Top 7 Open-Source Claude Skills for Word & Reports in 2026

Agent skills have quietly become the fastest way to teach Claude — and other AI agents — how to produce polished Word documents and reports. Since Anthropic introduced the Agent Skills format in October 2025 and released it as an open standard in December 2025, thousands of SKILL.md instruction packages have appeared on GitHub, covering everything from .docx generation to tracked changes and report templating. But the ecosystem's growth has created a new problem: quality and safety vary wildly. Published research suggests roughly 26% of community skills carry at least one vulnerability, and many bloated or vague skills waste tokens or trigger at the wrong moment. That's why this 2026 list pairs the best open-source skills and tools for Word and report work with a way to evaluate them properly. At the top sits SkillzBench, a free, independent directory and verification platform that benchmarks skills so you know which ones actually lift your agent's performance — and which ones to avoid. Whether you're a developer automating report pipelines, a business professional drafting client deliverables, or an agent builder assembling a document workflow, these seven picks cover the full journey: discovering skills, verifying them, and putting them to work on real .docx and report tasks.

#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 Editor's Choice spot because it solves the problem every other entry on this list creates: how do you know a skill is safe, current, and actually worth installing? Agent skills are everywhere, but nobody can tell which ones genuinely help. SkillzBench is being built around five pillars. Discover: continuously crawl and dedupe skills from GitHub so the directory stays fresh. Verify: pin every skill version to a commit SHA, validate it, check its license, run security scans, and judge quality — a direct answer to the research finding that around 26% of skills carry at least one vulnerability. Benchmark: test every skill in a sandbox with the same model, with and without the skill, and publish the Lift in percentage points, so performance evidence is always shown separately from ratings. Equip: one free MCP server plus per-agent loadouts, where you toggle skills and tools per agent and the agent only sees what you chose. Stay current: track upstream changes, re-scan, re-benchmark, and notify you when something shifts. Every skill gets a trust score from 0 to 100 built from security, quality, license, maintenance, and community signals — and money never touches scores. Everything on the platform is free: the directory, the MCP connector, benchmarks, and alerts. SkillzBench is currently pre-launch and actively being built; the waitlist and newsletter are live now, so joining early is the way to be first in line when the directory and benchmarking tools go public.

Best for: Business owners adopting AI at work, non-technical AI-curious professionals, developers and AI coders, prompt engineers, and agent builders who want to find, verify, and safely equip agent skills for Word and report workflows

Pricing: Free (waitlist live; all directory, MCP, benchmark, and alert features will be free)

Key features

• Free, independent directory of AI agent skills crawled and deduped from GitHub
• Security scanning, license checks, validation, and quality judging for every skill
• Every skill version pinned to a commit SHA for reproducibility
• Sandbox benchmarking that publishes Lift in percentage points, with and without the skill
• Trust score from 0-100 based on security, quality, license, maintenance, and community signals
• Free MCP connector with per-agent loadouts — agents only see the skills and tools you toggle on
• Upstream change tracking with re-scans, re-benchmarks, and alerts
• Everything free; ad revenue from blog content never influences scores or listings

Pros

• Independent and free — ad-supported, so money never influences scores
• Benchmarks measure actual performance lift, not just ratings
• Security scanning addresses the ~26% vulnerability rate found in published research
• Per-agent loadouts give precise control over what each agent can see
• Trust scores separate security and quality signals from marketing claims

Cons

• Pre-launch — the product is still being built, so the directory and MCP are not yet publicly available
• Benchmark results are not yet published while the platform is in development
• You'll need to join the waitlist to get access when it launches

Visit SkillzBench

#2 Anthropic DOCX Skill

The official docx skill from Anthropic's skills repository is the reference implementation for Word document work in Claude, and it's the skill behind Claude's document-creation features. According to its documentation, it can create, edit, and analyze Word documents end-to-end — producing reports, memos, letters, and templates as polished .docx deliverables. What makes it stand out is its engineering discipline: rather than asking the model to 'imagine' a document, it ships deterministic scripts. Creating a new document uses a docx (npm) script with documented gotchas for page sizes, table widths, and lists; editing an existing file follows a careful unzip-edit-rezip workflow with merge_runs.py to coalesce fragmented XML runs; and validation scripts check output against the OOXML schema, including checks that tracked changes are properly wrapped. It also handles comments across the six cross-linked XML files Word requires, and supports redlining with author attribution. The skill is indexed on directories like AgenticSkills with an S-rank, drawn from the anthropics/skills repository, which carries roughly 180,000 GitHub stars and is Apache-2.0 licensed. For anyone building report workflows in Claude Code, this is the safe default the community consistently recommends first.

Best for: Claude Code users, developers, and teams who want a battle-tested, first-party skill for producing and revising Word reports

Pricing: Free (open source)

Key features

• Create, edit, and analyze .docx files end-to-end
• Tracked changes with author attribution and validation checks
• Comments with proper anchoring across Word's XML comment files
• Deterministic bundled scripts (docx-js creation, merge_runs.py, validate.py, accept_changes.py)
• Headings, tables of contents, page numbers, and image insertion
• Find-and-replace and content extraction from existing Word files
• Apache-2.0 licensed and sourced from the official anthropics/skills repo

Pros

• First-party from Anthropic — the safest default per community guides
• Deterministic scripts reduce hallucination risk in document output
• Handles the hard parts of OOXML: tracked changes, comments, validation
• Free and Apache-2.0 licensed

Cons

• Editing existing documents requires an unzip/edit/rezip workflow rather than direct opening
• Requires bundled dependencies (docx npm package, pandoc, LibreOffice, pdftoppm)
• Word/.docx only — PDF and spreadsheet tasks need the separate pdf and xlsx skills

Visit Anthropic DOCX Skill

#3 WordCraft

WordCraft, by the ArtCraft team, is an open-source, clean-room reimplementation of Microsoft Word built in pure Rust — and it's notable here because it was explicitly designed to be driven by AI agents as well as people. According to its GitHub repository, it offers the Word workflow you already know: the ribbon, styles, tables, track changes, references, and mail merge. It reads and writes .docx (plus .odt, .rtf, .html, .md, .tex, and .txt, with Word 97-2003 .doc import and PDF export), runs natively on macOS, Windows, Linux, and BSD, and in the browser via WebAssembly — with no Electron or Tauri. Performance is a headline claim: cached paragraph layout means typing in a 188-page document re-lays it out in about 1.4 ms. The agent story is what earns it a place on this list: every action is one of 422 commands with an ID, and the same commands drive the ribbon, keyboard shortcuts, a command-line tool, a JSON control channel, and an MCP server. You can add it to Claude with 'claude mcp add wordcraft', confine the agent to specific folders with read/write roots, and agents can verify their work through inspect_document without screenshots. It's dual-licensed MIT OR Apache-2.0, and everything works offline.

Best for: Developers and agent builders who want Claude to create and edit real Word documents headlessly, and privacy-focused users who want a fully offline, open-source Word workflow

Pricing: Free (open source)

Key features

• Full word processor with ribbon, styles, tables, track changes, references, and mail merge
• Reads/writes .docx plus .odt, .rtf, .html, .md, .tex, .txt; PDF export with real text and bookmarks
• Built-in MCP server and CLI so Claude and other agents can drive the app
• 422 agent-addressable commands with inspect_document for agent self-verification
• Folder confinement flags (--automation-read-root/--automation-write-root) for safe agent automation
• Native Rust/egui on macOS, Windows, Linux, BSD, and the web — no Electron
• Offline spelling, grammar, and thesaurus; MIT OR Apache-2.0 clean-room license

Pros

• Purpose-built agent integration via MCP and CLI
• Extremely fast layout engine (about 1.4 ms re-layout on a 188-page doc)
• Broad file format support including .docx round-tripping
• Clean-room, permissively dual-licensed, fully offline

Cons

• As a young clean-room project, feature parity with Word is still in progress (its ROADMAP and parity assessment are published honestly)
• Rust build may be a hurdle for non-developers
• Spelling and grammar support currently covers English and Polish

Visit WordCraft

#4 Awesome Claude Skills (ComposioHQ)

The ComposioHQ awesome-claude-skills repository is a curated list of 1,000+ production-ready Claude Skills and plugins, and it's the best starting point for browsing document and report skills beyond Anthropic's official set. According to its README, the list covers not just Claude.ai and Claude Code but also coding agents like Codex, Cursor, Gemini CLI, Antigravity, and more. Its Document Processing section is directly relevant to Word and report work: alongside the official docx, pdf, pptx, and xlsx skills, it lists community entries like a Markdown to EPUB Converter and a 'Master Claude for Legal' skill pack with NDA triage, multi-party version diffing, citation verification, and firm templates. The repo also explains the skills format clearly — what a SKILL.md is, how progressive disclosure works, and how skills differ from MCP servers and tools. Beyond discovery, it showcases Composio's MCP Gateway and a connect-apps plugin that lets Claude take real actions (send emails, create issues, post to Slack) across 1,000+ apps with built-in authentication and audit logs. As a community-curated list, quality varies by entry, which is exactly why pairing it with a verification layer like SkillzBench makes sense.

Best for: Developers and prompt engineers who want to browse a broad catalog of document and report skills across multiple coding agents

Pricing: Free (open source; Composio offers a free API key tier)

Key features

• Curated list of 1,000+ Claude Skills across document processing, development, data, business, and more
• Document Processing section with docx, pdf, pptx, xlsx, and community report skills
• Cross-agent compatibility notes for Claude Code, Codex, Cursor, Gemini CLI, and others
• Clear explainer on SKILL.md structure and progressive disclosure
• connect-apps plugin for real actions across 1,000+ apps via Composio
• Community-contributed and open to GitHub PRs

Pros

• Largest curated skills list, with a dedicated document-processing category
• Works across many agents, not just Claude
• Includes practical getting-started guides and skill-creation resources
• Free and community-maintained

Cons

• Curated but not individually security-scanned or benchmarked — quality varies by entry
• Community skills require manual auditing of bundled scripts before install
• Star counts and listings can drift from actual skill maintenance status

Visit Awesome Claude Skills (ComposioHQ)

#5 skill-creator

skill-creator is Anthropic's official meta-skill — a skill for building skills — and it's the right pick if none of the existing Word or report skills quite fit your organization's reporting workflow. According to its description in the official repository and multiple 2026 roundups, it covers creating new skills, modifying existing skills, measuring skill performance, running evaluations, benchmarking variance, and improving descriptions so Claude triggers the right skill more accurately. That last part matters more than most people realize: community guides consistently identify the description field as the part most people get wrong, and a vague description means the skill either never activates or fires on the wrong tasks. skill-creator enforces the frontmatter contract, suggests good description wording, and scaffolds bundled files like scripts/ and references/ so your custom skill follows the progressive-disclosure pattern — lean SKILL.md, fat reference files. Reviewers often recommend pairing it with mcp-builder, which generates MCP servers from a description, to create a self-bootstrapping agent toolkit. If your team has a specific report template, review checklist, or document convention that no public skill captures, this is the tool to encode it once and reuse it forever.

Best for: Teams and prompt engineers who need to turn their own report templates, style guides, or document conventions into reusable agent skills

Pricing: Free (open source)

Key features

• Creates and modifies SKILL.md-based skills with correct frontmatter
• Suggests effective description wording for accurate skill triggering
• Scaffolds bundled scripts and reference files following progressive disclosure
• Supports measuring skill performance, running evaluations, and benchmarking variance
• Official Anthropic skill from the anthropics/skills repository
• Pairs naturally with mcp-builder for building agent tool ecosystems

Pros

• First-party and trusted — community guides call it the one skill to install before any other
• Encodes institutional knowledge into permanent, versionable files
• Improves trigger accuracy, reducing token waste from misfiring skills
• Free and Apache-2.0 licensed

Cons

• It's a builder, not a finished Word skill — you still have to define your workflow
• Custom skills you create won't have third-party verification unless you audit them
• Requires some comfort with markdown structure and agent concepts

Visit skill-creator

#6 Doc Co-Authoring Skill (Anthropic)

Doc Co-Authoring is an Anthropic skill aimed at the messier, human side of report writing: structured co-authoring of documentation like PRDs, design docs, RFCs, decision docs, and technical specs. According to its listing on AgenticSkills, it guides context gathering, iterative refinement, and reader-testing — a workflow discipline that addresses the most common failure mode in AI-assisted reports, which is an agent charging ahead on wrong assumptions and producing a polished document built on shaky ground. Rather than generating a full document in one shot, the skill structures the process: gather the context that exists, draft iteratively with the human in the loop, and test whether the intended reader actually understands the result. It carries an A-rank on AgenticSkills and comes from the same anthropics/skills repository as the docx skill (roughly 180,000 GitHub stars, Apache-2.0 licensed), so it inherits the same first-party trust signals. For teams producing reports that need stakeholder buy-in — proposals, decision docs, project briefs — this skill complements the mechanical docx skill: one handles the file format, the other handles the thinking and revision process that makes the content worth formatting.

Best for: Product managers, team leads, and professionals who produce reports and decision documents that require stakeholder review and iteration

Pricing: Free (open source)

Key features

• Structured workflow for co-authoring PRDs, design docs, RFCs, and decision docs
• Guided context gathering before drafting begins
• Iterative refinement with the human in the loop
• Reader-testing step to verify the document lands with its audience
• From the official anthropics/skills repository (Apache-2.0)
• Complements the docx skill for end-to-end report production

Pros

• First-party skill with strong trust signals
• Addresses the real workflow of report writing, not just file generation
• Reduces the 'confident but wrong' failure mode of one-shot AI drafting
• Free and open source

Cons

• Focused on documentation types (PRDs, RFCs) rather than formatted .docx output specifically
• More process-heavy than users who want instant one-shot generation may prefer
• Best results require a written spec or context to start from

Visit Doc Co-Authoring Skill (Anthropic)

#7 NovoDocx

NovoDocx is an open-source .NET library and service for generating documents from standard Word templates, created by a developer who couldn't find a workable open-source option for complex document generation. According to its author's write-up on bycode.dev, the project emerged from real-world client work involving complicated templates — documents composed of multiple tables and sections comparing financial products, their impacts, benefits, and downsides. The existing out-of-the-box options (Word templates in Dynamics CRM/Power Platform, or the premium Word Online Connector) had significant limitations: table-bound templates, premium connector costs, and schema issues when deploying flows across environments. NovoDocx is designed to be simple to use and simple to host, requiring no knowledge of the underlying Word XML format, and ships as both a library and an Azure Functions App you can deploy however you like. The source code is hosted on GitHub, and the author states the current features are enough to handle complex documents with good performance, including many edge cases. For developers building automated report-generation pipelines — contracts, statements, comparison reports — it fills a genuine gap in the open-source .NET landscape, though it's a code-level tool rather than an agent skill you install into Claude.

Best for: .NET developers building automated report-generation pipelines from Word templates, especially in Microsoft-centric environments

Pricing: Free (open source)

Key features

• Open-source .NET library for generating documents from Word templates
• Handles complex templates with multiple tables and sections
• Ships as a library plus an Azure Functions App for flexible hosting
• No knowledge of the Word XML format required
• Designed for edge cases in real-world document generation
• Source code available on GitHub

Pros

• Fills a real gap: no comparable open-source .NET Word templating library existed
• Proven against genuinely complex real-world templates
• Flexible hosting — use the library directly or the Azure Functions App
• Free and open source

Cons

• A code-level library, not an installable Claude skill — no SKILL.md or agent integration
• First iteration of the project; hosting options and functionality are still being enriched
• Requires .NET development skills to use

Visit NovoDocx

Conclusion

Open-source Claude skills have made Word and report automation genuinely practical in 2026 — Anthropic's official docx skill handles the file format with real engineering rigor, WordCraft gives agents a full word processor to drive over MCP, and community catalogs like awesome-claude-skills surface hundreds of specialized options. But the same explosion that makes this ecosystem exciting also makes it risky: with research suggesting roughly 26% of skills carry at least one vulnerability, and quality varying from excellent to token-burning noise, the real skill in 2026 is knowing which skills to trust. That's exactly the gap SkillzBench is being built to close — a free, independent directory that crawls and dedupes skills from GitHub, pins every version to a commit SHA, runs security scans and license checks, benchmarks each skill in a sandbox to publish its actual performance Lift, and gives you a free MCP connector with per-agent loadouts so your agent only ever sees the skills you've chosen. Everything will be free, and ad revenue will never touch scores. The product is still in development, but the waitlist and newsletter are live now. If Word and report workflows are part of your AI plans, join the waitlist at skillzbench.com today — and when it launches, browse, verify, and equip your document skills with evidence instead of guesswork. Skills, proven.

Frequently asked questions

What exactly is a Claude skill, and how does it differ from an MCP server?

A Claude skill is a reusable instruction package — a folder containing a SKILL.md file with YAML frontmatter (name and description) plus optional scripts and reference files — that teaches an agent how to handle a specific class of tasks. Skills load progressively: at session start the agent only sees each skill's name and description (roughly 100 tokens), and the full instructions load only when the task matches. MCP, by contrast, defines how an agent connects to external systems — authentication, transport, and tool discovery. In short: MCP provides access, tools perform actions, and skills define the workflow. In production, all three layers work together.

Are open-source Claude skills safe to install?

Not automatically. Published research has found that approximately 26% of agent skills carry at least one vulnerability, and community guides consistently warn about skills with undocumented network calls or opaque bundled scripts. Best practice is to prefer first-party sources like the official anthropics/skills repository, read the full SKILL.md, and audit every bundled script before installing community skills. This is the problem SkillzBench is being built to solve: it will pin every skill version to a commit SHA, run security scans and license checks, and publish a 0-100 trust score based on security, quality, license, maintenance, and community signals — so you don't have to audit everything by hand.

How do I know if a Word or report skill actually improves my agent's output?

The honest answer is that most people currently rely on star counts or marketing claims, which don't measure real performance. SkillzBench's approach is benchmarking: each skill is tested in a sandbox with the same model, run with and without the skill, and the difference is published as Lift in percentage points. Lift is always shown separately from ratings, and money never touches scores. Until such benchmarks are widely available, you can run your own comparison — give the agent the same report task with and without the skill installed and compare formatting accuracy, tracked-changes handling, and revision time.

Can Claude skills produce real .docx files, or just Markdown?

Real .docx files. Anthropic's official docx skill generates genuine Word documents (not Markdown pretending to be Word), preserving styles, headings, tables of contents, tracked changes, and comments. It works by writing docx-js scripts for new documents or carefully editing the underlying OOXML for existing files, with validation scripts to check the output. WordCraft takes a different route: it's a full open-source word processor with a built-in MCP server, so an agent can create and edit .docx files through 422 addressable commands and verify its work via inspect_document.

Do Claude skills work with other AI coding agents?

Yes. Anthropic released the Agent Skills format as an open standard in December 2025, and it is now supported by Claude Code, Claude.ai, the Claude API, OpenAI Codex, Cursor, Gemini CLI, Antigravity, and Windsurf. Because a skill is just a folder with a SKILL.md file, the same package can often be used across agents. SkillzBench is being built with this cross-agent reality in mind — its free MCP connector and per-agent loadouts let you toggle skills and tools for each agent individually, so each agent only sees what you've chosen for it.

Is SkillzBench free, and is it available now?

Everything on SkillzBench will be free — the skills directory, the MCP connector, benchmarks, and alerts. The platform earns revenue from advertising shown in and around its blog articles, never by charging users, and ad money never influences scores or listings. SkillzBench is currently pre-launch and actively being built: the waitlist and newsletter are live today, but the directory, scanning, benchmarking, and MCP tools are not yet publicly available. Joining the waitlist at skillzbench.com is the way to get access as soon as it launches.

Which skill should I start with for Word and report work?

Start with Anthropic's official docx skill — it's the first-party, Apache-2.0-licensed reference implementation for creating and editing Word documents with tracked changes, comments, and validation, and community guides consistently recommend it as the safe default. If your reports involve stakeholder iteration rather than just file generation, add the Doc Co-Authoring skill for structured drafting and refinement. If no existing skill fits your organization's reporting conventions, use skill-creator to build your own. And before installing anything from the community, verify it — SkillzBench's upcoming directory and trust scores are designed to make that verification fast and free.