---

name: docx description: Read, create and edit Word DOCX documents, including paragraphs, styles, tables, headers, figures and comments. Use for editable reports and document revisions, with PDF export and visual inspection when layout matters. license: MIT metadata: owner: svd-ai-lab/sim-skills source: https://github.com/svd-ai-lab/sim-skills/tree/main/docx

Word documents

Work on explicit paths and retain the source. Discover available Word or LibreOffice automation before choosing a rendering path. For document structure use python-docx, for example uv run --with python-docx script.py. This workflow uses local files and does not require an application account or a hosted API.

Read

Open the file with docx.Document(path). Inspect iter_inner_content() to retain body paragraph/table order, then inspect sections, headers, footers, styles and inline images as needed. Table cells can contain multiple paragraphs and nested tables. Extracting only document.paragraphs omits important content. Revision marks, text boxes, fields and drawings may require reading the OOXML archive or using Word/LibreOffice. State missing coverage rather than inventing content.

Create

Set the page size and section margins first. Define a small set of paragraph and heading styles; use genuine headings and list numbering rather than repeated manual formatting. Add captions and meaningful column labels. Fit tables and images within the usable page width. Use page breaks only when needed for the document structure. Save to a .docx file, then reopen it to verify text and table values.

Edit

Identify the exact paragraph, cell or run before modifying it. Preserve styles, hyperlinks and surrounding formatting. A word may be split across runs: inspect the run boundaries before replacing it. Assigning a whole paragraph's .text recreates its runs and loses their formatting; do that only when this loss is acceptable. Use targeted XML changes for unsupported structures after inspecting the actual markup, keeping namespaces and relationships valid. Do not claim tracked changes when making ordinary text replacements. Use document comments or an explicit change summary if revision markup is unavailable.

Render and verify

Use an available Word PDF export or LibreOffice conversion. With LibreOffice, use a separate temporary user profile to avoid interfering with an open session:

soffice -env:UserInstallation=file:///ABSOLUTE/TEMP/PROFILE --headless --convert-to pdf --outdir OUTPUT INPUT.docx

Pass each argument separately in a subprocess; construct the profile URI with Path.resolve().as_uri(). Use a unique output directory and check the process exit code and resulting PDF. Render that PDF with an available PDF renderer and inspect every changed page for overflow, orphaned headings, broken tables, missing glyphs and incorrect page breaks. Repeat after corrections. If no office renderer is available, do structural checks and clearly label visual review as not performed. Deliver the editable DOCX and the PDF when requested.

Reference: python-docx documentation.