What does this skill do, and when should you use it?
This is the docx skill bundled with the Open Cowork open-source desktop app, located at .claude/skills/docx/ in the repository. It treats a .docx file as a ZIP archive of XML files: reading uses pandoc, creation uses docx-js (JavaScript), and editing follows an unpack-XML → edit → repack workflow. The skill ships helper scripts (unpack.py, pack.py, comment.py) that preserve tracked changes, add comments, and auto-repair common XML issues. It works with any agent client that has shell and filesystem access, and is one of five skills in the app's Skills library (alongside pptx, pdf, xlsx, and skill-creator).
- Extracts text from .docx with pandoc (--track-changes=all), including tracked changes
- Generates new .docx files from scratch with the docx-js library: paragraphs, lists, tables, images, headers/footers, TOC, page breaks
- Unpacks existing documents to readable XML via unpack.py and repacks with pack.py (validation plus auto-repair)
- Marks insertions/deletions as tracked changes (<w:ins>/<w:del>) with author "Claude", and can reject or restore other authors' changes
- Adds comments and nested replies via comment.py
- Converts documents to PDF (soffice) and images (pdftoppm) for visual verification
- Legal or business professionals who need to revise a Word contract or report in place while preserving tracked changes
- Administrators or consultants generating well-structured reports (US Letter, heading styles, TOC, page numbers) at scale
- Reviewers who want an AI to add comments and replies to a Word document
- Authors who need to extract text from .docx including revision history for downstream processing
- Developers or automation pipelines converting documents to PDF/images for visual QA or delivery
- Locked-down environments that cannot install pandoc, Node.js, LibreOffice, Poppler, and Python — all are declared dependencies
- Users who need PPTX, PDF, or XLSX output — those are sibling skills in the same repository, not covered by this one
- Anyone wanting to work with Word files in a pure chat context with no shell or filesystem access — the skill is entirely command-line based
How do you install this skill?
- Shell / CLI
- Local filesystem
pandocNode.js (docx via npm install -g docx)LibreOffice (soffice)Poppler (pdftoppm)Python 3 (scripts/unpack.py, pack.py, comment.py)
The skill ships built into the Open Cowork app (.claude/skills/docx/) and can also be copied into any Agent Skills-compatible client's skills directory.
Open Cowork: bundled with the app; no separate installation needed.
Claude Code / other Agent Skills clients: copy the docx skill directory from the repo:
git clone https://github.com/OpenCoworkAI/open-cowork.git
cp -r open-cowork/.claude/skills/docx /path/to/your/skills/docxInstall the declared dependencies:
npm install -g docx
brew install pandoc
brew install --cask libreoffice
brew install popplerHow do you use this skill?
Once installed, send your agent any of these to trigger it:
- Read this report.docx and change every "30 days" to "60 days", marking the edits as tracked changes with a comment explaining why
- Create a new US Letter Word document with a title, two heading style levels, a footer with page numbers, and an automatic table of contents
- Extract all text from contract.docx, including tracked changes and comments, and output it as Markdown
- Convert this Word document to PDF and then to images so I can verify the table layout page by page
The skill is triggered automatically by the agent whenever a task involves .docx handling, based on the SKILL.md description. For new documents, the agent writes and runs docx-js JavaScript per the SKILL.md guidance; for editing existing documents it follows a fixed three-step flow:
# Step 1: Unpack
python3 scripts/unpack.py document.docx unpacked/# Step 2: Edit XML directly under unpacked/word/ (use the edit tool for string replacement, not Python scripts)
# Step 3: Repack (validates and auto-repairs)
python3 scripts/pack.py unpacked/ output.docx --original document.docx
# Read/extract text (with tracked changes)
pandoc --track-changes=all document.docx -o output.md# Convert to images for visual check
soffice --headless --convert-to pdf document.docx
pdftoppm -jpeg -r 150 document.pdf page
# Add comments and replies
python3 scripts/comment.py unpacked/ 0 "Comment text"
python3 scripts/comment.py unpacked/ 1 "Reply text" --parent 0Notable options: unpack.py accepts --merge-runs false to skip run merging; pack.py accepts --validate false to skip validation. Auto-repair handles out-of-range durableIds and missing xml:space attributes but not structural XML errors.
What are this skill's strengths and limitations?
- Script-supported three-step editing flow: unpacking merges runs and converts smart quotes; packing validates and auto-repairs common issues
- Explicit pitfall rules for docx-js (A4 default page size, dual table widths, ShadingType.CLEAR, PageBreak inside Paragraph, etc.)
- Complete tracked-changes XML patterns including edge cases like rejecting another author's insertion and restoring their deletion
- Dependencies are mature open-source tools (pandoc, LibreOffice, Poppler, docx-js)
- SKILL.md's license is marked "Proprietary. LICENSE.txt has complete terms", which differs from the repository's overall MIT license — review the LICENSE.txt in the skill directory before adopting
- Heavy dependency chain: pandoc, Node.js, LibreOffice, Poppler, and Python must all be pre-installed
- pack.py's auto-repair scope is limited; structurally malformed XML will still fail
- No independent test evidence beyond the Open Cowork repository itself; behavior outside Claude Code-style clients is documented but unverified
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Open Cowork DOCX Skill this page | 69 · Recommended | ★ 2.2k | 12d ago | MIT |
| DOCX Document Skill ✓ Anthropic · Official | 50 · Use with care | ★ 181k | 16d ago | — |
| DeepTutor DOCX Skill | 59 · Recommended | ★ 41k | 18d ago | Apache-2.0 |
| DOCX Document Processing Skill | 51 · Use with care | ★ 4.4k | 1mo ago | Apache-2.0 |
| PPTX Presentation Skill ✓ Anthropic · Official | 54 · Use with care | ★ 181k | 16d ago | — |
The Open Cowork README compares the hosting app as a whole against Claude Cowork and OpenClaw (remote control, GUI operation, etc.), but offers no alternative comparison for the docx skill itself; the pptx, pdf, and xlsx skills in the same repo are parallel skills, not competitors.
How did FollowSkills review this skill?
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.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
This review was completed under FSRS 1.0. Dimension scores are not converted; it is queued for re-review under FSRS 2.0.
See the full review method →