document-xlsx

这段内容是审核时保存的快照。外部来源才是完整且最新的版本。

---
name: document-xlsx
description: "Create/edit .xlsx spreadsheets with tables, formulas, charts, validation, and workbook automation. Use when asked to generate Excel reports, models, exports, or audit spreadsheets."
allowed-tools: Bash, Read, Write, Glob, Grep
compatibility: Claude Code + Codex. Uses runtime-specific allowed-tools / argument-hint fields.
version: "1.1"
last_validated: 2026-07-11
---

# Document XLSX Skill - Quick Reference

This skill enables creation, editing, inspection, and safe distribution of `.xlsx` workbooks. Use it for report exports, spreadsheet models, spreadsheet QA, workbook automation, and Excel-compatible deliverables.

Modern best practices (July 2026):
- Prefer Excel Tables over loose ranges.
- Separate inputs, calculations, and outputs.
- Treat spreadsheets as software: checks, owners, change control, and review loops.
- Treat untrusted workbooks as hostile: formulas, hyperlinks, external links, hidden content, and macros all need review.
- If workbooks are shared externally, include accessibility hygiene and run Excel's Accessibility Checker.

## Core Decision Rules (2026)

- First decide the runtime:
  local file generation, cloud workbook automation, or workbook audit/sanitization.
- Default to table-first exports:
  headers in row 1, frozen header row, autofilter, named table, bounded ranges.
- For native pivots:
  use Office Scripts or Excel automation; for headless exports prefer pre-computed summary tables.
- Libraries usually write formulas, but Excel calculates them when the file opens.
  If server-side computed values are required, calculate them in code and write values.
- `XlsxWriter` is write-only: it cannot open, read, or edit an existing `.xlsx` file.
  If the task is "edit this workbook" rather than "create a new one," reach for `openpyxl` (or ExcelJS in Node) instead — choosing `XlsxWriter` for an edit task is a common non-expert mistake that fails immediately.
- A formula written by `openpyxl` or `XlsxWriter` has no cached result until some calculation engine (Excel, LibreOffice headless, or a session-based tool such as xlwings) opens and recalculates the file.
  Reading that same file back with `openpyxl(..., data_only=True)` before any recalculation returns `None`, not the computed value — this looks like a bug but is expected behavior. If a downstream step (pandas, another script, an LLM) needs the number immediately, compute it in Python and write the literal value, or write both the formula and a plausible cached value only if you can guarantee it matches.
- ExcelJS is strong for workbook structure and styling, but it does not provide native chart generation; ExcelJS pivot-table support shipped as an experimental, limited feature only in recent 4.x releases — treat it as unstable and verify round-trip fidelity before relying on it in production.
- `openpyxl` can preserve VBA with `keep_vba=True`, but this skill does not author or execute macros.
- If ingesting untrusted workbooks with `openpyxl`, default to `keep_links=False` unless external links must be preserved.
- For very large exports (hundreds of thousands of rows or more), default `openpyxl` usage can balloon memory (a ~150MB source DataFrame has been observed using 2GB+ RAM with the default XML parser). Install `lxml` and use `Workbook(write_only=True)` for writing or `load_workbook(read_only=True)` for reading — both stream rather than build a full in-memory tree, and `lxml` alone materially cuts memory even outside those modes. Write-only workbooks can be saved exactly once; a second `save()` call raises `WorkbookAlreadySaved`, so batch all writes before saving.
- Row/column ceilings are fixed by the file format, not the library: 1,048,576 rows and 16,384 columns per worksheet. Exports approaching this need a pagination or multi-sheet strategy decided up front, not discovered at write time.

## Quick Reference

| Task | Tool/Library | Language | When to Use |
|------|--------------|----------|-------------