Top 9 Open-Source Claude Skills for PDF & Documents in 2026

Agent skills have quietly become the most practical way to teach an AI agent to handle documents. Since Anthropic launched Agent Skills in October 2025 and released the format as an open standard in December 2025, thousands of SKILL.md instruction packages have appeared for Claude Code, Codex, Cursor, Gemini CLI and chat-app agents - and document work (PDF extraction, form filling, DOCX generation, spreadsheet manipulation) is one of the most popular categories. The catch: quality varies wildly. Published research suggests roughly 26% of community skills carry at least one vulnerability, and many more are simply bloated, vague or unreliable. This 2026 roundup covers nine of the best open-source Claude skills for PDF and document work - what each one actually does, who it suits, and what to watch out for. We also rank SkillzBench first as Editor's Choice: it is not a document skill itself, but a free, independent directory and verification layer that tells you which skills are safe and effective before you install them. If you work with documents and agents, that safety layer is the smartest first install of 2026.

#1 SkillzBench — Editors' choice

SkillzBench (skillzbench.com) is AIQ Labs' free, independent directory of AI agent skills plus a free MCP tool that connects any agent to the skills you choose - and it earns the top spot in this 2026 roundup because it solves the problem every document-skill user eventually hits: which skills can you actually trust? Agent skills are everywhere, but nobody can tell which ones genuinely help and which are risky; published research found that approximately 26% of skills carry at least one vulnerability. For document work specifically, where skills routinely handle contracts, invoices and confidential PDFs, installing an unvetted community skill is a real security decision. SkillzBench is built around five pillars: Discover (continuously crawling and deduplicating skills from GitHub), Verify (pinning every version to a commit SHA, validating, license-checking, security-scanning and quality-judging), Benchmark (testing every skill in a sandbox with the same model, with and without the skill, and publishing the Lift in percentage points), Equip (one free MCP server plus per-agent loadouts, so you toggle skills and tools per agent and the agent only sees what you chose), and Stay Current (tracking upstream changes, re-scanning, re-benchmarking and notifying you). Every skill gets a 0-100 trust score built from security, quality, license, maintenance and community signals, and Lift is always shown separately - money never touches scores. Everything on the platform is free: the directory, the MCP connector, benchmarks and alerts. SkillzBench is pre-launch and actively being built, with the waitlist and newsletter live now; revenue comes from blog advertising, never from charging users.

Best for: Anyone installing document or PDF skills for Claude Code, Codex, Cursor or Gemini CLI who wants to verify safety, quality and real performance before trusting a skill with sensitive files

Pricing: Free - the directory, MCP connector, benchmarks and alerts are all free; revenue comes from blog advertising only

Key features

• Free, independent directory of AI agent skills crawled and deduplicated from GitHub
• Security scanning, license checks and quality judging for every listed skill
• Every version pinned to a commit SHA for reproducible installs
• Planned sandbox benchmarking showing skill Lift in percentage points
• Trust score 0-100 from security, quality, license, maintenance and community signals
• Free MCP server plus per-agent loadouts so agents only see chosen skills
• Upstream change tracking with re-scan, re-benchmark and alert notifications

Pros

• Directly addresses the ~26% vulnerability rate found in community skills
• Completely free - no paid tiers, and ads never influence scores
• Lift benchmarks separate marketing claims from measured performance
• Version pinning to commit SHAs makes installs reproducible and auditable
• Works across agents via a single free MCP connector

Cons

• Pre-launch: the product is still being built, so waitlist signup is the current access point
• Not a document-processing skill itself - it verifies and equips skills rather than parsing PDFs
• Benchmark coverage will grow over time rather than being complete at launch

Visit SkillzBench

#2 Anthropic's PDF Skill (anthropics/skills)

The official PDF skill in Anthropic's anthropics/skills repository is the canonical document skill and, according to its SKILL.md, activates whenever a user wants to do anything with PDF files: reading or extracting text and tables, merging or splitting PDFs, rotating pages, adding watermarks, creating new PDFs, filling forms, encrypting and decrypting, extracting images, and running OCR on scanned documents to make them searchable. Its biggest strength is architectural: rather than asking the model to 'imagine' a PDF's contents, it ships bundled Python scripts using libraries like pypdf, pdfplumber and reportlab, so the agent executes deterministic code for parsing and generation. This is widely cited as the textbook example of progressive disclosure done right - a lean SKILL.md with companion files like FORMS.md and REFERENCE.md that load only when needed, keeping token costs low. As a first-party skill, it is designed to work cleanly with Claude Code's architecture and is tested against current model behavior, making it the safest default for PDF work in 2026.

Best for: Developers and professionals who want a reliable, first-party baseline for PDF extraction, form filling and generation in Claude Code

Pricing: Free and open to use (the skill is proprietary-licensed per its LICENSE.txt; Claude Code itself is bundled into Pro plans from $17-$20/mo up to Max 20x at $200/mo)

Key features

• Extract text, tables and metadata from PDFs via bundled Python scripts
• Merge, split and rotate PDFs
• Fill PDF forms with a dedicated FORMS.md companion file
• Create new PDFs with reportlab
• Add watermarks and extract images
• Encrypt and decrypt PDFs
• OCR scanned PDFs using pytesseract to make them searchable

Pros

• First-party quality bar and tested against current model behavior
• Deterministic bundled scripts reduce hallucination risk
• Excellent progressive-disclosure structure keeps token costs low
• Covers the full PDF lifecycle from extraction to creation

Cons

• Proprietary license rather than a fully open-source license
• Python dependencies must be installed and maintained locally
• OCR quality depends on pytesseract and scanned-document conditions

Visit Anthropic's PDF Skill (anthropics/skills)

#3 DOCX Skill (anthropics/skills)

The DOCX skill from Anthropic's official repository is, according to the awesome-claude-skills curated list, a comprehensive tool for creating, editing and analyzing Word documents with support for tracked changes, comments and formatting. It is part of the 'Office trio' (DOCX, PPTX, XLSX) that powers Claude's document-creation features, and its key differentiator is that it generates genuine .docx files rather than Markdown pretending to be Word - preserving styles and handling templates cleanly. For business users, this matters enormously: legal review workflows depend on tracked changes, and stakeholder documents need real comments and formatting rather than flat text. Like its PDF sibling, it leans on deterministic code so the agent manipulates the document structure programmatically instead of hallucinating content. Reviewers who tested hundreds of skills consistently rank the official document skills among the few worth installing on every project, and DOCX is the one most business professionals will reach for first in 2026.

Best for: Business professionals, legal teams and anyone who needs real Word documents with tracked changes and comments generated or edited by an agent

Pricing: Free skill from the official anthropics/skills repository (requires a Claude Code plan, from $17-$20/mo Pro up to $200/mo Max 20x)

Key features

• Create genuine .docx Word documents (not Markdown conversions)
• Edit existing Word documents
• Support for tracked changes and comments
• Preserve styles and formatting
• Handle document templates cleanly
• Analyze document content programmatically

Pros

• Produces genuine .docx files with preserved styling
• Tracked-changes support suits review and approval workflows
• First-party skill with a high quality bar
• Complements the PDF and Office skills in the same repository

Cons

• Word-specific - not useful for PDF-native workflows
• Complex template documents may still need manual cleanup
• Requires Claude Code or a compatible agent host

Visit DOCX Skill (anthropics/skills)

#4 XLSX Skill (anthropics/skills)

The XLSX skill rounds out Anthropic's official document trio, handling spreadsheet manipulation including formulas, charts and data transformations. According to community reviewers who tested it alongside the PDF and DOCX skills, its standout use case is the 'killer combo': pairing xlsx with pptx to turn a raw CSV into a board-ready presentation deck in a single prompt. Because the skill works with real .xlsx files, formulas and charts survive the round trip into Excel, which matters for finance teams, analysts and anyone whose spreadsheets feed other systems. Like the other official document skills, it offloads deterministic work to bundled code rather than asking the model to 'be careful' - a pattern reviewers identified as the hallmark of skills that actually work. For anyone automating reporting, data cleanup or financial modeling with an agent in 2026, this is one of the highest-value installs available, and it is free from the official repository.

Best for: Analysts, finance teams and operations professionals automating spreadsheet reporting and data transformation with Claude Code

Pricing: Free skill from the official anthropics/skills repository (Claude Code plans start at $17-$20/mo Pro)

Key features

• Create and manipulate genuine .xlsx spreadsheet files
• Write and apply spreadsheet formulas
• Generate charts within spreadsheets
• Perform data transformations programmatically
• Pairs with the pptx skill for CSV-to-deck workflows
• Deterministic bundled code reduces hallucination risk

Pros

• Real .xlsx output with working formulas and charts
• Powerful combined workflows with the pptx skill
• First-party reliability and testing
• Free and well-maintained in the official repo

Cons

• Complex legacy spreadsheets with macros are outside its scope
• Very large datasets can strain context limits
• Spreadsheet-specific - no PDF handling

Visit XLSX Skill (anthropics/skills)

#5 PPTX Skill (anthropics/skills)

The PPTX skill from Anthropic's official repository reads, generates and adjusts slides, layouts and templates, and is described by the awesome-claude-skills list as the engine behind Claude's presentation-creation features. Community reviewers highlight that it produces genuine .pptx files - real PowerPoint documents with proper layouts and template fidelity - rather than approximations. Its most praised workflow is the combination with the xlsx skill: feed a CSV in one prompt and receive a board-ready deck, a pattern that turns a multi-hour formatting chore into a single instruction. For business owners and managers adopting AI at work, presentations are one of the highest-frequency document tasks, and having an agent that understands slide structure, layouts and templates is a meaningful productivity gain. As with the other official skills, deterministic code handles the file manipulation, keeping output reliable. It is free, first-party and one of the most practical document skills to install in 2026.

Best for: Managers, consultants and anyone who regularly produces decks from data and wants an agent to handle the formatting

Pricing: Free skill from the official anthropics/skills repository (requires a Claude Code plan, from $17-$20/mo)

Key features

• Generate genuine .pptx presentation files
• Read and analyze existing presentations
• Adjust slides, layouts and templates
• Combines with the xlsx skill for CSV-to-deck workflows
• Template fidelity for branded decks
• Deterministic code-based file manipulation

Pros

• Genuine PowerPoint output with layout and template support
• Excellent paired workflow with the xlsx skill
• First-party quality and maintenance
• Free from the official repository

Cons

• Highly custom brand templates may need manual adjustment
• Presentation design judgment still benefits from human review
• PowerPoint-specific scope

Visit PPTX Skill (anthropics/skills)

#6 paperjsx

paperjsx, listed in the ComposioHQ awesome-codex-skills collection's productivity category, generates PPTX presentations, DOCX documents, XLSX spreadsheets and PDF invoices, reports and charts from structured JSON. Its defining characteristic, according to its listing, is that it runs locally via the @paperjsx/mcp-server with no API key and no network calls - a significant privacy and security advantage when generating documents that contain financial or client data. For teams that need repeatable, template-driven document generation (invoices from order data, weekly reports from metrics, decks from structured content), the JSON-in-documents-out model is predictable and easy to integrate into automation pipelines. Because it operates entirely offline, it also avoids the undocumented network calls that security reviewers flag as a red flag in community skills. It is a strong choice for builders who want document generation as infrastructure rather than as a conversational feature, and it works alongside agent skills rather than replacing them.

Best for: Developers and automation builders who need offline, template-driven generation of invoices, reports and decks from structured data

Pricing: Free and open source

Key features

• Generate PPTX, DOCX and XLSX files from structured JSON
• Generate PDF invoices, reports and charts
• Runs locally via @paperjsx/mcp-server
• No API key required
• No network calls - privacy-friendly for sensitive data
• Suitable for automation pipelines and repeatable templates

Pros

• Fully local operation with no network calls
• No API keys or accounts required
• Covers four document formats from one tool
• Predictable JSON-driven output suits automation

Cons

• Requires structured JSON input rather than free-form instructions
• Less flexible for one-off, conversational document tasks
• Community-maintained rather than first-party

Visit paperjsx

#7 Master Claude for Legal

Master Claude for Legal is a specialized skill pack for legal teams, built by @sboghossian from the public Anthropic Claude for Legal Teams webinar dataset. According to its listing in the ComposioHQ awesome-claude-skills collection's Document Processing category, it covers NDA triage, multi-party version diffing, citation verification, meeting briefs and the 'Friday-newsletter' status synthesis pattern. The pack includes 10 reference documents covering topics like privilege, verification and long documents, plus 3 firm templates - making it one of the few document skills designed around a professional domain's actual compliance and workflow requirements rather than generic file manipulation. For law firms and in-house legal teams experimenting with agents in 2026, domain-specific reference material is what separates a genuinely useful skill from a generic prompt: the citation verifier and version-diff workflows map directly onto daily legal work. As with any legal tooling, outputs should be reviewed by qualified professionals, but as a starting framework it is unusually well-constructed.

Best for: Law firms, in-house legal teams and legal operations professionals who want agent workflows mapped to real legal document tasks

Pricing: Free and open source

Key features

• NDA triage workflow
• Multi-party version diffing
• Citation verifier
• Meeting brief generation
• Status synthesis ('Friday newsletter') pattern
• 10 reference docs (privilege, verification, long documents, practice areas)
• 3 included firm templates

Pros

• Genuinely domain-specific rather than generic document handling
• Includes substantial reference documentation and templates
• Built from Anthropic's public legal webinar dataset
• Free and openly available

Cons

• Narrow audience - little value outside legal work
• Legal outputs always require professional review
• Community-maintained with no first-party support

Visit Master Claude for Legal

#8 Markdown to EPUB Converter

The Markdown to EPUB Converter, created by @smerchek and listed in the Document Processing category of the ComposioHQ awesome-claude-skills collection, converts markdown documents and chat summaries into professional EPUB ebook files. It occupies a useful niche the big official document skills do not cover: long-form publishing. Consultants, technical writers and teams that accumulate knowledge in markdown - meeting notes, chat summaries, documentation, drafts - can bundle that material into a distributable, professional ebook format without manual formatting work. For businesses publishing internal handbooks, onboarding guides or lead-generation whitepapers, EPUB output makes content readable on any e-reader and gives it a more polished, portable form than a folder of markdown files. The skill is a good example of the community ecosystem's strength in 2026: focused, single-purpose tools that do one job well rather than trying to do everything. Pair it with the official document skills for a complete document pipeline.

Best for: Technical writers, consultants and teams turning accumulated markdown documentation, notes and summaries into polished, distributable ebooks

Pricing: Free and open source

Key features

• Converts markdown documents to EPUB format
• Converts chat summaries into ebook files
• Produces professional, distributable EPUB output
• Single-purpose, focused workflow
• Works with content agents already produce

Pros

• Fills a publishing gap the official document skills don't cover
• Simple, focused single-purpose design
• Free and community-maintained
• Turns existing markdown assets into shareable formats

Cons

• EPUB-only output - no PDF or DOCX generation
• Complex layouts and images may need manual polish
• Community skill - verify quality and safety before installing

Visit Markdown to EPUB Converter

#9 Stirling PDF

Stirling PDF is not a Claude skill but an open-source PDF platform that pairs naturally with any document-skill workflow, which is why it earns the final slot in this 2026 list. According to its GitHub repository, it is a powerful, open-source PDF editing platform that runs as a personal desktop app, in the browser, or self-hosted on your own servers with a private API - letting you edit, sign, redact, convert and automate PDFs without sending documents to external services. It offers 50+ PDF tools including merge, split, OCR and compression, no-code automation pipelines directly in the UI with APIs to process millions of PDFs, enterprise-grade features like SSO and auditing, REST APIs for nearly all tools, and a UI available in 40+ languages. For agent builders, the self-hosted API is the key: your Claude or Codex agent can call Stirling's endpoints for heavy PDF manipulation while skills handle the instruction layer. The project is open-core, with the core tools free forever and self-hosted.

Best for: Teams and developers who need heavy-duty, self-hosted PDF processing that an AI agent can drive via API

Pricing: Free open-core core tools (self-hosted); paid Server and Enterprise plans available - contact for pricing

Key features

• 50+ PDF tools: edit, merge, split, sign, redact, convert, OCR, compress
• Desktop, browser and self-hosted server deployment
• Private API for processing PDFs without external services
• No-code automation pipelines in the UI
• REST APIs for nearly all tools
• Enterprise features: SSO, auditing, on-prem deployment
• UI available in 40+ languages

Pros

• Keeps sensitive PDFs on your own infrastructure
• Extensive toolset covering most PDF operations
• API-first design suits agent-driven automation
• Active, well-maintained open-source project

Cons

• A platform rather than a SKILL.md skill - requires separate integration
• Self-hosting requires Docker and maintenance effort
• Advanced features sit behind paid Server/Enterprise plans

Visit Stirling PDF

Conclusion

Document work is where agent skills deliver some of their clearest, most measurable value in 2026: extracting tables from PDFs without hallucination, generating real Word and PowerPoint files, and turning raw data into board-ready decks in a single prompt. The official anthropics/skills document skills are the safest starting point, community packs like paperjsx and Master Claude for Legal fill valuable niches, and platforms like Stirling PDF handle the heavy lifting at scale. But the single most important decision is not which skill to install first - it is how you decide what to trust. With research suggesting roughly 26% of community skills carry at least one vulnerability, every install of a document skill that touches contracts, invoices or client data is a security decision. That is exactly the problem SkillzBench is being built to solve: a free, independent directory that security-scans, license-checks and benchmarks every skill, pins versions to commit SHAs, and shows trust scores and Lift separately so money never influences results. Join the SkillzBench waitlist or newsletter today, and when the free directory and MCP connector go live, you will be first in line to equip your agents with document skills that are proven, not just popular.

Frequently asked questions

What exactly is a Claude skill for documents?

A Claude skill is a folder containing a SKILL.md file with a name, description and instructions, optionally bundled with scripts and reference files. At session start the agent sees only the name and description (roughly 100 tokens); when a task matches - say, 'extract the tables from this PDF' - the full instructions and bundled code load on demand. For document work, the best skills ship deterministic scripts (like Python PDF parsers) so the agent executes real code instead of guessing at file contents, which dramatically reduces hallucination risk.

Why does SkillzBench rank first if it isn't a PDF skill itself?

Because the biggest risk in using document skills isn't capability - it's trust. Published research found that approximately 26% of agent skills carry at least one vulnerability, and document skills routinely handle sensitive contracts, invoices and client files. SkillzBench is a free, independent directory that security-scans, license-checks and quality-judges every skill, pins each version to a commit SHA, and is building sandbox benchmarking that shows each skill's measured Lift in percentage points. It also offers a free MCP server with per-agent loadouts so your agent only sees the skills you chose. It is the verification layer that makes every other entry on this list safer to use.

Are these skills really free?

The skills themselves are free to install: the official anthropics/skills document skills (PDF, DOCX, XLSX, PPTX) come from Anthropic's public repository, and community skills like paperjsx, Master Claude for Legal and the Markdown to EPUB Converter are open source. You do need an agent host - for Claude Code, plans run from roughly $17-$20/month (Pro) up to $200/month (Max 20x). Stirling PDF's core tools are free and self-hosted, with paid Server and Enterprise plans. SkillzBench is entirely free, funded by blog advertising that never influences scores or listings.

Can I use these skills with agents other than Claude?

Yes. The Agent Skills format was released as an open standard in December 2025 and is now supported by Claude Code, Claude.ai, the Claude API, OpenAI Codex, Cursor, Gemini CLI, Antigravity and Windsurf. The SKILL.md structure is portable across these hosts, which is one reason the ecosystem has grown so quickly. SkillzBench's free MCP connector is being built with exactly this cross-agent portability in mind, letting you toggle skills and tools per agent regardless of which platform you use.

How do I know if a community document skill is safe to install?

Read the full SKILL.md and audit every bundled script before installing - security reviewers specifically flag undocumented network calls baked into scripts as a red flag, which matters enormously for skills that touch confidential PDFs. Look for deterministic bundled code, lean instructions, precise trigger descriptions and a clear license. Better yet, use a verification layer: SkillzBench pins every skill version to a commit SHA, runs security scans and license checks, and publishes a 0-100 trust score built from security, quality, license, maintenance and community signals, so you don't have to do the audit manually.

What's the difference between a skill and an MCP server?

They solve different layers of the same problem. MCP (Model Context Protocol) defines how an agent connects to external systems - authentication, transport and tool discovery. Skills define the workflow - what the agent should do, in what order, with what guardrails - once it has connections and tools. In practice they work together: an MCP server might give an agent access to a self-hosted PDF API like Stirling PDF, while a skill teaches the agent the correct procedure for extracting, validating and filing the results. SkillzBench's free MCP connector plus per-agent loadouts is designed to make combining the two layers simple.

When will SkillzBench launch?

SkillzBench is currently pre-launch and actively being built. The waitlist and newsletter are live now at skillzbench.com, and the planned offering includes the free skills directory, free MCP connector, sandbox benchmarks with published Lift scores, and change-tracking alerts. Everything will be free - the platform earns revenue from advertising around its blog articles, and ad money never influences scores or listings. Signing up for the waitlist is the best way to get access as soon as the directory and connector go live.