docx

Use the docx npm library to build documents from Paragraph, Table, and ImageRun elements, then pack them into a .docx file. Set page size explicitly in DXA units, since the default is A4, and use numbering configs rather than literal bullet characters for lists.

Unzip the .docx, edit word/document.xml directly, then rezip the directory. The docx-js library cannot open existing files, so direct XML editing is required; run merge_runs.py first so fragmented text runs become searchable.

Wrap inserted runs in <w:ins> and deleted runs in <w:del> elements with w:id, w:author, and w:date attributes, using <w:delText> instead of <w:t> inside deletions. Validate with the --author flag to catch any edits missing tracked-change markup.

Word splits visible text across many <w:r> runs due to revision IDs and spell-check markers, so phrases rarely exist as contiguous strings. Run merge_runs.py to coalesce adjacent identically-formatted runs before searching.

Yes, comments require six cross-linked files: comments.xml, commentsExtended.xml, commentsIds.xml, commentsExtensible.xml, plus relationship and content-type entries. The comment.py script writes all of them and prints the marker XML to anchor the comment in document.xml.

Pandoc never joins paragraphs when accepting a deleted paragraph mark, and LibreOffice-based acceptance fails when a deleted paragraph is followed by an empty spacer. Both can leave stray empty bullets, so verify paragraph deletions directly in the XML.