docx

Use the docx npm library (docx-js) to build a Document object with paragraphs, tables, images, and styles, then write it with Packer.toBuffer. Always set the page size explicitly in DXA units since docx-js defaults to A4, and validate the output file afterward.

Unpack the .docx into its XML parts with the unpack script, edit the XML in word/document.xml directly, then repack with validation. This approach supports tracked changes, comments, and image insertion while preserving the original formatting.

Insert w:ins and w:del elements around the affected runs in document.xml, using w:delText instead of w:t inside deletions. Replace entire w:r blocks rather than nesting change tags inside runs, and copy the original run properties to preserve formatting.

Tables need dual widths: set columnWidths on the table and a matching width on every cell, all in DXA units that sum to the table width. Avoid WidthType.PERCENTAGE, which breaks in Google Docs, and use ShadingType.CLEAR instead of SOLID for cell shading.

Legacy .doc files can be converted to .docx with LibreOffice headless before editing, and .docx files can be converted to PDF or page images via soffice and pdftoppm. The skill does not edit PDFs, spreadsheets, or Google Docs.

Run the accept_changes.py script, which drives LibreOffice headless with a Basic macro to accept all revisions and save a clean document. LibreOffice must be installed on the system for this to work.