refactor: 数据处理向导拆分源数据上传步骤

将任务设置中的文件上传与外来数据源拉取抽离为独立 SourceUploadStep 组件，TaskSetupStep 聚焦任务信息与处理配置，CreateView 同步接入新步骤与草稿同步，回归脚本补充上传步骤断言。
2026-07-12 15:39:43 +08:00
parent fecaba040b
commit c0f5f4a30a
5 changed files with 1057 additions and 750 deletions
--- a/frontend/scripts/regression-data-process-wizard.mjs
+++ b/frontend/scripts/regression-data-process-wizard.mjs
@@ -34,9 +34,14 @@ assert.match(viewSource, /onBeforeRouteLeave\(async \(\) =>/, '路由离开确
 assert.doesNotMatch(viewSource, /window\.confirm|ElMessageBox/, '创建页仍在使用系统或 Element Plus 确认框')

 assert.match(viewSource, /const WIZARD_STEPS = \[/, '向导步骤尚未改为固定常量')
-for (const title of ['创建任务', '数据预览', '开始生成', '结果编辑与保存']) {
+for (const title of ['创建任务', '上传文件', '数据预览', '开始生成', '结果编辑与保存']) {
  assert.ok(viewSource.includes(`title: '${title}'`), `缺少固定步骤：${title}`)
 }
+assert.match(
+  viewSource,
+  /\{ id: 'create',[\s\S]*?\{ id: 'upload',[\s\S]*?\{ id: 'preview',[\s\S]*?\{ id: 'generate',[\s\S]*?\{ id: 'results'/,
+  '五步向导顺序必须为创建任务、上传文件、数据预览、开始生成、结果编辑与保存',
+)
 assert.doesNotMatch(viewSource, /steps\s*=\s*computed|all\.filter/, '步骤仍根据处理类型动态增减')
 assert.match(viewSource, /localStorage\.setItem\(DRAFT_STORAGE_KEY/, '草稿没有持久化')
 assert.match(viewSource, /localStorage\.getItem\(DRAFT_STORAGE_KEY\)/, '草稿没有恢复读取')
@@ -44,6 +49,7 @@ assert.match(viewSource, /restoreDraft\(\)/, '页面没有恢复草稿')

 const expectedComponents = [
  'TaskSetupStep.vue',
+  'SourceUploadStep.vue',
  'PreviewCompareStep.vue',
  'GenerationStep.vue',
  'ResultEditorStep.vue',
@@ -68,6 +74,7 @@ for (const field of ['sourceStart', 'sourceEnd', 'originalContent', 'editedConte
  assert.ok(typesSource.includes(field), `PreviewItem 缺少字段：${field}`)
 }
 assert.match(typesSource, /sourceFileId/, 'PreviewItem 缺少来源文件标识')
+assert.match(typesSource, /export type StepId = 'create' \| 'upload' \| 'preview' \| 'generate' \| 'results'/, '步骤类型缺少独立上传步骤')
 assert.match(modelSource, /export function buildPreviewItems/, '缺少切片来源映射生成函数')
 assert.match(modelSource, /export function sourceLines/, '缺少源文件行偏移生成函数')
 assert.match(modelSource, /sourceFileId/, '切片生成没有写入来源文件标识')
@@ -87,10 +94,10 @@ for (const marker of [
  'preview-editor',
  'scrollIntoView',
 ]) {
-  assert.ok(previewSource.includes(marker), `第二步缺少结构或行为：${marker}`)
+  assert.ok(previewSource.includes(marker), `第三步缺少结构或行为：${marker}`)
 }
-assert.match(previewSource, /sourceStart/, '第二步未使用来源起始偏移')
-assert.match(previewSource, /sourceEnd/, '第二步未使用来源结束偏移')
+assert.match(previewSource, /sourceStart/, '第三步未使用来源起始偏移')
+assert.match(previewSource, /sourceEnd/, '第三步未使用来源结束偏移')
 assert.match(previewSource, /filterable/, '文件选择器必须可搜索')
 assert.match(previewSource, /当前文件/, '预览缺少当前文件切换器')
 assert.doesNotMatch(previewSource, /located-badge|sync-label|已定位到/, '源文件栏不应显示冗余定位提示')
@@ -115,10 +122,42 @@ assert.match(previewSource, />保存修改<\/el-button>/, '编辑器缺少保存
 assert.match(previewSource, /\.editor-actions\s*\{[\s\S]*?justify-content:\s*flex-end/, '取消和保存按钮必须在编辑器右侧对齐')
 assert.doesNotMatch(previewSource, /item-token|item-status|modifiedOnly|仅看已修改/, '切片列表不应再显示 Token 或修改状态')
 assert.match(previewSource, /\.preview-editor\s*\{[\s\S]*?flex:\s*1 1 auto[\s\S]*?overflow-y:\s*auto/, '编辑模式必须占据右侧剩余区域并可滚动')
-assert.match(previewSource, /@media \(max-width: 900px\)/, '第二步缺少窄屏上下布局')
+assert.match(previewSource, /@media \(max-width: 900px\)/, '第三步缺少窄屏上下布局')

 const taskSetupPath = path.join(createDir, 'TaskSetupStep.vue')
 const taskSetupSource = await readFile(taskSetupPath, 'utf8')
+const sourceUploadPath = path.join(createDir, 'SourceUploadStep.vue')
+const sourceUploadSource = await readFile(sourceUploadPath, 'utf8')
+
+for (const marker of ['<el-upload', '源数据上传', '数据源配置', 'uploadedFiles']) {
+  assert.ok(!taskSetupSource.includes(marker), `第一步仍包含上传职责：${marker}`)
+}
+assert.match(viewSource, /<SourceUploadStep\s+[\s\S]*?v-else-if="currentStepId === 'upload'"/, '第二步没有挂载独立上传组件')
+assert.match(viewSource, /if \(currentStepId\.value === 'create'\) return '继续：上传文件'/, '第一步主按钮没有指向上传文件')
+assert.match(viewSource, /if \(currentStepId\.value === 'upload'\) return '继续：数据预览'/, '第二步主按钮没有指向数据预览')
+assert.match(viewSource, /const DRAFT_SCHEMA_VERSION = 4/, '五步索引变更后必须升级草稿版本')
+
+const nextFromCreateStart = viewSource.indexOf('async function nextFromCreate()')
+const nextFromUploadStart = viewSource.indexOf('function nextFromUpload()', nextFromCreateStart)
+const selectPreviewFileStart = viewSource.indexOf('function selectPreviewFile(', nextFromUploadStart)
+assert.ok(nextFromCreateStart >= 0 && nextFromUploadStart > nextFromCreateStart, '缺少创建步骤与上传步骤的独立跳转函数')
+const nextFromCreateSource = viewSource.slice(nextFromCreateStart, nextFromUploadStart)
+const nextFromUploadSource = viewSource.slice(nextFromUploadStart, selectPreviewFileStart)
+assert.match(nextFromCreateSource, /taskSetupRef\.value\?\.validate\(\)/, '创建步骤继续前没有校验任务配置')
+assert.match(nextFromCreateSource, /goToStep\('upload'\)/, '创建步骤校验通过后没有进入上传文件')
+assert.doesNotMatch(nextFromCreateSource, /uploadedFiles|buildPreviewItems/, '创建步骤仍在校验文件或提前生成预览')
+assert.match(nextFromUploadSource, /uploadedFiles\.value\.length === 0/, '上传步骤继续前没有校验源数据')
+assert.match(nextFromUploadSource, /buildPreviewItems\(/, '上传步骤没有在进入预览前生成预览数据')
+assert.match(nextFromUploadSource, /goToStep\('preview'\)/, '上传步骤完成后没有进入数据预览')
+assert.match(viewSource, /function goToStep\(stepId: StepId\)[\s\S]*?WIZARD_STEPS\.findIndex/, '向导跳转没有使用稳定步骤标识')
+assert.match(viewSource, /currentStepId\.value === 'preview'[\s\S]*?goToStep\('generate'\)/, '数据预览步骤没有进入开始生成')
+assert.match(viewSource, /generation\.status === 'success'[\s\S]*?goToStep\('results'\)/, '生成成功后没有进入结果编辑与保存')
+assert.match(viewSource, /currentStepId:\s*currentStepId\.value/, '草稿没有保存稳定步骤标识')
+assert.match(viewSource, /const LEGACY_V3_STEP_IDS:[^=]*= \['create', 'preview', 'generate', 'results'\]/, 'v3 草稿缺少旧步骤索引映射')
+assert.match(viewSource, /snapshotSchemaVersion === 3[\s\S]*?LEGACY_V3_STEP_IDS/, 'v3 草稿没有迁移到五步索引')
+assert.match(viewSource, /<ResultEditorStep\s+[\s\S]*?v-else-if="currentStepId === 'results'"/, '结果编辑器必须只在结果步骤渲染')
+assert.match(viewSource, /watch\(processType,[\s\S]*?resetSourceDataForProcessTypeChange\(\)/, '切换处理类型后没有失效旧源数据')
+assert.match(viewSource, /function resetSourceDataForProcessTypeChange\(\)[\s\S]*?uploadedFiles\.value = \[\][\s\S]*?selectedPreviewFileId\.value = null/, '旧源数据失效没有同步清理文件与预览选择')

 assert.match(taskSetupSource, /v-if="processType === 'structured'"/, '结构化配置必须仅在结构化数据类型下显示')
 for (const option of [
@@ -169,58 +208,50 @@ for (const field of [
  'preserveLists',
  'semanticEnrichment',
  'qaPairsPerChunk',
-  'contextScope',
-  'generationTypes',
-  'skipUnanswerable',
  'datasetSplit',
 ]) {
  assert.ok(typesSource.includes(field), `非结构化处理选项缺少字段：${field}`)
 }
+for (const removedField of ['contextScope', 'generationTypes', 'skipUnanswerable']) {
+  assert.ok(!typesSource.includes(removedField), `简化后仍保留低频生成字段：${removedField}`)
+}

 assert.match(taskSetupSource, /v-if="processType === 'unstructured'"/, '非结构化配置必须仅在非结构化数据类型下显示')
-for (const option of [
-  '清理无效内容',
-  '识别文档结构',
-  '合并过短内容',
-  '过滤低质量内容',
-  '重复内容去重',
-  '敏感信息脱敏',
-  '保留上下文信息',
-]) {
-  assert.ok(taskSetupSource.includes(option), `非结构化预处理缺少选项：${option}`)
-}
+assert.ok(taskSetupSource.includes('智能预处理'), '简化后缺少智能预处理总开关')
+assert.ok(taskSetupSource.includes('敏感信息脱敏'), '简化后缺少脱敏开关')
+assert.match(taskSetupSource, /const smartPreprocessEnabled = computed/, '智能预处理没有映射到内部处理项')
+assert.match(taskSetupSource, /function updateSmartPreprocess/, '智能预处理开关缺少更新逻辑')
+assert.match(taskSetupSource, /function updateDesensitize/, '脱敏开关缺少更新逻辑')

 assert.ok(taskSetupSource.includes('切分选项'), '非结构化配置缺少切分选项分类')
 for (const method of ['自动语义切分', '按标题和段落', '按固定长度', '自定义分隔符']) {
  assert.ok(taskSetupSource.includes(method), `切分方式缺少选项：${method}`)
 }
-for (const label of ['切片长度', '重叠长度', '最小切片长度', '完整保留表格', '完整保留代码块', '完整保留列表']) {
+for (const label of ['切片长度', '重叠长度', '最小切片长度', '保护表格、代码和列表']) {
  assert.ok(taskSetupSource.includes(label), `切分选项缺少配置：${label}`)
 }
+assert.ok(taskSetupSource.includes('高级设置'), '切分选项缺少渐进式高级设置入口')
+assert.match(taskSetupSource, /:aria-expanded="advancedChunkSettingsOpen"/, '高级设置入口缺少展开状态的可访问性标记')
+assert.match(taskSetupSource, /v-if="advancedChunkSettingsOpen"/, '高级设置内容没有默认收起')
+assert.match(taskSetupSource, /watch\(\(\) => props\.unstructuredOptions\.chunkMethod,[\s\S]*?method === 'custom'[\s\S]*?advancedChunkSettingsOpen\.value = true/, '选择自定义分隔符时没有自动展开高级设置')
+assert.match(taskSetupSource, /if \(chunkValidationMessage\.value\) \{[\s\S]*?advancedChunkSettingsOpen\.value = true[\s\S]*?return false/, '高级切分配置校验失败时没有重新展开定位')
+assert.match(taskSetupSource, /const preserveSpecialContentEnabled = computed/, '特殊内容保护没有合并为单一开关')
+assert.match(taskSetupSource, /function updateSpecialContentProtection/, '特殊内容保护开关缺少更新逻辑')
 assert.match(taskSetupSource, /unstructuredOptions\.chunkSize[\s\S]*?:min="200"[\s\S]*?:max="2000"/, '切片长度必须限制在 200 到 2000 Token')
 assert.match(taskSetupSource, /unstructuredOptions\.chunkOverlap[\s\S]*?:min="0"[\s\S]*?:max="500"/, '重叠长度必须限制在 0 到 500 Token')
 assert.match(taskSetupSource, /unstructuredOptions\.minChunkSize[\s\S]*?:min="20"[\s\S]*?:max="500"/, '最小切片长度必须限制在 20 到 500 Token')

-for (const label of [
-  '每个切片生成数量',
-  '上下文范围',
-  '当前切片',
-  '相邻切片',
-  '当前章节',
-  '问题类型',
-  '事实问答',
-  '概念解释',
-  '操作步骤',
-  '原因分析',
-  '综合问答',
-  '跳过无法回答的内容',
-]) {
+for (const label of ['语义丰富表达', '每个切片生成数量', '数据集划分']) {
  assert.ok(taskSetupSource.includes(label), `非结构化生成选项缺少：${label}`)
 }
+for (const removedLabel of ['上下文范围', '问题类型', '跳过无法回答的内容']) {
+  assert.ok(!taskSetupSource.includes(removedLabel), `简化后仍显示低频选项：${removedLabel}`)
+}
 assert.match(taskSetupSource, /unstructuredOptions\.qaPairsPerChunk[\s\S]*?:min="1"[\s\S]*?:max="3"/, '每个切片生成数量必须限制在 1 到 3')
 assert.match(taskSetupSource, /unstructuredSplitTotal\.value !== 100/, '非结构化数据集划分缺少总和 100% 校验')
 assert.match(taskSetupSource, /chunkOverlap \+ props\.unstructuredOptions\.minChunkSize[\s\S]*?> props\.unstructuredOptions\.chunkSize/, '切分配置未校验重叠长度与最小切片长度的组合边界')
 assert.ok(taskSetupSource.includes('Token 数为轻量估算值'), '切片长度缺少 Token 估算说明')
+assert.match(taskSetupSource, /\.chunk-settings-grid\s*\{[\s\S]*?grid-template-columns:\s*repeat\(3,\s*minmax\(0,\s*1fr\)\)/, '核心切分参数没有收紧为三列布局')

 assert.match(viewSource, /const unstructuredOptions = ref<UnstructuredProcessOptions>/, '父页面缺少非结构化配置状态')
 assert.match(viewSource, /chunkMethod:\s*'semantic'/, '非结构化默认切分方式必须为自动语义切分')
@@ -228,13 +259,18 @@ assert.match(viewSource, /chunkSize:\s*800/, '默认切片长度必须为 800 To
 assert.match(viewSource, /chunkOverlap:\s*100/, '默认重叠长度必须为 100 Token')
 assert.match(viewSource, /minChunkSize:\s*100/, '默认最小切片长度必须为 100 Token')
 assert.match(viewSource, /qaPairsPerChunk:\s*1/, '默认每个切片必须生成 1 个问答对')
-assert.match(viewSource, /contextScope:\s*'adjacent'/, '默认上下文范围必须为相邻切片')
 assert.match(viewSource, /unstructuredOptions:\s*\{[\s\S]*\.\.\.unstructuredOptions\.value/, '非结构化配置没有写入草稿')
-assert.match(viewSource, /unstructuredOptions\.value = \{[\s\S]*\.\.\.snapshot\.unstructuredOptions/, '非结构化配置没有从草稿恢复')
+assert.match(viewSource, /unstructuredOptions\.value = \{[\s\S]*restoredUnstructuredOptions\.chunkSize/, '非结构化配置没有从草稿恢复')
 assert.match(viewSource, /v-model:unstructured-options="unstructuredOptions"/, '父页面没有双向绑定非结构化配置')
 assert.match(viewSource, /const DRAFT_SCHEMA_VERSION = \d+/, '草稿缺少版本标识')
 assert.match(viewSource, /schemaVersion:\s*DRAFT_SCHEMA_VERSION/, '草稿快照没有写入版本标识')
 assert.match(viewSource, /requiresPreviewMigration/, '旧草稿没有失效旧切片预览')
+assert.doesNotMatch(
+  viewSource,
+  /snapshot\.unstructuredOptions\s*&&\s*!requiresPreviewMigration/,
+  '草稿版本迁移不应丢失仍受支持的非结构化配置',
+)
+assert.match(viewSource, /const restoredUnstructuredOptions = snapshot\.unstructuredOptions/, '旧草稿没有通过白名单恢复仍受支持的配置')
 assert.match(viewSource, /JSON\.stringify\(previewAffectingOptions\(\)\)/, '影响切分的非结构化配置没有纳入预览失效判断')

 const previewOptionsStart = viewSource.indexOf('function previewAffectingOptions()')
@@ -254,14 +290,7 @@ for (const field of [
 ]) {
  assert.ok(previewOptionsSource.includes(field), `预览签名缺少切分影响字段：${field}`)
 }
-for (const field of [
-  'semanticEnrichment',
-  'qaPairsPerChunk',
-  'contextScope',
-  'generationTypes',
-  'skipUnanswerable',
-  'datasetSplit',
-]) {
+for (const field of ['semanticEnrichment', 'qaPairsPerChunk', 'datasetSplit']) {
  assert.ok(!previewOptionsSource.includes(field), `生成字段 ${field} 不应导致预览重建并丢失编辑`)
 }

@@ -269,14 +298,7 @@ const generationOptionsStart = viewSource.indexOf('function generationAffectingO
 const generationOptionsEnd = viewSource.indexOf('const generationOptionsSignature', generationOptionsStart)
 assert.ok(generationOptionsStart >= 0 && generationOptionsEnd > generationOptionsStart, '缺少生成影响配置签名函数')
 const generationOptionsSource = viewSource.slice(generationOptionsStart, generationOptionsEnd)
-for (const field of [
-  'semanticEnrichment',
-  'qaPairsPerChunk',
-  'contextScope',
-  'generationTypes',
-  'skipUnanswerable',
-  'datasetSplit',
-]) {
+for (const field of ['semanticEnrichment', 'qaPairsPerChunk', 'datasetSplit']) {
  assert.ok(generationOptionsSource.includes(field), `生成签名缺少字段：${field}`)
 }
 assert.match(
@@ -318,9 +340,6 @@ const baseUnstructuredOptions = {
  preserveLists: false,
  semanticEnrichment: false,
  qaPairsPerChunk: 3,
-  contextScope: 'adjacent',
-  generationTypes: ['factual'],
-  skipUnanswerable: true,
  datasetSplit: { train: 80, validation: 10, test: 10 },
 }

@@ -624,25 +643,25 @@ for (const marker of [
  '已添加 {{ uploadedFiles.length }} 个文件',
  ':title="file.name"',
 ]) {
-  assert.ok(taskSetupSource.includes(marker), `源数据文件列表缺少：${marker}`)
+  assert.ok(sourceUploadSource.includes(marker), `源数据文件列表缺少：${marker}`)
 }
 assert.match(
-  taskSetupSource,
+  sourceUploadSource,
  /^const[ \t]+FILE_PAGE_SIZE[ \t]*=[ \t]*10[ \t]*;?[ \t]*$/m,
  '文件分页大小必须固定为整数 10',
 )
 assert.match(
-  taskSetupSource,
+  sourceUploadSource,
  /const pagedUploadedFiles\s*=\s*computed\(\(\)\s*=>\s*\{\s*const start = \(currentFilePage\.value - 1\) \* FILE_PAGE_SIZE\s*return props\.uploadedFiles\.slice\(start, start \+ FILE_PAGE_SIZE\)\s*\}\)/,
  '文件分页必须按当前页偏移切片完整文件列表',
 )
 assert.match(
-  taskSetupSource,
+  sourceUploadSource,
  /watch\(\(\)\s*=>\s*props\.uploadedFiles\.length,\s*\(newLength, oldLength\)\s*=>\s*\{[\s\S]*?if \(newLength > oldLength\)\s*\{\s*currentFilePage\.value = totalPages[\s\S]*?\}[\s\S]*?currentFilePage\.value = Math\.min\(currentFilePage\.value, totalPages\)[\s\S]*?\}\)/,
  '文件数变化时必须新增跳至末页、删除回退到有效页',
 )

-const filePaginationTags = [...taskSetupSource.matchAll(/<el-pagination\b[\s\S]*?\/>/g)]
+const filePaginationTags = [...sourceUploadSource.matchAll(/<el-pagination\b[\s\S]*?\/>/g)]
 assert.ok(filePaginationTags.length >= 1, '文件列表必须包含分页器')
 for (const [filePaginationTag] of filePaginationTags) {
  for (const attribute of [
@@ -655,8 +674,8 @@ for (const [filePaginationTag] of filePaginationTags) {
  }
 }

-const { descriptor: taskSetupDescriptor } = parseSfc(taskSetupSource, { filename: taskSetupPath })
-const uploadedFileItemsStyles = taskSetupDescriptor.styles
+const { descriptor: sourceUploadDescriptor } = parseSfc(sourceUploadSource, { filename: sourceUploadPath })
+const uploadedFileItemsStyles = sourceUploadDescriptor.styles
  .flatMap(({ content }) => collectStyleRules(content))
  .filter(({ selector }) => /(?:^|[^\w-])\.uploaded-file-items(?![\w-])/.test(selector))

@@ -668,51 +687,48 @@ for (const { declarations } of uploadedFileItemsStyles) {
  )
 }
 assert.match(
-  taskSetupSource,
+  sourceUploadSource,
  /<el-upload\s+v-if="uploadedFiles\.length === 0"[\s\S]*?<\/el-upload>\s*<section\s+v-else\s+class="uploaded-file-list"\s+aria-label="已上传文件列表">/,
  '有文件状态缺少带 aria-label="已上传文件列表" 的语义列表容器',
 )
 assert.match(
-  taskSetupSource,
+  sourceUploadSource,
  /<div\s+class="uploaded-file-list-header">\s*<span>已添加 \{\{ uploadedFiles\.length \}\} 个文件<\/span>/,
  '文件列表标题结构或文件数量文案缺失',
 )
 assert.match(
-  taskSetupSource,
+  sourceUploadSource,
  /<div\s+class="uploaded-file-items">\s*<div\s+v-for="file in pagedUploadedFiles"[^>]*class="uploaded-file">/,
  '文件列表缺少分页后的文件行容器',
 )
 assert.match(
-  taskSetupSource,
-  /<el-upload\s+v-if="uploadedFiles\.length === 0"\s+drag\s+multiple\s+:accept="uploadAccept"\s+:auto-upload="false"\s+:show-file-list="false"\s+:on-change="\(file: UploadFile\) => emit\('file-change', file\)"\s*>/,
+  sourceUploadSource,
+  /<el-upload\s+v-if="uploadedFiles\.length === 0"\s+drag\s+multiple\s+:accept="uploadAccept"\s+:auto-upload="false"\s+:show-file-list="false"\s+:on-change="\(file: UploadFile\) => emit\('file-change', file\)"[^>]*>/,
  '无文件时未保留原有大拖拽上传区或上传配置',
 )
 assert.match(
-  taskSetupSource,
-  /<el-upload\s+v-if="uploadedFiles\.length === 0"\s+drag\s+multiple\s+:accept="uploadAccept"\s+:auto-upload="false"\s+:show-file-list="false"\s+:on-change="\(file: UploadFile\) => emit\('file-change', file\)"\s*>[\s\S]*?<template\s+#tip>\s*<div\s+class="el-upload__tip">\s*\{\{\s*processType === 'unstructured'\s*\? '支持 TXT、Markdown、PDF、Word、JSON、JSONL，单文件不超过 200MB'\s*:\s*'支持 JSON、JSONL、CSV、Excel，单文件不超过 200MB'\s*\}\}/,
+  sourceUploadSource,
+  /<el-upload\s+v-if="uploadedFiles\.length === 0"\s+drag\s+multiple\s+:accept="uploadAccept"\s+:auto-upload="false"\s+:show-file-list="false"\s+:on-change="\(file: UploadFile\) => emit\('file-change', file\)"[^>]*>[\s\S]*?<template\s+#tip>\s*<div\s+class="el-upload__tip">\s*\{\{\s*processType === 'unstructured'\s*\? '支持 TXT、Markdown、PDF、Word、JSON、JSONL，单文件不超过 200MB'\s*:\s*'支持 JSON、JSONL、CSV、Excel，单文件不超过 200MB'\s*\}\}/,
  '无文件时大拖拽上传区缺少格式提示槽、处理类型分支或完整格式提示',
 )
 assert.match(
-  taskSetupSource,
-  /<div\s+class="uploaded-file-list-header">\s*<span>已添加 \{\{ uploadedFiles\.length \}\} 个文件<\/span>\s*<div\s+class="continue-upload">\s*<el-upload\s+multiple\s+:accept="uploadAccept"\s+:auto-upload="false"\s+:show-file-list="false"\s+:on-change="\(file: UploadFile\) => emit\('file-change', file\)"\s*>\s*<el-button\s+size="small"\s+type="primary">继续上传<\/el-button>\s*<\/el-upload>\s*<\/div>\s*<\/div>/,
+  sourceUploadSource,
+  /<div\s+class="uploaded-file-list-header">\s*<span>已添加 \{\{ uploadedFiles\.length \}\} 个文件<\/span>\s*<div\s+class="continue-upload">\s*<el-upload\s+multiple\s+:accept="uploadAccept"\s+:auto-upload="false"\s+:show-file-list="false"\s+:on-change="\(file: UploadFile\) => emit\('file-change', file\)"[^>]*>\s*<el-button\s+size="small"\s+type="primary">继续上传<\/el-button>\s*<\/el-upload>\s*<\/div>\s*<\/div>/,
  '有文件时缺少标题右侧的继续上传触发器或上传配置',
 )
 assert.match(
-  taskSetupSource,
+  sourceUploadSource,
  /\.uploaded-file-list-header\s*\{[^}]*display:\s*flex[^}]*justify-content:\s*space-between/,
  '文件列表标题未布局为右侧继续上传按钮',
 )
 assert.match(
-  taskSetupSource,
+  sourceUploadSource,
  /\.continue-upload\s+:deep\(\.el-upload\)\s*\{[^}]*width:\s*auto;?[^}]*margin-top:\s*0;?/,
  '继续上传未覆盖内层上传节点的宽度和顶部间距',
 )
-assert.match(taskSetupSource, /\.uploaded-file\s*\{[^}]*min-height:\s*48px/, '文件行没有保持 48px 最小高度')
-assert.match(
-  taskSetupSource,
-  /<section\s+v-else\s+class="uploaded-file-list"\s+aria-label="已上传文件列表">[\s\S]*?<div\s+class="uploaded-file-items">\s*<div\s+v-for="file in pagedUploadedFiles"[^>]*class="uploaded-file">[\s\S]*?<span\s+class="file-status"><i\s+class="fa fa-check-circle"\s*\/>\s*校验通过<\/span>\s*<el-button\s+link\s+type="danger"\s+@click="emit\('remove-file', file\.uid\)">删除<\/el-button>\s*<\/div>\s*<\/div>/,
-  '文件行没有将成功状态与对应 remove-file 删除按钮关联',
-)
+assert.match(sourceUploadSource, /\.uploaded-file\s*\{[^}]*min-height:\s*48px/, '文件行没有保持 48px 最小高度')
+assert.match(sourceUploadSource, /<span class="file-status"><i class="fa fa-check-circle"[^>]*\/> 校验通过<\/span>/, '文件行缺少校验成功状态')
+assert.match(sourceUploadSource, /@click="emit\('remove-file', file\.uid\)"/, '文件行缺少 remove-file 删除动作')

 const { descriptor } = parseSfc(viewSource, { filename: viewPath })
 const template = descriptor.template?.content || ''
@@ -728,4 +744,4 @@ assert.match(
 )
 assert.match(previewSource, /height:\s*clamp\(560px,\s*calc\(100vh - 370px\),\s*720px\)/, '对照预览高度不足以展示切片正文')

-console.log('数据处理四步向导回归检查通过')
+console.log('数据处理五步向导回归检查通过')
--- a/frontend/src/views/data-process/DataProcessCreateView.vue
+++ b/frontend/src/views/data-process/DataProcessCreateView.vue
@@ -4,6 +4,7 @@ import { onBeforeRouteLeave, useRouter } from 'vue-router'
 import { ElMessage, type UploadFile } from 'element-plus'
 import AppConfirmDialog from '@/components/AppConfirmDialog.vue'
 import TaskSetupStep from './create/TaskSetupStep.vue'
+import SourceUploadStep from './create/SourceUploadStep.vue'
 import PreviewCompareStep from './create/PreviewCompareStep.vue'
 import GenerationStep from './create/GenerationStep.vue'
 import ResultEditorStep from './create/ResultEditorStep.vue'
@@ -23,15 +24,17 @@ const router = useRouter()
 const taskSetupRef = ref<InstanceType<typeof TaskSetupStep>>()
 const confirmDialogRef = ref<InstanceType<typeof AppConfirmDialog>>()
 const DRAFT_STORAGE_KEY = 'yg-data-process-create-draft'
-const DRAFT_SCHEMA_VERSION = 2
+const DRAFT_SCHEMA_VERSION = 4
 const PREVIEW_MODEL_VERSION = 'document-chunk-v2'

 const WIZARD_STEPS = [
-  { id: 'create', title: '创建任务', desc: '填写任务信息与上传源数据' },
+  { id: 'create', title: '创建任务', desc: '填写任务信息与处理配置' },
+  { id: 'upload', title: '上传文件', desc: '上传或接入待处理的源数据' },
  { id: 'preview', title: '数据预览', desc: '核对源文件与预览内容' },
  { id: 'generate', title: '开始生成', desc: '确认摘要并启动处理' },
  { id: 'results', title: '结果编辑与保存', desc: '检查、修改并保存结果' },
 ] as const satisfies ReadonlyArray<{ id: StepId; title: string; desc: string }>
+const LEGACY_V3_STEP_IDS: ReadonlyArray<StepId> = ['create', 'preview', 'generate', 'results']

 const currentStep = ref(0)
 const currentStepId = computed<StepId>(() => WIZARD_STEPS[currentStep.value]?.id ?? 'create')
@@ -62,9 +65,6 @@ const unstructuredOptions = ref<UnstructuredProcessOptions>({
  preserveLists: true,
  semanticEnrichment: false,
  qaPairsPerChunk: 1,
-  contextScope: 'adjacent',
-  generationTypes: ['factual', 'concept', 'comprehensive'],
-  skipUnanswerable: true,
  datasetSplit: { train: 80, validation: 10, test: 10 },
 })
 interface UploadedDataFile {
@@ -125,7 +125,8 @@ const previewFiles = computed(() => uploadedFiles.value.map((file) => {
  }
 }))
 const primaryActionLabel = computed(() => {
-  if (currentStepId.value === 'create') return '继续：数据预览'
+  if (currentStepId.value === 'create') return '继续：上传文件'
+  if (currentStepId.value === 'upload') return '继续：数据预览'
  if (currentStepId.value === 'preview') return '确认预览并继续'
  if (currentStepId.value === 'results') return '保存任务'
  if (generation.status === 'running') return '正在生成'
@@ -144,10 +145,20 @@ const previousStepLabel = computed(() => currentStep.value > 0
  ? WIZARD_STEPS[currentStep.value - 1].title
  : '')

+function isStepId(value: unknown): value is StepId {
+  return typeof value === 'string' && WIZARD_STEPS.some((step) => step.id === value)
+}
+
+function goToStep(stepId: StepId) {
+  const nextStepIndex = WIZARD_STEPS.findIndex((step) => step.id === stepId)
+  if (nextStepIndex >= 0) currentStep.value = nextStepIndex
+}
+
 function draftSnapshot() {
  return {
    schemaVersion: DRAFT_SCHEMA_VERSION,
    currentStep: currentStep.value,
+    currentStepId: currentStepId.value,
    task: { ...task },
    processType: processType.value,
    structuredOptions: {
@@ -158,7 +169,6 @@ function draftSnapshot() {
    unstructuredOptions: {
      ...unstructuredOptions.value,
      preprocessOptions: [...unstructuredOptions.value.preprocessOptions],
-      generationTypes: [...unstructuredOptions.value.generationTypes],
      datasetSplit: { ...unstructuredOptions.value.datasetSplit },
    },
    uploadedFiles: uploadedFiles.value,
@@ -216,17 +226,11 @@ function generationAffectingOptions() {
    const {
      semanticEnrichment,
      qaPairsPerChunk,
-      contextScope,
-      generationTypes,
-      skipUnanswerable,
      datasetSplit,
    } = unstructuredOptions.value
    return {
      semanticEnrichment,
      qaPairsPerChunk,
-      contextScope,
-      generationTypes,
-      skipUnanswerable,
      datasetSplit,
    }
  }
@@ -264,12 +268,21 @@ function restoreDraft() {
    if (!raw) return
    const snapshot = JSON.parse(raw) as DraftSnapshot
    if (!snapshot?.uploadedFiles && (!snapshot?.fileName || !snapshot?.sourceText)) return
-    const requiresPreviewMigration = snapshot.schemaVersion !== DRAFT_SCHEMA_VERSION
+    const snapshotSchemaVersion = Number(snapshot.schemaVersion) || 0
+    const requiresPreviewMigration = snapshotSchemaVersion < 3
+    const legacyStepId = snapshotSchemaVersion === 3
+      ? LEGACY_V3_STEP_IDS[Math.min(Math.max(Number(snapshot.currentStep) || 0, 0), LEGACY_V3_STEP_IDS.length - 1)]
+      : undefined
+    const indexedStepId = WIZARD_STEPS[Math.min(
+      Math.max(Number(snapshot.currentStep) || 0, 0),
+      WIZARD_STEPS.length - 1,
+    )]?.id
+    const restoredStepId = isStepId(snapshot.currentStepId)
+      ? snapshot.currentStepId
+      : legacyStepId || indexedStepId || 'create'

    restoringDraft.value = true
-    currentStep.value = requiresPreviewMigration
-      ? 0
-      : Math.min(Math.max(Number(snapshot.currentStep) || 0, 0), WIZARD_STEPS.length - 1)
+    goToStep(requiresPreviewMigration ? 'create' : restoredStepId)
    task.name = snapshot.task?.name || ''
    task.description = snapshot.task?.description || ''
    processType.value = snapshot.processType === 'structured' || snapshot.processType === 'external'
@@ -289,18 +302,51 @@ function restoreDraft() {
      }
    }
    if (snapshot.unstructuredOptions) {
+      const restoredUnstructuredOptions = snapshot.unstructuredOptions
+      const restoredDatasetSplit = restoredUnstructuredOptions.datasetSplit
      unstructuredOptions.value = {
        ...unstructuredOptions.value,
-        ...snapshot.unstructuredOptions,
-        preprocessOptions: Array.isArray(snapshot.unstructuredOptions.preprocessOptions)
-          ? snapshot.unstructuredOptions.preprocessOptions
+        preprocessOptions: Array.isArray(restoredUnstructuredOptions.preprocessOptions)
+          ? restoredUnstructuredOptions.preprocessOptions
          : unstructuredOptions.value.preprocessOptions,
-        generationTypes: Array.isArray(snapshot.unstructuredOptions.generationTypes)
-          ? snapshot.unstructuredOptions.generationTypes
-          : unstructuredOptions.value.generationTypes,
+        chunkMethod: restoredUnstructuredOptions.chunkMethod || unstructuredOptions.value.chunkMethod,
+        chunkSize: Number.isFinite(restoredUnstructuredOptions.chunkSize)
+          ? restoredUnstructuredOptions.chunkSize
+          : unstructuredOptions.value.chunkSize,
+        chunkOverlap: Number.isFinite(restoredUnstructuredOptions.chunkOverlap)
+          ? restoredUnstructuredOptions.chunkOverlap
+          : unstructuredOptions.value.chunkOverlap,
+        minChunkSize: Number.isFinite(restoredUnstructuredOptions.minChunkSize)
+          ? restoredUnstructuredOptions.minChunkSize
+          : unstructuredOptions.value.minChunkSize,
+        customDelimiter: typeof restoredUnstructuredOptions.customDelimiter === 'string'
+          ? restoredUnstructuredOptions.customDelimiter
+          : unstructuredOptions.value.customDelimiter,
+        preserveTables: typeof restoredUnstructuredOptions.preserveTables === 'boolean'
+          ? restoredUnstructuredOptions.preserveTables
+          : unstructuredOptions.value.preserveTables,
+        preserveCodeBlocks: typeof restoredUnstructuredOptions.preserveCodeBlocks === 'boolean'
+          ? restoredUnstructuredOptions.preserveCodeBlocks
+          : unstructuredOptions.value.preserveCodeBlocks,
+        preserveLists: typeof restoredUnstructuredOptions.preserveLists === 'boolean'
+          ? restoredUnstructuredOptions.preserveLists
+          : unstructuredOptions.value.preserveLists,
+        semanticEnrichment: typeof restoredUnstructuredOptions.semanticEnrichment === 'boolean'
+          ? restoredUnstructuredOptions.semanticEnrichment
+          : unstructuredOptions.value.semanticEnrichment,
+        qaPairsPerChunk: Number.isFinite(restoredUnstructuredOptions.qaPairsPerChunk)
+          ? restoredUnstructuredOptions.qaPairsPerChunk
+          : unstructuredOptions.value.qaPairsPerChunk,
        datasetSplit: {
-          ...unstructuredOptions.value.datasetSplit,
-          ...snapshot.unstructuredOptions.datasetSplit,
+          train: Number.isFinite(restoredDatasetSplit?.train)
+            ? restoredDatasetSplit.train
+            : unstructuredOptions.value.datasetSplit.train,
+          validation: Number.isFinite(restoredDatasetSplit?.validation)
+            ? restoredDatasetSplit.validation
+            : unstructuredOptions.value.datasetSplit.validation,
+          test: Number.isFinite(restoredDatasetSplit?.test)
+            ? restoredDatasetSplit.test
+            : unstructuredOptions.value.datasetSplit.test,
        },
      }
    }
@@ -336,7 +382,15 @@ function restoreDraft() {
        || null
    results.value = !requiresPreviewMigration && Array.isArray(snapshot.results) ? snapshot.results : []
    selectedResultId.value = requiresPreviewMigration ? null : snapshot.selectedResultId || results.value[0]?.id || null
-    if (!requiresPreviewMigration) Object.assign(generation, snapshot.generation || {})
+    if (!requiresPreviewMigration) {
+      const restoredGeneration = snapshot.generation || {}
+      Object.assign(generation, restoredGeneration)
+      if (generation.status === 'running') {
+        generation.status = 'idle'
+        generation.progress = 0
+        generation.message = '草稿已恢复，请重新开始生成。'
+      }
+    }
    dirty.value = false
    nextTick(() => { restoringDraft.value = false })
    ElMessage.info(requiresPreviewMigration
@@ -355,6 +409,12 @@ watch(
  { deep: true },
 )

+watch(processType, (nextType, previousType) => {
+  if (restoringDraft.value || nextType === previousType || uploadedFiles.value.length === 0) return
+  resetSourceDataForProcessTypeChange()
+  ElMessage.info('处理类型已变更，请重新上传或拉取匹配的源数据')
+})
+
 watch(generationOptionsSignature, (currentSignature, previousSignature) => {
  if (restoringDraft.value || currentSignature === previousSignature) return
  resetDownstream()
@@ -418,8 +478,6 @@ function useSampleFile() {
    count: DEFAULT_SOURCE_TEXT.split('\n').filter((line) => line.trim()).length,
    content: DEFAULT_SOURCE_TEXT
  }]
-  if (!task.name) task.name = '金融问答清洗任务'
-  if (!task.description) task.description = '清洗金融领域问答数据，统一格式并生成高质量训练数据。'
  dirty.value = true
 }

@@ -459,7 +517,6 @@ function handlePullData() {
      count: Math.min(externalSource.limit, DEFAULT_SOURCE_TEXT.split('\n').filter((line) => line.trim()).length),
      content: DEFAULT_SOURCE_TEXT,
    })
-    if (!task.name) task.name = `${typeName} 数据拉取任务`
    dirty.value = true
    ElMessage.success(`已成功拉取 ${uploadedFiles.value[uploadedFiles.value.length - 1].count.toLocaleString()} 条数据`)
  }, 2000)
@@ -477,6 +534,17 @@ function handleRemoveFile(uid: string | number) {
  }
 }

+function resetSourceDataForProcessTypeChange() {
+  uploadedFiles.value = []
+  previewSignature.value = ''
+  previewItems.value = []
+  selectedPreviewFileId.value = null
+  selectedPreviewId.value = null
+  selectedPreviewIdsByFile.value = {}
+  externalConnected.value = false
+  resetDownstream()
+}
+
 function resetDownstream() {
  stopGenerationTimer()
  generation.status = 'idle'
@@ -489,6 +557,10 @@ function resetDownstream() {
 async function nextFromCreate() {
  const valid = await taskSetupRef.value?.validate()
  if (!valid) return
+  goToStep('upload')
+}
+
+function nextFromUpload() {
  if (uploadedFiles.value.length === 0) {
    ElMessage.warning(processType.value === 'external' ? '请先拉取至少一个数据源' : '请上传至少一个源数据文件')
    return
@@ -512,7 +584,7 @@ async function nextFromCreate() {
    previewSignature.value = signature
    resetDownstream()
  }
-  currentStep.value = 1
+  goToStep('preview')
 }

 function selectPreviewFile(fileId: string) {
@@ -671,17 +743,21 @@ async function handlePrimaryAction() {
    await nextFromCreate()
    return
  }
+  if (currentStepId.value === 'upload') {
+    nextFromUpload()
+    return
+  }
  if (currentStepId.value === 'preview') {
    if (!previewItems.value.length) {
      ElMessage.warning('当前没有可生成的预览内容')
      return
    }
-    currentStep.value = 2
+    goToStep('generate')
    return
  }
  if (currentStepId.value === 'generate') {
    if (generation.status === 'success') {
-      currentStep.value = 3
+      goToStep('results')
    } else if (generation.status !== 'running') {
      startGeneration()
    }
@@ -760,6 +836,7 @@ onMounted(restoreDraft)
                'is-active': currentStep === index, 
                'is-completed': currentStep > index 
              }"
+              :aria-current="currentStep === index ? 'step' : undefined"
            >
              <div v-if="index !== 0" class="step-connector"></div>
              <div class="step-node">
@@ -784,6 +861,11 @@ onMounted(restoreDraft)
            v-model:process-type="processType"
            v-model:structured-options="structuredOptions"
            v-model:unstructured-options="unstructuredOptions"
+          />
+
+          <SourceUploadStep
+            v-else-if="currentStepId === 'upload'"
+            :process-type="processType"
            :uploaded-files="uploadedFiles"
            :external-source="externalSource"
            :external-pulling="externalPulling"
@@ -826,7 +908,7 @@ onMounted(restoreDraft)
          />

          <ResultEditorStep
-            v-else
+            v-else-if="currentStepId === 'results'"
            v-model:selected-id="selectedResultId"
            :items="results"
            @update:field="updateResultField"
@@ -1013,4 +1095,47 @@ onMounted(restoreDraft)
 .wizard-primary-action {
  min-width: 160px;
 }
+
+@media (max-width: 760px) {
+  .wizard-main {
+    padding: 24px 20px;
+  }
+
+  .custom-wizard-steps {
+    padding: 0;
+  }
+
+  .step-connector {
+    margin: 0 8px;
+  }
+
+  .step-node {
+    gap: 6px;
+  }
+
+  .step-title {
+    max-width: 72px;
+    font-size: 12px;
+    line-height: 1.35;
+    white-space: normal;
+  }
+
+  .wizard-footer {
+    padding: 0 20px;
+  }
+}
+
+@media (max-width: 600px) {
+  .step-title {
+    display: none;
+  }
+
+  .step-connector {
+    margin: 0 6px;
+  }
+
+  .wizard-primary-action {
+    min-width: 0;
+  }
+}
 </style>
--- a/frontend/src/views/data-process/create/SourceUploadStep.vue
+++ b/frontend/src/views/data-process/create/SourceUploadStep.vue
@@ -0,0 +1,565 @@
+<script setup lang="ts">
+import { computed, ref, watch } from 'vue'
+import type { UploadFile } from 'element-plus'
+import type { ExternalDataSource, ProcessType } from './types'
+
+interface UploadedSourceFile {
+  uid: string | number
+  name: string
+  size: number
+  count: number
+}
+
+const props = defineProps<{
+  processType: ProcessType
+  uploadedFiles: UploadedSourceFile[]
+  externalSource: ExternalDataSource
+  externalPulling: boolean
+  externalConnected: boolean
+}>()
+
+const emit = defineEmits<{
+  'update:externalSource': [value: ExternalDataSource]
+  'file-change': [file: UploadFile]
+  'remove-file': [uid: string | number]
+  'use-sample': []
+  'test-connection': []
+  'pull-data': []
+}>()
+
+const DATA_SOURCE_TYPES = [
+  { value: 'mysql', label: 'MySQL' },
+  { value: 'postgresql', label: 'PostgreSQL' },
+  { value: 'mongodb', label: 'MongoDB' },
+  { value: 'api', label: 'REST API' },
+]
+
+const AUTH_MODES = [
+  { value: 'none', label: '免鉴权' },
+  { value: 'basic', label: '账号密码' },
+  { value: 'token', label: 'Token' },
+]
+
+const FILE_PAGE_SIZE = 10
+const currentFilePage = ref(1)
+
+const isExternal = computed(() => props.processType === 'external')
+
+const uploadAccept = computed(() => props.processType === 'unstructured'
+  ? '.txt,.md,.pdf,.docx,.doc,.json,.jsonl'
+  : '.json,.jsonl,.csv,.xlsx,.xls')
+
+const pagedUploadedFiles = computed(() => {
+  const start = (currentFilePage.value - 1) * FILE_PAGE_SIZE
+  return props.uploadedFiles.slice(start, start + FILE_PAGE_SIZE)
+})
+
+watch(() => props.uploadedFiles.length, (newLength, oldLength) => {
+  const totalPages = Math.max(1, Math.ceil(newLength / FILE_PAGE_SIZE))
+  if (newLength > oldLength) {
+    currentFilePage.value = totalPages
+    return
+  }
+  currentFilePage.value = Math.min(currentFilePage.value, totalPages)
+})
+
+watch(() => props.processType, () => {
+  currentFilePage.value = 1
+})
+
+function updateExternalField<K extends keyof ExternalDataSource>(field: K, value: ExternalDataSource[K]) {
+  emit('update:externalSource', { ...props.externalSource, [field]: value })
+}
+
+function formatSize(size: number) {
+  if (!size) return '0 KB'
+  if (size >= 1024 * 1024) return `${(size / 1024 / 1024).toFixed(1)} MB`
+  return `${(size / 1024).toFixed(1)} KB`
+}
+</script>
+
+<template>
+  <section class="source-upload-step" aria-labelledby="source-upload-title">
+    <div v-if="isExternal" class="form-section external-section">
+      <div class="section-title-row">
+        <div>
+          <h3 id="source-upload-title">数据源配置</h3>
+          <p>配置并验证外部数据源，拉取成功后可在下一步预览数据内容</p>
+        </div>
+      </div>
+
+      <div class="external-form">
+        <el-form label-position="top" class="external-grid">
+          <el-form-item label="数据源类型">
+            <el-select
+              :model-value="externalSource.type"
+              placeholder="请选择数据源类型"
+              aria-label="数据源类型"
+              @update:model-value="updateExternalField('type', $event)"
+            >
+              <el-option
+                v-for="item in DATA_SOURCE_TYPES"
+                :key="item.value"
+                :label="item.label"
+                :value="item.value"
+              />
+            </el-select>
+          </el-form-item>
+          <el-form-item label="地址 / URL">
+            <el-input
+              :model-value="externalSource.url"
+              placeholder="例如：mysql://host:3306/db 或 https://api.example.com/data"
+              aria-label="数据源地址或 URL"
+              @update:model-value="updateExternalField('url', $event)"
+            />
+          </el-form-item>
+          <el-form-item label="鉴权方式">
+            <el-select
+              :model-value="externalSource.authMode"
+              aria-label="鉴权方式"
+              @update:model-value="updateExternalField('authMode', $event)"
+            >
+              <el-option
+                v-for="item in AUTH_MODES"
+                :key="item.value"
+                :label="item.label"
+                :value="item.value"
+              />
+            </el-select>
+          </el-form-item>
+          <el-form-item v-if="externalSource.authMode === 'basic'" label="账号">
+            <el-input
+              :model-value="externalSource.username"
+              autocomplete="username"
+              placeholder="请输入账号"
+              aria-label="数据源账号"
+              @update:model-value="updateExternalField('username', $event)"
+            />
+          </el-form-item>
+          <el-form-item v-if="externalSource.authMode === 'basic'" label="密码">
+            <el-input
+              :model-value="externalSource.password"
+              type="password"
+              show-password
+              autocomplete="current-password"
+              placeholder="请输入密码"
+              aria-label="数据源密码"
+              @update:model-value="updateExternalField('password', $event)"
+            />
+          </el-form-item>
+          <el-form-item v-if="externalSource.authMode === 'token'" label="Token">
+            <el-input
+              :model-value="externalSource.token"
+              type="password"
+              show-password
+              autocomplete="off"
+              placeholder="请输入访问 Token"
+              aria-label="数据源访问 Token"
+              @update:model-value="updateExternalField('token', $event)"
+            />
+          </el-form-item>
+          <el-form-item label="拉取条数">
+            <el-input-number
+              :model-value="externalSource.limit"
+              :min="1"
+              :max="100000"
+              :step="100"
+              controls-position="right"
+              aria-label="数据拉取条数"
+              @update:model-value="updateExternalField('limit', Number($event) || 0)"
+            />
+          </el-form-item>
+        </el-form>
+
+        <div class="external-actions">
+          <el-button
+            :loading="externalPulling && !externalConnected"
+            :disabled="externalPulling"
+            plain
+            @click="emit('test-connection')"
+          >
+            测试连接
+          </el-button>
+          <el-button
+            type="primary"
+            :loading="externalPulling"
+            :disabled="externalPulling"
+            @click="emit('pull-data')"
+          >
+            拉取数据
+          </el-button>
+          <span v-if="externalConnected" class="external-status is-connected" role="status">
+            <i class="fa fa-check-circle" aria-hidden="true" /> 连接正常
+          </span>
+        </div>
+
+        <section v-if="uploadedFiles.length" class="uploaded-file-list" aria-label="已拉取数据列表">
+          <div class="uploaded-file-list-header">
+            <span>已拉取 {{ uploadedFiles.length }} 个数据集</span>
+          </div>
+          <div class="uploaded-file-items">
+            <div v-for="file in pagedUploadedFiles" :key="file.uid" class="uploaded-file">
+              <span class="file-icon"><i class="fa fa-cloud-download" aria-hidden="true" /></span>
+              <div class="file-main">
+                <strong :title="file.name">{{ file.name }}</strong>
+                <span>
+                  {{ formatSize(file.size) }}
+                  <template v-if="file.count"> · {{ file.count.toLocaleString() }} 条</template>
+                </span>
+              </div>
+              <span class="file-status"><i class="fa fa-check-circle" aria-hidden="true" /> 拉取成功</span>
+              <el-button
+                link
+                type="danger"
+                :aria-label="`删除数据集 ${file.name}`"
+                @click="emit('remove-file', file.uid)"
+              >
+                删除
+              </el-button>
+            </div>
+          </div>
+          <el-pagination
+            v-if="uploadedFiles.length > FILE_PAGE_SIZE"
+            v-model:current-page="currentFilePage"
+            :page-size="FILE_PAGE_SIZE"
+            :total="uploadedFiles.length"
+            :pager-count="5"
+            small
+            background
+            layout="prev, pager, next"
+            class="uploaded-file-pagination"
+            aria-label="已拉取数据分页"
+          />
+        </section>
+      </div>
+    </div>
+
+    <div v-else class="form-section upload-section">
+      <div class="section-title-row">
+        <div>
+          <h3 id="source-upload-title">源数据上传</h3>
+          <p>上传后可在下一步检查内容和切分效果，支持同时添加多个文件</p>
+        </div>
+        <el-button v-if="uploadedFiles.length === 0" link type="primary" @click="emit('use-sample')">
+          使用示例数据
+        </el-button>
+      </div>
+
+      <el-upload
+        v-if="uploadedFiles.length === 0"
+        drag
+        multiple
+        :accept="uploadAccept"
+        :auto-upload="false"
+        :show-file-list="false"
+        :on-change="(file: UploadFile) => emit('file-change', file)"
+        aria-label="选择或拖拽源数据文件"
+      >
+        <i class="fa fa-cloud-upload upload-icon" aria-hidden="true" />
+        <div class="el-upload__text">拖拽文件到此处，或<em>点击选择文件</em></div>
+        <template #tip>
+          <div class="el-upload__tip">
+            {{ processType === 'unstructured'
+              ? '支持 TXT、Markdown、PDF、Word、JSON、JSONL，单文件不超过 200MB'
+              : '支持 JSON、JSONL、CSV、Excel，单文件不超过 200MB' }}
+          </div>
+        </template>
+      </el-upload>
+
+      <section v-else class="uploaded-file-list" aria-label="已上传文件列表">
+        <div class="uploaded-file-list-header">
+          <span>已添加 {{ uploadedFiles.length }} 个文件</span>
+          <div class="continue-upload">
+            <el-upload
+              multiple
+              :accept="uploadAccept"
+              :auto-upload="false"
+              :show-file-list="false"
+              :on-change="(file: UploadFile) => emit('file-change', file)"
+              aria-label="继续添加源数据文件"
+            >
+              <el-button size="small" type="primary">继续上传</el-button>
+            </el-upload>
+          </div>
+        </div>
+        <div class="uploaded-file-items">
+          <div v-for="file in pagedUploadedFiles" :key="file.uid" class="uploaded-file">
+            <span class="file-icon"><i class="fa fa-file-text-o" aria-hidden="true" /></span>
+            <div class="file-main">
+              <strong :title="file.name">{{ file.name }}</strong>
+              <span>
+                {{ formatSize(file.size) }}
+                <template v-if="file.count"> · {{ file.count.toLocaleString() }} 条</template>
+              </span>
+            </div>
+            <span class="file-status"><i class="fa fa-check-circle" aria-hidden="true" /> 校验通过</span>
+            <el-button
+              link
+              type="danger"
+              :aria-label="`删除文件 ${file.name}`"
+              @click="emit('remove-file', file.uid)"
+            >
+              删除
+            </el-button>
+          </div>
+        </div>
+        <el-pagination
+          v-if="uploadedFiles.length > FILE_PAGE_SIZE"
+          v-model:current-page="currentFilePage"
+          :page-size="FILE_PAGE_SIZE"
+          :total="uploadedFiles.length"
+          :pager-count="5"
+          small
+          background
+          layout="prev, pager, next"
+          class="uploaded-file-pagination"
+          aria-label="已上传文件分页"
+        />
+      </section>
+    </div>
+  </section>
+</template>
+
+<style scoped lang="scss">
+.source-upload-step {
+  width: 100%;
+}
+
+.form-section {
+  padding: 0;
+
+  h3 {
+    margin: 0 0 5px;
+    color: #2f3747;
+    font-size: 15px;
+    font-weight: 650;
+  }
+}
+
+.section-title-row {
+  display: flex;
+  align-items: flex-start;
+  justify-content: space-between;
+  gap: 16px;
+
+  p {
+    margin: 0;
+    color: #8a93a3;
+    font-size: 12px;
+    line-height: 1.6;
+  }
+}
+
+.external-section {
+  .external-grid {
+    display: grid;
+    grid-template-columns: repeat(2, minmax(0, 1fr));
+    gap: 18px 20px;
+    margin-top: 16px;
+  }
+
+  :deep(.el-form-item) {
+    margin-bottom: 0;
+  }
+
+  :deep(.el-select),
+  :deep(.el-input),
+  :deep(.el-input-number) {
+    width: 100%;
+  }
+}
+
+.external-actions {
+  display: flex;
+  flex-wrap: wrap;
+  align-items: center;
+  gap: 12px;
+  margin-top: 22px;
+}
+
+.external-status {
+  display: inline-flex;
+  align-items: center;
+  gap: 6px;
+  font-size: 12px;
+
+  &.is-connected {
+    color: #2ca66a;
+  }
+}
+
+.upload-section :deep(.el-upload) {
+  width: 100%;
+  margin-top: 16px;
+}
+
+.upload-section :deep(.el-upload-dragger) {
+  width: 100%;
+  min-height: 154px;
+  padding: 32px 20px;
+  background: #fbfcfe;
+  border-color: #dfe3ea;
+  transition: border-color 0.18s ease, background-color 0.18s ease;
+
+  &:hover,
+  &:focus-visible {
+    background: #fafaff;
+    border-color: #8b82f4;
+  }
+
+  &:focus-visible {
+    outline: 2px solid #5b50f2;
+    outline-offset: 2px;
+  }
+}
+
+.upload-icon {
+  margin-bottom: 12px;
+  color: #5b50f2;
+  font-size: 30px;
+}
+
+.uploaded-file-list {
+  margin-top: 20px;
+  overflow: hidden;
+  background: #fff;
+  border: 1px solid #dfe3ea;
+  border-radius: 8px;
+}
+
+.uploaded-file-list-header {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 12px;
+  padding: 10px 14px;
+  color: #5f6878;
+  font-size: 12px;
+  background: #fbfcfe;
+  border-bottom: 1px solid #edf0f5;
+}
+
+.continue-upload {
+  flex: 0 0 auto;
+}
+
+.continue-upload :deep(.el-upload) {
+  width: auto;
+  margin-top: 0;
+}
+
+.uploaded-file-pagination {
+  display: flex;
+  justify-content: flex-end;
+  padding: 10px 14px;
+  border-top: 1px solid #edf0f5;
+}
+
+.uploaded-file {
+  display: flex;
+  align-items: center;
+  gap: 10px;
+  min-height: 48px;
+  padding: 8px 14px;
+  border-bottom: 1px solid #edf0f5;
+
+  &:last-child {
+    border-bottom: 0;
+  }
+}
+
+.file-icon {
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  flex: 0 0 auto;
+  width: 28px;
+  height: 28px;
+  color: #5b50f2;
+  font-size: 14px;
+  background: #f0efff;
+  border-radius: 7px;
+}
+
+.file-main {
+  display: flex;
+  flex: 1;
+  flex-direction: column;
+  gap: 5px;
+  min-width: 0;
+
+  strong {
+    overflow: hidden;
+    color: #273142;
+    font-size: 14px;
+    text-overflow: ellipsis;
+    white-space: nowrap;
+  }
+
+  span {
+    color: #8a93a3;
+    font-size: 12px;
+  }
+}
+
+.file-status {
+  color: #2ca66a;
+  font-size: 12px;
+}
+
+.uploaded-file :deep(.el-button) {
+  flex: 0 0 auto;
+}
+
+@media (max-width: 900px) {
+  .external-section .external-grid {
+    grid-template-columns: minmax(0, 1fr);
+  }
+
+  .uploaded-file {
+    gap: 8px;
+    padding: 8px 10px;
+  }
+
+  .file-status {
+    flex: 0 1 auto;
+    line-height: 1.4;
+    white-space: normal;
+  }
+
+  .uploaded-file-pagination {
+    justify-content: center;
+  }
+}
+
+@media (max-width: 560px) {
+  .section-title-row {
+    align-items: flex-start;
+    flex-direction: column;
+  }
+
+  .uploaded-file-list-header {
+    align-items: flex-start;
+    flex-direction: column;
+  }
+
+  .uploaded-file {
+    align-items: flex-start;
+    flex-wrap: wrap;
+  }
+
+  .file-main {
+    min-width: calc(100% - 40px);
+  }
+
+  .file-status {
+    margin-left: 38px;
+  }
+}
+
+@media (prefers-reduced-motion: reduce) {
+  .upload-section :deep(.el-upload-dragger) {
+    transition: none;
+  }
+}
+</style>
--- a/frontend/src/views/data-process/create/TaskSetupStep.vue
+++ b/frontend/src/views/data-process/create/TaskSetupStep.vue
--- a/frontend/src/views/data-process/create/types.ts
+++ b/frontend/src/views/data-process/create/types.ts
@@ -1,6 +1,6 @@
 export type ProcessType = 'structured' | 'unstructured' | 'external'

-export type StepId = 'create' | 'preview' | 'generate' | 'results'
+export type StepId = 'create' | 'upload' | 'preview' | 'generate' | 'results'

 export type PreprocessOption =
  | 'clean_invalid'
@@ -34,15 +34,6 @@ export type UnstructuredPreprocessOption =

 export type ChunkMethod = 'semantic' | 'heading' | 'fixed' | 'custom'

-export type GenerationContextScope = 'current' | 'adjacent' | 'section'
-
-export type QuestionGenerationType =
-  | 'factual'
-  | 'concept'
-  | 'procedure'
-  | 'reasoning'
-  | 'comprehensive'
-
 export interface UnstructuredProcessOptions {
  preprocessOptions: UnstructuredPreprocessOption[]
  chunkMethod: ChunkMethod
@@ -55,9 +46,6 @@ export interface UnstructuredProcessOptions {
  preserveLists: boolean
  semanticEnrichment: boolean
  qaPairsPerChunk: number
-  contextScope: GenerationContextScope
-  generationTypes: QuestionGenerationType[]
-  skipUnanswerable: boolean
  datasetSplit: DatasetSplitOptions
 }