agentworkshop 0.7.54 → 0.7.59

package/server/services/workshop/agents/industrial/recipe-propose-tools.ts

@@ -0,0 +1,366 @@
+/**
+ * 整包方案工具(recipe_propose)—— 产线 Co-Pilot P2「诊断工程师」的结构化下发面(计划 §5.4)。
+ *
+ * 语义:Agent 提交 1~3 套候选参数包 → 服务端逐参数预检(镜像 recipe 真实执行限界)→
+ * 一张结构化审批单(kind='dcw',nodeId=`recipe-propose:<recipeId>`,payload.schemaVersion=1)
+ * → 人类在产线页审批卡**选定一套**批准(choice)→ 走既有 applyRecipe(overrides,trial) 管线
+ * 整批下发 → 回执逐参数映射并固定携带 runId(供后续轮次效果回访)。
+ *
+ * 铁律落点:
+ *  - recipe 级整包无论绑定 mode 永远挂审批单(本工具不查 binding mode,任何模式都审批);
+ *  - AML 优化循环活动期间硬禁下发(时间互斥,自动降级建议报告);
+ *  - 预检先于审批:越界参数在审批单生成前剔除,「批准必被拒包」不可达;
+ *  - 无依据(basis)与经验引用(exp_ref)的参数不得入包;
+ *  - 多方案批准未携带有效 choice 一律按拒绝收敛(fail-closed,见 tool-approvals.normalizeRecipeProposeDecision)。
+ */
+import { agentBadgeLabel } from '../agent-badge'
+import { getToolApprovals } from '../tool-approvals'
+import { getAgentNodeBindingRepo } from '../node-bindings.repo'
+import { securityRecipeDispatchTimeoutMs } from '../../settings'
+import { assertRecipeOperation } from './recipe-gate'
+import { getDcwController } from '../../dcw/dcw-controller'
+import { assertRecipeOpInterval } from '../../dcw/recipe-op-anchor'
+import { getDcwLineRepo } from '../../dcw/dcw-line.repo'
+import { recordOps } from '../../ops/ops'
+import { evaluateRecipeParamLimits } from '../../dcw/dcw-controller/write'
+import { amlActivityForLine } from './aml-activity'
+import { agentOpsScope } from './ops-tools'
+
+/** 候选包上限(防审批卡被刷屏;1~3 套) */
+const MAX_PACKAGES = 3
+
+/** 入参:单参数声明(snake_case;from 允许缺省 = 服务端自取节点当前值) */
+export interface RecipeProposeParamInput {
+  node_id?: string
+  from?: number | string
+  to?: number | string
+  unit?: string
+  basis?: string
+  exp_ref?: string
+}
+
+/** 入参:单套候选包 */
+export interface RecipeProposePackageInput {
+  name?: string
+  rationale?: string
+  params?: RecipeProposeParamInput[]
+}
+
+/** 审批单 payload 内的参数行(契约:schemaVersion=1;前端审批卡按此渲染) */
+export interface RecipeProposePayloadParam {
+  nodeId: string
+  paramName: string
+  from: number | null
+  to: number
+  unit: string
+  basis: string
+  exp_ref: string
+  preflight: { ok: boolean, reason?: string }
+}
+
+/** 审批单 payload(结构化审批卡与回执映射的共同事实源) */
+export interface RecipeProposePayload {
+  schemaVersion: 1
+  recipeId: string
+  lineId: string
+  packages: Array<{ name: string, rationale: string, params: RecipeProposePayloadParam[] }>
+}
+
+/** 参数行的单行渲染(审批单 detail 文本降级 + 预检明细共用) */
+function paramLine(p: RecipeProposePayloadParam, nodeLabel: string): string {
+  const range = p.from != null ? `${p.from}→${p.to}` : `→${p.to}`
+  const pf = p.preflight.ok ? '预检通过' : `预检失败:${p.preflight.reason ?? '未知原因'}`
+  return `· ${nodeLabel} ${range}${p.unit}${p.preflight.ok ? '' : `【${pf},已从下发集剔除】`}`
+}
+
+/** 工具:recipe_propose —— 提交整包候选方案,挂一张结构化审批单等待人工裁决。 */
+export async function toolRecipePropose(agentId: string, args: {
+  recipe_id?: string
+  packages?: RecipeProposePackageInput[]
+  emergency?: boolean | string
+}): Promise<{ text: string, isError?: boolean }> {
+  // ---------- ① 参数校验(packages 1..3;每参数 node_id/to/basis/exp_ref 必填) ----------
+  const recipeId = String(args.recipe_id ?? '').trim()
+  if (!recipeId) return { text: 'recipe_id 必填(line_context 可看到当前配方 id)。', isError: true }
+  const rawPackages = Array.isArray(args.packages) ? args.packages.filter(p => p && typeof p === 'object') : []
+  if (rawPackages.length === 0) return { text: 'packages 必填:至少 1 套候选方案(每套 { name, rationale, params: [{ node_id, to, basis, exp_ref }] });整包审批的意义就是多方案一次裁决。', isError: true }
+  if (rawPackages.length > MAX_PACKAGES) return { text: `packages 最多 ${MAX_PACKAGES} 套候选方案(收到 ${rawPackages.length} 套);更多方案请分批提交。`, isError: true }
+
+  const problems: string[] = []
+  const packages: Array<{ name: string, rationale: string, params: RecipeProposePayloadParam[] }> = []
+  rawPackages.forEach((pkg, pi) => {
+    const name = String(pkg.name ?? '').trim()
+    if (!name) {
+      problems.push(`方案 ${pi + 1}:缺 name(为方案起名,供审批卡展示)`)
+      return
+    }
+    const rationale = String(pkg.rationale ?? '').trim()
+    const rawParams = Array.isArray(pkg.params) ? pkg.params.filter(p => p && typeof p === 'object') : []
+    if (rawParams.length === 0) {
+      problems.push(`方案「${name}」:params 为空(至少 1 条参数)`)
+      return
+    }
+    const params: RecipeProposePayloadParam[] = []
+    rawParams.forEach((p, x) => {
+      const nodeId = String(p.node_id ?? '').trim()
+      const to = Number(p.to)
+      const basis = String(p.basis ?? '').trim()
+      const expRef = String(p.exp_ref ?? '').trim()
+      const where = `方案「${name}」第 ${x + 1} 条参数`
+      if (!nodeId) problems.push(`${where}:缺 node_id`)
+      if (!Number.isFinite(to)) problems.push(`${where}:to 必须为数字(收到 ${String(p.to)})`)
+      if (p.from !== undefined && p.from !== null && String(p.from).trim() !== '' && !Number.isFinite(Number(p.from))) problems.push(`${where}:from 需为数字或缺省(缺省=服务端自取当前值)`)
+      if (!basis) problems.push(`${where}:缺 basis(为什么这样调;无依据的参数不得入包)`)
+      if (!expRef) problems.push(`${where}:缺 exp_ref(经验/证据引用;无引用的参数不得入包)`)
+      if (!nodeId || !Number.isFinite(to) || !basis || !expRef) return
+      params.push({
+        nodeId,
+        paramName: nodeId,
+        from: p.from === undefined || p.from === null || String(p.from).trim() === '' ? null : Number(p.from),
+        to,
+        unit: String(p.unit ?? '').trim(),
+        basis,
+        exp_ref: expRef,
+        preflight: { ok: true },
+      })
+    })
+    if (params.length > 0) packages.push({ name, rationale, params })
+    else problems.push(`方案「${name}」:全部 ${rawParams.length} 条参数均未通过必填校验(见上),该方案未入审`)
+  })
+  if (packages.length === 0) {
+    return { text: `参数校验失败,未提交审批:\n${problems.map(p => `- ${p}`).join('\n')}\n请补齐后重新提交(每参数必须带 basis 与 exp_ref)。`, isError: true }
+  }
+
+  // ---------- ② 配方存在性与产线归属(权限 mirror toolRecipeApply:节点授权线 ∪ 频道绑线 + 配方内至少一个授权节点) ----------
+  const recipe = getDcwController().listRecipes().find(r => r.id === recipeId)
+  if (!recipe) return { text: `配方 ${recipeId} 不存在。`, isError: true }
+  // 权限模型 v2:绑定 → 二级认证 → 运行门(整包审批本身即 HITL,门只做准入)
+  const gate = assertRecipeOperation(agentId, recipe, 'dispatch')
+  if (!gate.ok) return { text: gate.text, isError: true }
+  const scope = agentOpsScope(agentId)
+  const readableLineIds = scope ? [...new Set([...scope.lineIds, ...scope.boundLineIds])] : []
+  if (!scope || !recipe.lineId || !readableLineIds.includes(recipe.lineId)) {
+    return { text: `无权提交配方 ${recipeId} 的整包方案(该配方不在你负责的产线上)。`, isError: true }
+  }
+  // 权限模型 v2:持有该配方绑定(kind=recipe)即算"配方内授权";兼容历史 dcw 节点绑定
+  const bindings = getAgentNodeBindingRepo().byAgent(agentId)
+  const boundInRecipe = bindings.some(b => b.kind === 'recipe' && b.nodeId === recipeId)
+    || bindings.some(b => b.kind === 'dcw' && recipe.params.some(p => p.nodeId === b.nodeId))
+  if (!boundInRecipe) {
+    return { text: `无权提交配方 ${recipeId} 的整包方案(需持有该配方的绑定;找 lead 授予配方操作权)。`, isError: true }
+  }
+  const line = getDcwLineRepo().byId(recipe.lineId)
+
+  // ---------- ③ AML 活动硬闸(时间互斥:活动期间禁止整包下发,降级为建议报告) ----------
+  const activity = amlActivityForLine(recipe.lineId)
+  if (activity.active) {
+    return {
+      text: [
+        'AML 优化循环活动期间禁止整包下发,已降级为建议 —— 本次方案未提交审批、产线无任何变更。',
+        ...activity.reasons.map(r => `- ${r}`),
+        '请把候选方案改写为建议报告沉淀入知识库(标签:建议、产线),并提醒人类:AML 循环结束(aml_activity 不活动)后再重新提交 recipe_propose。时间互斥是防「双写者」的铁律。',
+      ].join('\n'),
+      isError: true,
+    }
+  }
+
+  // ---------- ④ 在飞去重(同一配方的 pending 单存在 → 软提示;拒绝/超时后自然解除) ----------
+  // 不分 agent(防跨成员对同一配方各挂一张单、双双获批造成重复下发);本工具只会创建
+  // 精确 nodeId = recipe-propose:<recipeId> 的单,等值判定即前缀语义。
+  const inflight = getToolApprovals().listPending().filter(a => a.nodeId === `recipe-propose:${recipeId}`)
+  if (inflight.length > 0) {
+    return {
+      text: `已有在飞方案审批单(${inflight.map(a => a.id).join(', ')}),等待人类裁决;同一配方同时只保留一张在飞单。拒绝/超时后该单收敛,即可吸收意见修订重提;请勿重复提交,可先在汇报中提醒人类尽快处置。`,
+    }
+  }
+
+  // ---------- ④b Recipe 级下发间隔卡控(提案早拒):距上次已批准下发不足 opIntervalMs 即拒,不建卡 ----------
+  try {
+    assertRecipeOpInterval(recipe, { emergency: args.emergency === true || args.emergency === 'true' })
+  }
+  catch (err) {
+    return { text: err instanceof Error ? err.message : String(err), isError: true }
+  }
+
+  // ---------- ⑤ 预检(镜像 recipe 真实执行限界:量程∩参数∩产品;步长/60s/保持窗本就不适用 recipe) ----------
+  const preflightNotes: string[] = []
+  for (const pkg of packages) {
+    pkg.params = pkg.params.filter((p) => {
+      const node = getDcwController().byId(p.nodeId)
+      const label = node?.name ?? p.nodeId
+      const fail = (reason: string): boolean => {
+        p.preflight = { ok: false, reason }
+        p.paramName = label
+        preflightNotes.push(`方案「${pkg.name}」${label}:${reason}(已剔除)`)
+        return false
+      }
+      if (!node) return fail('节点已删除,参数不可下发')
+      if (node.lineId !== recipe.lineId) return fail('节点已不在配方所在产线(解绑/改挂),参数不可下发')
+      if (!node.enabled) return fail('节点已停用,参数不可下发')
+      if (!recipe.params.some(rp => rp.nodeId === p.nodeId)) return fail('参数不在配方内(整包下发只能覆盖配方已有参数;新增参数请先走 recipe_update)')
+      p.paramName = node.name
+      if (p.from == null) {
+        // 服务端自取当前值:优先平台现值,其次读回值;都无则保持 null(审批卡显示 "?→to")
+        p.from = typeof node.value === 'number' ? node.value : (typeof node.readValue === 'number' ? node.readValue : null)
+      }
+      if (!p.unit) p.unit = node.unit ?? ''
+      // 只读限界谓词(skipRecipe:true = 配方下发路径真实语义)
+      const pf = evaluateRecipeParamLimits(node, p.to, { skipRecipe: true })
+      if (!pf.ok) return fail(pf.violations[0] ?? '写入限界校验失败')
+      return true
+    })
+  }
+  // 整包全失败 → 丢该包;全部包失效 → isError 附明细
+  const alivePackages = packages.filter(pkg => pkg.params.some(p => p.preflight.ok))
+  if (alivePackages.length === 0) {
+    return {
+      text: [
+        '预检失败:全部候选方案无一参数通过限界校验,未提交审批、产线无任何变更。「批准必被拒包」的方案不得入审:',
+        ...problems.map(p => `- 校验:${p}`),
+        ...preflightNotes.map(p => `- ${p}`),
+        '请按各条失败原因修订(越界值改到「节点量程∩参数基准∩产品限界」交集内;失效参数剔除或先恢复节点),再重新提交。',
+      ].join('\n'),
+      isError: true,
+    }
+  }
+
+  // ---------- ⑥ 提交结构化审批单(kind='dcw',nodeId=recipe-propose:<id>,payload.schemaVersion=1) ----------
+  const payload: RecipeProposePayload = {
+    schemaVersion: 1,
+    recipeId,
+    lineId: recipe.lineId,
+    packages: alivePackages.map(pkg => ({ ...pkg, params: pkg.params.filter(p => p.preflight.ok) })),
+  }
+  const detail = [
+    `整包方案审批:配方「${recipe.name}」v${recipe.version ?? 1}(产线 ${line?.name ?? recipe.lineId}),${alivePackages.length} 套候选方案。请在产线页结构化审批卡选择方案批准(或拒绝并附指导);未携带方案序号的批准按拒绝收敛。`,
+    ...alivePackages.map((pkg) => {
+      const rows = pkg.params.map((p) => {
+        const node = getDcwController().byId(p.nodeId)
+        return `  ${paramLine(p, node?.name ?? p.nodeId)} | 依据:${p.basis} | 引用:${p.exp_ref}`
+      })
+      return `方案「${pkg.name}」${pkg.rationale ? `:${pkg.rationale}` : ''}\n${rows.join('\n')}`
+    }),
+    ...(preflightNotes.length > 0 ? [`预检剔除(不入下发集):`, ...preflightNotes.map(p => `  - ${p}`)] : []),
+  ].join('\n')
+
+  const approvalId = `recipe-propose:${recipeId}`
+  let decision
+  try {
+    // 挂起前先入审计(方案提交留痕;裁决与下发另录)
+    try {
+      recordOps({
+        actor: agentId, actorName: agentBadgeLabel(agentId), actorKind: 'agent',
+        action: 'recipe.propose', kind: 'recipe', targetKind: 'recipe', targetId: recipeId, recipeId, lineId: recipe.lineId,
+        summary: `提交整包方案审批(${alivePackages.length} 套/${alivePackages.reduce((n, p) => n + p.params.length, 0)} 参数,单 ${approvalId}):${alivePackages.map(p => p.name).join('、')}`,
+      })
+    }
+    catch { /* 审计失败不影响提交 */ }
+    decision = await getToolApprovals().request(agentId, approvalId, 'dcw', detail, {
+      title: `整包方案审批:${recipe.name}`,
+      timeoutMs: securityRecipeDispatchTimeoutMs(),
+      payload,
+    })
+  }
+  catch (err) {
+    return { text: `提交整包方案审批失败:${err instanceof Error ? err.message : String(err)}`, isError: true }
+  }
+
+  // ---------- ⑦a 拒绝/超时:软拒绝,comment 逐字回流,指引修订后重提(mirror recipe 下发拒绝口径) ----------
+  if (!decision.approved) {
+    return {
+      text: [
+        `方案未获批准(超时未批或人工拒绝),未做任何下发。人工意见:${decision.comment || '(无附言,仅不同意候选)'}`,
+        '请逐字吸收人工意见修订方案(每参数仍须带 basis 与 exp_ref;人类拒绝过的方案不得原样重提),然后重新提交 recipe_propose。',
+      ].join('\n'),
+    }
+  }
+
+  // ---------- ⑦b 批准(choice 已由 decide 链 fail-closed 归一):选定包整批下发 ----------
+  // 归一后的 decision.choice 必为合法下标(由 normalizeRecipeProposeDecision 保证;此处兜底防呆)
+  const chosenIdx = decision.choice ?? 0
+  const chosen = payload.packages[chosenIdx] ?? payload.packages[0]!
+  const overrides = chosen.params.map(p => ({ nodeId: p.nodeId, value: p.to }))
+  let run
+  try {
+    // mirror toolRecipeApply 实际下发路径:trial 语义(整批候选、不写配方版本),overrides 限配方内节点;
+    // Agent 下发族标记 + 审批单 id 透传(recipe 级下发间隔执行兜底与成功计锚)
+    run = await getDcwController().applyRecipe(recipeId, {
+      overrides,
+      trial: true,
+      agentDispatch: true,
+      agentApprovalId: decision.id,
+      emergency: args.emergency === true || args.emergency === 'true',
+    })
+  }
+  catch (err) {
+    // 响亮失败(典型:同线试验节拍未放行):方案已批准但产线未动 —— 指引稍后重试/直发,口径同 recipe_rollback
+    return {
+      text: [
+        `方案「${chosen.name}」已获批准,但整批下发启动失败:${err instanceof Error ? err.message : String(err)}`,
+        '产线未做任何变更;请等待节拍窗放行后用 dcw_control 逐节点直发本方案参数,或稍后重新提交 recipe_propose。',
+      ].join('\n'),
+      isError: true,
+    }
+  }
+  // ---------- ⑦c ACK 鉴定回执(2026-10-08):三段口径 + 未证实/失败一律 isError ----------
+  // 「ok」只是命令受理;回执按 设备证实(verified)/未证实(unverified)/失败(failed) 三段计,
+  // 只要有 unverified 或 failed 就 isError —— Agent 必须如实知道哪个参数未获设备确认。
+  const acks = run.ackSummary ?? { verified: run.results.filter(r => r.ok).length, unverified: 0, failed: run.results.filter(r => !r.ok).length, total: run.results.length }
+  const okN = run.results.filter(r => r.ok).length
+  // 批准下发留痕(mirror toolRecipeApply 的 recipe.apply 审计;附审批单与选定包溯源)
+  try {
+    recordOps({
+      actor: agentId, actorName: agentBadgeLabel(agentId), actorKind: 'agent',
+      action: 'recipe.apply', kind: 'recipe', targetKind: 'recipe', targetId: recipeId, recipeId, lineId: recipe.lineId,
+      level: acks.failed > 0 ? 'error' : acks.unverified > 0 ? 'warn' : 'info',
+      summary: `整包方案审批后下发(证实 ${acks.verified}/${acks.total},未证实 ${acks.unverified},失败 ${acks.failed},批次 ${run.id.slice(0, 8)},单 ${decision.id},选定方案「${chosen.name}」#${chosenIdx})`,
+      detail: { approvalId: decision.id, choice: chosenIdx, runId: run.id, ackSummary: acks },
+    })
+  }
+  catch { /* 审计失败不影响下发结果 */ }
+
+  // 逐参数回执:选定包参数对照 from→to;配方内其余参数随批下发(值未改动)
+  const chosenByNode = new Map(chosen.params.map(p => [p.nodeId, p]))
+  const ackTagOf = (r: typeof run.results[number]): string => {
+    if (!r.ok) return ''
+    if (r.verify?.verdict === 'verified' || r.ack === 'readback-verified') return ' [设备证实✓]'
+    return ' [未证实⚠ 仅链路受理]'
+  }
+  const rows = run.results.map((r) => {
+    const node = r.nodeId ? getDcwController().byId(r.nodeId) : undefined
+    const label = node?.name ?? r.templateRef ?? r.nodeId ?? '?'
+    const spec = r.nodeId ? chosenByNode.get(r.nodeId) : undefined
+    const change = spec ? `${spec.from ?? '?'}→${spec.to}` : `${r.value}(随批整发,本方案未改动)`
+    return `- ${label} ${change}${node?.unit ?? ''}${ackTagOf(r)}${r.ok ? '' : ` 被拒:${String(r.message).slice(0, 80)}`}`
+  })
+  const feedback = decision.comment ? `\n[人工反馈] ${decision.comment}` : ''
+  const runIdNote = `runId:${run.id}(后续轮次按此回访效果并回写经验)`
+  // 全失败:响亮失败(isError;统一回退指引,口径 mirror recipe_trial/recipe_rollback)
+  if (run.results.length > 0 && okN === 0) {
+    return {
+      text: [
+        `整包下发失败:方案「${chosen.name}」已获批准,但整批下发 0/${run.results.length} 参数成功(典型:节点写入保持窗/节流未放行)。${runIdNote}${feedback}`,
+        '产线仍在运行原参数 —— 请等待约 60s 后用 dcw_control 逐节点直发本方案参数,或用 recipe_rollback(recipe_id, dispatch=true) 统一回退到已知良好版本;并把失败原因写入汇报提醒人类。',
+      ].join('\n'),
+      isError: true,
+    }
+  }
+  // 部分未证实/失败(2026-10-08 ACK 鉴定铁律):一律 isError 如实上报,不得报成成功
+  if (acks.failed > 0 || acks.unverified > 0) {
+    return {
+      text: [
+        `整批下发 ACK 鉴定未全绿:方案「${chosen.name}」已获批并下发,但设备证实 ${acks.verified}/${acks.total},未证实 ${acks.unverified}(仅链路受理),失败 ${acks.failed}。${runIdNote}${feedback}`,
+        ...rows,
+        '「未证实」=命令已发到链路但设备侧未确认生效(mqtt/http 等无回读驱动或回读超时);「失败」=写入被拒。请:①逐条核查未证实参数的产线侧真实状态(daq_query 复测);②必要时用 dcw_control 对未证实参数单独重发;③把未证实/失败清单写入汇报提醒人类,劣化风险下用 recipe_rollback(recipe_id, dispatch=true) 统一回退。',
+      ].join('\n'),
+      isError: true,
+    }
+  }
+  return {
+    text: [
+      `方案「${chosen.name}」已获批准并整批下发,全部参数设备证实(${acks.verified}/${acks.total},批次 ${run.id.slice(0, 8)})。${runIdNote}${feedback}`,
+      ...rows,
+      '请等待工艺惯性后 daq_query 复测判读:有进步 → 效果回访时回写经验到知识库;劣化 → 立即 recipe_rollback(recipe_id, dispatch=true) 统一回退并通报。',
+    ].filter(Boolean).join('\n'),
+  }
+}

package/server/services/workshop/agents/industrial/twin-tools.ts

@@ -22,6 +22,7 @@ import { requestCalibration, DEFAULT_TWIN_CALIBRATION_POLICY } from '../../aml/t
 import { getTwinProviderRegistry } from '../../aml/twin/provider-registry'
 import { compileSceneDraft, discoverSceneNodes, type NodeSemanticInput } from '../../aml/twin/scene-builder'
 import { getDcwController } from '../../dcw/dcw-controller'
+import { getDcwRecipeRepo } from '../../dcw/dcw-recipe.repo'
 import { getDaqNodeRepo } from '../../daq/daq-node.repo'
 import { compileDeclarativeProvider } from '../../aml/twin/declarative-provider'
 import { validatePhysicsSpec } from '../../aml/twin/physics-spec'
@@ -228,17 +229,39 @@ function boundSceneNodes(agentId: string): NodeSemanticInput[] {
   const bindings = getAgentNodeBindingRepo().byAgent(agentId)
   const daq = getDaqNodeRepo()
   const dcw = getDcwController()
+  const recipes = getDcwRecipeRepo()
   const out: NodeSemanticInput[] = []
+  const seen = new Set<string>()
+  const pushDcw = (nodeId: string, paramWindow?: { min?: number, max?: number }, evidence: string[] = []) => {
+    if (seen.has(nodeId)) return
+    const node = dcw.byId(nodeId)
+    if (!node) return
+    seen.add(nodeId)
+    // 量程 = 节点量程 ∩ 配方参数工艺窗口(两者都有才收窄;窗口缺失取节点量程)
+    const min = paramWindow?.min != null && node.min != null ? Math.max(node.min, paramWindow.min) : (node.min ?? paramWindow?.min)
+    const max = paramWindow?.max != null && node.max != null ? Math.min(node.max, paramWindow.max) : (node.max ?? paramWindow?.max)
+    out.push({ nodeId: node.id, kind: 'dcw', name: node.name, physicalMeaning: node.semantics ?? node.name, unit: node.unit, min, max, maxStep: Math.max(Math.abs(max - min) * 0.02, 0.001), lineId: node.lineId, protocol: node.driver, writable: true, evidence: [...evidence, node.semantics ?? '', node.name, node.templateRef].filter(Boolean) })
+  }
   for (const binding of bindings) {
     if (binding.kind === 'daq') {
       const node = daq.byId(binding.nodeId)
       if (!node) continue
+      if (seen.has(node.id)) continue
+      seen.add(node.id)
       out.push({ nodeId: node.id, kind: 'daq', name: node.name, physicalMeaning: node.semantics ?? node.name, unit: node.unit, min: node.min, max: node.max, lineId: node.lineId, protocol: node.driver, evidence: [node.semantics ?? '', node.templateKey].filter(Boolean) })
     }
+    else if (binding.kind === 'recipe') {
+      // 权限模型 v2 配方展开(2026-10-05 闭环修复):配方绑定 = 配方参数面授权,
+      // 参数节点(dcw)以 kind='dcw'/writable 进入发现面 —— 否则 hybrid 频道
+      // (只有 daq+recipe 绑定)永远 SCENE_NO_CONTROLS,寻优闭环断在第一步。
+      const recipe = recipes.byId(binding.nodeId)
+      if (!recipe) continue
+      for (const param of recipe.params) {
+        pushDcw(param.nodeId, { min: param.min, max: param.max }, [`配方「${recipe.name}」v${recipe.version ?? 1} 参数(工艺窗口)`])
+      }
+    }
     else {
-      const node = dcw.byId(binding.nodeId)
-      if (!node) continue
-      out.push({ nodeId: node.id, kind: 'dcw', name: node.name, physicalMeaning: node.semantics ?? node.name, unit: node.unit, min: node.min, max: node.max, maxStep: Math.max(Math.abs((node.max ?? 1) - (node.min ?? 0)) * 0.02, 0.001), lineId: node.lineId, protocol: node.driver, writable: true, evidence: [node.semantics ?? '', node.name, node.templateRef].filter(Boolean) })
+      pushDcw(binding.nodeId)
     }
   }
   return out
@@ -381,13 +404,24 @@ export async function toolTwinSnapshotCreate(agentId: string, args: Record<strin
     const samples = args.auto_daq === true || suppliedSamples.length === 0
       ? await autoDaqSamples(agentId, scene, nowMs, freshnessMaxMs)
       : suppliedSamples
+    // 控制值服务端化(2026-10-05 方案 D):controls 缺省时由 dcw 节点现值/当前设定填充,
+    // 消除 Agent 自报基线;显式传入的 controls 仍被尊重(便于 what-if 假设检验)。
+    const suppliedControls = jsonArg<Record<string, number>>(args, 'controls', {}) ?? {}
+    const controls: Record<string, number> = { ...suppliedControls }
+    for (const control of scene.controls ?? []) {
+      const key = String(control.id ?? control.nodeId ?? '')
+      if (!key || controls[key] != null) continue
+      const node = control.nodeId ? getDcwController().byId(control.nodeId) : undefined
+      if (node && typeof node.value === 'number') controls[key] = node.value
+      else if (node && typeof node.readValue === 'number') controls[key] = node.readValue
+    }
     const snapshot = createTwinSnapshot({
       scene,
       frozenScene: authoritative.hybrid ? scene : undefined,
       channelId,
       createdBy: agentId,
       phase: String(args.phase ?? 'holding'),
-      controls: jsonArg<Record<string, number>>(args, 'controls', {}) ?? {},
+      controls,
       states: jsonArg<Record<string, number>>(args, 'states', {}) ?? {},
       disturbances: jsonArg<Record<string, number>>(args, 'disturbances', {}) ?? {},
       samples,