Anthropic ships four official Claude document skills, and most people are already using them without knowing it. Ask Claude on claude.ai for a spreadsheet, get back a real .xlsx with working formulas, and you just watched a skill run under the hood. The four are named exactly what you would guess: pdf, docx, pptx, and xlsx. They live in the public anthropics/skills repository, where they power document creation across Claude's products, and you can read the source yourself.
Here is the part worth knowing up front. On claude.ai and in the Claude API, these are pre-built skills you do not install at all. In Claude Code the story changes, and I will get to why.
What the official Claude document skills actually do
Each skill maps to one file format and handles the whole life of that format. Read it, create it, edit it.
The pdf skill is the broadest of the four. Per its SKILL.md, it extracts text and tables while preserving layout, merges and splits files, rotates pages, adds watermarks, fills forms, encrypts and decrypts, pulls out images, and runs OCR on scanned documents to make them searchable. Under the hood it reaches for pypdf, pdfplumber, reportlab, and pytesseract, plus command-line tools like qpdf and poppler-utils. When someone calls it the claude pdf skill, they mean a genuine PDF toolkit. It is not a thin wrapper that pastes text into a template.
The claude xlsx skill is the one I trust most for real work. Its SKILL.md handles .xlsx, .xlsm, .csv, and .tsv, and it pushes hard on writing actual Excel formulas rather than hardcoded numbers. A catch is baked into the skill itself. openpyxl writes formulas as strings without computing them, so the skill bundles a scripts/recalc.py that fires up LibreOffice to recalculate and then scans every cell for errors like #REF! and #DIV/0!. That recalc-and-verify step is what separates a spreadsheet that looks right from one whose totals actually came out right.
The claude docx skill creates and edits Word documents, handling formatting, tables, headers and footers, and image insertion. The claude pptx skill (the claude powerpoint skill, if you searched it that way) builds and edits presentations, slides, layouts, and speaker notes. Both treat Office files as the ZIP-and-XML structures they really are, which is why the edits hold up when you open the result in actual Microsoft Office.
One honest note here. Anthropic describes the four document skills in the repo as source-available, not open source. You can read them and learn from them. They are shared as a reference for what a production-grade skill looks like, and they happen to be some of the most complicated skills Anthropic has published.
How to install the Claude document skills
Where you are determines whether "install" even applies.
On claude.ai, you do nothing. Per Anthropic's Agent Skills overview, the pre-built PowerPoint, Excel, Word, and PDF skills "are already working behind the scenes when you create documents." Ask for a Word doc, get a Word doc. No toggle, no upload, as long as your plan has code execution.
In the Claude API, you reference a pre-built skill by its skill_id, say pptx or xlsx, inside the container parameter alongside the code execution tool. The API path needs three beta headers: code-execution-2025-08-25, skills-2025-10-02, and files-api-2025-04-14. The quickstart tutorial walks the exact request. Plan around one constraint here. API skills run with no network access and no runtime package installation, so you only get the pre-installed libraries.
Claude Code is where you actually install pdf skill files yourself. The repo is set up as a plugin marketplace. Register it, then pull the bundle:
/plugin marketplace add anthropics/skills
/plugin install document-skills@anthropic-agent-skills
That puts the document skills on your machine, and Claude Code discovers them from the filesystem the same way it finds any skill in ~/.claude/skills/ or a project's .claude/skills/. If the plugin route gives you trouble, or you just want to understand the folder-and-frontmatter mechanics behind all of this, the general walkthrough is in how to install a Claude skill.
Using them, and where they stop
You mostly do not invoke these by name. Each skill's YAML description tells Claude when it applies, so "extract the tables from this PDF and put them in a spreadsheet" pulls the pdf skill and the xlsx skill on its own. That is the progressive-disclosure design at work. Claude sees only the one-line description until your request matches it, and then it reads the full SKILL.md.
Two things tend to bite people. First, the API sandbox has no network and no package installs, so a skill that assumes a library is present can fail there while working fine in Claude Code, which has full local access. Second, the xlsx recalc step is not optional for formula-heavy files. Skip it and you can ship a spreadsheet whose totals never computed.
The four document skills are deliberately generic. They know the .docx format cold. What they do not know is your quarterly board deck's required section order, your firm's contract clause library, or the five validations your invoices have to pass before finance accepts them. That gap is the whole reason custom skills exist, and you can see the range of them in these Claude skill examples.
For a repeatable job that is specific to how you work, a custom skill beats prompting the generic one every single time. Knack builds that custom skill from a short interview and outputs it in the same open SKILL.md format Anthropic uses, so it drops straight into Claude Code, the Claude apps, Codex, Cursor, or Gemini CLI, right next to the document skills you already have. Lean on the official pdf and xlsx skills for everyday documents, then build your own for the workflow only you run.