务上下文则忽略报错),最后原样回复"codex 在线"。
+[03:25:29.004]   [事件] #32 a2a.artifact artifact(output)
+[03:25:29.004]   PASS  codex 回执到达(真实引擎回合) — 内容=
+[03:25:29.032]   PASS  人类→dsh 消息已投递 — status=200
+[03:25:29.891]   [事件] #35 a2a.message ?→?: 兼容性检查:请用一句话汇报你正在运行的引擎类型(如 dsh/ACP),然后原样回复"dsh 在线"。
+[03:25:39.365]   [事件] #38 a2a.artifact artifact(output)
+[03:25:39.365]   PASS  dsh 回执到达(真实引擎回合) — 内容=
+[03:25:39.372]   PASS  人类→codex(要求跨引擎发信)已投递 — status=200
+[03:25:39.937]   [事件] #41 a2a.message ?→?: 请使用 send_message_to_agent 工具向同事 worker-dsh 发送一句话:"跨引擎测试:请回复你的引擎类型",并设置 require_reply=true;发送完成后回复我"已发送"。
+[03:27:03.462]   [事件] #77 a2a.artifact artifact(output)
+[03:27:03.462]   PASS  codex 经工具跨引擎发信完成 — 内容=
+[03:29:03.828]   FAIL  dsh 侧收到 codex 来信
+[03:29:03.878]   PASS  团队作业(goal)已提交至 lead — taskId=2d2d0a82-1efd-4d8e-afb7-86fd4049e2ee state=SUBMITTED
+[03:29:04.736]   [事件] #81 task.status 温度工艺整定(数采+数控) → SUBMITTED
+[03:29:04.736]   [事件] #82 a2a.message ?→?: 
+[03:30:18.590]   [事件] #84 task.status 温度工艺整定(数采+数控) → WORKING
+[03:30:18.590]   [事件] #85 task.status 温度工艺整定(数采+数控) → WAITING
+[03:30:18.590]   [事件] #86 task.status 温度工艺整定(数采+数控) → ASSIGNED
+[03:30:18.590]   [事件] #88 task.status 温度工艺整定(数采+数控) → WORKING
+[03:31:50.861]   [事件] #91 a2a.message ?→?: 【去重协调】你的任务 b1b6ba41「温度工艺整定(数采+数控)」与我的父任务 2d2d0a82 是同一作业目标(我已于 03:29:49 将正式执行子任务 0396b351 派给 worker-dsh:daq_que
+[03:31:57.360]   [事件] #92 a2a.message ?→?: 【单一执行路径声明】你队列中的子任务 0396b351「温度整定执行:数采均值查询+审批写入175.0+复读确认」是本次温度整定的唯一正式执行任务,按其步骤执行即可。
+注意:worker-codex 名下有一个同目标任务
+[03:31:57.360]   [事件] #94 task.status 温度整定执行:数采均值查询+审批写入175.0+复读确认 → WORKING
+[03:32:31.864]   [WS/hitl.request] dcw-approval: DCW 下发审批
+[03:32:31.864]   [WS/hitl.request] dcw-approval: DCW 下发审批
+[03:32:31.864]   [WS/hitl.request] dcw-approval: DCW 下发审批
+[03:32:31.868]   PASS  数控写入触发 dcw-approval HITL(worker-dsh) — id=ap-8c34b592 detail=联调温度设定器(烘箱温度设定)设定 175℃,安全量程 150~200℃
+[03:32:31.869] 审批从提交到出现耗时 207.99s
+[03:32:31.869] 测试员批准审批(模拟用户确认)…
+[03:32:31.880]   [WS/hitl.resolved] dcw-approval outcome=answered
+[03:32:31.881]   PASS  HITL 批准应答成功(respond 200) — status=200 {"status":200,"code":0,"message":"ok","data":{"ok":true,"kind":"dcw-approval","id":"ap-8c34b592","ap
+[03:32:31.881]   [WS/hitl.resolved] dcw-approval outcome=answered
+[03:32:31.881]   [WS/hitl.resolved] dcw-approval outcome=answered
+[03:32:32.959]   [事件] #101 hitl.request dcw-approval:DCW 下发审批
+[03:32:32.959]   [事件] #102 hitl.resolved dcw-approval outcome=answered
+[03:32:43.503]   [事件] #105 a2a.message ?→?: 【温度整定执行汇报】任务 0396b351(三步:数采均值→审批写入175.0→复读确认)已完成,四项结果如下(数值均经标定、带时间,可复核):
+
+① 数采均值:最近5分钟窗口(2026-09-05T03:27~03:3
+[03:32:46.626]   [事件] #106 task.status 温度整定执行:数采均值查询+审批写入175.0+复读确认 → COMPLETED
+[03:32:55.013]   [事件] #112 a2a.artifact artifact(output)
+[03:32:59.279]   [事件] #115 a2a.message ?→?: 【收到,确认对齐】明白:0396b351 是本次温度整定的唯一正式执行任务。该任务我已按步骤完成并上报(COMPLETED,队列已空):审批写入 175.0℃ 一次成功,dcw_read 复读 ACT=SET=175℃ 
+[03:33:01.388]   [事件] #120 a2a.artifact artifact(output)
+[03:33:18.184]   [事件] #150 a2a.message ?→?: 【汇报已验收】四项结果收到并确认:①数采无数据(节点 offline/停线,如实记录未伪造,处理正确) ②审批放行 ③写入 175.0℃ 成功(opt-23a51812) ④复读 ACT=SET=175℃ 一致。任务 0
+[03:33:21.365]   [事件] #155 a2a.artifact artifact(output)
+[03:33:26.614]   [事件] #157 a2a.message ?→?: 【证据转发,请收口 b1b6ba41】唯一一次物理执行已由 worker-dsh 完成(正式子任务 0396b351,状态 COMPLETED),你任务判定标准所需的每一步证据如下:
+① 数采均值:daq_query 查
+[03:33:41.809]   [事件] #171 a2a.message ?→?: 两条均已知悉:① 0396b351 派发回执无待办(该任务已 COMPLETED,成果已验收);② 对齐确认收到——维持「唯一一次写入、不重复执行」纪律即可,无需补充材料。你的部分已收尾,后续若有复线复测再另行安排。
+[03:33:43.400]   [事件] #176 a2a.artifact artifact(output)
+[03:34:26.492]   [事件] #238 a2a.artifact artifact(output)
+[03:36:18.851]   [事件] #252 a2a.message ?→?: 任务「温度工艺整定(数采+数控)」进度 0% 已 360s 未变化,请确认是否仍在推进;若卡住请说明阻塞并请求协助
+[03:38:25.031]   [事件] #270 error {"code":"CODEX_ERROR","message":"[object Object]"}
+[03:38:25.031]   [事件] #271 task.status 温度工艺整定(数采+数控) → FAILED
+[03:38:54.565]   [WS/hitl.request] codex-approval: codex 命令审批:"C:\\WINDOWS\\System32\\WindowsPowerShell\\v1.0\\powershell.exe" -Command '$src='"'D:\\codes\\ABO\\AgentWorkShop\\data\\workshop.sqlite'; "'$dst="$PWD'"\\workshop-copy.sqlite\"; Copy-Item -LiteralPath
+[03:38:54.565]   [WS/hitl.request] codex-approval: codex 命令审批:"C:\\WINDOWS\\System32\\WindowsPowerShell\\v1.0\\powershell.exe" -Command '$src='"'D:\\codes\\ABO\\AgentWorkShop\\data\\workshop.sqlite'; "'$dst="$PWD'"\\workshop-copy.sqlite\"; Copy-Item -LiteralPath
+[03:38:54.565]   [WS/hitl.request] codex-approval: codex 命令审批:"C:\\WINDOWS\\System32\\WindowsPowerShell\\v1.0\\powershell.exe" -Command '$src='"'D:\\codes\\ABO\\AgentWorkShop\\data\\workshop.sqlite'; "'$dst="$PWD'"\\workshop-copy.sqlite\"; Copy-Item -LiteralPath
+[03:38:55.160]   [事件] #274 hitl.request codex-approval:codex 命令审批:"C:\\WINDOWS\\System32\\WindowsPowerShell\\v1.0\\powershell.exe" -Command '$src='"'D:\\codes\\ABO\\AgentWorkShop\\data\\workshop.sqlite'; "'$dst="$PWD'"\\workshop-copy.sqlite\"; Copy-Item -LiteralPath
+[03:39:35.331]   [事件] #277 task.status 温度工艺整定(数采+数控) → ASSIGNED
+[03:39:35.331]   [事件] #279 task.status 温度工艺整定(数采+数控) → WORKING
+[03:39:42.696]   [事件] #280 a2a.message ?→?: 【收口请求,只做簿记不做执行】背景:平台曾把同目标 goal 任务 b1b6ba41「温度工艺整定(数采+数控)」重复派给 worker-codex,其停滞后被结算 FAILED;父任务 2d2d0a82 在它未 set
+[03:40:22.373]   [WS/hitl.request] dcw-approval: DCW 下发审批
+[03:40:22.373]   [WS/hitl.request] dcw-approval: DCW 下发审批
+[03:40:22.373]   [WS/hitl.request] dcw-approval: DCW 下发审批
+[03:40:23.530]   [事件] #286 hitl.request dcw-approval:DCW 下发审批
+[03:42:11.972]   [事件] #291 task.status 温度工艺整定(数采+数控) → COMPLETED
+[03:42:11.972]   [事件] #292 a2a.artifact artifact(goal-summary)
+[03:42:11.972]   [事件] #293 task.status 温度工艺整定(数采+数控) → WORKING
+[03:42:11.972]   [事件] #294 a2a.message ?→?: b1b6ba41 已收口(complete_task 已调用,deliverable 采用 03:32:42 四项证据)。补充两点如实汇报:① 本回合我复核过现场——daq_query 近5分钟无数据(停线),dcw_r
+[03:42:23.565]   [事件] #298 a2a.artifact artifact(output)
+[03:42:26.704]   [事件] #300 a2a.message ?→?: 两点均已知悉:① b1b6ba41 收口确认,感谢复核(无数据/设定仍生效)且遵守了未再写入;② dw-c5cc2e42 上那条 175.0 待审批指令(审批超时遗留,值与现设定相同)——我会在父任务收口交付与班组时间线
+[03:42:30.871]   [事件] #303 a2a.artifact artifact(output)
+[03:42:35.112]   PASS  团队任务闭环 COMPLETED — state=COMPLETED
+[03:42:35.127]   PASS  写入生效:dcw_read 读回 175 附近 — 读取成功:联调温度设定器(烘箱温度设定)   PLC 读数(ACT): 175℃ @ 03:42:35   当前设定(SET): 175℃   对照: 读数与设定一致(设定已生效)   原始值(raw): 175
+[03:42:35.154]   PASS  超时场景消息已投递(下调 160) — status=200
+[03:42:35.162]   [事件] #305 task.status 温度工艺整定(数采+数控) → COMPLETED
+[03:42:35.162]   [事件] #306 a2a.message ?→?: 请用 dcw_control 工具将「联调温度设定器」下调至 160.0(hypothesis=超时路径验证)。该写入需要人工审批:请发起审批后耐心等待;若审批被拒绝或超时未批,请如实汇报等待与结果,不要重复发起。
+[03:42:35.174]   PASS  第二次写入触发审批 — id=ap-623e8813 expiresAt=2026-09-05T03:43:22.373Z
+[03:42:35.174] 测试员故意不批准 — 等待审批超时上限(security.hitl_timeout_ms,默认 180s)…
+[03:43:22.252]   [事件] #315 a2a.message ?→?: 需要你协调:测试员(非名册成员,我无法直接寻址回复)刚指示我将 dw-c5cc2e42 下调至 160.0(hypothesis=超时路径验证,走人工审批)。问题:我调用 dcw_control 被平台拒绝——该节点上仍
+[03:43:22.386]   [WS/hitl.resolved] dcw-approval outcome=expired
+[03:43:22.386]   [WS/hitl.resolved] dcw-approval outcome=expired
+[03:43:22.386]   [WS/hitl.resolved] dcw-approval outcome=expired
+[03:43:23.654]   [事件] #316 hitl.resolved dcw-approval outcome=expired
+[03:43:23.654]   PASS  审批到上限自动收敛(expired) — outcome=expired
+[03:43:23.658]   PASS  审批从待办清单移除
+[03:43:23.666]   PASS  拒绝后值未变(仍 ~175,非 160) — 读取成功:联调温度设定器(烘箱温度设定)   PLC 读数(ACT): 175℃ @ 03:43:23   当前设定(SET): 175℃   对照: 读数与设定一致(设定已生效)   原始值(raw): 175
+[03:43:52.390]   [事件] #318 a2a.message ?→?: 【升级确认】收到你的阻塞上报:dw-c5cc2e42 残留的 175.0 待审批指令使新的 dcw_control(160.0)被平台拒绝——这是节点级「单挂起审批」约束,你处理正确(未重复发起)。
+处置:我正通过决策弹
+[03:43:54.568]   [WS/hitl.resolved] codex-approval outcome=cancelled
+[03:43:54.568]   [WS/hitl.resolved] codex-approval outcome=cancelled
+[03:43:54.568]   [WS/hitl.resolved] codex-approval outcome=cancelled
+[03:43:55.747]   [事件] #319 hitl.resolved codex-approval outcome=cancelled
+[03:43:56.401]   [事件] #322 a2a.message ?→?: 收到,明确执行:收到你的 go 信号前不调用 dcw_control,仅以 dcw_read 保持对 dw-c5cc2e42 的设定/读数观察、被问及时汇报;收到 go 后仅重新发起一次 160.0(hypothesis
+[03:43:58.008]   [WS/hitl.request] omp-dialog: dw-c5cc2e42「联调温度设定器」残留一条 175.0 待审批指令(此前审批超时遗留,值与当前设定相同),它已阻塞 worker-dsh 按您的指示发起 160.0 下调(平台拒绝同节点并发挂起审批)。请选择处置方式:
+[03:43:58.008]   [WS/hitl.request] omp-dialog: dw-c5cc2e42「联调温度设定器」残留一条 175.0 待审批指令(此前审批超时遗留,值与当前设定相同),它已阻塞 worker-dsh 按您的指示发起 160.0 下调(平台拒绝同节点并发挂起审批)。请选择处置方式:
+[03:43:58.008]   [WS/hitl.request] omp-dialog: dw-c5cc2e42「联调温度设定器」残留一条 175.0 待审批指令(此前审批超时遗留,值与当前设定相同),它已阻塞 worker-dsh 按您的指示发起 160.0 下调(平台拒绝同节点并发挂起审批)。请选择处置方式:
+[03:43:58.400]   [事件] #323 hitl.request omp-dialog:dw-c5cc2e42「联调温度设定器」残留一条 175.0 待审批指令(此前审批超时遗留,值与当前设定相同),它已阻塞 worker-dsh 按您的指示发起 160.0 下调(平台拒绝同节点并发挂起审批)。请选择处置方式:
+[03:43:59.867]   [事件] #325 a2a.artifact artifact(output)
+[03:45:51.597]   [WS/hitl.resolved] omp-dialog outcome=cancelled
+[03:45:51.597]   [WS/hitl.resolved] omp-dialog outcome=cancelled
+[03:45:51.597]   [WS/hitl.resolved] omp-dialog outcome=cancelled
+[03:45:52.854]   [事件] #329 hitl.resolved omp-dialog outcome=cancelled
+[03:46:05.125]   [WS/hitl.request] omp-dialog: dw-c5cc2e42「联调温度设定器」残留一条 175.0 待审批指令(审批超时遗留,值与当前设定相同),平台的防堆积机制已拒绝 worker-dsh 发起您的 160.0 下调(非对 160.0 的否决;当前 ACT=SET=175℃ 未变,dsh 已按规挂起)。请处置:
+[03:46:05.125]   [WS/hitl.request] omp-dialog: dw-c5cc2e42「联调温度设定器」残留一条 175.0 待审批指令(审批超时遗留,值与当前设定相同),平台的防堆积机制已拒绝 worker-dsh 发起您的 160.0 下调(非对 160.0 的否决;当前 ACT=SET=175℃ 未变,dsh 已按规挂起)。请处置:
+[03:46:05.125]   [WS/hitl.request] omp-dialog: dw-c5cc2e42「联调温度设定器」残留一条 175.0 待审批指令(审批超时遗留,值与当前设定相同),平台的防堆积机制已拒绝 worker-dsh 发起您的 160.0 下调(非对 160.0 的否决;当前 ACT=SET=175℃ 未变,dsh 已按规挂起)。请处置:
+[03:46:06.561]   [事件] #345 hitl.request omp-dialog:dw-c5cc2e42「联调温度设定器」残留一条 175.0 待审批指令(审批超时遗留,值与当前设定相同),平台的防堆积机制已拒绝 worker-dsh 发起您的 160.0 下调(非对 160.0 的否决;当前 ACT=SET=175℃ 未变,dsh 已按规挂起)。请处置:
+[03:48:23.717]   FAIL  worker 回执汇报超时/拒绝结果 — 300s 未回
+[03:48:23.717] REST 事件历史共 350 帧;WS 实时收到 51916 帧
+[03:48:23.717] 帧分布: {"agent.member":4,"a2a.message":19,"agent.status":40,"agent.delta":178,"agent.status.message":65,"a2a.artifact":12,"agent.message":4,"task.status":13,"task.progress":2,"hitl.request":5,"hitl.resolved":4,"memory.saved":3,"error":1}
+[03:48:23.718]   PASS  事件流覆盖关键类型(task/status/message/hitl) — {"agent.member":4,"a2a.message":19,"agent.status":40,"agent.delta":178,"agent.status.message":65,"a2a.artifact":12,"agent.message":4,"task.status":13,"task.prog

package/scripts/_dbg-registry-smoke.ts

@@ -0,0 +1,14 @@
+/**
+ * registry 冒烟:全部 harness 注册与能力面(调试脚本)。
+ * 运行:npx tsx scripts/_dbg-registry-smoke.ts
+ */
+async function main() {
+  const m = await import('../server/services/workshop/agents/registry')
+  const metas = m.harnessMetas()
+  console.log('harnesses:', metas.map(h => h.id).join(','))
+  for (const h of metas) {
+    console.log(`  ${h.id}: caps=${JSON.stringify(h.capabilities)}`)
+  }
+  void main
+}
+void main()

package/scripts/_dbg-sim-keeper.mjs

@@ -0,0 +1,100 @@
+/**
+ * 模拟工况看护:每 5s 探测各模拟器端口,死亡则以 detached 方式重启。
+ *   node scripts/_dbg-sim-keeper.mjs           (常驻;Ctrl+C 退出,不影响已拉起的模拟器)
+ *   node scripts/_dbg-sim-keeper.mjs --once    (单轮自愈后退出;供 schtasks 每分钟调度)
+ * 附带 dev server(:3000)看护:死亡且锁 pid 不存活 → 清陈旧锁后重启。
+ */
+import net from 'node:net'
+import { spawn, spawnSync } from 'node:child_process'
+import { openSync, existsSync, readFileSync, unlinkSync } from 'node:fs'
+import { dirname, resolve } from 'node:path'
+import { fileURLToPath } from 'node:url'
+
+/** Windows 可靠的 pid 存活探测(process.kill(pid,0) 在 Windows 对死 pid 常误报存活) */
+function pidAlive(pid) {
+  if (!Number.isInteger(pid) || pid <= 0) return false
+  const out = spawnSync('tasklist', ['/FI', `PID eq ${pid}`, '/FO', 'CSV', '/NH'], { encoding: 'utf-8' })
+  const line = (out.stdout ?? '').trim()
+  return line.length > 0 && !/INFO:\s*no tasks/i.test(line)
+}
+
+const repo = resolve(dirname(fileURLToPath(import.meta.url)), '..')
+const out = resolve(repo, '.AgentWorkShop', 'data')
+const ONCE = process.argv.includes('--once')
+
+const SIMS = [
+  { name: 'mqtt-http', port: 18830, cmd: 'scripts/dev-protocol-simulators.mjs', env: { MQTT_SIM_PORT: '18830', HTTP_SIM_PORT: '1889' }, log: 'sim-mqtt-http.log' },
+  { name: 'http-sim', port: 1889, cmd: null },
+  { name: 'modbus-tcp', port: 1502, cmd: 'scripts/dev-modbus-simulator.mjs', env: {}, log: 'sim-modbus-tcp.log' },
+  { name: 'opcua', port: 4840, cmd: 'scripts/dev-opcua-simulator.mjs', env: { OPCUA_SIM_PORT: '4840' }, log: 'sim-opcua.log' },
+  { name: 'plc', port: 15040, cmd: 'scripts/dev-plc-simulator.mjs', env: {}, args: ['--port', '15040'], log: 'sim-plc.log' },
+  { name: 'rtu', port: 15030, cmd: 'scripts/_rtu-mini-slave.mjs', env: {}, log: 'sim-rtu.log' },
+]
+
+function portUp(port) {
+  return new Promise((resolveDone) => {
+    const s = net.connect(port, '127.0.0.1')
+    const t = setTimeout(() => { s.destroy(); resolveDone(false) }, 1200)
+    s.on('connect', () => { clearTimeout(t); s.destroy(); resolveDone(true) })
+    s.on('error', () => { clearTimeout(t); resolveDone(false) })
+  })
+}
+
+function launch(sim) {
+  if (!sim.cmd) return
+  const fd = openSync(resolve(out, sim.log), 'a')
+  const child = spawn(process.execPath, [resolve(repo, sim.cmd), ...(sim.args ?? [])], {
+    cwd: repo,
+    env: { ...process.env, NO_PROXY: '127.0.0.1,localhost', no_proxy: '127.0.0.1,localhost', ...sim.env },
+    detached: true,
+    stdio: ['ignore', fd, fd],
+  })
+  child.unref()
+  console.log(`[keeper ${new Date().toISOString()}] ${sim.name} down → respawn pid=${child.pid}`)
+}
+
+async function tick() {
+  for (const sim of SIMS) {
+    const up = await portUp(sim.port)
+    if (!up) launch(sim)
+  }
+  // dev server 看护:down → 锁 pid 活着(启动中)等它;锁 pid 死/无锁 → 清锁重启
+  const devUp = await portUp(3000)
+  if (!devUp) {
+    const lockPath = resolve(repo, '.AgentWorkShop', '.runtime', 'aw.lock')
+    let lockPid = 0
+    try {
+      if (existsSync(lockPath)) {
+        const lock = JSON.parse(readFileSync(lockPath, 'utf-8'))
+        lockPid = Number(lock.pid ?? 0)
+      }
+    }
+    catch { /* 无锁/坏锁 */ }
+    if (lockPid && pidAlive(lockPid)) {
+      // 僵尸 guard:占锁不服务(worker 被外部击杀后 guard 悬挂)→ 强杀进程树再重启
+      spawnSync('taskkill', ['/PID', String(lockPid), '/T', '/F'])
+      console.log(`[keeper ${new Date().toISOString()}] dev-server 僵尸 guard(pid=${lockPid})已强杀`)
+    }
+    try { unlinkSync(lockPath) } catch { /* 无锁 */ }
+    const fd = openSync(resolve(out, 'dev-server.log'), 'a')
+    const child = spawn(process.execPath, ['bin/aw.mjs', 'dev'], {
+      cwd: repo,
+      env: { ...process.env, NO_PROXY: '127.0.0.1,localhost', no_proxy: '127.0.0.1,localhost' },
+      detached: true,
+      stdio: ['ignore', fd, fd],
+    })
+    child.unref()
+    console.log(`[keeper ${new Date().toISOString()}] dev-server down(锁已清)→ respawn pid=${child.pid}`)
+  }
+}
+
+const sleep = ms => new Promise(r => setTimeout(r, ms))
+console.log(`[keeper] watching(once=${ONCE}):`, SIMS.map(s => `${s.name}:${s.port}`).join(' '), '+ dev:3000')
+if (ONCE) {
+  try { await tick() } catch (err) { console.error('[keeper] tick error:', err?.message ?? err) }
+} else {
+  for (;;) {
+    try { await tick() } catch (err) { console.error('[keeper] tick error:', err?.message ?? err) }
+    await sleep(5000)
+  }
+}

package/scripts/_dbg-team-drill.mjs

@@ -0,0 +1,173 @@
+/**
+ * 四引擎班组精简演练(真实服务 + 模拟工况数据):
+ *  lead=omp;worker-dsh=数采(daq_query);worker-codex=数控写入(dcw_control,HITL)+复读;
+ *  worker-opencode=数控读(dcw_read)。单 goal 任务,审批到达即批准,最小 token 消耗。
+ * 运行:node scripts/_dbg-team-drill.mjs
+ */
+const BASE = process.env.AW_BASE ?? 'http://127.0.0.1:3000'
+let failures = 0
+const check = (name, ok, detail = '') => {
+  console.log(`  ${ok ? 'PASS' : 'FAIL'}  ${name}${detail ? ` — ${detail}` : ''}`)
+  if (ok) passed += 1
+  else failures += 1
+}
+let passed = 0
+const sleep = (ms) => new Promise(r => setTimeout(r, ms))
+const log = (l) => console.log(`[${new Date().toISOString().slice(11, 19)}] ${l}`)
+
+const reg = await fetch(`${BASE}/api/workshop/users/register`, { method: 'POST', headers: { 'content-type': 'application/json' }, body: JSON.stringify({ name: 'drill-' + Math.random().toString(36).slice(2, 8) }) }).then(r => r.json())
+const T = reg?.data?.token
+if (!T) { console.error('注册失败(服务未启动?)'); process.exit(1) }
+const sleep2 = (ms) => new Promise(r => setTimeout(r, ms))
+const api = async (method, path, body, attempt = 0) => {
+  try {
+    const res = await fetch(`${BASE}${path}`, { method, headers: { 'content-type': 'application/json', authorization: `Bearer ${T}` }, body: body !== undefined ? JSON.stringify(body) : undefined })
+    return await res.json().catch(() => ({}))
+  }
+  catch (e) {
+    if (attempt < 3) {
+      await sleep2(1500)
+      return await api(method, path, body, attempt + 1)
+    }
+    throw e
+  }
+}
+
+// ── WS 录制流(事件持久化由首个订阅触发)──
+const { createRequire } = await import('node:module')
+const { WebSocket } = createRequire(import.meta.url)('D:/codes/ABO/AgentWorkShop/node_modules/.pnpm/ws@8.21.3/node_modules/ws')
+
+const seen = []
+let lastSeq = 0
+let ws = null
+function startWs(channelId) {
+  ws = new WebSocket(BASE.replace(/^http/, 'ws') + '/api/workshop/ws')
+  ws.on('open', () => ws.send(JSON.stringify({ type: 'sub', channelId, token: T })))
+  ws.on('message', (raw) => {
+    try {
+      const f = JSON.parse(String(raw))
+      if (f.type === 'hitl.request') log(`[WS] HITL ${f.payload?.kind}: ${String(f.payload?.detail ?? f.payload?.title ?? '').slice(0, 90)}`)
+      if (f.type === 'hitl.resolved') log(`[WS] HITL resolved outcome=${f.payload?.outcome}`)
+      if (f.type === 'task.status') log(`[WS] task ${f.payload?.state} ${(f.payload?.title ?? '').slice(0, 40)}`)
+    }
+    catch { /* ignore */ }
+  })
+  ws.on('error', () => {})
+}
+async function pollEvents(channelId) {
+  const res = await api('GET', `/api/workshop/channels/${channelId}/events?limit=500`)
+  for (const f of (res?.data?.items ?? [])) {
+    if (Number(f.seq ?? 0) > lastSeq) {
+      lastSeq = Number(f.seq)
+      seen.push(f)
+    }
+  }
+}
+async function waitUntil(cond, timeoutMs, intervalMs = 1500) {
+  const deadline = Date.now() + timeoutMs
+  while (Date.now() < deadline) {
+    try {
+      if (await cond()) return true
+    }
+    catch { /* 瞬态网络/重载抖动,按条件未满足继续轮询 */ }
+    await sleep(intervalMs)
+  }
+  return false
+}
+
+// ── 1. 产线 + 模拟工况节点 ──
+const line = await api('POST', '/api/workshop/dcw/lines', { name: '四引擎演练线' })
+const dcw = await api('POST', '/api/workshop/dcw', { templateRef: 'dcw-temp-sp', name: '演练温度设定器', driver: 'mock', lineId: line?.data?.line?.id })
+const daq = await api('POST', '/api/workshop/daq', { templateRef: 'daq-temp-tc', name: '演练温度采集', driver: 'mock', lineId: line?.data?.line?.id, intervalMs: 1000 })
+const dcwId = dcw?.data?.node?.id
+const daqId = daq?.data?.node?.id
+check('产线+数控节点+数采节点(mock 工况)创建', !!dcwId && !!daqId, `dcw=${dcwId} daq=${daqId}`)
+
+// ── 2. 四引擎班组 ──
+const ch = await api('POST', '/api/workshop/channels', {
+  name: '四引擎班组演练',
+  leadAgent: { name: 'lead-omp', harness: 'omp', config: { provider: 'zhipu-coding-plan', model: 'glm-5.2' } },
+})
+const cid = ch?.data?.channelId
+check('创建班组(lead=omp)', !!cid)
+startWs(cid)
+
+async function addWorker(name, harness, config) {
+  const w = await api('POST', `/api/workshop/channels/${cid}/agents`, { name, harness, role: 'worker', config })
+  return w?.data?.id ?? null
+}
+const dsh = await addWorker('worker-dsh', 'dsh', { promptTimeoutMs: 480_000 })
+// codex 需 per-agent CODEX_HOME(种子化全局凭据/网关配置;impl 追加 [mcp_servers.aw] 平台工具面)
+const { mkdtempSync, copyFileSync, existsSync } = await import('node:fs')
+const { tmpdir, homedir } = await import('node:os')
+const { join } = await import('node:path')
+const codexHome = mkdtempSync(join(tmpdir(), 'aw-drill-codex-'))
+for (const f of ['auth.json', 'config.toml', 'cc-switch-model-catalog.json']) {
+  try { if (existsSync(join(homedir(), '.codex', f))) copyFileSync(join(homedir(), '.codex', f), join(codexHome, f)) } catch { /* ignore */ }
+}
+const codex = await addWorker('worker-codex', 'codex', { codexHome, approvalPolicy: 'on-request', sandbox: 'workspace-write', promptTimeoutMs: 480_000, systemPromptPrefix: '执行要直接:按指令调用平台工具并汇报,不探索文件系统。' })
+const ocTmp = (await import('node:fs')).mkdtempSync((await import('node:os')).tmpdir() + '/aw-drill-oc-')
+const ocCfg = (await import('node:fs')).mkdtempSync((await import('node:os')).tmpdir() + '/aw-drill-occ-')
+try { (await import('node:fs')).mkdirSync(ocTmp + '/opencode', { recursive: true }); (await import('node:fs')).copyFileSync((await import('node:os')).homedir() + '/.local/share/opencode/auth.json', ocTmp + '/opencode/auth.json') } catch { /* ignore */ }
+const opencode = await addWorker('worker-opencode', 'opencode', { promptTimeoutMs: 480_000, model: 'zhipuai-coding-plan/glm-5.3-flash', dataDir: ocTmp, configDir: ocCfg, baseUrl: BASE })
+check('三 worker 入组(dsh/codex/opencode)', !!dsh && !!codex && !!opencode, `dsh=${dsh?.slice(0, 8)} codex=${codex?.slice(0, 8)} oc=${opencode?.slice(0, 8)}`)
+
+// ── 3. 绑定工业节点 ──
+await api('POST', '/api/workshop/agent-tools/bindings', { agentId: dsh, nodeId: daqId, kind: 'daq', mode: 'auto' })
+await api('POST', '/api/workshop/agent-tools/bindings', { agentId: codex, nodeId: dcwId, kind: 'dcw', mode: 'manual' })
+await api('POST', '/api/workshop/agent-tools/bindings', { agentId: opencode, nodeId: dcwId, kind: 'dcw', mode: 'manual' })
+check('绑定完成(dsh↔DAQ auto;codex/opencode↔DCW manual)', true)
+
+// ── 4. 团队作业(单 goal,三步,指令点对点到人)──
+const task = await api('POST', `/api/workshop/channels/${cid}/tasks`, {
+  title: '四引擎数采数控联调',
+  mode: 'goal',
+  description: [
+    '按以下分工执行,每步只做一次,不要额外发挥:',
+    '1. worker-dsh:用 daq_query 工具查询你绑定的温度采集节点最近 5 分钟数据,把样本数与均值汇报给 lead;',
+    '2. worker-codex:用 dcw_control 工具把你绑定的「演练温度设定器」写入 172.0(会触发人工审批,等待批准后再继续),然后用 dcw_read 复读并把读数汇报给 lead;',
+    '3. worker-opencode:用 dcw_read 工具读取你绑定的「演练温度设定器」当前设定值,汇报给 lead;',
+    '4. lead:三项结果齐后调用 complete_task 汇总。',
+  ].join('\n'),
+})
+const taskId = task?.data?.id
+check('团队作业 goal 已提交', !!taskId, `taskId=${taskId?.slice(0, 8)}`)
+
+// ── 5. HITL:审批到达即批准 ──
+const approvalsDone = new Set()
+const approved = await waitUntil(async () => {
+  const p = await api('GET', '/api/workshop/hitl/pending')
+  const items = (p?.data?.items ?? []).filter(i => i.kind === 'dcw-approval' && !approvalsDone.has(i.id))
+  for (const item of items) {
+    log(`审批到达: ${String(item.detail ?? '').slice(0, 80)} → 批准`)
+    const r = await api('POST', '/api/workshop/hitl/respond', { kind: 'dcw-approval', id: item.id, confirmed: true, comment: '演练批准' })
+    approvalsDone.add(item.id)
+    if (r?.data?.ok) log(`已批准 ${item.id}`)
+  }
+  return approvalsDone.size > 0 && (pollEvents(), seen.some(f => f.type === 'hitl.resolved' && f.payload?.outcome === 'answered'))
+}, 600_000)
+check('数控写入审批触发并批准', approved, `批准数=${approvalsDone.size}`)
+
+// ── 6. 任务闭环 + 写入落地 ──
+const done = await waitUntil(async () => {
+  const t = await api('GET', `/api/workshop/tasks/${taskId}`)
+  const st = t?.data?.state
+  if (st === 'COMPLETED' || st === 'FAILED' || st === 'CANCELED') return st
+  await pollEvents(cid)
+  return null
+}, 900_000)
+check('团队任务闭环 COMPLETED', done === 'COMPLETED', `state=${done}`)
+
+const read = await api('POST', '/api/workshop/agent-tools/invoke', { agentId: codex, tool: 'dcw_read', args: { node_id: dcwId } })
+const rb = String(read?.data?.result?.text ?? '')
+check('写入物理生效(读回 ~172)', /17[12]/.test(rb), rb.replace(/\n/g, ' ').slice(0, 110))
+
+await pollEvents(cid)
+const byType = {}
+for (const f of seen) byType[f.type] = (byType[f.type] ?? 0) + 1
+console.log('事件帧分布:', JSON.stringify(byType))
+check('事件流覆盖(task/message/hitl)', (byType['task.status'] ?? 0) > 0 && (byType['a2a.message'] ?? 0) > 0)
+
+console.log(`\n━━━ 演练结论:PASS=${passed} FAIL=${failures} ━━━`)
+try { ws?.close() } catch { /* ignore */ }
+process.exit(failures === 0 ? 0 : 1)

package/scripts/dev-guard.mjs

@@ -29,6 +29,10 @@ const { existsSync, readFileSync } = await import('node:fs')
 const { dirname, resolve } = await import('node:path')
 const { fileURLToPath: u2f } = await import('node:url')
 
+// dev 进程时间输出统一本地时区(重入实例同样生效;nitro worker 由 00-local-time 插件覆盖)
+const { installLocalIso } = await import('../shared/local-time.mjs')
+installLocalIso()
+
 const root = resolve(dirname(u2f(import.meta.url)), '..')
 
 // --- 预载 .env（与 start.mjs 同策略；已导出环境变量优先） ---
@@ -43,9 +47,13 @@ if (existsSync(envPath)) {
 
 // --- 共享配置引擎：计算有效开发端口/主机 ---
 const { loadEffective } = await import('../shared/config/engine.mjs')
+// 模式感知路径单一入口:settings/锁根必须与 start/CLI 同源,
+// 否则"检出内无 .AgentWorkShop"场景下 dev 写 data/、start 锁 home,互斥与设置双双失效。
+const { resolveRunMode } = await import('../shared/config/home.mjs')
+const rmDev = resolveRunMode({ cwd: process.cwd(), packageRoot: root, env: process.env })
 const eff = loadEffective({
-  configPath: resolve(root, 'config.yml'),
-  settingsPath: resolve(root, 'data', 'runtime-settings.json'),
+  configPath: rmDev.configPath,
+  settingsPath: rmDev.settingsPath,
   env: process.env,
   mode: 'dev',
 })
@@ -76,7 +84,7 @@ console.log(`[config] worker PORT env -> ${effectivePort}`)
 // --- 单实例互斥 + 端口占用显式报错(hardening ST-1/ST-4):不再静默换端口 ---
 {
   const { acquireLock, checkPort } = await import('../shared/config/single-instance.mjs')
-  const lock = acquireLock(resolve(root, '.AgentWorkShop'), { mode: 'dev', port: effectivePort })
+  const lock = acquireLock(rmDev.configRoot, { mode: 'dev', port: effectivePort })
   if (!lock.ok) {
     const h = lock.holder
     console.error(`✖ 已有实例在运行(pid=${h.pid}${h.port ? `,端口=${h.port}` : ''},启动于 ${h.startedAt ?? '未知'},模式=${h.mode ?? '未知'})`)

package/scripts/dev-protocol-simulators.mjs

@@ -8,6 +8,12 @@
 import net from 'node:net'
 import http from 'node:http'
 import { createRequire } from 'node:module'
+import { appendFileSync } from 'node:fs'
+import { join } from 'node:path'
+import { installLocalIso } from '../shared/local-time.mjs'
+
+// 模拟器日志时间统一本地时区(覆盖 node-opcua 等库内部的 toISOString 输出)
+installLocalIso()
 
 const reqRoot = createRequire(import.meta.url)
 // pnpm 严格布局:mqtt-packet 是 mqtt 的传递依赖,从 mqtt 包入口路径解析
@@ -31,6 +37,14 @@ const clients = new Set()
 
 function brokerPublish(topic, payload) {
   const frame = mqttPacket.generate({ cmd: 'publish', topic, payload, qos: 0 })
+  console.log(`[mqtt-sim] route topic=${topic} clients=${clients.size} filters=[${[...clients].map(c => c.filters.join('|')).join(' ; ')}]`)
+  // 控制下行审计:broker 真实收到的路由落盘(供 e2e 做确定性验收,不依赖订阅端时序)
+  if (topic.startsWith('aw/sim/setpoint')) {
+    try {
+      appendFileSync(join(process.env.AW_SIM_AUDIT_DIR ?? '.AgentWorkShop/data', 'sim-setpoint-routes.log'), `${new Date().toISOString()} ${topic} ${payload.toString()}\n`)
+    }
+    catch { /* 审计失败不影响路由 */ }
+  }
   for (const c of clients) {
     if (c.filters.some(f => topicMatch(f, topic))) {
       try {
@@ -45,7 +59,20 @@ const mqttServer = net.createServer((sock) => {
   const client = { sock, filters: [] }
   clients.add(client)
   const parser = mqttPacket.parser({ protocolVersion: 4 })
-  sock.on('data', d => parser.parse(d))
+  sock.on('data', (d) => {
+    // 畸形帧(截断/坏头字节)只弃帧断该链,不许杀掉整个模拟工况
+    try {
+      parser.parse(d)
+    }
+    catch (err) {
+      console.error('[mqtt-sim] 弃帧(解析异常):', err instanceof Error ? err.message : err)
+      try {
+        sock.destroy()
+      }
+      catch { /* 已断 */ }
+      clients.delete(client)
+    }
+  })
   sock.on('error', () => clients.delete(client))
   sock.on('close', () => clients.delete(client))
   parser.on('packet', (packet) => {