feat(web): AI 工作台意图规划与规划思考模型

- 新增 workbenchAiIntentPlannerModel，基于 LLM function_call 解析建单/草稿/提交意图，区分 model 与 rule_fallback 来源
- 新增 workbenchAiPlanningThinkingModel 合并规划思考事件流，按 eventId 去重合并
- application gate/preview 模型接入意图规划，usePersonalWorkbenchAiMode/useWorkbenchAiStewardFlow/useWorkbenchAiActionRouter 链路适配，支持上下文提交
- steward 服务与 stewardPlanModel 适配新动作结构，receipt-folder-view 微调样式
- 新增 intent-planner-model/application-context-submit/steward-actions-service 测试，更新 gate-model/action-router/plan-message-copy/fast-preview 测试

web/src/composables/workbenchAiMode/usePersonalWorkbenchAiMode.js

@@ -35,7 +35,20 @@ import { useWorkbenchAiMessageActions } from './useWorkbenchAiMessageActions.js'
 import { useWorkbenchAiMessageExpansion } from './useWorkbenchAiMessageExpansion.js'
 import { useWorkbenchAiSessionCommands } from './useWorkbenchAiSessionCommands.js'
 import { useWorkbenchAiStewardFlow } from './useWorkbenchAiStewardFlow.js'
-import { isReimbursementCreationIntent } from './workbenchAiApplicationGateModel.js'
+import {
+  isReimbursementCreationIntent
+} from './workbenchAiApplicationGateModel.js'
+import {
+  buildRuleFallbackWorkbenchAiIntentPlan,
+  normalizeWorkbenchAiIntentPlan,
+  resolveExecutableTravelApplicationPlan,
+  shouldRequestWorkbenchAiIntentPlan
+} from './workbenchAiIntentPlannerModel.js'
+import {
+  buildInitialModelPlanningThinkingEvents,
+  buildModelPlanningProgressSchedule,
+  mergeWorkbenchAiThinkingEvents
+} from './workbenchAiPlanningThinkingModel.js'
 
 const AI_SEARCH_CONVERSATION_ID = 'ai-search'
 const INLINE_ANSWER_STREAM_CHUNK_SIZE = 6
@@ -251,11 +264,16 @@ export function usePersonalWorkbenchAiMode(props, emit) {
     applicationFlow,
     assistantDraft,
     attachmentFlow,
+    conversationMessages,
+    createInlineMessage,
     emit,
     expenseFlow,
     focusAiModeInput,
     hasInlineAttachmentOcrDetails,
+    persistCurrentConversation,
+    replaceInlineMessage,
     resolveLatestInlineUserPrompt,
+    scrollInlineConversationToBottom,
     selectedFiles,
     startInlineConversation,
     toast,
@@ -560,6 +578,131 @@ export function usePersonalWorkbenchAiMode(props, emit) {
     return false
   }
 
+  function isModelPlannedReimbursementTask(modelPlan = {}) {
+    const tasks = Array.isArray(modelPlan?.tasks) ? modelPlan.tasks : []
+    return tasks.some((task) => {
+      const taskType = String(task?.task_type || task?.taskType || '').trim()
+      const assignedAgent = String(task?.assigned_agent || task?.assignedAgent || '').trim()
+      return taskType === 'reimbursement' || assignedAgent === 'reimbursement_assistant'
+    })
+  }
+
+  function updateModelPlanningThinkingEvent(messageId, event) {
+    const message = conversationMessages.value.find((item) => item.id === messageId)
+    if (!message) {
+      return
+    }
+    const currentPlan = message.stewardPlan || {}
+    message.stewardPlan = {
+      ...currentPlan,
+      streamStatus: 'streaming',
+      thinkingEvents: mergeWorkbenchAiThinkingEvents(resolveInlineThinkingEvents(message), [event])
+    }
+    persistCurrentConversation()
+    scrollInlineConversationToBottom({ force: inlineConversationAutoScrollPinned.value })
+  }
+
+  function startModelPlanningProgressUpdates(messageId) {
+    const timerIds = buildModelPlanningProgressSchedule().map(({ delayMs, event }) => (
+      globalThis.setTimeout(() => {
+        updateModelPlanningThinkingEvent(messageId, event)
+      }, delayMs)
+    ))
+    return () => {
+      timerIds.forEach((timerId) => globalThis.clearTimeout(timerId))
+    }
+  }
+
+  function startModelPlanningConversation(cleanPrompt, entry = {}) {
+    if (conversationId.value === AI_SEARCH_CONVERSATION_ID) {
+      conversationId.value = ''
+      conversationMessages.value = []
+      activeConversationTitle.value = ''
+    }
+    activateInlineConversation({
+      title: entry.label || cleanPrompt.slice(0, 18) || '新对话'
+    })
+    inlineConversationAutoScrollPinned.value = true
+    conversationMessages.value.push(createInlineMessage('user', cleanPrompt))
+    assistantDraft.value = ''
+    removeWorkbenchDateTag()
+    closeWorkbenchDatePicker()
+    filesFlow.clearAiModeFiles()
+    const pendingMessage = createInlineMessage('assistant', '正在识别意图，准备拆解申请、报销和附件任务。', {
+      pending: true,
+      stewardPlan: {
+        streamStatus: 'streaming',
+        thinkingEvents: buildInitialModelPlanningThinkingEvents()
+      }
+    })
+    conversationMessages.value.push(pendingMessage)
+    scrollInlineConversationToBottom()
+    persistCurrentConversation()
+    return pendingMessage
+  }
+
+  function startModelPlannedApplicationPreview(travelApplicationRequest, plannerPendingMessage = null) {
+    void applicationFlow.startAiApplicationPreview(
+      travelApplicationRequest.expenseType,
+      travelApplicationRequest.expenseTypeLabel,
+      travelApplicationRequest.sourceText,
+      {
+        userMessage: travelApplicationRequest.sourceText,
+        pushUserMessage: !plannerPendingMessage,
+        pendingMessageId: plannerPendingMessage?.id,
+        ontologyFields: travelApplicationRequest.ontologyFields,
+        autoSubmit: travelApplicationRequest.autoSubmit,
+        autoSaveDraft: travelApplicationRequest.autoSaveDraft
+      }
+    )
+  }
+
+  async function executeModelPlannedWorkbenchIntent(cleanPrompt, entry = {}, files = []) {
+    let intentPlan = null
+    let modelPlan = null
+    const plannerPendingMessage = startModelPlanningConversation(cleanPrompt, entry)
+    const stopPlanningProgressUpdates = startModelPlanningProgressUpdates(plannerPendingMessage.id)
+    sending.value = true
+    try {
+      modelPlan = await stewardFlow.resolveInlineExecutionPlan(cleanPrompt, entry, files, {
+        pendingMessageId: plannerPendingMessage.id
+      })
+      intentPlan = normalizeWorkbenchAiIntentPlan(modelPlan, { prompt: cleanPrompt })
+    } catch (error) {
+      console.warn('AI mode intent planner failed, using local fallback:', error)
+      const rulePlan = buildRuleFallbackWorkbenchAiIntentPlan(cleanPrompt)
+      const ruleRequest = resolveExecutableTravelApplicationPlan(rulePlan)
+      if (ruleRequest) {
+        sending.value = false
+        startModelPlannedApplicationPreview(ruleRequest, plannerPendingMessage)
+        return
+      }
+    } finally {
+      stopPlanningProgressUpdates()
+      sending.value = false
+    }
+
+    const travelApplicationRequest = resolveExecutableTravelApplicationPlan(intentPlan)
+    if (travelApplicationRequest) {
+      startModelPlannedApplicationPreview(travelApplicationRequest, plannerPendingMessage)
+      return
+    }
+
+    if (isModelPlannedReimbursementTask(modelPlan) || isReimbursementCreationIntent(cleanPrompt)) {
+      replaceInlineMessage(plannerPendingMessage.id, createInlineMessage('assistant', '已识别为报销任务，正在进入报销流程。', {
+        id: plannerPendingMessage.id,
+        stewardPlan: {
+          streamStatus: 'completed',
+          thinkingEvents: resolveInlineThinkingEvents(plannerPendingMessage).map((item) => ({ ...item, status: 'completed' }))
+        }
+      }))
+      void expenseFlow.startAiReimbursementAssociationGate(cleanPrompt, entry.label || cleanPrompt)
+      return
+    }
+
+    void stewardFlow.requestInlineAssistantReply(cleanPrompt, entry, files, { pendingMessageId: plannerPendingMessage.id })
+  }
+
   function handleAiAnswerMarkdownClick(event) {
     const target = event?.target
     const link = target?.closest?.('a[href^="#ai-open-document-detail:"], a[href^="#ai-open-application-detail:"]')
@@ -594,6 +737,11 @@ export function usePersonalWorkbenchAiMode(props, emit) {
       return
     }
 
+    if (shouldRequestWorkbenchAiIntentPlan(cleanPrompt)) {
+      void executeModelPlannedWorkbenchIntent(cleanPrompt, entry, files)
+      return
+    }
+
     if (isReimbursementCreationIntent(cleanPrompt)) {
       void expenseFlow.startAiReimbursementAssociationGate(cleanPrompt, entry.label || '我要报销')
       return