下载地址
https://www.zhengdianxue.com/wp-content/uploads/skills/org-1lzygkpw--spreadsheet-first-aid.zip将提示词发送给你的 AI 安装该 skills
请根据 https://skillhub.cn/install/skillhub.md，安装 @org-1lzygkpw/spreadsheet-first-aid。自动诊断并清洗 CSV、TSV、XLS 和 XLSX 表格，处理重复行、空值、乱表头、隐藏空格、乱码、日期格式和数据类型混乱等问题。先生成问题报告，再保留原文件输出干净副本，避免订单号、手机号等编号被错误转换。
阅读全文 （展开查看完整说明）
概述
Spreadsheet First Aid
Diagnose first, explain the rules, then create a copy. Never overwrite the source by default, silently drop rows, or coerce identifiers to numbers.
Workflow
- Identify format, worksheets, header row, data range, encoding, and delimiter. Prefer an available spreadsheet tool for Excel; use
scripts/spreadsheet_aid.pyfor CSV or TSV. - Report row and column counts, blank or duplicate headers, inconsistent widths, exact duplicate rows, missingness, mixed types, invalid dates, and identifiers at risk of losing leading zeros.
- Separate proposed changes into safe automatic fixes, changes needing business confirmation, and prohibited automatic changes.
- When authorized, create
<source>_cleaned, retain column order, and record change counts. - Reopen the output and verify shape, encoding or worksheets, formulas, and key identifiers.
Safety
- Never overwrite the source unless the user explicitly requests it and a backup exists.
- Never guess missing amounts, dates, states, or categories.
- Treat customer IDs, order IDs, phone numbers, and postal codes as text by default.
- State whether deduplication uses whole rows or a business key. If the key is unclear, report candidates only.
- Inventory formulas, merged cells, hidden rows or columns, and validation rules before editing Excel.
CSV Commands
Create a privacy-preserving profile:
python scripts/spreadsheet_aid.py profile "<input.csv>" --report "<report.json>"Preview impact:
python scripts/spreadsheet_aid.py clean "<input.csv>" "<output.csv>" --dry-runCreate a clean copy after confirmation:
python scripts/spreadsheet_aid.py clean "<input.csv>" "<output.csv>" --trim --normalize-empty --dedupeRead [references/cleaning-rules.md](references/cleaning-rules.md) for detailed decisions.
Delivery
Return the issue summary, applied rules, change counts, output path, and unresolved questions. Never report success without auditable counts.
