Guide
Best agent Skills for document processing
The best agent Skill for document processing is whichever matches the file type, and all four defaults are Anthropic-official: pdf for PDFs, docx for Word, xlsx for Excel, pptx for PowerPoint. Each ships tested scripts so the agent runs a deterministic parser instead of improvising file manipulation. pdf is the one most teams reach for first.
Pick by output format, not input. The difference these Skills make is deterministic handling of the edge cases plain python-docx or pdfplumber get wrong — encrypted PDFs, tracked changes, merged cells, embedded charts — so the same prompt produces the same file every run.
Top pick
pdfAnthropic-built. Extracts text and tables, fills forms, and merges PDFs using bundled Python scripts, so the agent runs a tested parser instead of hallucinating page contents.
Best when: Anything PDF — extraction, form-filling, batch merge or redaction.
Ranked
The shortlist
Ranked by fit for the common case, not by raw popularity. Each is a catalogued, verified Skill — open the skill page for the composite score, license, and full install commands.
| Skill | Source | Best when | Verified agents |
|---|---|---|---|
| Official | Anything PDF — extraction, form-filling, batch merge or redaction. | Claude Code, Claude.ai, Claude API, Cursor, Codex | |
| docx | Official | Generating or redlining Word documents, especially with track-changes. | Claude Code, Claude.ai, Claude API |
| xlsx | Official | The document is an Excel workbook with formulas or pivots. | Claude Code, Claude.ai, Claude API |
| pptx | Official | Producing or extracting from slide decks at scale. | Claude Code, Claude.ai, Claude API |
| doc-coauthoring | Official | Collaborative editing where the agent must not clobber human edits. | Claude Code, Claude.ai |
See every entry, ranked by composite score, in the Document Processing category.
Why these
Pick by pick
- 1pdf
Anthropic-built. Extracts text and tables, fills forms, and merges PDFs using bundled Python scripts, so the agent runs a tested parser instead of hallucinating page contents.
- 2docx
Anthropic-built. Creates and edits Word documents with formatting, tables, comments, and tracked changes — the edge cases improvised python-docx usually breaks on.
- 3xlsx
Anthropic-built. Produces real Excel with formulas and charts rather than a flat CSV — the right tool when the document itself is a spreadsheet.
- 4pptx
Anthropic-built. Generates PowerPoint decks with layouts, charts, and speaker notes from a prose outline, and parses existing decks for content.
- 5doc-coauthoring
Anthropic-built. Multi-author editing patterns — suggestion mode and conflict resolution — for when the agent is one of several voices on a document.
FAQ
Common questions
What is the best Claude skill for document processing?
There is no single best — pick by the file. pdf handles PDF extraction, form-filling, and merging; docx handles Word with tracked changes; xlsx handles Excel with formulas and pivots; pptx handles PowerPoint decks. All four are built by Anthropic, ship bundled scripts, and run natively in Claude Code, Claude.ai, and the Claude API. pdf is the most-reached-for because PDF extraction and form-filling are the most common document tasks.
Are these better than asking the model to write a Python script?
For one-off tasks, prompting works. For anything reproducible — same input, same output, every time — a Skill is meaningfully more reliable. It bundles tested scripts and handles the format edge cases the model routinely gets wrong: encrypted PDFs, .docx tracked changes, .xlsx merged cells, embedded images. It also avoids prompt-to-prompt drift where the generated script changes each run.
Which agents support these document Skills?
Anthropic's pdf, docx, pptx, and xlsx Skills run natively in Claude Code, Claude.ai (Pro/Max/Team/Enterprise), and the Claude API. pdf is additionally verified on Cursor and Codex. We list only verified-working agents on each skill page.
Do these skills handle OCR and scanned documents?
The pdf Skill extracts text and tables from PDFs that contain a text layer. Pure scanned-image PDFs need an OCR step first; the file manipulation itself runs locally in the agent sandbox, while cloud OCR requires network access. For the underlying spec and how bundled scripts load, see the anatomy of a SKILL.md guide.
Other surfaces
Same job, different tooling
- Reach documents with an MCP server →
Top MCPs ranks the productivity MCP servers — Notion and more — that give an agent access to the docs and knowledge bases where content lives, before these Skills convert it to a real file.
Related
More best-skill guides
- Best Claude Code & agent Skills, by use case
The hub — the best Skill for every developer job, in one place.
- Best agent Skills for code review
The best agent Skills for reviewing code before merge — structured, severity-tagged reviews and security-first passes across Claude Code, Cursor, and Codex.
- Best agent Skills for security & auditing
The best agent Skills for security review and vulnerability detection — firm-grade static analysis (CodeQL, Semgrep), Kubernetes hardening, and real-exploit validation.
- Best agent Skills for testing & QA
The best agent Skills for writing and running tests — Playwright web-app testing, browser automation patterns, and test-driven-development workflows.
- Best agent Skills for DevOps & infrastructure
The best agent Skills for DevOps work — CI/CD, containers, blue-green and canary deploys, Terraform, and Kubernetes hardening across AWS, GCP, and Azure.
New to Skills? What is a Claude skill · How to install · Skills vs subagents vs MCP vs plugins