Productivity & Collaboration docxword-documentstracked-changespandocdocx-jsooxmldocument-editinglibreoffice

Open Cowork DOCX Skill

Lets an AI create, edit, and analyze Word (.docx) files with tracked changes, comments, formatting preservation, and text extraction.

FollowSkills review · FSRS-1.0
Recommended
69/ 100 5-point scale 3.5 / 5
This review was completed under FSRS 1.0. Dimension scores are not converted; it is queued for re-review under FSRS 2.0.
1 2 3 4 5 6
1Utility16 / 20 · 4.0/5

This is a bundled copy of Anthropic's official docx-handling skill (the license statement matches Anthropic's own 'Proprietary, see LICENSE.txt'). The content itself is trustworthy, but note this is a third-party repo's bundled copy of an official skill, which may lag the official source.

2Reliability15 / 20 · 3.8/5
3Safety18 / 25 · 3.6/5
4Evidence6 / 15 · 2.0/5
5Usability8 / 10 · 4.0/5
6Maintenance6 / 10 · 3.0/5
Evidence confidence:Low Reviewed Jul 17, 2026
Review evidence [1]
See the full review method →

What does this skill do, and when should you use it?

This is the docx skill bundled with the Open Cowork desktop app (.claude/skills/docx/SKILL.md), which teaches an AI to work with Word documents professionally. It treats .docx as a ZIP archive of XML files: new documents are generated with the docx-js library, while existing ones are unpacked, edited as raw XML, and repacked. The skill covers tracked changes, comments, headers/footers, tables, images, and tables of contents, and encodes many rules for avoiding common rendering bugs. Open Cowork itself is MIT-licensed open source, but the SKILL.md is marked Proprietary with terms in LICENSE.txt — a licensing nuance worth checking before reuse.

Reading/analysis: extracts text (including tracked changes) via pandoc --track-changes=all, or unpacks raw XML with scripts/unpack.py. Creating documents: writes JavaScript against the globally installed docx npm package, handling page size, heading styles, proper numbered/bulleted lists, dual-width tables, images, page breaks, TOCs, and headers/footers. Editing: unpack → directly edit XML under unpacked/word/ (authoring tracked changes as 'Claude', using smart-quote XML entities) → repack with scripts/pack.py, which validates and auto-repairs issues like out-of-range durableIds and missing xml:space. It can also convert documents to images via LibreOffice + pdftoppm, and add or reply to comments with scripts/comment.py.

  1. Knowledge workers who need polished Word reports, contracts, or proposals generated from scratch, complete with headings, tables, and a TOC
  2. Professionals reviewing someone else's document who want minimal, format-preserving edits marked as tracked changes (e.g., changing '30 days' to '60 days')
  3. Teams in multi-round review cycles who need comments and nested replies added to documents
  4. Users who want to extract text (with change history) from batches of .docx files for analysis or archiving
  5. Anyone who needs to render .docx files to PDF or images for preview, distribution, or visual checks

What are this skill's strengths and limitations?

Pros
  • Covers both creation and editing paths: docx-js for new files, direct OOXML editing for existing ones, rather than a single approach
  • Tracked-change handling is notably thorough, including nesting scenarios like rejecting another author's insertion or restoring their deletion
  • Encodes many hard-won rules (dual table widths, ShadingType.CLEAR, PageBreak inside Paragraph) that prevent common rendering failures
  • The pack step validates and auto-repairs (durableId, xml:space), reducing risk when hand-editing XML
Limitations
  • SKILL.md is marked Proprietary (terms in LICENSE.txt), which differs from the repo's MIT license — check terms before redistributing
  • Long dependency chain: pandoc, the docx npm package, LibreOffice, Poppler, and Python 3; missing any one disables part of the functionality
  • Editing existing documents relies on the AI editing raw XML; despite validation, complex documents may still need human review when things break
  • No test suite or accuracy benchmarks are provided in the source; real-world reliability is unverified

How do you install this skill?

The skill ships with the Open Cowork desktop app: install Open Cowork and the skill lives at .claude/skills/docx/, ready to use. Open Cowork can be installed via Homebrew (brew tap OpenCoworkAI/tap && brew install --cask --no-quarantine open-cowork), from the Releases page (.exe/.dmg installers), or built from source (git clone, then npm install && npm run rebuild && npm run dev). To use it standalone in another Agent Skills-compatible client, copy the docx folder (SKILL.md plus scripts/) into that client's skills directory — the source does not document standalone installation steps beyond this.

How do you use this skill?

In Open Cowork, pick a workspace folder and issue natural-language prompts, e.g., 'Read financial_report.csv and create a Word summary with tracked changes' or 'Open this .docx and add a comment suggesting we remove section 3'. The skill picks the right path automatically: docx-js for new files, unpack-edit-repack for existing ones. Ensure dependencies are present first: pandoc (extraction), the global docx npm package (creation, npm install -g docx), LibreOffice and Poppler (PDF/image conversion), and Python 3 (for the unpack/pack/comment scripts).

How does this skill compare with similar options?

The Open Cowork README positions its Skills system alongside official Claude Cowork (Open Cowork adds remote control and GUI operation), but the source makes no direct comparison to other standalone Word-processing tools.

FAQ

Does using this skill cost anything on its own?
The skill is bundled free with Open Cowork, which is MIT-licensed; however, the SKILL.md file itself is marked Proprietary (see LICENSE.txt). Your real cost is the AI model API usage you configure.
Will editing a tracked-changes contract corrupt the original formatting?
The editing workflow is built for this: unpacking merges adjacent runs, original <w:rPr> formatting is preserved in tracked-change runs, and edits are minimal. pack.py validates, but auto-repair cannot fix malformed XML or missing relationships, so edge cases may still need manual checking.
What dependencies does it need?
Full functionality requires pandoc, the globally installed docx npm package, LibreOffice, Poppler (pdftoppm), and Python 3. You can skip some for partial use — e.g., no docx npm package needed if you never create new documents.
Can it run outside Open Cowork?
Yes. The SKILL.md is standard Agent Skills format (frontmatter is only name and description), and the instructions are generic shell/filesystem guidance. Copy the folder into Claude Code, Codex, or a similar client and install the script dependencies; a bare model API would need the host to supply an execution environment.

More skills from this repository

All from OpenCoworkAI/open-cowork

Related skills