优化: AI Chat 渲染与无障碍 + 文件查看 + 同步 IO 收口

- 渲染: 结果/过程分离(气泡只放最终结果,过程轮叙述+工具卡收进轨迹卡默认收起可展开——治『流式后正文消失』且不再 N 气泡打扰)
- 渲染 P0: v-memo 定向 dep(只有真正渲染流式块的 item 失效,其余取占位) + renderItems 桶 memo(签名覆盖全消费输入,item 引用稳定) + 指纹 watch 降为 O(1) toolEpoch
- TopBar 三处 O(n) messages computed 合并单次遍历;md 缓存改有界 LRU(除满即全清抖动);useTurnTrace 补 never 穷尽 + default;MessageItem 模板去重
- 无障碍(四批累计): 可点容器键盘化(含 role/tabindex/keydown/aria-expanded) + icon-only aria-label 约 60 处 + 装饰图标 aria-hidden + 52 个滚动容器 overscroll-behavior:contain + 设置开关聚焦可见环 + outline 零补偿站点补 :focus-visible
- UX 走查项: 灵感 IDEA-10/11、知识 KNOW-6/7/9、项目详情 O8、审计 SA-4、设置 SA-6/26/27/28(脏表单/冗余 save/clamp/删除二次确认)
- 文件查看专项: 切工程状态残留 F5/F6、FileTree 去重与展开反馈 F7/F15/F17、切文件头部残留 F18、R 状态着色 F19;FilePreview diff 文件头(---/+++)误当增删渲染修复 + 大纲行高亮死样式修复 + 图片失败文案 i18n 化
- 同步 IO 收口(异步上下文内阻塞): unattended 看板读写改 tokio::fs、fetch_url 探活改 spawn_blocking、http/generate_image 清理改 tokio::fs
- 子代理 P3 批量: 任务书钉扎上限/截断文案区分父子语境/环境画像日期/temperature 可覆盖/model 回退上报/lang 透传
- crate 清理: df-execute·df-types 注释与格式、df-ideas 过时注释、df-nodes subflow 夸大描述、df-project 死变量

src/views/AuditLog.vue

@@ -82,7 +82,10 @@
                 <span v-else class="decided-none">—</span>
               </td>
               <td class="col-args"><code class="brief">{{ r.arguments_brief || '—' }}</code></td>
-              <td class="col-result"><code class="brief">{{ r.result_brief || '—' }}</code></td>
+              <td class="col-result">
+                <span v-if="hasOffloadMark(r.result_brief)" class="offload-chip">{{ t('aiTool.resultOffloaded') }}</span>
+                <code v-else class="brief">{{ r.result_brief || '—' }}</code>
+              </td>
             </tr>
             <tr v-if="expandedId === r.id" class="audit-detail-row">
               <td colspan="7">
@@ -93,7 +96,18 @@
                   </div>
                   <div class="detail-section">
                     <span class="detail-label">{{ t('auditLog.col.result') }}</span>
-                    <pre class="detail-pre">{{ r.result_brief || '—' }}</pre>
+                    <!-- 外置句柄(D2a):显「结果已外置」+ 展开经 IPC 读回原文全文 -->
+                    <template v-if="offloadHandle">
+                      <OffloadedResultBar
+                        :handle="offloadHandle"
+                        :expanded="offloadExpanded"
+                        :loading="offloadLoading"
+                        :error="offloadError"
+                        @toggle="toggleOffload"
+                      />
+                      <pre v-if="offloadResolved != null" class="detail-pre">{{ offloadResolved }}</pre>
+                    </template>
+                    <pre v-else class="detail-pre">{{ r.result_brief || '—' }}</pre>
                   </div>
                 </div>
               </td>
@@ -175,6 +189,8 @@ import { invoke } from '@tauri-apps/api/core'
 import { formatRelative, formatDate } from '@/utils/time'
 import { usePersistedRef } from '@/composables/usePersistedRef'
 import { aiApi } from '@/api/ai'
+import { hasOffloadMark, useOffloadedResult } from '@/composables/ai/useOffloadedResult'
+import OffloadedResultBar from '@/components/OffloadedResultBar.vue'
 import type { AiConversationSummary } from '@/api/types'
 
 const { t } = useI18n()
@@ -255,6 +271,20 @@ function toggleExpand(id: string) {
   expandedId.value = expandedId.value === id ? null : id
 }
 
+// 展开行的外置句柄详情(D2a):同一视图只展开一行,读回状态复用单实例;
+// 切行时 reset 防串上一行的读回结果/错误。句柄来源=后端截断的 result_brief。
+const expandedRecord = computed(() => records.value.find(r => r.id === expandedId.value) ?? null)
+const {
+  handle: offloadHandle,
+  expanded: offloadExpanded,
+  loading: offloadLoading,
+  error: offloadError,
+  resolved: offloadResolved,
+  toggle: toggleOffload,
+  reset: resetOffload,
+} = useOffloadedResult(() => expandedRecord.value?.result_brief ?? null)
+watch(expandedId, () => resetOffload())
+
 // 审批→执行 耗时(秒级,毫秒差太精细)
 function durationLabel(requested: string, executed: string): string {
   const ms = new Date(executed).getTime() - new Date(requested).getTime()
@@ -356,24 +386,32 @@ async function loadConversations() {
     convLoading.value = false
   }
 }
+// 对话详情请求序号守卫(对齐上方 _loadSeq 模式):快速切换对话时,
+// 旧对话 export 响应晚到不覆盖新对话消息;折叠时自增使在途请求作废
+let _convSeq = 0
 async function toggleConv(id: string) {
   if (expandedConv.value === id) {
+    _convSeq++ // 折叠:作废在途请求,避免其回填残留
     expandedConv.value = null
     convMessages.value = []
+    convLoading.value = false
     return
   }
+  const seq = ++_convSeq
   expandedConv.value = id
   convLoading.value = true
   convError.value = ''
   try {
     const json = await aiApi.exportConversation(id, 'json')
+    if (seq !== _convSeq) return // 旧响应丢弃,只取最新
     const parsed = JSON.parse(json) as ConvMessage[] ?? []
     convMessages.value = parsed || []
   } catch (e) {
+    if (seq !== _convSeq) return // 旧请求的失败不污染当前视图
     convError.value = String(e)
     convMessages.value = []
   } finally {
-    convLoading.value = false
+    if (seq === _convSeq) convLoading.value = false
   }
 }
 
@@ -578,6 +616,18 @@ onMounted(() => {
   white-space: pre-wrap;
 }
 
+/* D2a 外置句柄:结果列显标记替代截断 JSON(点行展开看详情并可读回全文) */
+.offload-chip {
+  display: inline-block;
+  font-size: 10px;
+  font-weight: 600;
+  padding: 1px 7px;
+  border-radius: var(--df-radius-xs);
+  background: var(--df-warning-bg);
+  color: var(--df-warning);
+  white-space: nowrap;
+}
+
 /* 风险 badge */
 .risk-badge {
   display: inline-block;