doc-office

doc-office — Word, Excel, PowerPoint

The deliverable is a .docx, .xlsx, or .pptx. Reach for the right library; prefer templating over hand-building a document element by element.

Word

• Templated (same shape each time — a letter, a report) → docxtpl: Jinja2 inside a real .docx template the user designed in Word. Styling stays in the template, data in code.
• From scratch → python-docx: headings, tables, images. Keep to named styles, don't hand-format every run.

Excel

• openpyxl — read+write: formulas, styles, conditional formatting, charts, named ranges.
• XlsxWriter — write-only, but richer formatting/charts and faster on big sheets (constant_memory mode).
• pandas to_excel for a quick data dump; drop to openpyxl/XlsxWriter for formatting and formulas.
• Keep numbers as real formulas (=SUM(...)), not pre-computed strings — the user expects a live model. Set column widths, freeze the header row, format currency/dates, add a total row.

PowerPoint

• python-pptx — slides from data: pick a layout, fill placeholders, add tables/charts/images. Reuse a branded template (Presentation("template.pptx")) instead of the default theme.

Quality gate

• Open the file in the real app (or LibreOffice headless) — valid-looking OOXML a viewer rejects is not "done".
• Named styles + a template beat inline formatting for anything the user will maintain.

Pitfalls

• python-docx can't "update" a table of contents — Word rebuilds it on open (tell the user to press F9).
• openpyxl silently drops charts/macros it doesn't understand on round-trip — don't blindly read+write a macro workbook (.xlsm); use the right tool or keep a copy.
• Merged cells + formulas fight — set the value on the top-left cell only.

---

Learned from the community's document skills (incl. Anthropic's official DOCX/XLSX/PPTX skills) — re-implemented as guidance over open libraries, not vendored. See docs/ECOSYSTEM.md. Pairs with doc-pdf.