Word documents

Use python-docx for DOCX creation and ordinary edits. Follow an explicit user or applicable AGENTS.md requirement for a project environment or another library. Otherwise call load_workspace_dependencies and execute the returned Python path with its bundled libraries. Do not install packages or discover a system Python for the default workflow. If the tool is unavailable, use an already configured environment and report a missing dependency only when it prevents the requested operation.

Keep source scripts, intermediate files, and final documents in the task workspace; the runtime and this skill directory are read-only resources. Use the user's requested language and preserve an existing document's design unless a redesign is requested.

Create and edit

For existing files, inspect paragraphs, runs, tables, sections, headers, and footers before changing the affected content. Save to a new file unless the user requests an in-place edit. Replacing a paragraph's .text destroys its run formatting; change the relevant runs when formatting must survive. Reconstructing the whole document can lose features outside python-docx's supported editing API.

Use paragraph styles for headings and body text. Size tables for the section that contains them, and account for merged cells and nested tables. Chinese, Japanese, and Korean text may need an explicit w:eastAsia font assignment in addition to run.font.name; font names alone do not establish glyph availability or rendered appearance.

For a new document, use the selected Python executable:

from docx import Document

document = Document()
document.add_heading("Project report", level=0)
document.add_paragraph("Summary", style="Heading 1")
document.add_paragraph("The requested findings go here.")
document.save("report.docx")

python-docx does not paginate or render documents. Do not represent ordinary replacement, colored text, or comments as tracked changes. When real revisions or unsupported OOXML features matter, preserve their package parts and verify the requested operation rather than silently discarding them.

Check and deliver

Run the shared checker with the selected Python executable; <skill-directory> is this loaded skill's resource base:

<python> <skill-directory>/../scripts/check_office.py <document.docx> --out <checks.json>

It checks ZIP/XML integrity and internal relationships, and reports paragraphs, logical table dimensions, and sections. Optional --contains TEXT arguments assert required text. A successful structural check does not verify pagination, clipping, fonts, or visual appearance. Compare the summary and reopened document with the user's request, including unchanged content that matters to an edit.

Render for a requested image/PDF deliverable or an actionable layout check. Before generating images only for inspection, establish that the current model accepts images. If image input is unavailable, finish structural and content checks and state that visual layout was not inspected; do not generate unreadable previews, retry read_image, or choose a second model.

Use the bundled LibreOffice binaries unless the user explicitly asks not to use them. Without that explicit opt-out, never search for or invoke system LibreOffice, soffice, an application-installed binary, or a downloaded replacement. If the bundled CLI is unavailable or fails, report that failure instead of substituting another executable.

Use the libreofficeKit.node and libreofficeKit.cli absolute paths supplied in the loaded skill's Installed LibreOffice Kit section. Run Node with the CLI entry as its first argument, then capabilities --json once. For the bundled runtime, these paths work from the task directory; do not search PATH, guess a package directory, use npx, or install a renderer. Quote each path separately; PowerShell also requires the & invocation operator. The examples below use <node> and <cli> for those two returned paths. Direct rendering avoids an intermediate PDF:

"<node>" "<cli>" render --input report.docx --output-dir preview-v1 --pages 1,3 --dpi 144
"<node>" "<cli>" convert --input report.docx --output report.pdf

Choose pages relevant to the requested edit; inspect all pages when whole-document layout is required. Batch selected pages in one call and use the returned manifest's page count, image paths, and missing-font diagnostics. Output directories and converted files must be new. Reuse images for the same saved document; render again only after a source change, for an unrendered page, or to diagnose a concrete failure. Check page breaks, clipped text, headings, and table widths. Do not repeatedly export unchanged documents or infer visual overflow solely from guessed text geometry.

If this deployment explicitly disables the CLI but render_document is provided, use it without pages first and read its returned images only on a ready result; status: "skipped" ends visual checking. If neither renderer is available or rendering fails, preserve the usable source and report the inspection limit. Do not search for COM automation, install conversion packages, or build another rendering pipeline for routine QA. LibreOffice pagination can differ from Microsoft Word. Once the requested content and applicable checks pass, deliver the document.

Call present({"files":[{"path":"report.docx"}]}) with the actual final DOCX path. It exposes the current source file without copying or preserving its bytes, so keep that file in place and do not present temporary QA reports unless requested. If present is unavailable, provide the final workspace path using the session's supported file delivery method.