agentworkshop 0.7.33 → 0.7.34

package/.output/server/package.json

@@ -1,6 +1,6 @@
 {
   "name": "agentworkshop-prod",
-  "version": "0.7.33",
+  "version": "0.7.34",
   "type": "module",
   "private": true,
   "dependencies": {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "agentworkshop",
-  "version": "0.7.33",
+  "version": "0.7.34",
   "type": "module",
   "description": "AgentWorkShop 软件开发系统 - 基于 Nuxt 4 的配置驱动运行范式",
   "license": "PolyForm-Noncommercial-1.0.0",

package/scripts/_dbg-harness-cancel-repro.mjs

@@ -0,0 +1,81 @@
+/**
+ * codex/dsh 任务终态复现:单引擎 × 单任务,保留现场,输出任务状态时间线/消息/事件/
+ * 工具面,判定「谁在何时把任务置为 CANCELED」。
+ * 用法:NO_PROXY='*' AW_BASE=http://127.0.0.1:3001 AW_E2E_TOKEN=<tok> HARNESS=codex node scripts/_dbg-harness-cancel-repro.mjs
+ */
+const BASE = process.env.AW_BASE ?? 'http://127.0.0.1:3001'
+const TOKEN = process.env.AW_E2E_TOKEN ?? ''
+const HARNESS = process.env.HARNESS ?? 'codex'
+const TAG = `cx${Math.random().toString(36).slice(2, 6)}`
+const sleep = (ms) => new Promise(r => setTimeout(r, ms))
+
+async function api(method, path, body) {
+  const res = await fetch(BASE + path, {
+    method,
+    headers: { authorization: `Bearer ${TOKEN}`, 'content-type': 'application/json' },
+    body: body === undefined ? undefined : JSON.stringify(body),
+  })
+  return res.json()
+}
+
+const tpl = await api('POST', '/api/workshop/agents', { name: `${HARNESS}-repro-${TAG}`, harness: HARNESS })
+const agentId = tpl?.data?.id
+console.log(`template=${agentId} harness=${HARNESS}`)
+const ch = await api('POST', '/api/workshop/channels', { name: `${HARNESS}-repro-${TAG}` })
+const channelId = ch?.data?.channelId ?? ch?.data?.id ?? ch?.data?.channel?.id
+// lead(渠道须有 lead 才能收任务;mock 引擎足够)
+const leadTpl = await api('POST', '/api/workshop/agents', { name: `${HARNESS}-repro-lead-${TAG}`, harness: 'mock' })
+await api('POST', `/api/workshop/channels/${channelId}/agents`, { agentId: leadTpl?.data?.id, role: 'lead' })
+await api('POST', `/api/workshop/channels/${channelId}/agents`, { agentId, role: 'worker' })
+const inst = await api('GET', `/api/workshop/channels/${channelId}/agents`)
+const instId = (inst?.data ?? []).find(a => a.harness === HARNESS)?.id ?? agentId
+console.log(`channel=${channelId} instance=${instId}`)
+
+// 工具面:complete_task 是否存在
+const tools = await api('GET', `/api/workshop/agent-tools/list?agentId=${instId}`)
+const names = (tools?.data ?? tools)?.tools?.map(t => t?.name) ?? (tools?.data ?? tools)?.map?.(t => t?.name) ?? []
+console.log('tool surface:', JSON.stringify(names))
+
+const t = await api('POST', `/api/workshop/channels/${channelId}/tasks`, {
+  title: `repro-${TAG}:读最近日志并回报`,
+  description: '用 ops_log 查看最近 3 条日志,然后调用 complete_task 交付一句 LOGCHECK-OK 开头的总结。',
+})
+const taskId = t?.data?.task?.id ?? t?.data?.id ?? t?.data?.taskId
+console.log('task=', taskId, JSON.stringify(t ?? {}).slice(0, 200))
+
+// 轮询:每 5s 记录状态,直到终态(最长 8 分钟),记录状态变化时间点
+const timeline = []
+let last = ''
+const t0 = Date.now()
+while (Date.now() - t0 < 8 * 60_000) {
+  const me = await api('GET', `/api/workshop/tasks/${taskId}`)
+  const st = me?.data?.state
+  if (st && st !== last) {
+    timeline.push(`t+${Math.round((Date.now() - t0) / 1000)}s ${st}`)
+    console.log(`  t+${Math.round((Date.now() - t0) / 1000)}s -> ${st}`)
+    last = st
+  }
+  if (['COMPLETED', 'FAILED', 'CANCELED'].includes(st)) break
+  await sleep(5000)
+}
+
+// 现场:任务详情 / 频道消息 / 事件
+const task = await api('GET', `/api/workshop/tasks/${taskId}`)
+const msgs = await api('GET', `/api/workshop/channels/${channelId}/messages?limit=50`)
+const evs = await api('GET', `/api/workshop/channels/${channelId}/events?limit=120`)
+const fs = await import('node:fs')
+fs.writeFileSync(`.e2e-shots/repro-${HARNESS}-task.json`, JSON.stringify(task?.data ?? task, null, 1))
+fs.writeFileSync(`.e2e-shots/repro-${HARNESS}-msgs.json`, JSON.stringify(msgs?.data ?? msgs, null, 1))
+fs.writeFileSync(`.e2e-shots/repro-${HARNESS}-events.json`, JSON.stringify(evs?.data ?? evs, null, 1))
+console.log('timeline:', timeline.join(' | '))
+const blob = JSON.stringify(task?.data ?? {}) + JSON.stringify(msgs?.data ?? {}) + JSON.stringify(evs?.data ?? {})
+console.log('LOGCHECK-OK in transcript:', blob.includes('LOGCHECK-OK'))
+console.log('implicit-complete marker:', blob.includes('隐式收口'))
+// 消息里的角色/内容前 3 条(找 agent 最后说了什么)
+const list = msgs?.data ?? []
+for (const m of list.slice(-3)) {
+  const who = m?.role ?? m?.metadata?.['x-aw-from-label'] ?? '?'
+  const text = (m?.parts ?? []).map(p => p?.text ?? '').join(' ').slice(0, 160)
+  console.log(`  [${who}] ${text}`)
+}
+console.log(`现场已存 .e2e-shots/repro-${HARNESS}-*.json(channel=${channelId} 保留未删)`)

package/scripts/_dbg-opslog-agent-e2e.mjs

@@ -140,6 +140,7 @@ async function main() {
     ok(Boolean(tasks[e.key]), `[${e.key}] 任务下发`)
   }
   const states = {}
+  const lastSeen = {}
   const pollTask = async (id) => {
     for (let i = 0; i < 3; i++) {
       try {
@@ -151,11 +152,13 @@ async function main() {
     return undefined
   }
   const deadline = Date.now() + 20 * 60_000
+  const t0 = Date.now()
   while (Date.now() < deadline && Object.keys(states).length < ENGINES.length) {
     for (const e of ENGINES) {
       if (states[e.key]) continue
       const st = await pollTask(tasks[e.key])
-      if (st && ['COMPLETED', 'FAILED', 'CANCELED'].includes(st)) states[e.key] = st
+      if (st && st !== lastSeen[e.key]) console.log(`  · t+${Math.round((Date.now()-t0)/1000)}s [${e.key}] ${st}`)
+      if (st && ['COMPLETED', 'FAILED', 'CANCELED'].includes(st)) { states[e.key] = st; lastSeen[e.key] = st }
     }
     if (Object.keys(states).length < ENGINES.length) await sleep(6000)
   }
@@ -171,7 +174,7 @@ async function main() {
   }
 
   // ── 清理 ──
-  for (const m of Object.values(members)) await api('DELETE', `/api/workshop/channels/${m.channelId}?purge=1`, { token }).catch(() => {})
+  if (process.env.KEEP) console.log('KEEP=1 保留现场: ' + JSON.stringify(Object.fromEntries(Object.entries(members).map(([k,v])=>[k,v.channelId]))))
   for (const e of [...ENGINES, { key: 'mock' }]) {
     const list = await api('GET', '/api/workshop/agents', { token })
     const tpl = (list.data ?? []).find(x => x.name === `ops-${e.key}-${TAG}`)

package/scripts/production-closed-loop-e2e.mjs

@@ -178,9 +178,10 @@ let autoRunId = ''
   touchReload()
   console.log('  · 已启用 auto_diag(规则:节点 ' + freshNode + ' > 0)并触发热重载')
 
-  // 等待窗 4 分钟:重装载(~20s)+ 下一采样拍 + 上传/启动;mock 引擎可能在窗内就跑完,
-  // 所以 running/completed 均算命中
-  for (let i = 0; i < 48 && !autoRunId; i++) {
+  // 等待窗 8 分钟:插件宿主热重载(~20-40s,含全部插件重装载)+ 下一采样拍(采样间隔
+  // 最长 60s+)+ 自动上传/发起;实测共享环境下整链路可达 4 分钟以上(48×5s 曾在
+  // 触发前 0.4s 超时误报)。mock 引擎可能在窗内就跑完,running/completed 均算命中
+  for (let i = 0; i < 96 && !autoRunId; i++) {
     await sleep(5000)
     const runs = await api('GET', '/api/plugins/diag-bridge/runs', { token: userToken })
     const hit = (runs.json?.runs ?? []).find(x => x.source === 'auto' && !preAuto.has(x.runId) && (x.status === 'running' || x.status === 'completed'))
@@ -291,7 +292,9 @@ console.log('\n── Stage E 知识沉淀验证 ──')
   ok(count > 0, '经验库可读', `count=${count}`)
   ok(hasOpt, '经验库含 optimization 类目(优化方案在册)')
 
-  const s3 = await api('GET', `/api/plugins/rag-bridge/search?q=${encodeURIComponent('闭环优化 设定 判定')}&top_k=3`, { token: userToken })
+  // 检索词须与刚沉淀经验的词法重叠(stage1 为 BM25 关键字检索;查询与语料无词法重叠时
+  // rag 侧有意跳过全局向量回退,避免低分噪声 → 0 命中)。用标题/标签中的确证词。
+  const s3 = await api('GET', `/api/plugins/rag-bridge/search?q=${encodeURIComponent('优化方案 压力设定器 keep production-e2e')}&top_k=3`, { token: userToken })
   ok(s3.json?.success === true && (s3.json?.results?.length ?? 0) > 0, 'two-stage 检索命中闭环优化知识', `results=${s3.json?.results?.length}`)
 }
 

package/server/services/workshop/agents/base-agent.ts

@@ -111,6 +111,15 @@ export abstract class BaseAgentImpl implements AgentInterface {
 
   // ===== run:消息分流(平台语义,全引擎一致) =====
 
+  /**
+   * 装配期注入工作区(AgentRuntime 构造时调用):使 host 工具在首个 run() 之前
+   * 即可直调 —— 否则 REST agent-tools/invoke 在实例未处理过任何消息时会得到
+   * 「workspace 未就绪」(实测生产闭环 Stage D 的 kb_store 偶发因此失败)。
+   */
+  attachWorkspace(ws: AgentRunContext['workspace']): void {
+    if (ws && !this.workspace) this.workspace = ws
+  }
+
   async* run(request: AgentRunRequest, ctx: AgentRunContext): AsyncIterable<AgentEvent> {
     if (!this.workspace) this.workspace = ctx.workspace
     const kind = request.message.metadata?.['x-aw-task-kind']