dsh-data-cleaning-agent 0.8.9 → 0.8.11

package/CHANGELOG.md

@@ -2,6 +2,25 @@
 
 本文件记录 `dsh-data-cleaning-agent` 的版本变更。格式遵循 [Keep a Changelog](https://keepachangelog.com/zh-CN/1.1.0/)，版本号遵循 [SemVer](https://semver.org/lang/zh-CN/)。
 
+## [Unreleased]
+
+## [0.8.11] - 2026-09-07
+
+### Fixed
+- 补全结果按已确认映射填入原列空白单元格，保留非空值和列顺序；工作台预览、CSV/XLSX 和异常清单一致，不再为已映射列追加「（补全）」副列。首次调用及候选续跑均保护原表英文键的非空值。
+- 字段映射覆盖当前 128 个可补全字段的标准中文名、英文键和明确别名；移除未知首列强制映射企业名称的兜底。
+- 参照 AI 填表的推荐确认方式，为多义表头提供推荐、折叠分组选择和搜索；重复目标需人工选列，保留已有选择及输出字段范围。
+- CSV/XLSX 重复表头生成唯一列键，防止法定代表人等同名列相互覆盖。旧任务已丢失的原值需重新导入原文件恢复。
+
+## [0.8.10] - 2026-09-07
+
+### Changed
+- 依据共享 DSH-UX-001 v1.3.0 / UX-39，为自有右侧工作台补充左侧连续拖拽分隔条、方向键和 Home/End 调宽、双击复位、拖动取消、手动宽度恢复；中央会话按实际面板宽度同步让位。
+- 按宿主会话区域实际可用宽度约束上下限，空间不足时全屏降级；关闭、切换或卸载恢复自有布局变量并清理监听，不改变 Host 任务、QCC 调用或发送机制。
+
+### Tests
+- 补充实际拖拽、取消、展开转手动、键盘调宽、关闭重开、宿主区域尺寸变化及让位清理回归；发布前通过 214 项检查测试与 10 组隔离 Chromium UI 回归。
+
 ## [0.8.9] - 2026-09-07
 
 ### Changed

package/README.en.md

@@ -2,9 +2,9 @@
 
 > 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: **0.8.9** (stable release)
+> Current source version / 当前源码版本: **0.8.11** (stable release)
 
-This release aligns the workbench stage menu with the tender agent: five equal-width columns in one row, line icons above short labels, and a pale-blue selection with a blue underline. Narrow layouts stay within the panel. All 128 fields, existing workflows, and native DSH input and send behavior are preserved.
+This release adds automatic mapping for all 128 fields using standard labels, field keys, and explicit aliases, with searchable grouped choices and confirmation for ambiguous or duplicate columns. Enrichment fills empty cells in mapped original columns and preserves nonempty values; selected fields without a mapped column are appended. Duplicate CSV/XLSX headers retain independent values. Preview and exports follow the same rules. QCC-blue menus, resizable workbench, and native DSH input/send behavior remain unchanged. Existing download artifacts are not rewritten.
 
 [![CI](https://github.com/duhu2000/dsh-data-cleaning-agent/actions/workflows/ci.yml/badge.svg)](https://github.com/duhu2000/dsh-data-cleaning-agent/actions/workflows/ci.yml)
 [![npm](https://img.shields.io/npm/v/dsh-data-cleaning-agent)](https://www.npmjs.com/package/dsh-data-cleaning-agent)

package/README.md

@@ -2,9 +2,9 @@
 
 > 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: **0.8.9**（正式版本）
+> 当前源码版本 / Current source version: **0.8.11**（正式版本）
 
-本版将右侧阶段菜单对齐招投标工作台：单行五个等宽分栏、上图标下短标题，选中项采用浅蓝底和蓝色底线；窄屏不换行、不横向溢出。保留企查查蓝界面、全部 128 个字段和原有工作流，不替换 DSH 原生输入和发送机制。
+本版为全部 128 个字段加入标准名、英文键及明确别名自动映射，多义表头和重复列提供可确认推荐与分组搜索。补全结果优先写入已映射原列的空白单元格，保留非空原值；无对应原列的已选字段才新增列。CSV/XLSX 重复表头独立保留，预览与导出规则一致。保留企查查蓝菜单、可调宽工作台及 DSH 原生输入和发送机制。历史下载文件不自动改写。
 
 [![CI](https://github.com/duhu2000/dsh-data-cleaning-agent/actions/workflows/ci.yml/badge.svg)](https://github.com/duhu2000/dsh-data-cleaning-agent/actions/workflows/ci.yml)
 [![npm](https://img.shields.io/npm/v/dsh-data-cleaning-agent)](https://www.npmjs.com/package/dsh-data-cleaning-agent)

package/docs/USER-GUIDE.md

@@ -45,6 +45,8 @@ bash <(curl -fsSL https://raw.githubusercontent.com/duhu2000/dsh-data-cleaning-a
 4. **清洗补全**：执行本地确定性清洗；需要 QCC 时先零调用估算，再由当前用户确认使用自己的账号额度；
 5. **下载数据**：生成结果 CSV/XLSX 与异常清单 CSV/XLSX，后续可从任务历史恢复下载。
 
+0.8.11 起的原列回填规则：规则确认时同时检查「字段映射」和「输出字段」。已选字段优先填入已映射原列的空白单元格，非空原值保持不变，原表没有对应映射的已选字段才新增列。地址、网址、开业时间等推荐需确认含义；重复列需明确选定一列。未返回的数据不编造。结果预览、XLSX/CSV 及异常清单遵循同一规则；历史已生成文件不会自动改写。
+
 输入框左上角的「提示词生成」提供三种名单录入方式：
 
 - **粘贴名单**：每行一个企业名称或统一社会信用代码；

package/lib/artifacts.js

@@ -8,6 +8,7 @@
 import { createHash, randomUUID } from 'node:crypto';
 import XLSX from 'xlsx';
 import { FIELD_LABELS, normalizeFieldSelection } from './workflow-contract.js';
+import { projectCompletionResult } from './engine.js';
 
 const ROOT = '.dsh-data-cleaning-artifacts/v1';
 const MAX_ARTIFACT_BYTES = 32 * 1024 * 1024;
@@ -218,12 +219,17 @@ export class WorkflowArtifactStore {
 
   async createBundle(taskId, input = {}) {
     assertId(taskId, 'taskId');
-    const rows = normalizeRows(input.rows);
-    const headers = normalizeHeaders(input.headers, rows, input.fieldSelection);
+    const projected = projectCompletionResult({
+      rows: normalizeRows(input.rows), headers: input.headers || [],
+      mappings: input.mappings || [], fieldSelection: normalizeFieldSelection(input.fieldSelection),
+    });
+    const rows = projected.rows;
+    const headers = normalizeHeaders(projected.headers, rows);
     if (!headers.length) throw new ArtifactError('DC_ARTIFACT_HEADERS_REQUIRED', 'At least one export column is required.', 400);
     const exceptions = input.exceptionRows === undefined
       ? deriveExceptionRows(rows)
-      : normalizeRows(input.exceptionRows);
+      : projectCompletionResult({ rows: normalizeRows(input.exceptionRows), headers: input.headers || [],
+        mappings: input.mappings || [], fieldSelection: normalizeFieldSelection(input.fieldSelection) }).rows;
     const exceptionHeaders = normalizeHeaders([...headers, '_exception_reason'], exceptions);
     const baseName = safeFilePart(input.baseName, '数据清洗补全结果');
     const timestamp = this.nowFn();

package/lib/client.js

@@ -77,6 +77,123 @@ window.__ModuleLoader__.load({
         'aria-hidden': true, focusable: false }, h('path', { d: STAGE_ICON_PATHS[kind] }));
     }
 
+    function workbenchWidthBounds(viewport, available = viewport) {
+      const mobile = viewport <= 760 || available < 740;
+      return { mobile, min: mobile ? viewport : 320, max: mobile ? viewport : Math.max(320, available - 420) };
+    }
+
+    // 自有抽屉的可逆兼容适配；不改 Better Sidebar 或其他插件的面板偏好。
+    function installWorkbenchSizing(drawer, { preferredWidth, expanded, onWidthChange }) {
+      const doc = drawer.ownerDocument;
+      const view = doc.defaultView;
+      const handle = drawer.querySelector('.dcAgentResizeHandle');
+      if (!handle || !view) return () => {};
+      const reservations = new Map();
+      let requested = expanded ? 980 : preferredWidth;
+      let current = 0;
+      let drag = null;
+      const boundsForView = () => workbenchWidthBounds(view.innerWidth,
+        doc.querySelector('[data-conversation-scroll]')?.getBoundingClientRect().width || view.innerWidth);
+      const reserve = () => {
+        const width = boundsForView().mobile ? 0 : drawer.getBoundingClientRect().width;
+        for (const el of doc.querySelectorAll('[data-conversation-scroll]')) {
+          if (!reservations.has(el)) {
+            reservations.set(el, {
+              value: el.style.getPropertyValue('--dc-agent-workbench-reserve'),
+              priority: el.style.getPropertyPriority('--dc-agent-workbench-reserve'),
+            });
+            sizeObserver?.observe(el, { box: 'border-box' });
+          }
+          if (el.style.getPropertyValue('--dc-agent-workbench-reserve') !== `${width}px`) {
+            el.style.setProperty('--dc-agent-workbench-reserve', `${width}px`);
+          }
+        }
+      };
+      const apply = (value) => {
+        const bounds = boundsForView();
+        const fallback = view.innerWidth <= 1040 ? Math.min(420, view.innerWidth * .48) : Math.min(460, view.innerWidth * .42);
+        current = Math.round(Math.min(bounds.max, Math.max(bounds.min, Number.isFinite(value) ? value : fallback)));
+        drawer.dataset.narrow = String(bounds.mobile);
+        drawer.style.setProperty('--dc-agent-panel-width', `${current}px`);
+        handle.setAttribute('aria-valuemin', String(bounds.min));
+        handle.setAttribute('aria-valuemax', String(bounds.max));
+        handle.setAttribute('aria-valuenow', String(current));
+        handle.setAttribute('aria-valuetext', `${current} 像素`);
+        reserve();
+      };
+      const end = (cancel = false) => {
+        if (!drag) return;
+        const previous = drag;
+        drag = null;
+        drawer.classList.remove('is-resizing');
+        if (handle.hasPointerCapture?.(previous.id)) handle.releasePointerCapture(previous.id);
+        if (cancel) apply(requested);
+        else { requested = current; onWidthChange(current); }
+      };
+      const down = (event) => {
+        if (event.button !== 0 || drag || boundsForView().mobile) return;
+        event.preventDefault();
+        handle.focus();
+        drag = { id: event.pointerId, startX: event.clientX, startWidth: current };
+        handle.setPointerCapture(event.pointerId);
+        drawer.classList.add('is-resizing');
+      };
+      const move = (event) => {
+        if (drag?.id !== event.pointerId) return;
+        apply(drag.startWidth + drag.startX - event.clientX);
+      };
+      const up = (event) => { if (drag?.id === event.pointerId) end(); };
+      const cancel = () => end(true);
+      const keydown = (event) => {
+        if (event.key === 'Escape' && drag) {
+          event.preventDefault(); event.stopPropagation(); cancel(); return;
+        }
+        if (drag || boundsForView().mobile) return;
+        const bounds = boundsForView();
+        const delta = event.shiftKey ? 48 : 16;
+        const value = { ArrowLeft: current + delta, ArrowRight: current - delta, Home: bounds.min, End: bounds.max }[event.key];
+        if (value === undefined) return;
+        event.preventDefault(); event.stopPropagation();
+        requested = value; apply(value); onWidthChange(current);
+      };
+      const reset = () => { cancel(); requested = null; apply(null); onWidthChange(null); };
+      const resize = () => { cancel(); apply(requested); };
+      const sizeObserver = typeof view.ResizeObserver === 'function' ? new view.ResizeObserver(resize) : null;
+      handle.addEventListener('pointerdown', down);
+      handle.addEventListener('pointermove', move);
+      handle.addEventListener('pointerup', up);
+      handle.addEventListener('pointercancel', cancel);
+      handle.addEventListener('lostpointercapture', cancel);
+      handle.addEventListener('keydown', keydown);
+      handle.addEventListener('dblclick', reset);
+      view.addEventListener('resize', resize);
+      view.addEventListener('blur', cancel);
+      // 会话挂载点可被宿主替换；只对当前存在的滚动区设置自有让位变量。
+      const observer = typeof view.MutationObserver === 'function' ? new view.MutationObserver(reserve) : null;
+      observer?.observe(doc.body, { childList: true, subtree: true });
+      apply(requested);
+      return () => {
+        cancel();
+        observer?.disconnect();
+        sizeObserver?.disconnect();
+        handle.removeEventListener('pointerdown', down);
+        handle.removeEventListener('pointermove', move);
+        handle.removeEventListener('pointerup', up);
+        handle.removeEventListener('pointercancel', cancel);
+        handle.removeEventListener('lostpointercapture', cancel);
+        handle.removeEventListener('keydown', keydown);
+        handle.removeEventListener('dblclick', reset);
+        view.removeEventListener('resize', resize);
+        view.removeEventListener('blur', cancel);
+        drawer.style.removeProperty('--dc-agent-panel-width');
+        delete drawer.dataset.narrow;
+        for (const [el, previous] of reservations) {
+          if (previous.value) el.style.setProperty('--dc-agent-workbench-reserve', previous.value, previous.priority);
+          else el.style.removeProperty('--dc-agent-workbench-reserve');
+        }
+      };
+    }
+
     // 搜索仅改变可见字段，绝不清除已选字段或改变 Host 的字段目录。
     function visibleCatalogFields(group, query) {
       const needle = String(query || '').trim().toLowerCase();
@@ -166,12 +283,13 @@ window.__ModuleLoader__.load({
 }
 
 .dcAgentWorkbench {
+  position: relative;
   box-sizing: border-box;
   display: flex;
   flex-direction: column;
-  width: min(460px, 42vw);
+  width: var(--dc-agent-panel-width, min(460px, 42vw));
   max-width: calc(100vw - 420px);
-  min-width: 360px;
+  min-width: 320px;
   height: 100vh;
   border: 1px solid var(--dc-border);
   border-radius: 12px 0 0 12px;
@@ -181,7 +299,14 @@ window.__ModuleLoader__.load({
   overflow: hidden;
   pointer-events: auto;
 }
-.dcAgentWorkbench.is-expanded { width: min(980px, calc(100vw - 72px)); }
+.dcAgentWorkbench.is-expanded { width: var(--dc-agent-panel-width, min(980px, calc(100vw - 420px))); }
+.dcAgentResizeHandle { position: absolute; left: 0; top: 0; bottom: 0; width: 8px; z-index: 5; cursor: col-resize; touch-action: none; outline: none; }
+.dcAgentResizeHandle::after { content: ''; position: absolute; left: 0; top: 0; bottom: 0; width: 3px; background: transparent; }
+.dcAgentResizeHandle:hover::after, .dcAgentResizeHandle:focus-visible::after, .is-resizing .dcAgentResizeHandle::after { background: var(--dc-action); }
+.dcAgentWorkbench.is-resizing { user-select: none; }
+.dcAgentWorkbench[data-narrow="true"] { width: 100vw; max-width: none; min-width: 0; border-radius: 0; }
+.dcAgentWorkbench[data-narrow="true"] .dcAgentResizeHandle { display: none; }
+.dcAgentOverlay:has(.dcAgentWorkbench[data-narrow="true"]) { background: rgba(8, 10, 14, 0.22); pointer-events: auto; }
 
 .dcAgentCapabilities {
   display: flex;
@@ -819,6 +944,11 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
 .dcAgentMappingRow { display: grid; grid-template-columns: minmax(0, 1fr) 24px minmax(180px, 1fr); align-items: center; gap: 8px; margin-top: 7px; font-size: 12px; }
 .dcAgentMappingRow > span { overflow: hidden; text-overflow: ellipsis; white-space: nowrap; }
 .dcAgentMappingRow > b { text-align: center; color: light-dark(#8792a3, #8290a5); }
+.dcAgentMappingPicker summary { cursor: pointer; overflow-wrap: anywhere; }
+.dcAgentMappingPicker[open] { padding: 8px; border: 1px solid var(--dc-border); border-radius: 8px; }
+.dcAgentMappingPicker input, .dcAgentMappingPicker select { width: 100%; margin-top: 8px; max-width: 100%; min-width: 0; }
+.dcAgentMappingRow .dcAgentButton { max-width: 100%; white-space: normal; overflow-wrap: anywhere; }
+.dcAgentMappingRow small { display: block; margin: 4px 0; overflow-wrap: anywhere; }
 .dcAgentRulesGrid { display: grid; grid-template-columns: repeat(2, minmax(0, 1fr)); gap: 8px; }
 .dcAgentFieldGroup { display: grid; gap: 7px; margin-top: 12px; }
 .dcAgentFieldGroupHead { display: flex; align-items: flex-start; justify-content: space-between; gap: 10px; }
@@ -847,24 +977,25 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
 @media (min-width: 1041px) {
   body:has(.dcAgentWorkbench) [data-conversation-scroll] {
     box-sizing: border-box;
-    padding-right: min(460px, 42vw);
+    padding-right: var(--dc-agent-workbench-reserve, min(460px, 42vw));
   }
 }
 @media (min-width: 761px) and (max-width: 1040px) {
   body:has(.dcAgentWorkbench) [data-conversation-scroll] {
     box-sizing: border-box;
-    padding-right: min(420px, 48vw);
+    padding-right: var(--dc-agent-workbench-reserve, min(420px, 48vw));
   }
 }
 @media (min-width: 761px) {
   body:has(.dcAgentWorkbench.is-expanded) [data-conversation-scroll] {
     box-sizing: border-box;
-    padding-right: min(980px, calc(100vw - 420px));
+    padding-right: var(--dc-agent-workbench-reserve, min(980px, calc(100vw - 420px)));
   }
 }
 @media (max-width: 760px) {
   .dcAgentWorkbench, .dcAgentWorkbench.is-expanded { width: 100vw; max-width: none; min-width: 0; border-radius: 0; }
   .dcAgentOverlay { background: rgba(8, 10, 14, 0.22); pointer-events: auto; }
+  .dcAgentResizeHandle { display: none; }
   .dcAgentQccBadge, .dcAgentJobsPill { display: none; }
   .dcAgentCandidate { grid-template-columns: 1fr; }
   .dcAgentCapabilities { justify-content: flex-start; padding-inline: 12px; }
@@ -877,7 +1008,7 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
   .dcAgentMappingRow > b { display: none; }
 }
 @media (min-width: 761px) and (max-width: 1040px) {
-  .dcAgentWorkbench { width: 420px; max-width: 48vw; min-width: 360px; }
+  .dcAgentWorkbench { width: var(--dc-agent-panel-width, 420px); }
   .dcAgentQccBadge, .dcAgentJobsPill { display: none; }
 }
 @media (max-width: 360px) {
@@ -1302,30 +1433,128 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
       return RESULT_FIELD_LABELS.get(String(key ?? '')) ?? String(key ?? '');
     }
 
-    const MAPPING_PATTERNS = [
-      ['company_name', /^(name|company|company_name|企业名称|公司名称|单位名称)$/i],
-      ['credit_no', /^(credit_no|creditCode|unified_credit_code|统一社会信用代码|信用代码)$/i],
-      ['reg_no', /^(reg_no|registration_no|注册号)$/i],
-      ['phone', /^(phone|mobile|tel|telephone|联系电话|手机号码|手机号)$/i],
-      ['registered_address', /^(address|registered_address|注册地址|企业地址|公司地址)$/i],
-    ];
-
-    function guessMappings(headers) {
-      const output = [];
-      const usedTargets = new Set();
-      for (const sourceField of Array.isArray(headers) ? headers : []) {
-        const match = MAPPING_PATTERNS.find(([targetField, pattern]) => !usedTargets.has(targetField) && pattern.test(String(sourceField)));
-        if (match) {
-          output.push({ sourceField: String(sourceField), targetField: match[0] });
-          usedTargets.add(match[0]);
+    const MAPPING_ALIASES = {
+      company_name: ['name', 'company', '公司名称', '单位名称', '企业全称', '公司全称', '主体标识'],
+      credit_no: ['creditCode', 'unified_credit_code', '信用代码', '社会信用代码', '统一信用代码', 'USCC'],
+      reg_no: ['registration_no', '工商注册号'],
+      legal_rep: ['法定代表', '法人代表', 'legalRepresentative', 'legal_person'],
+      reg_capital: ['注册资金'], establish_date: ['成立时间', '企业成立日期', 'establishmentDate'],
+      reg_status: ['工商登记状态'], registered_address: ['企业注册地址', '公司注册地址'],
+      phone: ['mobile', 'tel', 'telephone', '电话', '手机号码', '手机号'],
+      contact_official_website: ['官网', '企业官网', '官方网站地址'],
+      contact_preferred_email: ['首选电子邮箱'],
+    };
+    // 语义不完全等同：只推荐，绝不冒充精确匹配或发起调用。
+    const MAPPING_HINTS = {
+      地址: ['registered_address', 'mailing_address', 'invoice_address'],
+      address: ['registered_address', 'mailing_address'],
+      企业地址: ['registered_address', 'mailing_address'], 公司地址: ['registered_address', 'mailing_address'],
+      网址: ['contact_official_website'], website: ['contact_official_website'],
+      开业时间: ['establish_date'], 开业日期: ['establish_date'],
+      行业: ['industry_category', 'qcc_industry'],
+      法人: ['legal_rep'], 邮箱: ['contact_preferred_email'], email: ['contact_preferred_email'],
+    };
+    function normalizeMappingLabel(value) {
+      return String(value ?? '').normalize('NFKC').toLowerCase().replace(/[\s_\-·]/g, '');
+    }
+    function mappingRecommendations(headers, fields = [...ENRICHMENT_OPTIONS, ...INPUT_ONLY_MAPPING_OPTIONS]) {
+      const rows = (Array.isArray(headers) ? headers : []).map((sourceField) => {
+        const label = normalizeMappingLabel(String(sourceField).replace(/（重复列 \d+）$/, ''));
+        let candidates = fields.filter(([id, name]) => [id, name].some((value) => normalizeMappingLabel(value) === label));
+        let reason = '标准字段名匹配';
+        if (!candidates.length) {
+          candidates = fields.filter(([id]) => (MAPPING_ALIASES[id] || []).some((value) => normalizeMappingLabel(value) === label));
+          reason = '常用别名匹配';
+        }
+        const exact = candidates.length === 1;
+        if (!candidates.length) {
+          const hints = Object.entries(MAPPING_HINTS).find(([key]) => normalizeMappingLabel(key) === label)?.[1] || [];
+          candidates = hints.flatMap((key) => fields.filter(([id]) => id === key));
+          reason = candidates.length ? '含义需确认，点击推荐后采用' : '未识别表头，请手动选择';
+        }
+        return { sourceField: String(sourceField), candidates, automatic: exact, reason };
+      });
+      for (const row of rows) {
+        if (row.automatic && rows.filter((other) => other.automatic && other.candidates[0][0] === row.candidates[0][0]).length > 1) {
+          row.conflict = true;
+          row.reason = '多列推荐同一字段，请选择其中一列；其他列保留原值';
         }
       }
-      if (!output.length && Array.isArray(headers) && headers[0]) {
-        output.push({ sourceField: String(headers[0]), targetField: 'company_name' });
+      return rows.map((row) => ({ ...row, automatic: row.automatic && !row.conflict }));
+    }
+    function guessMappings(headers, fields) {
+      return mappingRecommendations(headers, fields).filter((row) => row.automatic)
+        .map((row) => ({ sourceField: row.sourceField, targetField: row.candidates[0][0] }));
+    }
+    function mergeRecommendedMappings(headers, mappings, fields) {
+      const output = [...mappings];
+      for (const item of guessMappings(headers, fields)) {
+        if (!output.some((current) => current.sourceField === item.sourceField || current.targetField === item.targetField)) output.push(item);
       }
       return output;
     }
 
+    function MappingPicker({ sourceField, mappings, groups, recommendation, onChange }) {
+      const [query, setQuery] = react.useState('');
+      const value = mappings.find((item) => item.sourceField === sourceField)?.targetField || '';
+      const occupied = (id) => mappings.some((item) => item.sourceField !== sourceField && item.targetField === id);
+      const fields = groups.flatMap(([, , items]) => items);
+      const label = fields.find(([id]) => id === value)?.[1] || '不参与匹配 / 补全';
+      const matches = (id, name) => normalizeMappingLabel([id, name, ...(MAPPING_ALIASES[id] || [])].join(' ')).includes(normalizeMappingLabel(query));
+      return h('div', { className: 'dcAgentMappingRow' },
+        h('span', { title: sourceField }, sourceField), h('b', { 'aria-hidden': 'true' }, '→'),
+        h('div', { style: { minWidth: 0 } },
+          h('details', { className: 'dcAgentMappingPicker', onKeyDown: (event) => {
+            if (event.key === 'Escape' && event.currentTarget.open) {
+              event.preventDefault(); event.stopPropagation();
+              event.currentTarget.open = false; event.currentTarget.querySelector('summary')?.focus();
+            }
+          } },
+            h('summary', { className: 'dcAgentField', 'aria-label': `${sourceField} 当前映射` }, label),
+            h('input', { className: 'dcAgentField', value: query, placeholder: '搜索字段名称 / 英文键 / 别名', 'aria-label': `${sourceField} 搜索映射字段`, onChange: (event) => setQuery(event.target.value) }),
+            h('select', { className: 'dcAgentField', value, 'aria-label': `${sourceField} 字段映射`, onChange: (event) => {
+              onChange(sourceField, event.target.value);
+              const details = event.currentTarget.closest('details');
+              if (details) { details.open = false; details.querySelector('summary')?.focus(); }
+            } },
+              h('option', { value: '' }, '不参与匹配 / 补全'),
+              groups.map(([id, title, items]) => h('optgroup', { key: id, label: title }, items.filter(([key, name]) => key === value || matches(key, name)).map(([key, name]) =>
+                h('option', { key, value: key, disabled: occupied(key) }, `${name}${occupied(key) ? '（已被其他列使用）' : ''}`)))),
+            ),
+            query && !fields.some(([id, name]) => matches(id, name)) ? h('p', { className: 'dcAgentHint', role: 'status' }, '无匹配字段；当前选择保持不变。') : null,
+          ),
+          h('small', { className: 'dcAgentHint' }, value ? `已映射：${label}；可手动调整` : recommendation.reason),
+          !value && recommendation.candidates.length ? h('div', { className: 'dcAgentRow', 'aria-label': `${sourceField} 推荐映射` }, recommendation.candidates.map(([id, name]) =>
+            h('button', { key: id, type: 'button', className: 'dcAgentButton', disabled: occupied(id), onClick: () => onChange(sourceField, id) }, `推荐：${name}${occupied(id) ? '（已占用）' : ''}`))) : null,
+        ),
+      );
+    }
+
+    function projectCompletionResult({ rows = [], headers = [], mappings = [], fieldSelection = [] } = {}) {
+      const sourceHeaders = [...new Set(headers)];
+      const selected = [...new Set(fieldSelection)];
+      const pairs = mappings.filter((item) => sourceHeaders.includes(item.sourceField) && selected.includes(item.targetField)
+        && mappings.filter((other) => other.sourceField === item.sourceField).length === 1
+        && mappings.filter((other) => other.targetField === item.targetField).length === 1);
+      const hidden = new Set(pairs.filter((item) => item.sourceField !== item.targetField && !sourceHeaders.includes(item.targetField)).map((item) => item.targetField));
+      const blank = (value) => value == null || (typeof value === 'string' && value.trim() === '');
+      const output = rows.map((row) => {
+        const next = { ...row };
+        const status = row.qcc_match_status;
+        if (!status || ['enriched', 'exact'].includes(status)) {
+          for (const { sourceField, targetField } of pairs) {
+            if (blank(row[sourceField]) && !blank(row[targetField])) next[sourceField] = row[targetField];
+          }
+        }
+        for (const key of hidden) delete next[key];
+        return next;
+      });
+      return {
+        rows: output,
+        headers: [...new Set([...sourceHeaders, ...selected, ...output.flatMap((row) => Object.keys(row))])].filter((key) => !hidden.has(key)),
+      };
+    }
+
     function qualitySummaryFor(rows, mappings) {
       const list = Array.isArray(rows) ? rows : [];
       const anchors = (Array.isArray(mappings) ? mappings : []).filter((mapping) => ['company_name', 'credit_no', 'reg_no'].includes(mapping.targetField));
@@ -1487,6 +1716,7 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
         init: () => ({
           open: false,
           expanded: false,
+          preferredWidth: null,
           step: 'upload',
           busy: false,
           error: null,
@@ -1521,6 +1751,10 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
           },
           close: (draft) => { draft.open = false; },
           toggleExpanded: (draft) => { draft.expanded = !draft.expanded; },
+          setPreferredWidth: (draft, width) => {
+            draft.preferredWidth = Number.isFinite(width) ? width : null;
+            draft.expanded = false;
+          },
           setStep: (draft, step) => { draft.step = step; },
           setActiveSession: (draft, sessionId) => { draft.activeSessionId = sessionId ?? null; },
           setBusy: (draft, busy) => { draft.busy = busy; },
@@ -1532,7 +1766,8 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
           setTaskTitle: (draft, value) => { draft.taskTitle = String(value ?? ''); },
           setMappings: (draft, value) => { draft.mappings = Array.isArray(value) ? value : []; },
           setMapping: (draft, sourceField, targetField) => {
-            const mappings = draft.mappings.filter((item) => item.sourceField !== sourceField && item.targetField !== targetField);
+            if (targetField && draft.mappings.some((item) => item.sourceField !== sourceField && item.targetField === targetField)) return;
+            const mappings = draft.mappings.filter((item) => item.sourceField !== sourceField);
             if (targetField) mappings.push({ sourceField, targetField });
             draft.mappings = mappings;
           },
@@ -2968,6 +3203,7 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
       const open = useStore((state) => state.open);
       const step = useStore((state) => state.step);
       const expanded = useStore((state) => state.expanded);
+      const preferredWidth = useStore((state) => state.preferredWidth);
       const busy = useStore((state) => state.busy);
       const error = useStore((state) => state.error);
       const input = useStore((state) => state.input);
@@ -2993,6 +3229,13 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
       const activeSessionId = useStore((state) => state.activeSessionId);
       const [fieldQuery, setFieldQuery] = react.useState('');
 
+      react.useEffect(() => {
+        if (!open || typeof document === 'undefined') return undefined;
+        const drawer = document.querySelector('.dcAgentWorkbench');
+        if (!drawer) return undefined;
+        return installWorkbenchSizing(drawer, { preferredWidth, expanded, onWidthChange: actions.setPreferredWidth });
+      }, [open, expanded, preferredWidth, actions]);
+
       react.useEffect(() => {
         rootWorkbenchActions = actions;
         const handleOpen = (event) => {
@@ -3104,7 +3347,7 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
           view?.removeEventListener?.('resize', align);
           reset();
         };
-      }, [open, expanded]);
+      }, [open, expanded, preferredWidth]);
 
       // DSH 的 useStore 基于 React hooks；所有订阅必须在每次渲染时以相同顺序调用。
       // 把关闭态 guard 放在全部 useStore 之后，避免首次关闭、随后打开时触发
@@ -3653,6 +3896,7 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
         ? workflowContract.inputOnlyMappingFields.map((field) => [field.id, field.label])
         : INPUT_ONLY_MAPPING_OPTIONS;
       const targetFields = [...uiFieldGroups.flatMap(([, , fields]) => fields), ...inputOnlyMappingFields];
+      const mappingHints = mappingRecommendations(dataset?.headers || [], targetFields);
       const setWorkbenchFieldGroup = (fields, enabledGroup) => {
         const selected = new Set(fieldSelection);
         for (const [id] of fields) {
@@ -3703,19 +3947,13 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
           h('section', { className: 'dcAgentSection' },
             h('h3', null, '字段映射'),
             h('p', { className: 'dcAgentHint' }, '至少映射企业名称、统一社会信用代码或注册号其中一项。未映射列会原样保留，但不参与匹配。'),
-            (dataset?.headers || []).map((sourceField) => h('label', { className: 'dcAgentMappingRow', key: sourceField },
-              h('span', { title: sourceField }, sourceField),
-              h('b', { 'aria-hidden': 'true' }, '→'),
-              h('select', {
-                className: 'dcAgentField',
-                value: mappings.find((mapping) => mapping.sourceField === sourceField)?.targetField || '',
-                'aria-label': `${sourceField} 字段映射`,
-                onChange: (event) => actions.setMapping(sourceField, event.target.value),
-              },
-                h('option', { value: '' }, '不参与匹配 / 补全'),
-                targetFields.map(([id, label]) => h('option', { key: id, value: id }, label)),
-              ),
-            )),
+            h('p', { className: 'dcAgentHint' }, `已映射 ${mappings.length} / ${dataset?.headers?.length || 0} 列。标准字段名和明确别名自动匹配；多义表头、重复列请确认推荐。仅在本机匹配表头，不调用模型或企查查，不改变下方输出字段。`),
+            h('button', { type: 'button', className: 'dcAgentButton', onClick: () => actions.setMappings(mergeRecommendedMappings(dataset?.headers || [], mappings, targetFields)) }, '补充自动映射（保留已有选择）'),
+            mappingHints.map((recommendation, index) => h(MappingPicker, {
+              key: `${index}:${recommendation.sourceField}`, sourceField: recommendation.sourceField,
+              mappings, groups: [...uiFieldGroups, ['input-only', '输入校验字段', inputOnlyMappingFields]],
+              recommendation, onChange: actions.setMapping,
+            })),
           ),
           h('section', { className: 'dcAgentSection' },
             h('h3', null, '任务目标'),
@@ -3733,6 +3971,7 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
           ),
           h('section', { className: 'dcAgentSection' },
             h('h3', null, `输出字段 · 已选 ${fieldSelection.length}`),
+            h('p', { className: 'dcAgentHint' }, '补全结果优先填入已映射原列的空白单元格，保留非空原值；没有对应映射列的已选字段新增列。重复或多义列请先确认映射。'),
             h('div', { className: 'dcAgentRow', 'aria-label': '补全字段批次概览' },
               h('span', { className: 'dcAgentChip' }, '基础字段 30'),
               h('span', { className: 'dcAgentChip' }, '第一批 40'),
@@ -3844,7 +4083,7 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
               stat('失败', qccRun.summary?.failed ?? 0, (qccRun.summary?.failed ?? 0) > 0 ? 'bad' : null),
             ),
             Array.isArray(qccRun.rows) && qccRun.rows.length ? h(DatasetReview, {
-              key: qccRun.runId, rows: qccRun.rows, headers: runtime.headers,
+              key: qccRun.runId, ...projectCompletionResult({ rows: qccRun.rows, headers: runtime.headers, mappings: cachedTask?.mappings || mappings, fieldSelection: cachedTask?.fieldSelection || fieldSelection }),
               title: '企查查匹配补全结果', expectedCount: qccRun.summary?.totalRows,
               onError: actions.setError,
             }) : null,
@@ -3979,6 +4218,7 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
           'aria-modal': 'false',
           'aria-label': '数据清洗补全工作台',
           'data-session-id': activeSessionId ?? undefined,
+          id: 'dc-agent-workbench',
           onKeyDown: (event) => {
             if (event.key === 'Escape') {
               event.preventDefault();
@@ -3987,6 +4227,9 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
             }
           },
         },
+          h('div', { className: 'dcAgentResizeHandle', role: 'separator', tabIndex: 0,
+            'aria-label': '调整工作台宽度', 'aria-orientation': 'vertical', 'aria-controls': 'dc-agent-workbench',
+            title: '拖动调整宽度；左右方向键微调，Home/End 最小/最大，双击恢复默认' }),
           h('header', { className: 'dcAgentWbHeader' },
             h('div', { className: 'dcAgentWbTitle' },
               h('span', { className: 'dcAgentWbIcon', 'aria-hidden': 'true' }, h(DatabaseLogo, { size: 22 })),
@@ -4153,6 +4396,7 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
     exports.inject = inject;
     // 测试用纯函数，不构成 Host / DSH 稳定 API。
     exports.__testing = {
+      workbenchWidthBounds,
       DatabaseLogo,
       visibleCatalogFields,
       installCapabilityMount,
@@ -4170,6 +4414,10 @@ body:has(.dcAgentPromptBackdrop) { overflow: hidden; }
       isImagePathEntry,
       requestPromptImage,
       guessMappings,
+      projectCompletionResult,
+      mappingRecommendations,
+      mergeRecommendedMappings,
+      MappingPicker,
       deactivateCleaningSession,
       installSessionOwnershipBridge,
       installNewSessionBridge,

package/lib/engine.js

@@ -9,6 +9,47 @@
  * - 引擎层返回完整明细（供下载/入库），模型工具层只回摘要（安全边界在工具层）。
  */
 
+// 重复表头必须获得稳定、无碰撞的键，否则行对象会覆盖前一列原值。
+function uniqueHeaders(values) {
+  const bases = Array.from(values, (value, i) => String(value ?? '').trim() || `col_${i + 1}`);
+  const reserved = new Set(bases);
+  const used = new Set();
+  return bases.map((base) => {
+    let key = base;
+    let ordinal = 2;
+    // 原表也可能包含同名的重复列标记，不能覆盖它。
+    while (used.has(key) || (key !== base && reserved.has(key))) key = `${base}（重复列 ${ordinal++}）`;
+    used.add(key);
+    return key;
+  });
+}
+
+/** 原列补空投影：由已确认映射决定目的列，非空原值（包括 0/false）不覆盖。 */
+export function projectCompletionResult({ rows = [], headers = [], mappings = [], fieldSelection = [] } = {}) {
+  const sourceHeaders = [...new Set(headers)];
+  const selected = [...new Set(fieldSelection)];
+  const pairs = mappings.filter((item) => sourceHeaders.includes(item.sourceField) && selected.includes(item.targetField)
+    && mappings.filter((other) => other.sourceField === item.sourceField).length === 1
+    && mappings.filter((other) => other.targetField === item.targetField).length === 1);
+  const hidden = new Set(pairs.filter((item) => item.sourceField !== item.targetField && !sourceHeaders.includes(item.targetField)).map((item) => item.targetField));
+  const blank = (value) => value == null || (typeof value === 'string' && value.trim() === '');
+  const output = rows.map((row) => {
+    const next = { ...row };
+    const status = row.qcc_match_status;
+    if (!status || ['enriched', 'exact'].includes(status)) {
+      for (const { sourceField, targetField } of pairs) {
+        if (blank(row[sourceField]) && !blank(row[targetField])) next[sourceField] = row[targetField];
+      }
+    }
+    for (const key of hidden) delete next[key];
+    return next;
+  });
+  return {
+    rows: output,
+    headers: [...new Set([...sourceHeaders, ...selected, ...output.flatMap((row) => Object.keys(row))])].filter((key) => !hidden.has(key)),
+  };
+}
+
 /** 去掉 BOM、按 CSV/RFC4180 子集解析为 { headers, rows }。 */
 export function parseCsv(text) {
   const src = String(text ?? '').replace(/^\uFEFF/, '');
@@ -48,7 +89,7 @@ export function parseCsv(text) {
   const nonEmpty = records.filter((r) => r.some((c) => c.trim() !== ''));
   if (nonEmpty.length === 0) return { headers: [], rows: [] };
 
-  const headers = nonEmpty[0].map((h, i) => (String(h).trim() || `col_${i + 1}`));
+  const headers = uniqueHeaders(nonEmpty[0]);
   const rows = nonEmpty.slice(1).map((r) => {
     const o = {};
     headers.forEach((h, i) => { o[h] = r[i] === undefined ? '' : String(r[i]); });
@@ -72,7 +113,7 @@ export async function parseXlsx(buffer) {
   if (!sheetName) return { headers: [], rows: [] };
   const aoa = XLSX.utils.sheet_to_json(wb.Sheets[sheetName], { header: 1, defval: '', raw: false });
   if (!aoa.length) return { headers: [], rows: [] };
-  const headers = aoa[0].map((h, i) => (String(h ?? '').trim() || `col_${i + 1}`));
+  const headers = uniqueHeaders(aoa[0]);
   const rows = aoa.slice(1).map((r) => {
     const o = {};
     headers.forEach((h, i) => { o[h] = r[i] === undefined || r[i] === null ? '' : String(r[i]); });

package/lib/qcc-runs.js

@@ -229,7 +229,7 @@ export class G5RunStore {
       if (result.status === 'enriched') {
         record.rows[index] = {
           ...row,
-          ...result.fields,
+          ...Object.fromEntries(Object.entries(result.fields || {}).filter(([key]) => row[key] == null || (typeof row[key] === 'string' && row[key].trim() === ''))),
           qcc_match_status: 'enriched',
           qcc_source: 'qcc-mcp',
         };

package/lib/qcc.js

@@ -1027,7 +1027,7 @@ export class QccHostBridge {
       if (result?.status === 'enriched') {
         return {
           ...row,
-          ...result.fields,
+          ...Object.fromEntries(Object.entries(result.fields || {}).filter(([key]) => row[key] == null || (typeof row[key] === 'string' && row[key].trim() === ''))),
           qcc_match_status: 'enriched',
           qcc_source: 'qcc-mcp',
         };

package/lib/web.js

@@ -537,6 +537,7 @@ export function mountWebRoutes(wctx, { logger, report, TOOL_NAME, SKILL_NAME })
             rows: payload.rows,
             headers: payload.headers,
             fieldSelection: readyTask.fieldSelection,
+            mappings: readyTask.mappings,
             exceptionRows: payload.exceptionRows,
             baseName: payload.baseName || readyTask.title,
           });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-data-cleaning-agent",
-  "version": "0.8.9",
+  "version": "0.8.11",
   "description": "Clean, complete, and profile enterprise name lists in DeepSeek Harness — a data cleaning & completion agent plugin with local CSV/XLSX/JSON engine and optional Qichacha (QCC) MCP enrichment. Maintained by Qichacha/QCC.",
   "type": "module",
   "main": "lib/index.js",