feat: 增加 Agent Excel 批量测试

2026-08-12 17:00:50 +08:00
parent c4231f268b
commit 3faecab6ac
26 changed files with 1327 additions and 6 deletions
--- a/ai_knowledge_base_v2/.env.example
+++ b/ai_knowledge_base_v2/.env.example
@@ -32,6 +32,9 @@ TOPIC_SETTLEMENT_WORKER_ENABLED=true
 TOPIC_SETTLEMENT_POLL_SECONDS=2
 TOPIC_SETTLEMENT_STALE_MINUTES=30
 TOPIC_SETTLEMENT_MAX_ATTEMPTS=3
+AGENT_BATCH_WORKER_ENABLED=true
+AGENT_BATCH_POLL_SECONDS=2
+AGENT_BATCH_STALE_MINUTES=30

 # ============================================
 # 内网穿透（frpc sidecar）
--- a/ai_knowledge_base_v2/.gitignore
+++ b/ai_knowledge_base_v2/.gitignore
@@ -1 +1,2 @@
 .env
+outputs/
--- a/ai_knowledge_base_v2/apps/admin-web/src/App.vue
+++ b/ai_knowledge_base_v2/apps/admin-web/src/App.vue
@@ -759,6 +759,7 @@ async function clearFeishuCache() {
        <template v-if="activeMenu === 'prompt'">
          <AgentManagementView
            :preview-knowledge-id="previewKnowledgeId"
+            :can-batch="can('prompt.batch')"
            @consumed-preview="previewKnowledgeId = null"
          />
        </template>
--- a/ai_knowledge_base_v2/apps/admin-web/src/components/AgentBatchTestPanel.vue
+++ b/ai_knowledge_base_v2/apps/admin-web/src/components/AgentBatchTestPanel.vue
@@ -0,0 +1,225 @@
+<script setup lang="ts">
+import { ElMessage, ElMessageBox } from "element-plus";
+import { computed, onBeforeUnmount, onMounted, reactive, ref } from "vue";
+
+import { api } from "../services/api";
+import type { AgentBatchItem, AgentBatchJob, AgentBatchJobDetail } from "../types/api";
+import AdminPagination from "./AdminPagination.vue";
+
+const props = defineProps<{
+  config: Record<string, unknown>;
+  modelName: string;
+  knowledgeSummary: string;
+}>();
+
+const fileInput = ref<HTMLInputElement | null>(null);
+const selectedFile = ref<File | null>(null);
+const taskName = ref("");
+const creating = ref(false);
+const loading = ref(false);
+const exportingId = ref<number | null>(null);
+const jobs = ref<AgentBatchJob[]>([]);
+const pager = reactive({ page: 1, pageSize: 10, total: 0 });
+const detailOpen = ref(false);
+const detailLoading = ref(false);
+const selectedJob = ref<AgentBatchJobDetail | null>(null);
+const detailPager = reactive({ page: 1, pageSize: 20, total: 0 });
+let pollTimer: number | null = null;
+
+const hasRunningJob = computed(() => jobs.value.some((job) => ["pending", "running"].includes(job.status)));
+
+onMounted(() => void loadJobs());
+onBeforeUnmount(stopPolling);
+
+async function loadJobs(page = pager.page, pageSize = pager.pageSize, quiet = false) {
+  if (!quiet) loading.value = true;
+  try {
+    const result = await api.agentBatchJobs({ page, pageSize });
+    jobs.value = result.items;
+    Object.assign(pager, { page: result.page, pageSize: result.pageSize, total: result.total });
+    schedulePolling();
+    if (selectedJob.value && detailOpen.value) await loadDetail(selectedJob.value.id, detailPager.page, detailPager.pageSize, true);
+  } catch (error) {
+    if (!quiet) ElMessage.error(errorMessage(error, "批量测试任务加载失败"));
+  } finally {
+    if (!quiet) loading.value = false;
+  }
+}
+
+function schedulePolling() {
+  stopPolling();
+  if (!hasRunningJob.value) return;
+  pollTimer = window.setTimeout(() => void loadJobs(pager.page, pager.pageSize, true), 2500);
+}
+
+function stopPolling() {
+  if (pollTimer !== null) window.clearTimeout(pollTimer);
+  pollTimer = null;
+}
+
+function selectFile(event: Event) {
+  const input = event.target as HTMLInputElement;
+  selectedFile.value = input.files?.[0] || null;
+  if (selectedFile.value && !taskName.value.trim()) taskName.value = selectedFile.value.name.replace(/\.xlsx$/i, "");
+}
+
+function clearFile() {
+  selectedFile.value = null;
+  if (fileInput.value) fileInput.value.value = "";
+}
+
+async function downloadTemplate() {
+  try {
+    await api.downloadAgentBatchTemplate();
+  } catch (error) {
+    ElMessage.error(errorMessage(error, "模板下载失败"));
+  }
+}
+
+async function createJob() {
+  if (!selectedFile.value) return ElMessage.warning("请先选择填写好的 Excel 文件");
+  if (!String(props.config.modelId || "")) return ElMessage.warning("请先在调试配置中选择模型");
+  creating.value = true;
+  try {
+    const created = await api.createAgentBatch(selectedFile.value, { ...props.config, name: taskName.value.trim() });
+    ElMessage.success(`已创建批量测试任务，共 ${created.totalCount} 个问题`);
+    clearFile();
+    taskName.value = "";
+    await loadJobs(1, pager.pageSize);
+    await openDetail(created);
+  } catch (error) {
+    ElMessage.error(errorMessage(error, "批量测试任务创建失败"));
+  } finally {
+    creating.value = false;
+  }
+}
+
+async function openDetail(job: AgentBatchJob) {
+  detailOpen.value = true;
+  detailPager.page = 1;
+  await loadDetail(job.id, 1, detailPager.pageSize);
+}
+
+async function loadDetail(jobId: number, page = detailPager.page, pageSize = detailPager.pageSize, quiet = false) {
+  if (!quiet) detailLoading.value = true;
+  try {
+    const result = await api.agentBatchJob(jobId, { page, pageSize });
+    selectedJob.value = result;
+    Object.assign(detailPager, { page: result.items.page, pageSize: result.items.pageSize, total: result.items.total });
+  } catch (error) {
+    if (!quiet) ElMessage.error(errorMessage(error, "任务详情加载失败"));
+  } finally {
+    if (!quiet) detailLoading.value = false;
+  }
+}
+
+async function exportJob(job: AgentBatchJob) {
+  exportingId.value = job.id;
+  try {
+    await api.exportAgentBatch(job.id);
+    ElMessage.success("测试结果已导出");
+  } catch (error) {
+    ElMessage.error(errorMessage(error, "结果导出失败"));
+  } finally {
+    exportingId.value = null;
+  }
+}
+
+async function cancelJob(job: AgentBatchJob) {
+  try {
+    await ElMessageBox.confirm("取消后尚未开始的问题不会继续生成，已完成结果仍可导出。", "取消批量测试", {
+      type: "warning", confirmButtonText: "确认取消", cancelButtonText: "继续执行",
+    });
+  } catch {
+    return;
+  }
+  try {
+    await api.cancelAgentBatch(job.id);
+    ElMessage.success("批量测试已取消");
+    await loadJobs(pager.page, pager.pageSize);
+  } catch (error) {
+    ElMessage.error(errorMessage(error, "取消失败"));
+  }
+}
+
+function statusLabel(status: string) {
+  return ({ pending: "等待执行", running: "生成中", completed: "已完成", completed_with_errors: "部分失败", cancelled: "已取消", success: "成功", failed: "失败" } as Record<string, string>)[status] || status;
+}
+
+function statusType(status: string) {
+  if (["completed", "success"].includes(status)) return "success";
+  if (["completed_with_errors", "failed"].includes(status)) return "danger";
+  if (status === "running") return "primary";
+  if (status === "cancelled") return "info";
+  return "warning";
+}
+
+function formatDate(value?: string | null) {
+  return value ? new Date(value).toLocaleString("zh-CN", { hour12: false }) : "—";
+}
+
+function itemDuration(item: AgentBatchItem) {
+  return item.durationMs == null ? "—" : `${(item.durationMs / 1000).toFixed(1)} 秒`;
+}
+
+function errorMessage(error: unknown, fallback: string) {
+  return error instanceof Error ? error.message : fallback;
+}
+</script>
+
+<template>
+  <div class="agent-batch-panel">
+    <div class="agent-section-intro">
+      <div><h3>Excel 批量测试</h3><p>使用当前编辑框中的提示词、调试模型、知识库和生成参数创建配置快照，后台逐条生成答案。</p></div>
+      <el-button @click="downloadTemplate">下载导入模板</el-button>
+    </div>
+
+    <section class="agent-batch-create-card">
+      <div class="agent-batch-config-summary">
+        <span><small>模型</small><strong>{{ modelName }}</strong></span>
+        <span><small>知识库</small><strong>{{ knowledgeSummary }}</strong></span>
+        <span><small>单次上限</small><strong>200 个问题</strong></span>
+      </div>
+      <div class="agent-batch-upload-row">
+        <label class="agent-batch-name"><span>任务名称</span><input v-model="taskName" maxlength="160" placeholder="例如：8月主提示词回归测试" /></label>
+        <label class="agent-batch-file" :class="{ selected: selectedFile }">
+          <input ref="fileInput" type="file" accept=".xlsx" @change="selectFile" />
+          <span>{{ selectedFile?.name || "选择填写好的 Excel 文件" }}</span>
+          <small>{{ selectedFile ? `${(selectedFile.size / 1024).toFixed(1)} KB` : "仅支持 .xlsx，最大 5MB" }}</small>
+        </label>
+        <el-button v-if="selectedFile" @click="clearFile">移除</el-button>
+        <el-button type="primary" :loading="creating" :disabled="!selectedFile" @click="createJob">开始批量测试</el-button>
+      </div>
+    </section>
+
+    <div class="agent-batch-list-head"><div><h4>测试记录</h4><p>任务在后台执行，离开页面不会中断。</p></div><el-button :loading="loading" @click="loadJobs()">刷新</el-button></div>
+    <el-table v-loading="loading" :data="jobs" stripe>
+      <el-table-column label="任务" min-width="190"><template #default="{ row }"><strong>{{ row.name }}</strong><small class="agent-batch-cell-note">#{{ row.id }} · {{ formatDate(row.createdAt) }}</small></template></el-table-column>
+      <el-table-column label="配置快照" min-width="180"><template #default="{ row }"><span>{{ row.modelName }}</span><small class="agent-batch-cell-note">{{ row.knowledgeNames.join('、') }}</small></template></el-table-column>
+      <el-table-column label="进度" min-width="190"><template #default="{ row }"><el-progress :percentage="row.progress" :status="row.status === 'completed' ? 'success' : undefined" /><small class="agent-batch-cell-note">{{ row.processedCount }}/{{ row.totalCount }} · 成功 {{ row.successCount }} · 失败 {{ row.failedCount }}</small></template></el-table-column>
+      <el-table-column label="状态" width="108"><template #default="{ row }"><el-tag :type="statusType(row.status)">{{ statusLabel(row.status) }}</el-tag></template></el-table-column>
+      <el-table-column label="操作" width="190" fixed="right"><template #default="{ row }"><el-button link type="primary" @click="openDetail(row)">查看</el-button><el-button link type="primary" :loading="exportingId === row.id" @click="exportJob(row)">导出</el-button><el-button v-if="['pending','running'].includes(row.status)" link type="danger" @click="cancelJob(row)">取消</el-button></template></el-table-column>
+    </el-table>
+    <AdminPagination :page="pager.page" :page-size="pager.pageSize" :total="pager.total" @change="loadJobs" />
+  </div>
+
+  <el-dialog v-model="detailOpen" width="min(1100px, 94vw)" title="批量测试详情" destroy-on-close>
+    <div v-loading="detailLoading" class="agent-batch-detail">
+      <template v-if="selectedJob">
+        <div class="agent-batch-detail-head">
+          <div><strong>{{ selectedJob.name }}</strong><span>{{ selectedJob.modelName }} · {{ selectedJob.knowledgeNames.join('、') }}</span></div>
+          <el-tag :type="statusType(selectedJob.status)">{{ statusLabel(selectedJob.status) }}</el-tag>
+        </div>
+        <el-table :data="selectedJob.items.items" stripe max-height="520">
+          <el-table-column prop="externalNo" label="序号" width="76" />
+          <el-table-column prop="question" label="问题" min-width="230" show-overflow-tooltip />
+          <el-table-column label="答案" min-width="360"><template #default="{ row }"><div class="agent-batch-answer">{{ row.answer || row.errorMessage || statusLabel(row.status) }}</div></template></el-table-column>
+          <el-table-column label="状态" width="90"><template #default="{ row }"><el-tag size="small" :type="statusType(row.status)">{{ statusLabel(row.status) }}</el-tag></template></el-table-column>
+          <el-table-column label="耗时" width="92"><template #default="{ row }">{{ itemDuration(row) }}</template></el-table-column>
+        </el-table>
+        <AdminPagination :page="detailPager.page" :page-size="detailPager.pageSize" :total="detailPager.total" @change="(page, pageSize) => loadDetail(selectedJob!.id, page, pageSize)" />
+      </template>
+    </div>
+    <template #footer><el-button @click="detailOpen = false">关闭</el-button><el-button v-if="selectedJob" type="primary" :loading="exportingId === selectedJob.id" @click="exportJob(selectedJob)">导出 Excel</el-button></template>
+  </el-dialog>
+</template>
--- a/ai_knowledge_base_v2/apps/admin-web/src/components/AgentManagementView.vue
+++ b/ai_knowledge_base_v2/apps/admin-web/src/components/AgentManagementView.vue
@@ -5,11 +5,12 @@ import { computed, nextTick, onMounted, reactive, ref, watch } from "vue";
 import { api, streamDebugAgent } from "../services/api";
 import type { AdminUser, AgentRuntimeConfig, KnowledgeItem, ModelItem, PromptDetail, PromptHistoryItem, TopicSessionRecord } from "../types/api";
 import AgentGenerationParameters from "./AgentGenerationParameters.vue";
+import AgentBatchTestPanel from "./AgentBatchTestPanel.vue";
 import AgentResponseDepthControl from "./AgentResponseDepthControl.vue";
 import AdminPagination from "./AdminPagination.vue";
 import StreamingMarkdownMessage from "./StreamingMarkdownMessage.vue";

-const props = defineProps<{ previewKnowledgeId?: number | null }>();
+const props = withDefaults(defineProps<{ previewKnowledgeId?: number | null; canBatch?: boolean }>(), { canBatch: false });
 const emit = defineEmits<{ consumedPreview: [] }>();

 const loading = ref(false);
@@ -97,6 +98,19 @@ const selectedDebugTopicLabel = computed(() => {
  const topic = debugTopics.value.find((item) => item.id === agentForm.topicSessionId);
  return topic ? topic.title : `主题 #${agentForm.topicSessionId}`;
 });
+const batchConfig = computed(() => ({
+  promptContent: promptContent.value,
+  modelId: agentForm.modelId,
+  knowledgeIds: [...agentForm.knowledgeIds],
+  temperature: agentForm.temperature,
+  topP: agentForm.topP,
+  topK: agentForm.topK,
+  presencePenalty: agentForm.presencePenalty,
+  frequencyPenalty: agentForm.frequencyPenalty,
+  maxToken: agentForm.maxToken,
+  streamEnabled: agentForm.streamEnabled,
+  responseDepth: agentForm.responseDepth,
+}));

 onMounted(load);

@@ -481,7 +495,7 @@ function errorMessage(error: unknown, fallback: string) {
    </div>
  </div>

-  <section v-loading="loading" class="agent-workbench agent-workbench-v2">
+  <section v-loading="loading" class="agent-workbench agent-workbench-v2" :class="{ 'agent-batch-mode': activeConfigTab === 'batch' }">
    <div class="agent-config-panel agent-control-panel">
      <el-tabs v-model="activeConfigTab" class="agent-config-tabs">
        <el-tab-pane label="主提示词" name="prompt">
@@ -651,6 +665,14 @@ function errorMessage(error: unknown, fallback: string) {
          </el-form>
        </el-tab-pane>

+        <el-tab-pane v-if="props.canBatch" label="批量测试" name="batch" lazy>
+          <AgentBatchTestPanel
+            :config="batchConfig"
+            :model-name="selectedModelName"
+            :knowledge-summary="selectedKnowledgeSummary"
+          />
+        </el-tab-pane>
+
        <el-tab-pane label="历史版本" name="history">
          <div class="agent-section-intro"><div><h3>主提示词历史</h3><p>查看完整内容或恢复任意版本。恢复操作会作为一个新版本记录。</p></div></div>
          <div v-loading="historyLoading" class="prompt-history-list">
--- a/ai_knowledge_base_v2/apps/admin-web/src/services/api.ts
+++ b/ai_knowledge_base_v2/apps/admin-web/src/services/api.ts
@@ -8,6 +8,8 @@ import type {
  AgentDebugStreamComplete,
  AgentGenerationConfig,
  AgentRuntimeConfig,
+  AgentBatchJob,
+  AgentBatchJobDetail,
  AiLogRecord,
  ApiResponse,
  ChatDetail,
@@ -259,6 +261,19 @@ export const api = {
  agentRuntimeConfig: () => request<AgentRuntimeConfig>("/admin/agent/runtime-config"),
  saveAgentRuntimeConfig: (payload: AgentGenerationConfig) =>
    request<AgentRuntimeConfig>("/admin/agent/runtime-config", { method: "PUT", body: JSON.stringify(payload) }),
+  downloadAgentBatchTemplate: () => download("/admin/agent/batch/template", "Agent批量测试导入模板.xlsx"),
+  createAgentBatch: (file: File, config: Record<string, unknown>) => {
+    const formData = new FormData();
+    formData.append("file", file);
+    formData.append("configJson", JSON.stringify(config));
+    return upload<AgentBatchJob>("/admin/agent/batch/jobs", formData);
+  },
+  agentBatchJobs: (query: { page?: number; pageSize?: number } = {}) =>
+    request<PageResult<AgentBatchJob>>(`/admin/agent/batch/jobs${queryString(query)}`),
+  agentBatchJob: (id: number, query: { page?: number; pageSize?: number } = {}) =>
+    request<AgentBatchJobDetail>(`/admin/agent/batch/jobs/${id}${queryString(query)}`),
+  exportAgentBatch: (id: number) => download(`/admin/agent/batch/jobs/${id}/export`, `Agent批量测试结果_${id}.xlsx`),
+  cancelAgentBatch: (id: number) => request<AgentBatchJob>(`/admin/agent/batch/jobs/${id}/cancel`, { method: "POST", body: "{}" }),
  models: () => request<ModelItem[]>("/admin/model/list"),
  createModel: (payload: Record<string, unknown>) =>
    request<ModelItem>("/admin/model", { method: "POST", body: JSON.stringify(payload) }),
--- a/ai_knowledge_base_v2/apps/admin-web/src/styles.css
+++ b/ai_knowledge_base_v2/apps/admin-web/src/styles.css
@@ -975,6 +975,10 @@ textarea {
  align-items: start;
 }

+.agent-workbench.agent-batch-mode { grid-template-columns: minmax(0, 1fr); }
+.agent-workbench.agent-batch-mode > .agent-preview-panel,
+.agent-workbench.agent-batch-mode > .agent-trace-panel { display: none; }
+
 .agent-page-head {
  display: flex;
  align-items: flex-start;
@@ -1049,6 +1053,33 @@ textarea {
  margin-top: 16px;
 }

+.agent-batch-panel { display: grid; gap: 16px; }
+.agent-batch-create-card { display: grid; gap: 16px; padding: 16px; border: 1px solid #dfe8e5; border-radius: 10px; background: #f8fbfa; }
+.agent-batch-config-summary { display: grid; grid-template-columns: repeat(3, minmax(0, 1fr)); gap: 10px; }
+.agent-batch-config-summary span { min-width: 0; padding: 11px 12px; border: 1px solid #e1e9e6; border-radius: 8px; background: #fff; }
+.agent-batch-config-summary small, .agent-batch-config-summary strong { display: block; }
+.agent-batch-config-summary small { margin-bottom: 4px; color: #7a8c85; font-size: 11px; }
+.agent-batch-config-summary strong { overflow: hidden; color: #2b4038; font-size: 13px; text-overflow: ellipsis; white-space: nowrap; }
+.agent-batch-upload-row { display: flex; align-items: flex-end; gap: 10px; flex-wrap: wrap; }
+.agent-batch-name { display: grid; gap: 6px; width: 220px; color: #536860; font-size: 12px; }
+.agent-batch-name input { height: 40px; padding: 0 12px; border: 1px solid #d8e2df; border-radius: 6px; outline: none; background: #fff; color: #263832; }
+.agent-batch-name input:focus { border-color: #2f9479; box-shadow: 0 0 0 2px rgba(47, 148, 121, .12); }
+.agent-batch-file { position: relative; display: grid; min-width: 260px; flex: 1; gap: 3px; min-height: 40px; padding: 7px 12px; border: 1px dashed #b9cac4; border-radius: 7px; background: #fff; cursor: pointer; }
+.agent-batch-file.selected { border-style: solid; border-color: #62aa95; background: #f3faf7; }
+.agent-batch-file input { position: absolute; inset: 0; width: 100%; height: 100%; opacity: 0; cursor: pointer; }
+.agent-batch-file span { overflow: hidden; color: #315148; font-size: 13px; font-weight: 600; text-overflow: ellipsis; white-space: nowrap; }
+.agent-batch-file small { color: #82918c; font-size: 11px; }
+.agent-batch-list-head { display: flex; align-items: center; justify-content: space-between; gap: 12px; }
+.agent-batch-list-head h4, .agent-batch-list-head p { margin: 0; }
+.agent-batch-list-head h4 { margin-bottom: 4px; color: #263832; font-size: 15px; }
+.agent-batch-list-head p { color: #71827c; font-size: 12px; }
+.agent-batch-cell-note { display: block; margin-top: 4px; overflow: hidden; color: #7b8c86; font-size: 11px; text-overflow: ellipsis; white-space: nowrap; }
+.agent-batch-detail { min-height: 260px; }
+.agent-batch-detail-head { display: flex; align-items: flex-start; justify-content: space-between; gap: 14px; margin-bottom: 14px; padding: 13px 14px; border-radius: 8px; background: #f5f8f7; }
+.agent-batch-detail-head div { display: grid; gap: 4px; min-width: 0; }
+.agent-batch-detail-head span { overflow: hidden; color: #71827c; font-size: 12px; text-overflow: ellipsis; white-space: nowrap; }
+.agent-batch-answer { max-height: 104px; overflow: auto; white-space: pre-wrap; word-break: break-word; line-height: 1.6; }
+
 .agent-advanced-settings {
  margin-top: 16px;
  border: 1px solid #e2eae7;
@@ -2315,6 +2346,10 @@ textarea {
  .agent-debug-context-grid {
    grid-template-columns: 1fr;
  }
+
+  .agent-batch-config-summary { grid-template-columns: 1fr; }
+  .agent-batch-name, .agent-batch-file { width: 100%; min-width: 0; }
+  .agent-batch-upload-row > .el-button { flex: 1; }
 }

@media (max-width: 860px) {
--- a/ai_knowledge_base_v2/apps/admin-web/src/types/api.ts
+++ b/ai_knowledge_base_v2/apps/admin-web/src/types/api.ts
@@ -392,6 +392,44 @@ export interface AgentDebugStreamComplete {
  questionType?: string;
 }

+export type AgentBatchStatus = "pending" | "running" | "completed" | "completed_with_errors" | "cancelled";
+
+export interface AgentBatchJob {
+  id: number;
+  name: string;
+  originalFilename: string;
+  status: AgentBatchStatus;
+  totalCount: number;
+  processedCount: number;
+  successCount: number;
+  failedCount: number;
+  progress: number;
+  modelId: number;
+  modelName: string;
+  knowledgeIds: number[];
+  knowledgeNames: string[];
+  createdAt: string;
+  startedAt?: string | null;
+  finishedAt?: string | null;
+}
+
+export interface AgentBatchItem {
+  id: number;
+  rowNumber: number;
+  externalNo?: string | null;
+  question: string;
+  answer?: string | null;
+  status: "pending" | "running" | "success" | "failed" | "cancelled";
+  errorMessage?: string | null;
+  modelName?: string | null;
+  retrieveCount?: number | null;
+  durationMs?: number | null;
+}
+
+export interface AgentBatchJobDetail extends AgentBatchJob {
+  items: PageResult<AgentBatchItem>;
+}
+
 export interface KnowledgeVersion {
  id: number;
  knowledgeId: number;
--- a/ai_knowledge_base_v2/apps/backend/alembic/versions/0033_agent_batch_tests.py
+++ b/ai_knowledge_base_v2/apps/backend/alembic/versions/0033_agent_batch_tests.py
@@ -0,0 +1,74 @@
+"""add persistent Agent batch tests
+
+Revision ID: 0033_agent_batch_tests
+Revises: 0032_message_feedback
+"""
+
+from alembic import op
+import sqlalchemy as sa
+
+
+revision = "0033_agent_batch_tests"
+down_revision = "0032_message_feedback"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.create_table(
+        "sys_agent_batch_test",
+        sa.Column("id", sa.BigInteger(), primary_key=True, autoincrement=True),
+        sa.Column("name", sa.String(160), nullable=False),
+        sa.Column("original_filename", sa.String(255), nullable=False),
+        sa.Column("status", sa.String(20), nullable=False, server_default="pending"),
+        sa.Column("total_count", sa.Integer(), nullable=False, server_default="0"),
+        sa.Column("success_count", sa.Integer(), nullable=False, server_default="0"),
+        sa.Column("failed_count", sa.Integer(), nullable=False, server_default="0"),
+        sa.Column("processed_count", sa.Integer(), nullable=False, server_default="0"),
+        sa.Column("prompt_content", sa.Text(), nullable=False),
+        sa.Column("model_id", sa.BigInteger(), sa.ForeignKey("sys_model.id"), nullable=False),
+        sa.Column("model_name", sa.String(100), nullable=False),
+        sa.Column("knowledge_ids", sa.Text(), nullable=False),
+        sa.Column("knowledge_names", sa.Text(), nullable=False),
+        sa.Column("generation_config", sa.Text(), nullable=False),
+        sa.Column("created_by", sa.BigInteger(), sa.ForeignKey("sys_admin.id"), nullable=False),
+        sa.Column("started_at", sa.DateTime(), nullable=True),
+        sa.Column("finished_at", sa.DateTime(), nullable=True),
+        sa.Column("cancelled_at", sa.DateTime(), nullable=True),
+        sa.Column("created_at", sa.DateTime(), nullable=False, server_default=sa.func.now()),
+        sa.Column("updated_at", sa.DateTime(), nullable=False, server_default=sa.func.now()),
+    )
+    op.create_index("ix_agent_batch_test_creator_created", "sys_agent_batch_test", ["created_by", "created_at"])
+    op.create_index("ix_agent_batch_test_status_created", "sys_agent_batch_test", ["status", "created_at"])
+    op.create_table(
+        "sys_agent_batch_test_item",
+        sa.Column("id", sa.BigInteger(), primary_key=True, autoincrement=True),
+        sa.Column("batch_id", sa.BigInteger(), sa.ForeignKey("sys_agent_batch_test.id", ondelete="CASCADE"), nullable=False),
+        sa.Column("row_number", sa.Integer(), nullable=False),
+        sa.Column("external_no", sa.String(100), nullable=True),
+        sa.Column("question", sa.Text(), nullable=False),
+        sa.Column("answer", sa.Text(), nullable=True),
+        sa.Column("status", sa.String(20), nullable=False, server_default="pending"),
+        sa.Column("error_message", sa.Text(), nullable=True),
+        sa.Column("model_name", sa.String(100), nullable=True),
+        sa.Column("knowledge_ids", sa.Text(), nullable=True),
+        sa.Column("retrieve_count", sa.Integer(), nullable=True),
+        sa.Column("duration_ms", sa.Integer(), nullable=True),
+        sa.Column("attempt_count", sa.Integer(), nullable=False, server_default="0"),
+        sa.Column("max_attempts", sa.Integer(), nullable=False, server_default="2"),
+        sa.Column("next_run_at", sa.DateTime(), nullable=True),
+        sa.Column("locked_at", sa.DateTime(), nullable=True),
+        sa.Column("locked_by", sa.String(120), nullable=True),
+        sa.Column("started_at", sa.DateTime(), nullable=True),
+        sa.Column("finished_at", sa.DateTime(), nullable=True),
+        sa.Column("created_at", sa.DateTime(), nullable=False, server_default=sa.func.now()),
+        sa.Column("updated_at", sa.DateTime(), nullable=False, server_default=sa.func.now()),
+    )
+    op.create_index("ix_agent_batch_item_job_order", "sys_agent_batch_test_item", ["batch_id", "row_number"])
+    op.create_index("ix_agent_batch_item_claim", "sys_agent_batch_test_item", ["status", "next_run_at", "id"])
+    op.create_index("ix_agent_batch_item_stale", "sys_agent_batch_test_item", ["status", "locked_at"])
+
+
+def downgrade() -> None:
+    op.drop_table("sys_agent_batch_test_item")
+    op.drop_table("sys_agent_batch_test")
--- a/ai_knowledge_base_v2/apps/backend/app/api/admin_agent_batch.py
+++ b/ai_knowledge_base_v2/apps/backend/app/api/admin_agent_batch.py
@@ -0,0 +1,157 @@
+from __future__ import annotations
+
+import json
+from datetime import UTC, datetime
+from typing import Annotated
+from urllib.parse import quote
+
+from fastapi import APIRouter, Depends, File, Form, HTTPException, Query, UploadFile, status
+from fastapi.responses import StreamingResponse
+from sqlalchemy import func, select, update
+from sqlalchemy.orm import Session
+
+from app.api.pagination import page_result
+from app.core.database import get_db
+from app.core.dependencies import get_current_admin
+from app.core.responses import api_success
+from app.models.admin import Admin
+from app.models.agent_batch import AgentBatchTest, AgentBatchTestItem
+from app.services.admin_service import OperationLogService
+from app.services.agent_batch_test_service import AgentBatchTestService, item_dict, job_dict
+
+
+router = APIRouter()
+
+
+@router.get("/agent/batch/template")
+def download_batch_template(
+    current_admin: Admin = Depends(get_current_admin),
+) -> StreamingResponse:
+    return StreamingResponse(
+        AgentBatchTestService.template_workbook(),
+        media_type="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
+        headers={"Content-Disposition": "attachment; filename*=UTF-8''Agent_batch_test_template.xlsx"},
+    )
+
+
+@router.post("/agent/batch/jobs")
+def create_batch_job(
+    file: Annotated[UploadFile, File(...)],
+    configJson: Annotated[str, Form(...)],
+    db: Session = Depends(get_db),
+    current_admin: Admin = Depends(get_current_admin),
+) -> dict:
+    try:
+        config = json.loads(configJson)
+    except json.JSONDecodeError as exc:
+        raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="批量测试配置格式不正确") from exc
+    if not isinstance(config, dict):
+        raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="批量测试配置格式不正确")
+    rows = AgentBatchTestService.parse_questions(file)
+    job = AgentBatchTestService.create_job(
+        db,
+        admin=current_admin,
+        filename=file.filename or "Agent批量测试.xlsx",
+        rows=rows,
+        config=config,
+    )
+    OperationLogService.write(db, admin_id=current_admin.id, module="agent", action="batch_create", target_id=job.id)
+    db.commit()
+    return api_success(job_dict(job))
+
+
+@router.get("/agent/batch/jobs")
+def list_batch_jobs(
+    page: int = Query(default=1, ge=1),
+    pageSize: int = Query(default=10, ge=10, le=100),
+    db: Session = Depends(get_db),
+    current_admin: Admin = Depends(get_current_admin),
+) -> dict:
+    total = db.scalar(select(func.count(AgentBatchTest.id))) or 0
+    rows = list(
+        db.scalars(
+            select(AgentBatchTest)
+            .order_by(AgentBatchTest.created_at.desc(), AgentBatchTest.id.desc())
+            .offset((page - 1) * pageSize)
+            .limit(pageSize)
+        )
+    )
+    return api_success(page_result([job_dict(row) for row in rows], total=total, page=page, page_size=pageSize))
+
+
+@router.get("/agent/batch/jobs/{job_id}")
+def get_batch_job(
+    job_id: int,
+    page: int = Query(default=1, ge=1),
+    pageSize: int = Query(default=20, ge=10, le=100),
+    db: Session = Depends(get_db),
+    current_admin: Admin = Depends(get_current_admin),
+) -> dict:
+    job = _job_or_404(db, job_id)
+    total = db.scalar(select(func.count(AgentBatchTestItem.id)).where(AgentBatchTestItem.batch_id == job.id)) or 0
+    items = list(
+        db.scalars(
+            select(AgentBatchTestItem)
+            .where(AgentBatchTestItem.batch_id == job.id)
+            .order_by(AgentBatchTestItem.row_number.asc(), AgentBatchTestItem.id.asc())
+            .offset((page - 1) * pageSize)
+            .limit(pageSize)
+        )
+    )
+    return api_success({**job_dict(job), "items": page_result([item_dict(item) for item in items], total=total, page=page, page_size=pageSize)})
+
+
+@router.get("/agent/batch/jobs/{job_id}/export")
+def export_batch_job(
+    job_id: int,
+    db: Session = Depends(get_db),
+    current_admin: Admin = Depends(get_current_admin),
+) -> StreamingResponse:
+    job = _job_or_404(db, job_id)
+    items = list(
+        db.scalars(
+            select(AgentBatchTestItem)
+            .where(AgentBatchTestItem.batch_id == job.id)
+            .order_by(AgentBatchTestItem.row_number.asc(), AgentBatchTestItem.id.asc())
+        )
+    )
+    filename = quote(f"Agent批量测试结果_{job.id}.xlsx")
+    OperationLogService.write(db, admin_id=current_admin.id, module="agent", action="batch_export", target_id=job.id)
+    db.commit()
+    return StreamingResponse(
+        AgentBatchTestService.export_workbook(job, items),
+        media_type="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
+        headers={"Content-Disposition": f"attachment; filename*=UTF-8''{filename}"},
+    )
+
+
+@router.post("/agent/batch/jobs/{job_id}/cancel")
+def cancel_batch_job(
+    job_id: int,
+    db: Session = Depends(get_db),
+    current_admin: Admin = Depends(get_current_admin),
+) -> dict:
+    job = _job_or_404(db, job_id)
+    if job.status not in {"pending", "running"}:
+        raise HTTPException(status_code=status.HTTP_409_CONFLICT, detail="当前任务已结束，不能取消")
+    now = datetime.now(UTC).replace(tzinfo=None)
+    job.status = "cancelled"
+    job.cancelled_at = now
+    job.finished_at = now
+    db.add(job)
+    db.execute(
+        update(AgentBatchTestItem)
+        .where(AgentBatchTestItem.batch_id == job.id, AgentBatchTestItem.status == "pending")
+        .values(status="cancelled", finished_at=now, next_run_at=None)
+    )
+    OperationLogService.write(db, admin_id=current_admin.id, module="agent", action="batch_cancel", target_id=job.id)
+    db.commit()
+    db.refresh(job)
+    return api_success(job_dict(job))
+
+
+def _job_or_404(db: Session, job_id: int) -> AgentBatchTest:
+    job = db.get(AgentBatchTest, job_id)
+    if job is None:
+        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="批量测试任务不存在")
+    return job
--- a/ai_knowledge_base_v2/apps/backend/app/api/router.py
+++ b/ai_knowledge_base_v2/apps/backend/app/api/router.py
@@ -7,6 +7,7 @@ from app.api import (
    admin_auth,
    admin_content_generation,
    admin_agent_records,
+    admin_agent_batch,
    admin_dashboard,
    admin_entitlements,
    admin_knowledge,
@@ -38,6 +39,7 @@ api_router.include_router(admin_management.router, prefix="/admin", tags=["admin
 guard = [Depends(enforce_admin_access)]
 api_router.include_router(admin_content_generation.router, prefix="/admin", tags=["admin-content-generation"], dependencies=guard)
 api_router.include_router(admin_agent_records.router, prefix="/admin", tags=["admin-agent-records"], dependencies=guard)
+api_router.include_router(admin_agent_batch.router, prefix="/admin", tags=["admin-agent-batch"], dependencies=guard)
 api_router.include_router(admin_dashboard.router, prefix="/admin", tags=["admin"], dependencies=guard)
 api_router.include_router(admin_entitlements.router, prefix="/admin", tags=["admin-entitlements"], dependencies=guard)
 api_router.include_router(admin_users.router, prefix="/admin", tags=["admin"], dependencies=guard)
--- a/ai_knowledge_base_v2/apps/backend/app/core/config.py
+++ b/ai_knowledge_base_v2/apps/backend/app/core/config.py
@@ -80,6 +80,9 @@ class Settings(BaseSettings):
    topic_settlement_poll_seconds: int = 2
    topic_settlement_stale_minutes: int = 30
    topic_settlement_max_attempts: int = 3
+    agent_batch_worker_enabled: bool = True
+    agent_batch_poll_seconds: int = 2
+    agent_batch_stale_minutes: int = 30
    bootstrap_admin_username: str = ""
    bootstrap_admin_password: str = ""
    bootstrap_admin_name: str = "系统管理员"
--- a/ai_knowledge_base_v2/apps/backend/app/core/dependencies.py
+++ b/ai_knowledge_base_v2/apps/backend/app/core/dependencies.py
@@ -100,6 +100,8 @@ def enforce_admin_access(
        permission = "users.delete" if method == "DELETE" else ("users.view" if method == "GET" else ("users.create" if method == "POST" and (path in {"user", "user/import", "user/import/excel"}) else "users.edit"))
    elif path.startswith("knowledge"):
        permission = "knowledge.delete" if method == "DELETE" else ("knowledge.view" if method == "GET" else ("knowledge.publish" if path.endswith("open-status") or path.endswith("lifecycle") else "knowledge.edit"))
+    elif path.startswith("agent/batch"):
+        permission = "prompt.batch"
    elif path.startswith("prompt") or path.startswith("agent/"):
        permission = "prompt.view" if method == "GET" else "prompt.edit"
    elif path.startswith("model"):
--- a/ai_knowledge_base_v2/apps/backend/app/main.py
+++ b/ai_knowledge_base_v2/apps/backend/app/main.py
@@ -15,6 +15,7 @@ from app.services.secret_service import SecretService
 from app.services.maintenance_service import MaintenanceService
 from app.services.periodic_report_worker import PeriodicReportWorker
 from app.services.topic_settlement_worker import TopicSettlementWorker
+from app.services.agent_batch_test_worker import AgentBatchTestWorker


@asynccontextmanager
@@ -26,12 +27,13 @@ async def lifespan(app: FastAPI):
    maintenance_task = asyncio.create_task(MaintenanceService.run_forever())
    periodic_report_task = asyncio.create_task(PeriodicReportWorker.run_forever())
    topic_settlement_task = asyncio.create_task(TopicSettlementWorker.run_forever())
+    agent_batch_task = asyncio.create_task(AgentBatchTestWorker.run_forever())
    try:
        yield
    finally:
-        for task in (maintenance_task, periodic_report_task, topic_settlement_task):
+        for task in (maintenance_task, periodic_report_task, topic_settlement_task, agent_batch_task):
            task.cancel()
-        for task in (maintenance_task, periodic_report_task, topic_settlement_task):
+        for task in (maintenance_task, periodic_report_task, topic_settlement_task, agent_batch_task):
            try:
                await task
            except asyncio.CancelledError:
--- a/ai_knowledge_base_v2/apps/backend/app/models/init.py
+++ b/ai_knowledge_base_v2/apps/backend/app/models/init.py
@@ -1,4 +1,5 @@
 from app.models.admin import Admin, Role
+from app.models.agent_batch import AgentBatchTest, AgentBatchTestItem
 from app.models.ai_config import ContentGenerationConfig, ModelConfig, Prompt, SystemConfig
 from app.models.base import Base
 from app.models.chat import ChatMessage, ChatSession, TopicSession
@@ -29,6 +30,8 @@ from app.models.user import User

 __all__ = [
    "Admin",
+    "AgentBatchTest",
+    "AgentBatchTestItem",
    "AiRequestLog",
    "Base",
    "ChatMessage",
--- a/ai_knowledge_base_v2/apps/backend/app/models/agent_batch.py
+++ b/ai_knowledge_base_v2/apps/backend/app/models/agent_batch.py
@@ -0,0 +1,71 @@
+from __future__ import annotations
+
+from datetime import datetime
+
+from sqlalchemy import BigInteger, DateTime, ForeignKey, Index, Integer, String, Text, func
+from sqlalchemy.orm import Mapped, mapped_column
+
+from app.models.base import Base
+
+
+PRIMARY_KEY_TYPE = BigInteger().with_variant(Integer, "sqlite")
+
+
+class AgentBatchTest(Base):
+    __tablename__ = "sys_agent_batch_test"
+    __table_args__ = (
+        Index("ix_agent_batch_test_creator_created", "created_by", "created_at"),
+        Index("ix_agent_batch_test_status_created", "status", "created_at"),
+    )
+
+    id: Mapped[int] = mapped_column(PRIMARY_KEY_TYPE, primary_key=True, autoincrement=True)
+    name: Mapped[str] = mapped_column(String(160), nullable=False)
+    original_filename: Mapped[str] = mapped_column(String(255), nullable=False)
+    status: Mapped[str] = mapped_column(String(20), default="pending", nullable=False)
+    total_count: Mapped[int] = mapped_column(Integer, default=0, nullable=False)
+    success_count: Mapped[int] = mapped_column(Integer, default=0, nullable=False)
+    failed_count: Mapped[int] = mapped_column(Integer, default=0, nullable=False)
+    processed_count: Mapped[int] = mapped_column(Integer, default=0, nullable=False)
+    prompt_content: Mapped[str] = mapped_column(Text, nullable=False)
+    model_id: Mapped[int] = mapped_column(ForeignKey("sys_model.id"), nullable=False)
+    model_name: Mapped[str] = mapped_column(String(100), nullable=False)
+    knowledge_ids: Mapped[str] = mapped_column(Text, nullable=False)
+    knowledge_names: Mapped[str] = mapped_column(Text, nullable=False)
+    generation_config: Mapped[str] = mapped_column(Text, nullable=False)
+    created_by: Mapped[int] = mapped_column(ForeignKey("sys_admin.id"), nullable=False)
+    started_at: Mapped[datetime | None] = mapped_column(DateTime, nullable=True)
+    finished_at: Mapped[datetime | None] = mapped_column(DateTime, nullable=True)
+    cancelled_at: Mapped[datetime | None] = mapped_column(DateTime, nullable=True)
+    created_at: Mapped[datetime] = mapped_column(DateTime, server_default=func.now(), nullable=False)
+    updated_at: Mapped[datetime] = mapped_column(DateTime, server_default=func.now(), onupdate=func.now(), nullable=False)
+
+
+class AgentBatchTestItem(Base):
+    __tablename__ = "sys_agent_batch_test_item"
+    __table_args__ = (
+        Index("ix_agent_batch_item_job_order", "batch_id", "row_number"),
+        Index("ix_agent_batch_item_claim", "status", "next_run_at", "id"),
+        Index("ix_agent_batch_item_stale", "status", "locked_at"),
+    )
+
+    id: Mapped[int] = mapped_column(PRIMARY_KEY_TYPE, primary_key=True, autoincrement=True)
+    batch_id: Mapped[int] = mapped_column(ForeignKey("sys_agent_batch_test.id", ondelete="CASCADE"), nullable=False)
+    row_number: Mapped[int] = mapped_column(Integer, nullable=False)
+    external_no: Mapped[str | None] = mapped_column(String(100), nullable=True)
+    question: Mapped[str] = mapped_column(Text, nullable=False)
+    answer: Mapped[str | None] = mapped_column(Text, nullable=True)
+    status: Mapped[str] = mapped_column(String(20), default="pending", nullable=False)
+    error_message: Mapped[str | None] = mapped_column(Text, nullable=True)
+    model_name: Mapped[str | None] = mapped_column(String(100), nullable=True)
+    knowledge_ids: Mapped[str | None] = mapped_column(Text, nullable=True)
+    retrieve_count: Mapped[int | None] = mapped_column(Integer, nullable=True)
+    duration_ms: Mapped[int | None] = mapped_column(Integer, nullable=True)
+    attempt_count: Mapped[int] = mapped_column(Integer, default=0, nullable=False)
+    max_attempts: Mapped[int] = mapped_column(Integer, default=2, nullable=False)
+    next_run_at: Mapped[datetime | None] = mapped_column(DateTime, nullable=True)
+    locked_at: Mapped[datetime | None] = mapped_column(DateTime, nullable=True)
+    locked_by: Mapped[str | None] = mapped_column(String(120), nullable=True)
+    started_at: Mapped[datetime | None] = mapped_column(DateTime, nullable=True)
+    finished_at: Mapped[datetime | None] = mapped_column(DateTime, nullable=True)
+    created_at: Mapped[datetime] = mapped_column(DateTime, server_default=func.now(), nullable=False)
+    updated_at: Mapped[datetime] = mapped_column(DateTime, server_default=func.now(), onupdate=func.now(), nullable=False)
--- a/ai_knowledge_base_v2/apps/backend/app/services/admin_permission_service.py
+++ b/ai_knowledge_base_v2/apps/backend/app/services/admin_permission_service.py
@@ -12,7 +12,7 @@ PERMISSION_TREE = [
    {"code": "users", "name": "用户管理", "children": [{"code": "users.view", "name": "查看用户"}, {"code": "users.create", "name": "新增/导入"}, {"code": "users.edit", "name": "编辑/权益续期"}, {"code": "users.delete", "name": "删除用户"}]},
    {"code": "entitlements", "name": "权益管理", "children": [{"code": "entitlements.view", "name": "查看权益"}, {"code": "entitlements.edit", "name": "编辑权益"}]},
    {"code": "knowledge", "name": "知识库管理", "children": [{"code": "knowledge.view", "name": "查看知识库"}, {"code": "knowledge.edit", "name": "新增/编辑/同步"}, {"code": "knowledge.publish", "name": "开放/归档"}, {"code": "knowledge.delete", "name": "删除知识库"}]},
-    {"code": "prompt", "name": "Agent 管理", "children": [{"code": "prompt.view", "name": "查看 Agent"}, {"code": "prompt.edit", "name": "编辑/测试 Agent"}]},
+    {"code": "prompt", "name": "Agent 管理", "children": [{"code": "prompt.view", "name": "查看 Agent"}, {"code": "prompt.edit", "name": "编辑/单条测试 Agent"}, {"code": "prompt.batch", "name": "批量测试/导出"}]},
    {"code": "models", "name": "模型管理", "children": [{"code": "models.view", "name": "查看模型"}, {"code": "models.edit", "name": "新增/编辑/测试"}, {"code": "models.delete", "name": "删除模型"}]},
    {"code": "content-generation", "name": "内容生成", "children": [{"code": "content-generation.view", "name": "查看配置"}, {"code": "content-generation.edit", "name": "编辑/测试配置"}]},
    {"code": "configs", "name": "系统配置", "children": [{"code": "configs.view", "name": "查看配置"}, {"code": "configs.edit", "name": "修改配置"}]},
--- a/ai_knowledge_base_v2/apps/backend/app/services/agent_batch_test_service.py
+++ b/ai_knowledge_base_v2/apps/backend/app/services/agent_batch_test_service.py
@@ -0,0 +1,274 @@
+from __future__ import annotations
+
+import json
+from datetime import UTC, datetime
+from io import BytesIO
+
+from fastapi import HTTPException, UploadFile, status
+from openpyxl import Workbook, load_workbook
+from openpyxl.styles import Alignment, Font, PatternFill
+from openpyxl.worksheet.table import Table, TableStyleInfo
+from sqlalchemy import func, select
+from sqlalchemy.orm import Session
+
+from app.models.admin import Admin
+from app.models.agent_batch import AgentBatchTest, AgentBatchTestItem
+from app.models.ai_config import ModelConfig
+from app.models.knowledge import Knowledge
+
+
+MAX_BATCH_QUESTIONS = 200
+MAX_QUESTION_LENGTH = 2000
+TEMPLATE_HEADERS = ["序号", "问题"]
+
+
+class AgentBatchTestService:
+    @staticmethod
+    def parse_questions(file: UploadFile) -> list[dict]:
+        filename = file.filename or ""
+        if not filename.lower().endswith(".xlsx"):
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="请上传 .xlsx 格式的 Excel 文件")
+        content = file.file.read()
+        if not content:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Excel 文件为空")
+        if len(content) > 5 * 1024 * 1024:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Excel 文件不能超过 5MB")
+        try:
+            workbook = load_workbook(BytesIO(content), read_only=True, data_only=False)
+        except Exception as exc:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Excel 文件解析失败，请重新下载模板填写") from exc
+        sheet = workbook.active
+        if sheet.max_row > 1000:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Excel 行数异常，请删除多余空行后重试")
+        headers = [str(value or "").strip() for value in next(sheet.iter_rows(min_row=1, max_row=1, values_only=True), ())]
+        header_map = {name: index for index, name in enumerate(headers) if name}
+        if "问题" not in header_map:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="模板缺少必填列：问题")
+        rows: list[dict] = []
+        seen: set[str] = set()
+        for row_number, values in enumerate(sheet.iter_rows(min_row=2, max_row=min(sheet.max_row, 1000), values_only=True), start=2):
+            raw_question = _value_at(values, header_map["问题"])
+            if isinstance(raw_question, str) and raw_question.startswith("="):
+                raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail=f"第 {row_number} 行问题不能使用 Excel 公式")
+            question = _cell_text(raw_question)
+            external_no = _cell_text(_value_at(values, header_map.get("序号"))) or str(row_number - 1)
+            if not question:
+                continue
+            if len(question) > MAX_QUESTION_LENGTH:
+                raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail=f"第 {row_number} 行问题超过 {MAX_QUESTION_LENGTH} 字")
+            normalized = " ".join(question.split())
+            if normalized in seen:
+                raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail=f"第 {row_number} 行问题与前面重复")
+            seen.add(normalized)
+            rows.append({"rowNumber": row_number, "externalNo": external_no[:100], "question": question})
+            if len(rows) > MAX_BATCH_QUESTIONS:
+                raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail=f"单次最多导入 {MAX_BATCH_QUESTIONS} 个问题")
+        if not rows:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Excel 中没有可测试的问题")
+        return rows
+
+    @staticmethod
+    def create_job(
+        db: Session,
+        *,
+        admin: Admin,
+        filename: str,
+        rows: list[dict],
+        config: dict,
+    ) -> AgentBatchTest:
+        prompt_content = str(config.get("promptContent") or "").strip()
+        if not prompt_content:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="主提示词不能为空")
+        try:
+            model_id = int(config.get("modelId"))
+        except (TypeError, ValueError) as exc:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="请选择批量测试模型") from exc
+        model = db.get(ModelConfig, model_id)
+        if model is None:
+            raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="批量测试模型不存在")
+        knowledge_ids = _positive_int_list(config.get("knowledgeIds"))
+        if knowledge_ids:
+            found = list(db.scalars(select(Knowledge).where(Knowledge.id.in_(knowledge_ids))))
+            if len(found) != len(knowledge_ids):
+                raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="所选知识库中包含已删除的数据")
+            knowledge_names = [item.name for item in sorted(found, key=lambda item: knowledge_ids.index(item.id))]
+        else:
+            knowledge_names = ["全部正式开放知识库"]
+        generation_config = {
+            "temperature": config.get("temperature"),
+            "topP": config.get("topP"),
+            "topK": config.get("topK"),
+            "presencePenalty": config.get("presencePenalty"),
+            "frequencyPenalty": config.get("frequencyPenalty"),
+            "maxToken": config.get("maxToken") or 8192,
+            "streamEnabled": int(config.get("streamEnabled", 1)),
+            "reasoningVisible": 0,
+            "responseDepth": int(config.get("responseDepth", 35)),
+        }
+        job = AgentBatchTest(
+            name=str(config.get("name") or filename.rsplit(".", 1)[0] or "Agent 批量测试")[:160],
+            original_filename=filename[:255],
+            total_count=len(rows),
+            prompt_content=prompt_content,
+            model_id=model.id,
+            model_name=model.display_name or model.model_name,
+            knowledge_ids=json.dumps(knowledge_ids, ensure_ascii=False),
+            knowledge_names=json.dumps(knowledge_names, ensure_ascii=False),
+            generation_config=json.dumps(generation_config, ensure_ascii=False),
+            created_by=admin.id,
+        )
+        db.add(job)
+        db.flush()
+        db.add_all([
+            AgentBatchTestItem(
+                batch_id=job.id,
+                row_number=row["rowNumber"],
+                external_no=row["externalNo"],
+                question=row["question"],
+            )
+            for row in rows
+        ])
+        db.flush()
+        return job
+
+    @staticmethod
+    def template_workbook() -> BytesIO:
+        workbook = Workbook()
+        sheet = workbook.active
+        sheet.title = "批量测试问题"
+        sheet.append(TEMPLATE_HEADERS)
+        sheet.append([1, "请填写需要测试的问题，保留一行一个问题。"])
+        sheet.append([2, "例如：遇到情绪焦虑时，可以先做什么？"])
+        _style_sheet(sheet, widths=(12, 72), table_ref="A1:B3", table_name="AgentBatchQuestionTable")
+        note = workbook.create_sheet("填写说明")
+        note.append(["填写规则", "说明"])
+        note.append(["必填列", "问题"])
+        note.append(["单次数量", f"最多 {MAX_BATCH_QUESTIONS} 个问题"])
+        note.append(["问题长度", f"每个问题最多 {MAX_QUESTION_LENGTH} 字"])
+        note.append(["重复问题", "同一文件内不允许重复"])
+        note.append(["不要修改", "不要修改工作表名称和表头；可以删除示例行后填写"])
+        _style_sheet(note, widths=(18, 72), table_ref="A1:B6", table_name="AgentBatchInstructions")
+        stream = BytesIO()
+        workbook.save(stream)
+        stream.seek(0)
+        return stream
+
+    @staticmethod
+    def export_workbook(job: AgentBatchTest, items: list[AgentBatchTestItem]) -> BytesIO:
+        workbook = Workbook()
+        sheet = workbook.active
+        sheet.title = "批量测试结果"
+        headers = ["序号", "问题", "答案", "状态", "失败原因", "实际模型", "召回数量", "耗时(秒)"]
+        sheet.append(headers)
+        status_labels = {"success": "成功", "failed": "失败", "cancelled": "已取消", "pending": "等待中", "running": "生成中"}
+        for item in items:
+            sheet.append([
+                item.external_no or item.row_number - 1,
+                _excel_safe(item.question),
+                _excel_safe(item.answer or ""),
+                status_labels.get(item.status, item.status),
+                _excel_safe(item.error_message or ""),
+                _excel_safe(item.model_name or job.model_name),
+                item.retrieve_count,
+                round(item.duration_ms / 1000, 2) if item.duration_ms is not None else None,
+            ])
+        end_row = max(2, len(items) + 1)
+        _style_sheet(sheet, widths=(12, 42, 80, 12, 36, 22, 12, 12), table_ref=f"A1:H{end_row}", table_name="AgentBatchResultTable")
+        for row in sheet.iter_rows(min_row=2, max_row=end_row):
+            row[1].alignment = Alignment(vertical="top", wrap_text=True)
+            row[2].alignment = Alignment(vertical="top", wrap_text=True)
+            row[4].alignment = Alignment(vertical="top", wrap_text=True)
+        summary = workbook.create_sheet("任务信息")
+        summary.append(["项目", "内容"])
+        summary.append(["任务名称", job.name])
+        summary.append(["任务状态", job.status])
+        summary.append(["问题总数", job.total_count])
+        summary.append(["成功", job.success_count])
+        summary.append(["失败", job.failed_count])
+        summary.append(["测试模型", job.model_name])
+        summary.append(["知识库", "、".join(json.loads(job.knowledge_names or "[]"))])
+        summary.append(["创建时间", job.created_at])
+        summary.append(["完成时间", job.finished_at])
+        _style_sheet(summary, widths=(20, 86), table_ref="A1:B10", table_name="AgentBatchSummary")
+        summary.column_dimensions["B"].width = 86
+        stream = BytesIO()
+        workbook.save(stream)
+        stream.seek(0)
+        return stream
+
+
+def job_dict(job: AgentBatchTest) -> dict:
+    return {
+        "id": job.id,
+        "name": job.name,
+        "originalFilename": job.original_filename,
+        "status": job.status,
+        "totalCount": job.total_count,
+        "processedCount": job.processed_count,
+        "successCount": job.success_count,
+        "failedCount": job.failed_count,
+        "progress": round(job.processed_count * 100 / job.total_count) if job.total_count else 0,
+        "modelId": job.model_id,
+        "modelName": job.model_name,
+        "knowledgeIds": json.loads(job.knowledge_ids or "[]"),
+        "knowledgeNames": json.loads(job.knowledge_names or "[]"),
+        "createdAt": job.created_at,
+        "startedAt": job.started_at,
+        "finishedAt": job.finished_at,
+    }
+
+
+def item_dict(item: AgentBatchTestItem) -> dict:
+    return {
+        "id": item.id,
+        "rowNumber": item.row_number,
+        "externalNo": item.external_no,
+        "question": item.question,
+        "answer": item.answer,
+        "status": item.status,
+        "errorMessage": item.error_message,
+        "modelName": item.model_name,
+        "retrieveCount": item.retrieve_count,
+        "durationMs": item.duration_ms,
+    }
+
+
+def _style_sheet(sheet, *, widths: tuple[int, ...], table_ref: str, table_name: str) -> None:
+    sheet.freeze_panes = "A2"
+    sheet.sheet_view.showGridLines = False
+    header_fill = PatternFill("solid", fgColor="176B57")
+    for cell in sheet[1]:
+        cell.fill = header_fill
+        cell.font = Font(color="FFFFFF", bold=True)
+        cell.alignment = Alignment(vertical="center")
+    sheet.row_dimensions[1].height = 28
+    for index, width in enumerate(widths, start=1):
+        sheet.column_dimensions[chr(64 + index)].width = width
+    table = Table(displayName=table_name, ref=table_ref)
+    table.tableStyleInfo = TableStyleInfo(name="TableStyleMedium4", showRowStripes=True, showFirstColumn=False, showLastColumn=False)
+    sheet.add_table(table)
+
+
+def _cell_text(value) -> str:
+    if value is None:
+        return ""
+    return str(value).strip()
+
+
+def _value_at(row: tuple, index: int | None):
+    return row[index] if index is not None and index < len(row) else None
+
+
+def _positive_int_list(value) -> list[int]:
+    if not isinstance(value, list):
+        return []
+    try:
+        result = list(dict.fromkeys(int(item) for item in value if int(item) > 0))
+    except (TypeError, ValueError) as exc:
+        raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="知识库参数格式不正确") from exc
+    return result
+
+
+def _excel_safe(value: str) -> str:
+    text = str(value or "")
+    return f"'{text}" if text.lstrip().startswith(("=", "+", "-", "@")) else text
--- a/ai_knowledge_base_v2/apps/backend/app/services/agent_batch_test_worker.py
+++ b/ai_knowledge_base_v2/apps/backend/app/services/agent_batch_test_worker.py
@@ -0,0 +1,227 @@
+from __future__ import annotations
+
+import asyncio
+import json
+import logging
+import os
+import socket
+import time
+import uuid
+from datetime import UTC, datetime, timedelta
+
+from sqlalchemy import func, or_, select
+from sqlalchemy.orm import Session
+
+from app.core.config import get_settings
+from app.core.database import SessionLocal
+from app.models.admin import Admin
+from app.models.agent_batch import AgentBatchTest, AgentBatchTestItem
+from app.schemas.admin import AgentDebugRequest
+from app.services.agent_debug_service import AgentDebugService
+
+
+logger = logging.getLogger(__name__)
+
+
+class AgentBatchTestWorker:
+    """Persistent, restart-safe queue for Agent batch questions."""
+
+    @classmethod
+    async def run_forever(cls) -> None:
+        settings = get_settings()
+        if not settings.agent_batch_worker_enabled:
+            logger.info("agent batch test worker disabled")
+            return
+        worker_id = f"{socket.gethostname()}:{os.getpid()}:{uuid.uuid4().hex[:8]}"
+        while True:
+            try:
+                processed = await cls.run_once(worker_id)
+            except Exception:
+                processed = False
+                logger.exception("agent batch test worker iteration failed")
+            await asyncio.sleep(0 if processed else max(1, settings.agent_batch_poll_seconds))
+
+    @classmethod
+    async def run_once(cls, worker_id: str) -> bool:
+        now = _now()
+        with SessionLocal() as db:
+            recovered_batch_ids = cls.recover_stale_items(db, now=now)
+            db.commit()
+            for batch_id in recovered_batch_ids:
+                cls.refresh_job(db, batch_id)
+        with SessionLocal() as db:
+            item_id = cls.claim_next(db, worker_id=worker_id, now=now)
+        if item_id is None:
+            return False
+        await cls.execute_claimed(item_id=item_id, worker_id=worker_id)
+        return True
+
+    @staticmethod
+    def claim_next(db: Session, *, worker_id: str, now: datetime | None = None) -> int | None:
+        current = now or _now()
+        item = db.scalar(
+            select(AgentBatchTestItem)
+            .join(AgentBatchTest, AgentBatchTest.id == AgentBatchTestItem.batch_id)
+            .where(
+                AgentBatchTest.status.in_(("pending", "running")),
+                AgentBatchTestItem.status == "pending",
+                AgentBatchTestItem.attempt_count < AgentBatchTestItem.max_attempts,
+                or_(AgentBatchTestItem.next_run_at.is_(None), AgentBatchTestItem.next_run_at <= current),
+            )
+            .order_by(AgentBatchTestItem.batch_id.asc(), AgentBatchTestItem.row_number.asc())
+            .with_for_update(skip_locked=True)
+            .limit(1)
+        )
+        if item is None:
+            db.rollback()
+            return None
+        job = db.get(AgentBatchTest, item.batch_id)
+        item.status = "running"
+        item.attempt_count += 1
+        item.locked_at = current
+        item.locked_by = worker_id
+        item.started_at = current
+        if job is not None and job.status == "pending":
+            job.status = "running"
+            job.started_at = current
+            db.add(job)
+        db.add(item)
+        db.commit()
+        return item.id
+
+    @classmethod
+    async def execute_claimed(cls, *, item_id: int, worker_id: str) -> None:
+        with SessionLocal() as db:
+            item = db.get(AgentBatchTestItem, item_id)
+            if item is None or item.status != "running" or item.locked_by != worker_id:
+                return
+            job = db.get(AgentBatchTest, item.batch_id)
+            admin = db.get(Admin, job.created_by) if job is not None else None
+            if job is None or admin is None:
+                cls._finish_item(db, item, error="批量任务或创建管理员不存在")
+                return
+            if job.status == "cancelled":
+                item.status = "cancelled"
+                item.finished_at = _now()
+                item.locked_at = None
+                item.locked_by = None
+                db.add(item)
+                db.commit()
+                return
+            payload = _payload_from(job, item)
+            started = time.monotonic()
+            try:
+                result = await AgentDebugService.run_complete(payload, db, admin)
+                duration_ms = round((time.monotonic() - started) * 1000)
+                item.answer = result["answer"]
+                item.model_name = result.get("modelName") or job.model_name
+                item.knowledge_ids = json.dumps(result.get("knowledgeIds") or [], ensure_ascii=False)
+                item.retrieve_count = result.get("retrieveCount")
+                item.duration_ms = duration_ms
+                item.status = "success"
+                item.error_message = None
+                item.finished_at = _now()
+                item.next_run_at = None
+                item.locked_at = None
+                item.locked_by = None
+                db.add(item)
+                db.commit()
+            except Exception as exc:
+                db.rollback()
+                item = db.get(AgentBatchTestItem, item_id)
+                if item is None:
+                    return
+                item.duration_ms = round((time.monotonic() - started) * 1000)
+                if item.attempt_count < item.max_attempts:
+                    item.status = "pending"
+                    item.next_run_at = _now() + timedelta(seconds=15 * item.attempt_count)
+                    item.finished_at = None
+                else:
+                    item.status = "failed"
+                    item.next_run_at = None
+                    item.finished_at = _now()
+                item.error_message = (str(exc) or "Agent 生成失败")[:2000]
+                item.locked_at = None
+                item.locked_by = None
+                db.add(item)
+                db.commit()
+            cls.refresh_job(db, job.id)
+
+    @staticmethod
+    def refresh_job(db: Session, batch_id: int) -> None:
+        job = db.get(AgentBatchTest, batch_id)
+        if job is None or job.status == "cancelled":
+            return
+        counts = dict(
+            db.execute(
+                select(AgentBatchTestItem.status, func.count(AgentBatchTestItem.id))
+                .where(AgentBatchTestItem.batch_id == batch_id)
+                .group_by(AgentBatchTestItem.status)
+            ).all()
+        )
+        job.success_count = counts.get("success", 0)
+        job.failed_count = counts.get("failed", 0)
+        job.processed_count = job.success_count + job.failed_count
+        if job.processed_count >= job.total_count:
+            job.status = "completed" if job.failed_count == 0 else "completed_with_errors"
+            job.finished_at = _now()
+        else:
+            job.status = "running"
+        db.add(job)
+        db.commit()
+
+    @staticmethod
+    def recover_stale_items(db: Session, *, now: datetime | None = None) -> set[int]:
+        current = now or _now()
+        stale_before = current - timedelta(minutes=max(5, get_settings().agent_batch_stale_minutes))
+        items = list(
+            db.scalars(
+                select(AgentBatchTestItem)
+                .where(
+                    AgentBatchTestItem.status == "running",
+                    AgentBatchTestItem.locked_at.is_not(None),
+                    AgentBatchTestItem.locked_at < stale_before,
+                )
+                .order_by(AgentBatchTestItem.id.asc())
+                .limit(100)
+                .with_for_update(skip_locked=True)
+            )
+        )
+        affected_batch_ids: set[int] = set()
+        for item in items:
+            affected_batch_ids.add(item.batch_id)
+            item.status = "pending" if item.attempt_count < item.max_attempts else "failed"
+            item.error_message = "任务执行进程中断，系统已自动恢复" if item.status == "pending" else "任务执行进程连续中断"
+            item.next_run_at = current if item.status == "pending" else None
+            item.finished_at = current if item.status == "failed" else None
+            item.locked_at = None
+            item.locked_by = None
+            db.add(item)
+        return affected_batch_ids
+
+    @classmethod
+    def _finish_item(cls, db: Session, item: AgentBatchTestItem, *, error: str) -> None:
+        item.status = "failed"
+        item.error_message = error
+        item.finished_at = _now()
+        item.locked_at = None
+        item.locked_by = None
+        db.add(item)
+        db.commit()
+        cls.refresh_job(db, item.batch_id)
+
+
+def _payload_from(job: AgentBatchTest, item: AgentBatchTestItem) -> AgentDebugRequest:
+    generation = json.loads(job.generation_config or "{}")
+    return AgentDebugRequest(
+        promptContent=job.prompt_content,
+        modelId=job.model_id,
+        knowledgeIds=json.loads(job.knowledge_ids or "[]"),
+        question=item.question,
+        history=[],
+        **generation,
+    )
+
+
+def _now() -> datetime:
+    return datetime.now(UTC).replace(tzinfo=None)
--- a/ai_knowledge_base_v2/apps/backend/app/services/agent_debug_service.py
+++ b/ai_knowledge_base_v2/apps/backend/app/services/agent_debug_service.py
@@ -283,3 +283,20 @@ class AgentDebugService:
        except Exception as exc:
            db.rollback()
            yield {"type": "error", "message": str(exc) or "Agent 调试失败"}
+
+    @classmethod
+    async def run_complete(cls, payload: AgentDebugRequest, db: Session, current_admin: Admin) -> dict:
+        """Consume the same routed debug stream used by the UI and return one complete answer."""
+        answer_parts: list[str] = []
+        completion: dict = {}
+        async for event in cls.stream(payload, db, current_admin):
+            if event.get("type") == "content":
+                answer_parts.append(str(event.get("content") or ""))
+            elif event.get("type") == "complete":
+                completion = event
+            elif event.get("type") == "error":
+                raise RuntimeError(str(event.get("message") or "Agent 生成失败"))
+        answer = "".join(answer_parts).strip()
+        if not answer:
+            raise RuntimeError("模型未返回正式回答")
+        return {"answer": answer, **completion}
--- a/ai_knowledge_base_v2/apps/backend/tests/test_admin_permissions.py
+++ b/ai_knowledge_base_v2/apps/backend/tests/test_admin_permissions.py
@@ -13,6 +13,7 @@ def test_super_admin_has_all_permissions() -> None:
    admin = Admin(id=1, username="root", password="unused", name="root", status=1, is_super_admin=1)
    assert permissions_for(admin) == ALL_PERMISSION_CODES
    assert {"feedback.view", "feedback.detail", "feedback.export", "feedback.delete"} <= ALL_PERMISSION_CODES
+    assert "prompt.batch" in ALL_PERMISSION_CODES


 def test_role_permissions_are_restricted_to_catalog() -> None:
--- a/ai_knowledge_base_v2/apps/backend/tests/test_agent_batch_tests.py
+++ b/ai_knowledge_base_v2/apps/backend/tests/test_agent_batch_tests.py
@@ -0,0 +1,125 @@
+from io import BytesIO
+import json
+
+import pytest
+from fastapi import HTTPException, UploadFile
+from openpyxl import Workbook, load_workbook
+from sqlalchemy import create_engine, select
+from sqlalchemy.orm import Session
+from sqlalchemy.pool import StaticPool
+
+from app.models import Base
+from app.models.admin import Admin
+from app.models.agent_batch import AgentBatchTestItem
+from app.models.ai_config import ModelConfig
+from app.services.agent_batch_test_service import AgentBatchTestService, MAX_BATCH_QUESTIONS
+
+
+def _database() -> Session:
+    engine = create_engine(
+        "sqlite:///:memory:",
+        connect_args={"check_same_thread": False},
+        poolclass=StaticPool,
+    )
+    Base.metadata.create_all(engine)
+    return Session(engine)
+
+
+def _admin() -> Admin:
+    return Admin(id=1, username="admin", password="hash", name="系统管理员", status=1)
+
+
+def _model() -> ModelConfig:
+    return ModelConfig(
+        id=1,
+        provider="mock",
+        display_name="批量测试模型",
+        api_type="openai_compatible",
+        model_name="batch-model",
+        api_url="",
+        api_key="",
+        auth_type="bearer",
+        enabled=1,
+    )
+
+
+def test_template_can_be_downloaded_and_imported() -> None:
+    stream = AgentBatchTestService.template_workbook()
+    workbook = load_workbook(stream, data_only=True)
+
+    assert workbook.sheetnames == ["批量测试问题", "填写说明"]
+    assert [cell.value for cell in workbook["批量测试问题"][1]] == ["序号", "问题"]
+    assert workbook["填写说明"]["B3"].value == f"最多 {MAX_BATCH_QUESTIONS} 个问题"
+
+    upload = UploadFile(filename="批量测试.xlsx", file=BytesIO(stream.getvalue()))
+    rows = AgentBatchTestService.parse_questions(upload)
+    assert len(rows) == 2
+    assert rows[0]["externalNo"] == "1"
+    assert "测试的问题" in rows[0]["question"]
+
+
+def test_create_job_snapshots_config_and_export_keeps_failed_rows() -> None:
+    with _database() as db:
+        admin = _admin()
+        model = _model()
+        db.add_all([admin, model])
+        db.commit()
+        job = AgentBatchTestService.create_job(
+            db,
+            admin=admin,
+            filename="回归问题.xlsx",
+            rows=[
+                {"rowNumber": 2, "externalNo": "Q-1", "question": "第一个问题"},
+                {"rowNumber": 3, "externalNo": "Q-2", "question": "第二个问题"},
+            ],
+            config={
+                "name": "提示词回归测试",
+                "promptContent": "只根据知识库回答",
+                "modelId": model.id,
+                "knowledgeIds": [],
+                "temperature": 0.2,
+                "maxToken": 8192,
+                "responseDepth": 35,
+            },
+        )
+        items = list(db.scalars(select(AgentBatchTestItem).order_by(AgentBatchTestItem.row_number)))
+        items[0].status = "success"
+        items[0].answer = "第一个答案"
+        items[0].model_name = "批量测试模型"
+        items[0].retrieve_count = 3
+        items[0].duration_ms = 1250
+        items[1].status = "failed"
+        items[1].error_message = "供应商超时"
+        job.status = "completed_with_errors"
+        job.processed_count = 2
+        job.success_count = 1
+        job.failed_count = 1
+        db.commit()
+
+        assert json.loads(job.generation_config)["maxToken"] == 8192
+        assert json.loads(job.knowledge_names) == ["全部正式开放知识库"]
+
+        exported = AgentBatchTestService.export_workbook(job, items)
+        workbook = load_workbook(exported, data_only=True)
+        sheet = workbook["批量测试结果"]
+        assert sheet["C2"].value == "第一个答案"
+        assert sheet["D2"].value == "成功"
+        assert sheet["D3"].value == "失败"
+        assert sheet["E3"].value == "供应商超时"
+        assert sheet.freeze_panes == "A2"
+
+
+def test_import_rejects_excel_formula_question() -> None:
+    workbook = Workbook()
+    sheet = workbook.active
+    sheet.append(["序号", "问题"])
+    sheet.append([1, '=HYPERLINK("https://example.com","问题")'])
+    stream = BytesIO()
+    workbook.save(stream)
+    stream.seek(0)
+
+    with pytest.raises(HTTPException) as exc:
+        AgentBatchTestService.parse_questions(UploadFile(filename="危险问题.xlsx", file=stream))
+
+    assert exc.value.status_code == 400
+    assert "不能使用 Excel 公式" in exc.value.detail
--- a/ai_knowledge_base_v2/design-qa.md
+++ b/ai_knowledge_base_v2/design-qa.md
@@ -1,3 +1,20 @@
+# Agent 批量测试页面验收（2026-08-12）
+
+- 实现截图：`development_records/ui-evidence/agent-batch-test-2026-08-12.png`
+- 验收地址：`http://127.0.0.1:8080/`
+- 页面状态：系统管理员已登录 / Agent 管理 / 批量测试 / 2 条真实问题已完成
+- 视口：1280 × 720 CSS px；页面 `scrollWidth=clientWidth=1280`。
+- 批量测试页切换为 1012px 全宽工作区，隐藏与该任务无关的单条调试预览和检索追踪，上传区、配置快照、进度表格和操作按钮无需横向滚动。
+- 已真实下载并解析模板，工作表为“批量测试问题、填写说明”，表头为“序号、问题”，冻结首行且包含格式化表格。
+- 已真实上传模板并创建 2 条问题任务；状态从等待执行、生成中、50% 更新到已完成，最终成功 2、失败 0。
+- 已验证任务详情显示每条问题、完整答案、状态与耗时；结果导出包含问题、答案、失败原因、实际模型、召回数量和耗时。
+- 浏览器控制台错误和警告：0；无横向滚动。
+- `npm --prefix apps/admin-web run build`：通过；后端相关测试：31 passed。
+
+final result: passed
+
+---
+
 # 反馈管理日期筛选控件修复验收（2026-08-11）

 ## 对照范围
--- a/ai_knowledge_base_v2/development_records/ui-evidence/agent-batch-test-2026-08-12.png
+++ b/ai_knowledge_base_v2/development_records/ui-evidence/agent-batch-test-2026-08-12.png
--- a/ai_knowledge_base_v2/docker-compose.dev.yml
+++ b/ai_knowledge_base_v2/docker-compose.dev.yml
@@ -57,6 +57,9 @@ services:
      TOPIC_SETTLEMENT_POLL_SECONDS: "2"
      TOPIC_SETTLEMENT_STALE_MINUTES: "30"
      TOPIC_SETTLEMENT_MAX_ATTEMPTS: "3"
+      AGENT_BATCH_WORKER_ENABLED: "true"
+      AGENT_BATCH_POLL_SECONDS: "2"
+      AGENT_BATCH_STALE_MINUTES: "30"
      JWT_SECRET_KEY: local-dev-secret-change-before-production
      MOCK_SMS_ENABLED: "true"
      MOCK_SMS_CODE: "123456"
--- a/ai_knowledge_base_v2/docker-compose.prod.yml
+++ b/ai_knowledge_base_v2/docker-compose.prod.yml
@@ -56,6 +56,9 @@ services:
      TOPIC_SETTLEMENT_POLL_SECONDS: ${TOPIC_SETTLEMENT_POLL_SECONDS:-2}
      TOPIC_SETTLEMENT_STALE_MINUTES: ${TOPIC_SETTLEMENT_STALE_MINUTES:-30}
      TOPIC_SETTLEMENT_MAX_ATTEMPTS: ${TOPIC_SETTLEMENT_MAX_ATTEMPTS:-3}
+      AGENT_BATCH_WORKER_ENABLED: ${AGENT_BATCH_WORKER_ENABLED:-true}
+      AGENT_BATCH_POLL_SECONDS: ${AGENT_BATCH_POLL_SECONDS:-2}
+      AGENT_BATCH_STALE_MINUTES: ${AGENT_BATCH_STALE_MINUTES:-30}
      JWT_SECRET_KEY: ${JWT_SECRET_KEY:?必须配置 JWT_SECRET_KEY}
      CONFIG_ENCRYPTION_KEY: ${CONFIG_ENCRYPTION_KEY:?必须配置 CONFIG_ENCRYPTION_KEY}
      MOCK_SMS_ENABLED: "false"