Anthropic docx, xlsx, pptx, pdf Skills Compared (2026)
A. Frans
Published October 19, 2026
Table of Contents
- 01The four document skills side by side
- 02How to install them in Claude Code
- 03docx: the one for contracts and redlines
- 04xlsx: the one that checks its own math
- 05pptx: the one with built-in visual QA
- 06pdf: the one for extraction and forms
- 07What none of the four will do for you
- 08Which one should you install first?
- 09FAQ
Anthropic publishes four document skills in its anthropics/skills repo: docx, xlsx, pptx and pdf. They're the same skills that power file creation in the Claude apps, and you can install all four into Claude Code with two commands. They are not interchangeable, though. Each one leans on a different toolchain, has a different system dependency that trips people up, and is good at a different slice of office work.
I read through all four SKILL.md files to pull out what each one does, what it needs installed, and where it's likely to break. If you only install one, the answer is near the bottom.
The four document skills side by side
| docx | xlsx | pptx | ||
|---|---|---|---|---|
| Files | Word .docx (plus legacy .doc via conversion) | .xlsx, .xlsm, .csv, .tsv | .pptx, .potx templates | .pdf |
| Creates new files with | docx npm library (docx-js) | openpyxl | pptxgenjs | reportlab |
| Edits existing files by | Unzipping and editing the XML directly | openpyxl + pandas | Unzipping and editing slide XML | pypdf, qpdf |
| Standout feature | Tracked changes and comments | Formula recalculation with error checks | Render-and-inspect visual QA | Form filling and OCR for scans |
| Heavy dependency | LibreOffice, pandoc, Poppler | LibreOffice | LibreOffice, Poppler | Poppler, Tesseract (for OCR) |
| bestaifor page | docx-word | xlsx-spreadsheet | pptx-powerpoint | pdf-tools |
How to install them in Claude Code
The four skills ship together as one plugin. In a Claude Code session:
/plugin marketplace add anthropics/skills
/plugin install document-skills@anthropic-agent-skills
You can also run /plugin, choose Browse and install plugins, pick anthropic-agent-skills, then document-skills. Paid claude.ai plans already have these skills built in, and the Claude API exposes them as pre-built skills too.
Then install the system tools. On macOS:
brew install --cask libreoffice
brew install poppler pandoc tesseract qpdf
On Ubuntu or Debian:
sudo apt install libreoffice poppler-utils pandoc tesseract-ocr qpdf
Windows users should install LibreOffice from its site and get Poppler through a package manager like Scoop or Chocolatey. Our Windows setup guide for Claude Code skills covers the PATH issues you'll hit.
License and security note
Unlike most skills in that repo, which are Apache 2.0, the four document skills are source-available, not open source. You can read and use them, but check the license file before copying their code into your own product.
On security, these are about as trustworthy as third-party skills get: Anthropic wrote them and runs them in production. They do execute Python and Node scripts on your machine, so read the scripts/ folder before running them in a sensitive repo, the same as any skill. Our skill audit checklist takes about ten minutes.
docx: the one for contracts and redlines
The docx skill splits work into two paths. New documents are built with docx-js, a JavaScript library. Existing documents are never opened with docx-js at all, because it can't read existing files. Instead the skill unzips the .docx, edits word/document.xml directly, and zips it back up, with a validator script that checks the XML against the schema and repairs common issues.
That raw-XML route is what makes its best feature possible: real tracked changes. Insertions and deletions are written as proper <w:ins> and <w:del> revisions, so when you open the file in Word you see a normal redline you can accept or reject. It can also add Word comments anchored to specific text.
Gotchas from the SKILL.md worth knowing: page size defaults to A4 (set US Letter explicitly if you need it), and lists must use Word's numbering config rather than typed bullet characters.
Best for: lawyers, editors and anyone who sends documents back with suggested edits. See our full list for lawyers for tools that pair with it.
xlsx: the one that checks its own math
The xlsx skill uses openpyxl for workbooks with formulas and formatting, and pandas for bulk data. Its key rule is "use formulas, never hardcoded results": totals should be live formulas that recalculate when an input changes, not numbers Claude computed and pasted in.
Openpyxl writes formulas without calculating them, so the skill runs a recalc.py script that uses LibreOffice to compute every cell and report back. If the result is errors_found, the skill won't ship the file. The SKILL.md is candid that a clean recalc only proves the formulas evaluate, not that they're right, and tells Claude to spot-check two or three against expected values.
It also applies standard financial-model color coding unless you say otherwise: blue text for hardcoded inputs, black for formulas, green for links to other sheets, red for links to other files, and yellow fill for key assumptions.
Best for: analysts, accountants and finance teams. Our finance professionals list and the xlsx skill vs Excel MCP vs Google Sheets MCP comparison go deeper.
pptx: the one with built-in visual QA
New decks are generated with pptxgenjs. Templates and existing decks get the same unzip-edit-rezip treatment as docx, with helper scripts to duplicate and clean up slides and a thumbnail script that renders a labeled grid of every slide so Claude can pick a layout.
What sets it apart is the render-and-inspect loop. After building a deck, the skill converts it to images through LibreOffice and Poppler and reviews every slide. The first thing it checks for is text overflowing or cut off at a box edge, the most common way AI-built slides look broken.
That loop is also why pptx is the slowest of the four and the most sensitive to a missing LibreOffice install. Without the renderer, Claude can still write the file but can't look at it.
Best for: consultants, founders and anyone producing recurring decks from a company template. If you'd rather not touch PowerPoint at all, Gamma and Beautiful.ai are the non-skill alternatives.
pdf: the one for extraction and forms
The pdf skill is a toolkit rather than a single workflow. It uses pypdf for merging, splitting, rotating, watermarks and passwords; pdfplumber for pulling text and tables (with an example that exports tables straight to Excel); reportlab for generating new PDFs; and qpdf and pdftotext from the command line.
Two features earn it a place: form filling, which has its own separate instructions file, and OCR for scanned documents, which converts each page to an image and runs Tesseract on it. One gotcha it warns about: reportlab's built-in fonts lack Unicode subscript and superscript characters, which render as black boxes.
Best for: researchers, admin-heavy roles and anyone processing invoices or forms. Our best agent skills for PDF processing roundup compares it with community options like PDF Reader MCP.
What none of the four will do for you
These skills manipulate files. They don't connect to anything. There's no live link to Google Docs, Google Sheets, OneDrive or SharePoint, so a cloud document has to be downloaded first and uploaded again afterward. If your team works in Google Workspace, an MCP server for Sheets or Drive is the better fit, and the comparison linked in the xlsx section covers the trade-offs.
They also can't fix a missing system tool on their own. When LibreOffice isn't installed, xlsx can still write formulas but can't verify them, and pptx can still build slides but can't look at them. That's easy to miss in a long session, so check for a recalc or render step in the output before trusting a file.
And none of them make judgment calls about content. The xlsx skill checks that formulas evaluate, the pptx skill checks that text fits, the docx skill checks that the XML is valid. Whether the numbers, slides or clauses are right is still your job.
Which one should you install first?
Install the whole plugin. The four skills come as one package anyway, and skills cost almost nothing until they're used, because Claude only reads a skill's name and description until a task calls for it (explained in our progressive disclosure guide).
If you're deciding where to spend setup time, though, start with xlsx. It needs only LibreOffice, its formula-recalc check catches the errors that make AI spreadsheets dangerous, and spreadsheets are where most people's repetitive office work lives. Add Poppler next, which gets docx previews, pptx visual QA and PDF text extraction working in one install.
The one I'd be most careful with is docx in a legal setting. Tracked changes are a real advantage, but a redline you didn't read line by line is still a redline you're responsible for.
FAQ
Are Anthropic's docx, xlsx, pptx and pdf skills free?
Yes, they're free to use in Claude Code. They're source-available rather than open source, so check the license before reusing their code in your own product. Paid claude.ai plans include them by default.How do I install the document skills in Claude Code?
Run/plugin marketplace add anthropics/skills, then /plugin install document-skills@anthropic-agent-skills. Install LibreOffice and Poppler so the rendering and recalculation scripts work.
Why does the xlsx skill need LibreOffice?
Openpyxl saves formulas without calculated values. The skill'srecalc.py script uses LibreOffice to compute every formula and report errors before the file is delivered.
Can the docx skill do tracked changes?
Yes. It writes insertions and deletions as native Word revisions, so they appear as a normal redline you can accept or reject in Word. It can also add anchored comments.Can the pdf skill read scanned PDFs?
Yes. It converts each page to an image and runs Tesseract OCR on it, which requirespytesseract, pdf2image and the Tesseract binary to be installed.Share this article
⚙Related Tools
📄Related Articles
Get More AI Tool Guides
New comparisons and guides every week. Join thousands of professionals staying ahead of the AI curve.