dsh-data-cleaning-agent 0.9.2 → 0.9.3

package/CHANGELOG.md

@@ -4,6 +4,12 @@
 
 ## [Unreleased]
 
+## [0.9.3] - 2026-09-10
+
+- 图片向导保留独立缩略图与来源元数据，识别前显示待识别而非表格零行。
+- 明确发送授权后，图片高层工具沿用确认版本自动完成识别、体检、匹配、补全及新 XLSX 交付；不再要求第二次发送。歧义、超限及过期配置阻断企业查询，重复调用保持幂等。
+- 文件预览优先复用当前会话的侧栏预览 Tab，下载及无侧栏浏览器回退保持独立。
+
 ## [0.9.2] - 2026-09-10
 
 - Better Sidebar 改为可选 peer；缺失侧栏不阻断基础安装、原生会话和本地清洗/补全/画像工具。

package/README.en.md

@@ -10,7 +10,7 @@ Related agents: [数据清洗补全](https://github.com/duhu2000/dsh-data-cleani
 
 > A data cleaning & completion agent plugin for DeepSeek Harness: local CSV/XLSX/JSON engine plus optional Qichacha (QCC) MCP enterprise-data enrichment. Initiated and maintained by the Qichacha (QCC) team.
 >
-> Current source version / 当前源码版本: **0.9.2** (stable release)
+> Current source version / 当前源码版本: **0.9.3** (stable release)
 
 This release unifies spreadsheet previews, column mapping, and enrichment scope across the prompt wizard and workbench. Mapping statuses use color and text; mapped fields automatically enter the scope, with optional extra fields collapsed. Multiple confirmed output columns may share a non-identity field while preserving non-empty values. It fixes saved drafts resetting the scope to five defaults and detects stale Host versions. Preparing a draft does not query QCC; sending starts execution. Fully restart DSH after upgrading. Existing download artifacts are not rewritten.
 

package/README.md

@@ -9,7 +9,7 @@
 数据清洗补全智能体：面向 Excel/CSV/JSON 企业名单，提供数据清洗、表格清洗、清洗补全、去重、企业数据补全与字段补全，支持企查查 MCP 和结果导出。
 
 ```sh
-dsh plugin --profile web add dsh-data-cleaning-agent@0.9.2
+dsh plugin --profile web add dsh-data-cleaning-agent@0.9.3
 ```
 
 请先满足下文的 DSH、连接器及侧边栏依赖要求；安装后完整停止并重启对应 Profile。
@@ -27,7 +27,7 @@ dsh plugin --profile web add dsh-data-cleaning-agent@0.9.2
 
 > 在 DeepSeek Harness 中清洗、补全、画像企业名单数据的智能体插件：本地 CSV/XLSX/JSON 引擎 + 可选企查查（Qichacha/QCC）MCP 企业数据补全，由企查查（Qichacha/QCC）团队发起并维护。
 >
-> 当前源码版本 / Current source version: **0.9.2**（正式版本）
+> 当前源码版本 / Current source version: **0.9.3**（正式版本）
 
 本版上传表格后展示完整清单及空白数量，用绿色、琥珀色、红色及文字区分已通过、待确认和未匹配映射。已映射字段自动纳入补全范围，额外字段默认折叠；多个原列可经确认使用同一非主体字段，分别补空并保留非空原值。修复 Host 草稿重置为默认五字段的问题，增加前后端版本错配提示。生成说明不查询，发送才执行；升级后需完整重启 DSH，历史下载文件不自动改写。
 

package/lib/client.js

@@ -1190,7 +1190,7 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
       const entryCount = Number.isFinite(config.entryCount) ? config.entryCount : entries.length;
       const source = mode === 'image'
         ? config.imageState === 'completed'
-          ? `图片${config.fileName ? `「${config.fileName}」` : ''}，已由企查查智能文档解析识别并人工核验 ${entryCount} 条主体标识`
+          ? `图片${config.fileName ? `「${config.fileName}」` : ''}，已由企查查智能文档解析识别 ${entryCount} 条主体标识`
           : `图片${config.fileName ? `「${config.fileName}」` : ''}，已安全暂存在本机 Host，等待企查查智能文档解析识别`
         : mode === 'excel'
           ? `本地表格${config.fileName ? `「${config.fileName}」` : ''}，已解析 ${entryCount} 条主体标识，完整数据已载入数据清洗补全工作台`
@@ -1200,7 +1200,7 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
         `输入来源：${source}。`,
       ];
       if (mode === 'image' && config.imageCommandId && config.imageState !== 'completed') {
-        lines.push('请先通过企查查智能文档解析提取图片中的企业全称、18 位统一社会信用代码或注册号；模糊字符不得猜测。识别结果回传数据清洗补全工作台供我核验，然后沿用本任务已经选择的匹配规则、清洗要求和补全字段。');
+        lines.push('本次发送即授权使用本人企查查账号额度，按向导已确认范围执行图片识别、质量体检、主体匹配和字段补全，生成新 Excel。识别后自动沿用已选规则与字段，不重复确认或要求二次发送；仅识别歧义、多候选或权限异常时暂停，模糊字符不得猜测。');
         lines.push(`安全图片凭证：${config.imageCommandId}。请只调用一次 data_cleaning_extract_image_companies，参数只传 commandId；该高层工具会在 Host 内调用 qcc-document-mcp 的 parse_document / get_parse_result，不要直接调用底层 mcp__qcc-document* 工具。`);
       }
       if (entries.length) {
@@ -1486,7 +1486,7 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
       const base = task.source?.fileName?.replace(/\.[^.]+$/, '') || (task.source?.type === 'image' ? '图片名单' : '粘贴名单');
       const date = task.createdAt ? new Date(task.createdAt) : null;
       const time = date && Number.isFinite(date.getTime()) ? date.toLocaleString() : '';
-      return `${base}｜${task.source?.rowCount ?? 0} 条${time ? '｜' + time : ''}`;
+      return `${base}｜${task.source?.type === 'image' && task.state === 'draft' ? '待识别' : `${task.source?.rowCount ?? 0} 条`}${time ? '｜' + time : ''}`;
     }
     function cacheWorkflowTask(actions, sessionId, task) {
       if (!task) return null;
@@ -1511,6 +1511,7 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
           fieldSelection: draft.fieldSelection || DEFAULT_ENRICHMENT_KEYS,
           mappings: draft.mappings || [],
           matchRules: draft.matchRules,
+          ...(draft.source?.type === 'image' ? { source: draft.source } : {}),
         });
         workflowTaskCreationBySession.set(key, pending);
         const clearPending = () => {
@@ -1590,6 +1591,13 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
       const host = await requestJson('/data-cleaning/api/g5/capabilities');
       if (host?.workflowExecutionVersion !== 1) throw new Error('页面与 Host 版本不一致。请保存任务后升级并完整重启 DSH；当前 Host 不支持自动生成结果文件。');
       let task = await ensureWorkflowTask(actions, sessionId);
+      if (expected?.mode === 'image' && expected.imageState === 'prepared') {
+        if (JSON.stringify([...(task.fieldSelection || [])].sort()) !== JSON.stringify([...(expected.enrichmentKeys || [])].sort())) throw new Error('图片任务字段配置尚未保存，请重试。');
+        const bound = await requestJson(`/data-cleaning/api/images/commands/${encodeURIComponent(expected.imageCommandId)}`, 'PATCH', {
+          taskId: task.id, expectedRevision: task.revision, executeOnSend: true,
+        });
+        return { ...bound.command, prompt: buildTaskPrompt(expected) };
+      }
       if (expected) {
         const sameSet = (a, b) => JSON.stringify([...(a || [])].sort()) === JSON.stringify([...(b || [])].sort());
         if (!sameSet(task.fieldSelection, expected.enrichmentKeys)
@@ -2438,6 +2446,8 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
       const [imageFileName, setImageFileName] = react.useState('');
       const [imageSizeBytes, setImageSizeBytes] = react.useState(0);
       const [imageAttachment, setImageAttachment] = react.useState(null);
+      // 独立预览不依赖原生附件的 blob URL；发送前移除附件可能撤销该 URL。
+      const [imagePreviewUrl, setImagePreviewUrl] = react.useState('');
       const [imageCommand, setImageCommand] = react.useState(null);
       const [imageDropActive, setImageDropActive] = react.useState(false);
       const [imageLightbox, setImageLightbox] = react.useState(false);
@@ -2466,6 +2476,28 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
             const command = response?.command;
             if (!command) return;
             setImageCommand((current) => current?.commandId === commandId ? { ...current, ...command } : current);
+            if (command.taskId) {
+              const actions = sessionWorkbenchController?.actionsFor(sessionId);
+              if (actions) {
+                await refreshHostedTask(actions, sessionId, command.taskId, null, true);
+                if (command.state === 'running') requestWorkbenchOpen(command.qccCommand ? 'match' : 'upload', sessionId);
+                if (command.qccCommand) {
+                  const status = await requestJson(`/data-cleaning/api/g5/commands/${encodeURIComponent(command.qccCommand.commandId)}`);
+                  actions.setCommandProgress?.({ taskId: command.taskId, ...(status.command?.progress || {}) });
+                }
+              }
+              if (command.state === 'completed') {
+                const recognized = command.result?.entries || [];
+                setEntries(recognized); setEntryCount(recognized.length);
+                const result = entriesToDataset(recognized);
+                const runtime = runtimeFor(command.taskId);
+                runtime.rows = result.rows; runtime.sourceRows = result.rows; runtime.headers = result.headers;
+                runtime.source = { type: 'image', fileName: imageFileName, sizeBytes: imageSizeBytes };
+                if (actions) actions.setDataset(result);
+                setOpen(false);
+                return;
+              }
+            }
             if (command.state === 'completed') {
               const recognized = Array.isArray(command.result?.entries) ? command.result.entries : [];
               setEntries(recognized);
@@ -2504,6 +2536,7 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
           requestJson(`/data-cleaning/api/images/commands/${encodeURIComponent(imageCommand.commandId)}`, 'DELETE').catch(() => {});
         }
         setImageAttachment(null);
+        setImagePreviewUrl('');
         setImageCommand(null);
         setImageFileName('');
         setImageSizeBytes(0);
@@ -2536,11 +2569,13 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
           attached = Array.isArray(images) ? images[0] : null;
           if (!attached?.id) throw new Error('图片未能加入当前对话。');
           const staged = await stagePromptImage(file);
+          const previewUrl = `data:${file.type};base64,${await imageFileBase64(file)}`;
           if (imageAttachment && typeof removeImage === 'function') removeImage(sessionId, imageAttachment);
           if (imageCommand?.commandId) {
             requestJson(`/data-cleaning/api/images/commands/${encodeURIComponent(imageCommand.commandId)}`, 'DELETE').catch(() => {});
           }
           setImageAttachment(attached);
+          setImagePreviewUrl(previewUrl);
           setImageFileName(file.name);
           setImageSizeBytes(Number(file.size ?? 0));
           setImageCommand(staged.command);
@@ -2650,9 +2685,10 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
         setBusy(true);
         try {
           let prompt = buildTaskPrompt(config);
-          if (mode !== 'image' || imageCommand?.state === 'completed') {
+          if (mode !== 'image' || ['prepared', 'completed'].includes(imageCommand?.state)) {
             const command = await requestPreparedWorkflow(sessionId, config);
-            prompt = command.prompt + '\n\n' + buildTaskPrompt(config);
+            prompt = mode === 'image' && imageCommand?.state === 'prepared' ? command.prompt : command.prompt + '\n\n' + buildTaskPrompt(config);
+            if (mode === 'image' && imageCommand?.state === 'prepared') setImageCommand(command);
           }
           inputActions.setDraft(prompt);
           const updatedInput = getInputSnapshot?.(sessionId);
@@ -2840,16 +2876,16 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
               },
             },
               h('p', null, '可直接在此粘贴图片，或拖入 / 选择 PNG、JPEG、WebP。图片会在对话框显示缩略图，点击可放大。'),
-              imageAttachment ? h('div', { className: 'dcAgentPromptImagePreview' },
+              imagePreviewUrl ? h('div', { className: 'dcAgentPromptImagePreview' },
                 h('button', {
                   type: 'button', className: 'dcAgentPromptImageThumb',
                   'aria-label': `放大查看 ${imageFileName}`,
                   onClick: () => setImageLightbox(true),
-                }, imageAttachment.previewUrl ? h('img', { src: imageAttachment.previewUrl, alt: imageFileName }) : '图片'),
+                }, h('img', { src: imagePreviewUrl, alt: imageFileName })),
                 h('div', { className: 'dcAgentPromptImageMeta' },
                   h('b', null, imageFileName),
                   h('small', null, imageCommand?.state === 'completed'
-                    ? `已识别 ${entryCount} 条，下一步逐条核验`
+                    ? imageCommand.result?.workflowResult ? `已识别 ${entryCount} 条，任务结果见右侧工作台` : `已识别 ${entryCount} 条，下一步逐条核验`
                     : imageCommand?.state === 'running'
                       ? '识别中…'
                       : imageCommand?.state === 'failed'
@@ -2962,8 +2998,8 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
                   : wizardStep === 4 ? '回填到对话框' : '下一步'),
           ),
         ) : null,
-        imageLightbox && imageAttachment?.previewUrl ? h(ImagePreview, {
-          src: imageAttachment.previewUrl, alt: imageFileName, sessionId, onClose: () => setImageLightbox(false),
+        imageLightbox && imagePreviewUrl ? h(ImagePreview, {
+          src: imagePreviewUrl, alt: imageFileName, sessionId, onClose: () => setImageLightbox(false),
         }) : null,
       );
     }
@@ -3026,7 +3062,7 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
         ),
         meta.hint ? h('div', { className: 'dcAgentToolCardHint' }, meta.hint) : null,
         summary ? h('div', { className: 'dcAgentToolCardBody' }, summary) : null,
-        toolName === 'data_cleaning_qcc_run' ? h('div', { className: 'dcAgentArtifactList' },
+        ['data_cleaning_qcc_run', 'data_cleaning_extract_image_companies'].includes(toolName) ? h('div', { className: 'dcAgentArtifactList' },
           [...summary.matchAll(/\[([^\]]+)\]\(((?:https?:\/\/[a-zA-Z0-9.:[\]-]+)?\/data-cleaning\/api\/workflow\/tasks\/[a-zA-Z0-9%_-]+\/artifacts\/[a-zA-Z0-9%_-]+(?:\?preview=1)?)\)/g)].filter(match => match[2].startsWith('/') || (typeof location !== 'undefined' && new URL(match[2]).origin === location.origin)).map((match, index) =>
             h('a', { key: index, className: 'dcAgentButton', href: match[2], target: '_blank', rel: 'noopener noreferrer' }, match[1]))) : null,
       );
@@ -4174,17 +4210,17 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
           ) : null,
           h('div', { className: 'dcAgentWbBody' },
             cachedTask && step !== 'history' ? h('section', { className: 'dcAgentSection', 'aria-label': '当前任务进展', 'aria-live': 'polite' },
-              h('b', null, WORKFLOW_STATE_LABELS[cachedTask.state] || cachedTask.state),
+              h('b', null, cachedTask.source?.type === 'image' && cachedTask.state === 'draft' ? '图片已暂存，待识别' : WORKFLOW_STATE_LABELS[cachedTask.state] || cachedTask.state),
               h('h3', null, workflowDisplayName(cachedTask)),
               h('details', null, h('summary', null, '任务详情'), h('p', { className: 'dcAgentHint' }, `任务编号：${cachedTask.id}`)),
-              h('p', null, `${cachedTask.source?.type === 'image' ? '图片识别' : cachedTask.source?.type === 'text' ? '粘贴名单' : '上传表格'} · ${cachedTask.source?.fileName || ''} · ${cachedTask.source?.rowCount || 0} 条 · ${cachedTask.fieldSelection?.length || 0} 个补全字段`),
+              h('p', null, `${cachedTask.source?.type === 'image' ? '图片识别' : cachedTask.source?.type === 'text' ? '粘贴名单' : cachedTask.source ? '上传表格' : '待选择数据来源'}${cachedTask.source?.fileName ? ' · ' + cachedTask.source.fileName : ''} · ${cachedTask.source?.type === 'image' && cachedTask.state === 'draft' ? '条目数待识别' : `${cachedTask.source?.rowCount || 0} 条`} · ${cachedTask.fieldSelection?.length || 0} 个补全字段`),
               commandProgress?.taskId === cachedTask.id && Number.isFinite(commandProgress.totalUnique) ? h('p', null, `已处理主体 ${commandProgress.completedUnique}/${commandProgress.totalUnique}（去重后，含未匹配或失败，不代表补全成功）`) : null,
               commandProgress?.taskId === cachedTask.id && Number.isFinite(commandProgress.totalUnique) && commandProgress.totalUnique > 0 ? h('progress', {
                 max: commandProgress.totalUnique, value: Math.min(commandProgress.totalUnique, Math.max(0, commandProgress.completedUnique || 0)),
                 'aria-label': '主体处理进度', style: { width: '100%', accentColor: cachedTask.state === 'completed' ? '#16845b' : cachedTask.state === 'failed' ? '#c62828' : cachedTask.state === 'partial' ? '#a66a00' : '#087dcc' },
               }) : null,
               h('p', { style: { color: qccFailedCount > 0 ? '#c62828' : Number(qccReviewCount) > 0 ? '#a66a00' : qccEnrichedCount > 0 ? '#16845b' : 'inherit' } },
-                Number.isFinite(qccEnrichedCount) ? `已处理并补全 ${qccEnrichedCount} 条 · 待核验 ${Number.isFinite(qccReviewCount) ? qccReviewCount : '统计中'} 条 · 失败 ${qccFailedCount} 条（按输入行统计；未披露字段保留空白）` : '结果统计中，处理结束后展示补全与待核验数量'),
+                cachedTask.source?.type === 'image' && cachedTask.state === 'draft' ? '请发送对话中的任务说明开始图片识别，识别后核验企业名单。' : Number.isFinite(qccEnrichedCount) ? `已处理并补全 ${qccEnrichedCount} 条 · 待核验 ${Number.isFinite(qccReviewCount) ? qccReviewCount : '统计中'} 条 · 失败 ${qccFailedCount} 条（按输入行统计；未披露字段保留空白）` : '结果统计中，处理结束后展示补全与待核验数量'),
             ) : null,
             error ? h('div', { className: 'dcAgentError', role: 'alert' }, error) : null,
             step !== 'history' ? h('details', { className: 'dcAgentSection' },
@@ -4320,6 +4356,38 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
         }
       };
       let adapter;
+      let previewService = null;
+      const previewTabId = 'dsh-data-cleaning-agent:artifact-preview';
+      const previewUrls = new Map();
+      const previewListeners = new Set();
+      function ArtifactPreviewTab(props) {
+        const url = react.useSyncExternalStore(listener => {
+          previewListeners.add(listener);
+          return () => previewListeners.delete(listener);
+        }, () => previewUrls.get(props.scope.sessionId) || '', () => '');
+        return url ? h('iframe', { src: url, title: '清洗结果 Excel 预览',
+          sandbox: 'allow-same-origin allow-downloads',
+          style: { display: 'block', width: '100%', height: '100%', minHeight: '320px', border: 0 } })
+          : h('p', null, '请点击清洗结果文件的「预览 / 打开」。');
+      }
+      const previewClick = event => {
+        if (disposed || event.defaultPrevented || event.button !== 0 || event.ctrlKey || event.metaKey || event.shiftKey || event.altKey) return;
+        const anchor = event.target?.closest?.('a[href]');
+        if (!anchor || !previewService) return;
+        let url;
+        try { url = new URL(anchor.href, location.href); } catch { return; }
+        if (url.origin !== location.origin || url.search !== '?preview=1'
+          || !/^\/data-cleaning\/api\/workflow\/tasks\/dcw-[a-zA-Z0-9-]+\/artifacts\/dca-[a-zA-Z0-9-]+$/.test(url.pathname)) return;
+        const sessionId = previewService.getSnapshot().sessionId;
+        if (!sessionId || !previewService.isTabEnabled(previewTabId)) return;
+        previewUrls.set(sessionId, url.pathname + url.search);
+        try { previewService.openTab({ type: previewTabId }, { sessionId }); }
+        catch { return; } // Native browser link remains usable if the provider cannot open.
+        event.preventDefault();
+        event.stopPropagation();
+        for (const listener of previewListeners) listener();
+      };
+      host.addEventListener?.('click', previewClick, true);
       function SessionWorkbenchTab(props) {
         const instance = instanceFor(props.scope.sessionId);
         const useStore = pick => react.useSyncExternalStore(instance.subscribe,
@@ -4340,7 +4408,14 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
         const next = createCleaningSidebarAdapter(service, adapterOptions);
         adapter.dispose();
         adapter = next;
+        const unregisterPreview = next.compatible ? service.registerTab({
+          id: previewTabId, title: '清洗结果预览', order: 31, single: true,
+          icon: size => h(DatabaseLogo, { size }), component: ArtifactPreviewTab,
+        }) : () => {};
+        previewService = next.compatible ? service : null;
         return () => {
+          unregisterPreview();
+          if (previewService === service) previewService = null;
           next.dispose();
           if (!disposed && adapter === next) adapter = createCleaningSidebarAdapter(null, adapterOptions);
         };
@@ -4401,7 +4476,7 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
           const { sessionId, config, resolve, reject } = event.detail;
           const actions = controller.actionsFor(sessionId);
           queueWorkflowOperation(sessionId, () => stageConfirmedWorkflow(actions, sessionId, config))
-            .then(command => { if (!disposed) watchHostedCommand(actions, sessionId, command); resolve(command); }, reject);
+            .then(command => { if (!disposed && command.commandId?.startsWith('dcq-')) watchHostedCommand(actions, sessionId, command); resolve(command); }, reject);
         },
       };
       const bindings = Object.entries(listeners).map(([name, handler]) => {
@@ -4418,6 +4493,8 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
         disposed = true;
         commandWatchGeneration += 1;
         for (const remove of bindings) remove();
+        host.removeEventListener?.('click', previewClick, true);
+        previewUrls.clear(); previewListeners.clear();
         adapter.dispose();
         disconnectFixture();
         void providerFiber?.dispose();

package/lib/image-intake.js

@@ -260,11 +260,16 @@ function publicRecord(record, now = Date.now()) {
     updatedAt: record.updatedAt,
     expiresInMs: Math.max(0, record.expiresAt - now),
     result: record.result,
+    taskId: record.workflow?.taskId,
+    qccCommand: record.qccCommand,
     error: record.error,
   });
 }
 
 function safeFailure(error) {
+  if (['WorkflowError', 'QccBridgeError'].includes(error?.name)) {
+    return { code: error.code, message: error.message };
+  }
   if (error instanceof ImageIntakeError && error.code !== 'DC_IMAGE_PROVIDER_FAILED') {
     return { code: error.code, message: error.message };
   }
@@ -281,6 +286,7 @@ export class ImageIntakeStore {
     ttlMs = IMAGE_LIMITS.ttlMs,
     pollMs = 500,
     maxPolls = 60,
+    continueWorkflow = null,
   } = {}) {
     if (!tools || typeof tools.get !== 'function' || typeof tools.execute !== 'function') {
       throw new TypeError('ImageIntakeStore requires ctx.tools get/execute');
@@ -291,6 +297,7 @@ export class ImageIntakeStore {
     this.pollMs = pollMs;
     this.maxPolls = maxPolls;
     this.records = new Map();
+    this.continueWorkflow = continueWorkflow;
   }
 
   toolDefinition(name) {
@@ -506,7 +513,11 @@ export class ImageIntakeStore {
         entries,
         entryCount: entries.length,
         truncated: entries.length >= IMAGE_LIMITS.maxEntries,
+        needsReview: /[�？?]|无法辨认|识别不清|待确认/.test(text),
       };
+      if (record.workflow && this.continueWorkflow) {
+        record.result.workflowResult = await this.continueWorkflow(record, record.result, exec);
+      }
       record.state = 'completed';
       record.error = null;
       record.updatedAt = new Date(this.clock()).toISOString();
@@ -559,7 +570,7 @@ export function serializeImageExtractionPrompt(command) {
 export function registerImageIntakeTool(tools, store) {
   return tools.register({
     name: TOOL_IMAGE_EXTRACT,
-    description: 'Parse one already-staged local company-list image through QCC qcc-document-mcp. Call only when a visible data-cleaning prompt supplies a dci-* commandId. The Host owns the temporary image, calls parse_document once, and polls get_parse_result only while the returned task is processing.',
+    description: 'Execute one staged company-list image task using a visible dci-* commandId. The Host parses the image once. If the wizard bound an explicitly confirmed execute-on-send workflow, the same Agent-owned call continues quality checks, entity matching, selected-field enrichment and XLSX delivery. Ambiguous recognition or entity candidates require review; never call underlying MCP tools separately or repeat the task.',
     parameters: {
       type: 'object',
       additionalProperties: false,
@@ -575,12 +586,16 @@ export function registerImageIntakeTool(tools, store) {
           entries: { type: 'array', items: { type: 'string' } },
           entryCount: { type: 'integer' },
           truncated: { type: 'boolean' },
+          needsReview: { type: 'boolean' },
+          workflowResult: { type: 'object', additionalProperties: true },
         },
         required: ['commandId', 'entries', 'entryCount', 'truncated'],
       },
       render: (_args, value) => [{
         type: 'text',
-        text: `企查查智能文档解析已识别图片企业名单：${value.entryCount} 条，已同步回数据清洗补全工作台等待核验。请在「导入与核验」查看原始清单全部行或下载全量 CSV/JSON，核对后进入「规则与体检」确认规则并生成质量体检，再到「主体匹配」估算调用量并确认账号额度，点击「生成可编辑任务说明」自动回填中央对话框。无需手工复制 dcq- 凭证或重新粘贴名单；识别名单不代表已匹配，尚未查询时不要编造匹配摘要。`,
+        text: value.workflowResult
+          ? `图片识别及已确认任务已执行，识别 ${value.entryCount} 条。真实执行统计：${JSON.stringify(value.workflowResult)}。${value.workflowResult.reviewReason || '无需重复确认规则或再次发送；仅按真实异常提示处理。'}\n${(value.workflowResult.artifacts || []).map(a => `${String(a.fileName).replace(/[^a-zA-Z0-9\u4e00-\u9fff._ -]/g, '_')}：[预览 / 打开](${a.url}?preview=1) · [下载 Excel](${a.url})`).join('\n')}`
+          : `图片已识别 ${value.entryCount} 条，等待核验。此图片未绑定已确认的补全任务，请在「导入与核验」核验后配置后续处理；识别名单不代表已匹配。`,
       }],
     },
     async execute(args, exec) {

package/lib/image-workflow.js

@@ -0,0 +1,31 @@
+import { WorkflowError } from './workflow.js';
+
+// The wizard binds its confirmed revision. Execution reuses the original Agent token.
+export function createImageWorkflow({ getWorkflow, execution, commands }) {
+  return async (record, result, exec) => {
+    const store = await getWorkflow();
+    let task = await store.require(record.workflow.taskId);
+    if (task.revision !== record.workflow.revision || task.state !== 'draft') {
+      throw new WorkflowError('DC_WORKFLOW_REVISION_CONFLICT', '图片任务配置已变更，请重新生成说明；未执行企业查询。', 409);
+    }
+    const rows = result.entries.map(value => ({ 主体标识: value }));
+    const mappings = [{ sourceField: '主体标识', targetField: 'company_name' }];
+    task = await store.recordUpload(task.id, { expectedRevision: task.revision,
+      source: { type: 'image', fileName: record.fileName, sizeBytes: record.sizeBytes,
+        rowCount: rows.length, columnCount: 1, headers: ['主体标识'] } });
+    task = await store.updateDraft(task.id, { expectedRevision: task.revision, mappings });
+    if (result.truncated || result.needsReview || rows.length > 100) {
+      return { taskId: task.id, deliveryState: 'review_required', artifacts: [],
+        reviewReason: rows.length > 100 || result.truncated ? '识别名单超出单批范围，请核验并拆分。' : '识别文本含模糊字符或待确认提示，请核验名单。' };
+    }
+    task = await store.confirmRules(task.id, { expectedRevision: task.revision, mappings });
+    task = await store.recordQuality(task.id, { expectedRevision: task.revision,
+      summary: { total: rows.length, valid: rows.length, missingAnchor: 0,
+        duplicates: rows.length - new Set(result.entries).size } });
+    const payload = await execution.prepare({ taskId: task.id, expectedRevision: task.revision,
+      kind: 'enrich', rows, headers: ['主体标识'], nameField: '主体标识' });
+    const command = commands.prepare(payload, { artifactOrigin: record.workflow.artifactOrigin });
+    record.qccCommand = { commandId: command.commandId, taskId: task.id };
+    return commands.run(command.commandId, exec);
+  };
+}

package/lib/web.js

@@ -17,6 +17,7 @@ import { publicWorkflowContract } from './workflow-contract.js';
 import { DataCleaningWorkflowStore, WorkflowError } from './workflow.js';
 import { ArtifactError, WorkflowArtifactStore } from './artifacts.js';
 import { createWorkflowExecution } from './workflow-execution.js';
+import { createImageWorkflow } from './image-workflow.js';
 import {
   ImageIntakeError,
   ImageIntakeStore,
@@ -304,10 +305,11 @@ export function mountWebRoutes(wctx, { logger, report, TOOL_NAME, SKILL_NAME })
   const skills = wctx.skills;
   const disposers = [];
   const qccBridge = new QccHostBridge({ tools, logger });
-  const imageIntake = new ImageIntakeStore({ tools });
   const g5Runs = new G5RunStore();
   const execution = createWorkflowExecution({ getWorkflow: () => getWorkflow(), artifacts: wctx.fs ? new WorkflowArtifactStore({ fs: wctx.fs }) : null });
   const qccCommands = new QccCommandStore({ bridge: qccBridge, runs: g5Runs, lifecycle: execution });
+  const imageIntake = new ImageIntakeStore({ tools,
+    continueWorkflow: createImageWorkflow({ getWorkflow: () => getWorkflow(), execution, commands: qccCommands }) });
   if (typeof tools?.register === 'function') {
     disposers.push(registerQccCommandTool(tools, qccCommands));
     disposers.push(registerImageIntakeTool(tools, imageIntake));
@@ -417,6 +419,26 @@ export function mountWebRoutes(wctx, { logger, report, TOOL_NAME, SKILL_NAME })
           command: imageIntake.status(decodeURIComponent(rest[0])),
         });
       }
+      if (rest.length === 1 && req.method === 'PATCH') {
+        const payload = JSON.parse((await readBody(req)).toString('utf8'));
+        const record = imageIntake.require(decodeURIComponent(rest[0]));
+        const task = await (await getWorkflow()).require(payload.taskId);
+        if (payload.executeOnSend !== true || record.state !== 'prepared' || task.state !== 'draft'
+          || task.revision !== payload.expectedRevision || task.source?.type !== 'image'
+          || task.source.fileName !== record.fileName || !task.fieldSelection.length) {
+          throw new ImageIntakeError('DC_IMAGE_CONFIG_CONFLICT', '图片或任务配置已变更，或未选择补全字段，请重新确认。', 409);
+        }
+        let artifactOrigin = '';
+        const requestHost = req.headers?.host;
+        if (typeof requestHost === 'string' && /^[a-zA-Z0-9.:[\]-]+$/.test(requestHost)) {
+          try {
+            const origin = new URL(req.headers.origin || `${req.socket?.encrypted ? 'https' : 'http'}://${requestHost}`);
+            if (['http:', 'https:'].includes(origin.protocol) && origin.host === requestHost) artifactOrigin = origin.origin;
+          } catch { /* Use relative artifact links. */ }
+        }
+        record.workflow = { taskId: task.id, revision: task.revision, artifactOrigin };
+        return writeJson(res, 200, { ok: true, command: imageIntake.status(record.commandId) });
+      }
       if (rest.length === 1 && req.method === 'DELETE') {
         await imageIntake.remove(decodeURIComponent(rest[0]));
         return writeJson(res, 200, { ok: true, marker: 'data-cleaning-image-intake-v1', removed: true });

package/lib/workflow.js

@@ -149,7 +149,7 @@ export class DataCleaningWorkflowStore {
       fieldSelection: draft.fieldSelection,
       mappings: draft.mappings,
       matchRules: draft.matchRules,
-      source: null,
+      source: input.source?.type === 'image' ? sanitizeSource({ ...input.source, rowCount: 0 }) : null,
       qualitySummary: null,
       matchSummary: null,
       enrichmentSummary: null,
@@ -218,7 +218,8 @@ export class DataCleaningWorkflowStore {
         mappings: input.mappings ?? record.mappings,
         matchRules: input.matchRules ?? record.matchRules,
       });
-      return draft;
+      return { ...draft, ...(record.state === 'draft' && input.source?.type === 'image'
+        ? { source: sanitizeSource({ ...input.source, rowCount: 0 }) } : {}) };
     });
   }
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-data-cleaning-agent",
-  "version": "0.9.2",
+  "version": "0.9.3",
   "description": "Data cleaning and data enrichment for CSV/XLSX/JSON enterprise lists in DeepSeek Harness, including spreadsheet cleaning, deduplication, profiling, optional Qichacha MCP and exports.",
   "type": "module",
   "main": "lib/index.js",