Word documents

Use python-docx for new documents and ordinary edits. For a task-specific generator, run uv run --no-project --with 'python-docx>=1.2' script.py. Bundled scripts declare their own dependencies; run them with uv run --no-project. Resolve paths below relative to this skill, not the user's working directory.

Workflow

Confirm content, audience, page size, output and template. Inspect the original's paragraphs, runs, tables, headers and sections. Save to a new path; the bundled tools refuse overwriting originals and existing outputs.
Use built-in headings, named styles, real lists and table column widths. For editing, preserve formatting by changing run.text or a targeted table cell, not assigning whole paragraph.text values. Read document features for images, fields, TOC, headers/footers and templates.
For comments or redlining, read review edits, then use scripts/edit.py. Its tracked replacement handles a unique plain-text phrase even across differently formatted runs; it rejects fields, bookmarks and structural revisions it cannot safely handle.
Check content with Document(output) and run scripts/office.py validate. Render every page and inspect it. Fix truncation, bad pagination, missing images and tables running off the page; re-render changed output into a new directory.

uv run --no-project <skill>/scripts/edit.py replace input.docx tracked.docx "Old wording" "New wording" --author "Reviewer"
uv run --no-project <skill>/scripts/edit.py comment input.docx annotated.docx "Target phrase" "Please verify this" --author "Reviewer"
uv run --no-project <skill>/scripts/edit.py accept tracked.docx accepted.docx
uv run --no-project <skill>/scripts/office.py validate accepted.docx
uv run --no-project <skill>/scripts/office.py render accepted.docx word-preview