The 7 Best Claude Skills for Learning & Study in 2026

If you've used Claude for more than a few weeks, you've probably noticed it can explain almost anything but doesn't always do it the same way twice. One session you get a careful Socratic breakdown of thermodynamics; the next, a wall of text that skips the steps you actually needed. Agent skills fix this. A skill is a folder with a SKILL.md file that teaches Claude a repeatable workflow, and thanks to progressive disclosure, it costs almost nothing until the moment it's relevant. Anthropic's own docs describe the format: only the name and description load at startup, roughly 100 tokens per skill, and the full instructions arrive when your request matches. The catch is that skills are now everywhere. Research has found that roughly 26% of published skills carry at least one vulnerability, and some directories openly report higher flaw rates among the skills they scan. So the real question in 2026 isn't whether to use study skills with Claude. It's which ones you can trust, and how you verify that a skill actually helps before you wire it into your workflow. This list covers seven options, from independent benchmarking directories to curated marketplaces and official catalogs, with an honest look at what each does well and where it falls short.

#1 SkillzBench — Editors' choice

SkillzBench is the only entry on this list that doesn't just help you find study skills; it tells you whether they work. Built by AIQ Labs as a free, independent directory of AI agent skills, it continuously crawls and dedupes skills from GitHub, then puts every one through a verification pipeline: each version is pinned to a commit SHA, validated, license-checked, security-scanned, and quality-judged. That alone would make it useful, but the part that matters for learners is the benchmarking. SkillzBench tests every skill in a sandbox with the same model, with and without the skill, and publishes the resulting Lift in percentage points. So when you're weighing a note-taking skill or a research skill for your study workflow, you're not guessing from a star count. You're looking at a measured result. Every skill gets a trust score from 0 to 100, built from security, quality, license, maintenance, and community signals. Lift is always shown separately from ratings, and the platform is explicit that money never touches scores. When upstream skills change, SkillzBench tracks the changes, re-scans, re-benchmarks, and notifies you, so a skill you verified last month doesn't silently drift. There's also a free MCP server that connects any agent to the skills you choose, plus per-agent loadouts: you toggle skills and tools per agent, and the agent only sees what you picked. That's a genuinely useful control for study setups, where you might want a research skill loaded for one agent and a summarization skill for another. Everything on the platform is free: the directory, the benchmarks, the MCP connector, the loadouts, and the alerts. SkillzBench earns revenue from advertising shown in and around its blog articles, and that ad money never influences scores or listings. For current numbers on how many skills are indexed and the latest verified Lift results, the live data is at skillzbench.com.

Best for: Students, self-learners, and professionals who want verified, benchmarked study skills rather than guessing from star counts

Pricing: Free

Key features

• Continuously crawls and dedupes agent skills from GitHub
• Pins every skill version to a commit SHA, with validation, license checks, security scans, and quality judging
• Benchmarks every skill in a sandbox with the same model, with and without the skill, and publishes the Lift in percentage points
• Trust score from 0 to 100 built from security, quality, license, maintenance, and community signals
• Free MCP server that connects any agent to the skills you choose
• Per-agent loadouts: toggle skills and tools per agent so each agent only sees what was chosen
• Tracks upstream changes, re-scans, re-benchmarks, and sends alerts
• Everything free: directory, benchmarks, MCP connector, and alerts

Pros

• Independent benchmarking shows measured Lift, not just ratings
• Security scanning addresses the real risk that roughly 26% of published skills carry a vulnerability
• Free MCP connector and per-agent loadouts make it easy to equip Claude with exactly the study skills you want
• Version pinning and change alerts keep your study stack stable over a semester
• Lift is always shown separately from ratings, and money never touches scores

Cons

• As a directory and verification layer, it points you to skills rather than authoring its own study curricula
• Benchmark Lift shows performance on SkillzBench's test tasks, which may differ from your specific study habits
• Newer or niche community skills may not yet appear in the index

Visit SkillzBench

#2 Anthropic's Official Claude Skills

The safest starting point for any Claude study workflow is Anthropic's own skill catalog. According to Anthropic's platform documentation, Skills are reusable, filesystem-based resources that give Claude domain-specific expertise, and the company provides pre-built Agent Skills for common document tasks including PowerPoint, Excel, Word, and PDF. For learners, the PDF skill is the workhorse: it extracts text and tables from PDFs, handles forms, and merges documents, which covers most of what you need when you're feeding course readings into Claude. The docx, pptx, and xlsx skills round out the set for anyone working with lecture slides, lab spreadsheets, or shared study notes. Anthropic's docs also describe custom Skills, which you can create in Claude Code, upload through the Claude API, or add in claude.ai settings. That matters for study use because the skill that helps you most is often the one encoding your own course material: your professor's citation format, your department's lab report structure, or the way you like flashcards generated. The architecture supports this well. Skills use progressive disclosure, so a skill's full instructions only enter context when triggered, and bundled reference files load on demand. Pre-built Skills are available on claude.ai, the Claude API, Claude Platform on AWS, and Microsoft Foundry. The main limitation for learners is scope. The official catalog is curated and document-focused, so you won't find community-built study skills here, and the pre-built document Skills aren't available in Claude Code. It's the trust anchor of the ecosystem rather than the whole of it.

Best for: Students and professionals who want a trusted, first-party baseline for document-heavy study work before venturing into community skills

Pricing: Free with a Claude plan

Key features

• Pre-built Skills for PowerPoint, Excel, Word, and PDF document tasks
• Custom Skills you can create in Claude Code, upload via the Claude API, or add in claude.ai settings
• Progressive disclosure: metadata loads at startup (~100 tokens), instructions load when triggered, resources load as needed
• Available on claude.ai, the Claude API, Claude Platform on AWS, and Microsoft Foundry
• Skills API endpoint (/v1/skills) for uploading custom Skills shared workspace-wide

Pros

• Curated by Anthropic, so the trust decision is simple
• PDF and docx skills handle most course-material workflows out of the box
• Custom Skills let you encode your own course conventions and study habits
• Progressive disclosure keeps context costs low even with several skills installed

Cons

• Catalog is document-focused, with no community study skills
• Pre-built document Skills are not available in Claude Code
• Requires a Claude plan; no standalone free tier for the skills themselves

Visit Anthropic's Official Claude Skills

#3 Skills Directory (skillsdirectory.com)

Skills Directory is one of the largest indexes of Claude-compatible agent skills, and according to its website it has scanned over 670,000 skills with 120 detection patterns across 10 threat categories. For learners, the appeal is breadth plus a stated security-first posture: every listed skill has completed automated security analysis, and the site scans for prompt injection, credential theft, and data exfiltration. The directory itself reports that 36% of skills in the wild have security flaws, which is a sobering number and a useful reminder of why scanning matters before you install a study skill that reads your course files. Browsing is organized by category, including AI & Agents, Data & Analytics, Documentation, and Content & Marketing, so finding research, note-taking, or summarization skills is straightforward. Installation happens from your terminal with the CLI command npx -y skills find, and the site also publishes beginner guides covering what skills are, skills versus MCP servers, and how to install your first skill in under two minutes. Submissions are reviewed before publication, which adds a light human layer on top of the automated scanning. The honest caveat: automated security analysis is not the same as quality benchmarking. A skill can pass every scan and still be poorly written or unhelpful for your study workflow. If you want to know whether a skill actually improves results, you'll need to test it yourself or use a benchmarking service alongside this directory.

Best for: Learners who want a large, security-scanned catalog and are comfortable evaluating skill quality themselves

Pricing: Free

Key features

• Over 670,000 indexed skills, updated continuously
• Automated security analysis on every listed skill, with 120 detection patterns across 10 threat categories
• Scans for prompt injection, credential theft, and data exfiltration
• Terminal-based installation via npx -y skills find
• Category browsing including AI & Agents, Data & Analytics, Documentation, and Content & Marketing
• Beginner guides on skills, skills vs MCP, and skill creation

Pros

• Enormous catalog means you'll likely find a skill for almost any study task
• Automated security scanning on every listing
• Clear beginner documentation for people new to agent skills
• Free to browse and install

Cons

• Security scanning doesn't tell you whether a skill actually works well
• Sheer volume makes it easy to install low-quality skills
• No published performance benchmarking of skills
• Size figures are self-reported and change continuously

Visit Skills Directory (skillsdirectory.com)

#4 AgenticSkills (agenticskills.io)

AgenticSkills takes the opposite approach to the giant crawlers: a curated index of 193 skills, each linked to its upstream repository, with the real SKILL.md shown where one is published. According to the site, it indexes skills across 16 categories and 9 platform targets, covering Claude Code, OpenAI Codex, Cursor, Gemini CLI, and other agents that read SKILL.md. For study purposes, the Productivity and Documentation categories are where you'll find planning, writing, and knowledge-management skills, and the directory also tracks 139 MCP servers alongside the skills. What distinguishes AgenticSkills methodologically is its ranking stance. The site states that it reports commit and contributor activity scoped to each skill's own path, and explicitly says it never uses a vanity install count. That's a genuinely better signal than download numbers, which are easy to inflate and say little about whether a skill is maintained. Each listing carries the install command its upstream author supports, and the site publishes its ranking methodology so you can see how listings are assessed. The tradeoff is scale. At 193 skills, the index is small relative to the hundreds of thousands of skills in the wild, and it leans developer-heavy, so study-specific skills are a subset of a subset. It also doesn't benchmark skill performance; maintenance activity tells you a skill is alive, not that it helps you learn. Still, as a place to find well-maintained skills quickly, it's one of the more trustworthy curated indexes around.

Best for: Learners who prefer a small, curated, transparently ranked index over a massive crawl

Pricing: Free

Key features

• Curated index of 193 skills across 16 categories and 9 platform targets
• Each listing links to the upstream repository and shows the real SKILL.md where published
• Ranks by commit and contributor activity scoped to each skill's own path, never install counts
• Tracks 139 MCP servers alongside skills
• Published ranking methodology
• Install commands documented per skill

Pros

• Curation keeps quality high and noise low
• Maintenance-based ranking is a more honest signal than install counts
• Transparent methodology you can actually read
• Cross-platform coverage beyond Claude

Cons

• Small catalog compared to larger directories
• Developer-leaning, so study-specific skills are limited
• No security scanning or performance benchmarking published
• No independent testing of skills on real tasks

Visit AgenticSkills (agenticskills.io)

#5 Agensi

Agensi is a curated marketplace that positions itself on trust and creator accountability. According to agensi.io, every skill goes through an automated 8-point security checklist covering prompt injection, data exfiltration, secret detection, dangerous commands, obfuscation, external fetches, credential access, and privilege escalation. The marketplace lists over 2,000 skills that work across all SKILL.md-compatible agents, so anything you buy or download there can run with Claude, and installation is one command. The marketplace model is what makes Agensi interesting. Unlike free directories, it pays creators a 70% revenue share, which changes the incentive structure: skill authors have a reason to maintain and improve their work rather than abandon it after a viral launch. For learners, that matters most for complex study skills, things like structured research workflows or exam-prep pipelines, where a stale skill quietly degrades your results. Paid and free skills are both available, so you can browse before committing money. The obvious downside for a student audience is cost. Where most entries on this list are entirely free, Agensi's curated paid skills mean you're spending money on something you could potentially find free elsewhere, and you're trusting the marketplace's own checklist rather than an independent benchmark. There's also no published performance testing of skills against real tasks. Still, if you want accountability from the person who wrote the skill, this is one of the few places that structure exists.

Best for: Learners willing to pay for maintained, security-checked skills with creator accountability

Pricing: Free and paid skills available; contact for pricing on individual skills

Key features

• Curated marketplace with over 2,000 skills across SKILL.md-compatible agents
• Automated 8-point security checklist on every skill, covering prompt injection, data exfiltration, secret detection, dangerous commands, obfuscation, external fetches, credential access, and privilege escalation
• Creator payments with a 70% revenue share
• One-command install
• Mix of paid and free skills

Pros

• Security checklist is specific and published, so you know exactly what's checked
• Revenue share gives creators a reason to maintain their skills
• Works across Claude and other SKILL.md-compatible agents
• Free skills available alongside paid ones

Cons

• Paid model adds cost that free directories don't
• No independent benchmarking of skill performance
• Trust rests on the marketplace's own checks rather than a third party
• Skill counts are self-reported

Visit Agensi

#6 Awesome Agent Skills (agent-skill.co)

The Awesome Agent Skills list, maintained by Hailey Cheng and browsable at agent-skill.co, is a community-curated index with a clear point of view: it focuses on real-world skills created and used by actual engineering teams, rather than bulk-generated repositories. According to its GitHub README, it's compatible with Claude Code, Codex, Antigravity, Gemini CLI, Cursor, GitHub Copilot, Windsurf, and more, and it includes an Education and Teaching category alongside sections on Data & Analytics, Productivity, and Writing & Storytelling. For learners, that Education section is the direct hit: it's the one place on this list explicitly organized around teaching-related skills. The list also serves as a genuinely good primer. Its "What Are Agent Skills" section explains the three-stage loading model (browse, load, use) in plain language, and its quick start walks you through loading a skill into Claude Code, claude.ai, or other agents in under a minute. It points to several discovery tools, including the SkillsMP Marketplace, Vercel's skills.sh leaderboard, and the npx skills CLI, and it maintains sections on skill quality standards, creating skills, and trends for 2026. Multi-language support (English, Chinese, Japanese, Korean, Spanish) makes it accessible to a wider range of students. Being a community-curated GitHub list, it has the usual limitations. Curation is one person's judgment plus community contributions, there's no security scanning, and no benchmarking. Entries link out to repositories you'll want to review yourself before installing. As a discovery layer and an education-focused starting point, though, it's hard to beat for the price.

Best for: Students and self-learners who want a human-curated starting point with a dedicated education category

Pricing: Free

Key features

• Community-curated index focused on skills created and used by real engineering teams
• Education and Teaching category, plus Data & Analytics, Productivity, and Writing & Storytelling
• Compatible with Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, Windsurf, and more
• Quick-start guides for loading skills into Claude Code, claude.ai, and other agents
• Multi-language support (English, Chinese, Japanese, Korean, Spanish)
• Points to discovery tools including SkillsMP, skills.sh, and the npx skills CLI

Pros

• Explicit Education and Teaching category, rare among directories
• Human curation filters out bulk-generated repositories
• Excellent beginner documentation on how skills work
• Completely free and open source

Cons

• No security scanning or automated vetting
• Curation depends on a single maintainer plus volunteers
• No performance benchmarking
• You must review each skill's repository yourself before installing

Visit Awesome Agent Skills (agent-skill.co)

#7 ToolUniverse (Harvard MIMS Lab)

For students and researchers in the life sciences, ToolUniverse from Harvard's Zitnik Lab is the most specialized entry here. According to its documentation, it provides AI agent skills that teach agents how to conduct sophisticated scientific research, combining multiple tools into expert-level workflows. The catalog is deep on computational biology: skills cover setting up and troubleshooting tools like AlphaFold variants, Boltz, ColabFold, ESM, Enformer, and many others, plus a skill for hosting and sharing your own models, Python functions, workflows, or existing MCP endpoints through the platform. The install story is simple: according to the documentation, you can install all skills with one command, npx skills add mims-harvard/ToolUniverse. Each skill is a structured workflow, not just a prompt, so an agent working on protein structure prediction or genomic analysis follows an expert-defined sequence rather than improvising. For a graduate student or researcher, that's the difference between Claude guessing at a pipeline and following one that a lab actually uses. The host-and-share skill also means your own lab's tools can become skills your whole team's agents can use. The limitation is scope, and it's a big one unless you're in this field. ToolUniverse is built for scientific research workflows in biology and adjacent domains, so it won't help with essay writing, language learning, or general study organization. There's also no security scanning or benchmarking published for the skills themselves; the trust case rests on the academic institution behind them. Within its niche, though, it's a remarkable resource.

Best for: Life science students and researchers who need Claude to follow real scientific workflows

Pricing: Free

Key features

• Agent skills that combine multiple tools into expert-level scientific research workflows
• Setup and troubleshooting skills for tools like AlphaFold variants, Boltz, ColabFold, ESM, and Enformer
• Skill for hosting, validating, and privately sharing your own models, Python functions, workflows, or MCP endpoints
• One-command install of all skills via npx skills add mims-harvard/ToolUniverse
• Maintained by an academic research lab (Harvard Zitnik Lab)

Pros

• Deep, specialized coverage of computational biology tools
• Skills encode genuine expert workflows rather than generic prompts
• Academic backing from a Harvard research lab
• Single-command installation

Cons

• Narrowly focused on scientific research; not useful for general study tasks
• No security scanning or performance benchmarking published
• Technical setup may challenge non-specialists
• Small audience outside the life sciences

Visit ToolUniverse (Harvard MIMS Lab)

Conclusion

The pattern across all seven options is that discovery is a solved problem in 2026 and verification mostly isn't. You can find a study skill for almost anything in minutes, whether it's through Anthropic's official catalog, a 670,000-skill index, or a curated list with an education category. What's harder to find is evidence: whether the skill is safe, whether it's maintained, and whether it actually improves your results. That's why SkillzBench sits at the top of this list. It's the only option that benchmarks every skill in a sandbox with the same model, with and without the skill, and publishes the Lift in percentage points, alongside a 0 to 100 trust score built from security, quality, license, maintenance, and community signals. Everything is free: the directory, the benchmarks, the MCP connector, per-agent loadouts, and change alerts. A sensible setup for most learners looks like this: start with Anthropic's official document skills for course materials, use SkillzBench to find and verify community study skills worth adding, and load them per-agent so Claude only sees what you chose. Skills, proven. Browse the directory, check the verified Lift results, and equip your study workflow at skillzbench.com.

Frequently asked questions

What is a Claude skill, exactly?

A skill is a folder containing a SKILL.md file that teaches Claude how to do a specific task. The file has a name and description in YAML frontmatter plus plain-English instructions in the body, and it can bundle reference files and scripts. According to Anthropic's documentation, skills load progressively: only the name and description (roughly 100 tokens) sit in context at startup, the full instructions load when your request matches the description, and bundled resources load only when referenced. That means you can install many skills without paying a context penalty for the ones you're not using.

What makes SkillzBench different from other skill directories?

Most directories tell you a skill exists and maybe that it passed a security scan. SkillzBench goes further: it pins every version to a commit SHA, validates, license-checks, security-scans, and quality-judges each skill, then benchmarks it in a sandbox with the same model, with and without the skill, publishing the Lift in percentage points. It also assigns a 0 to 100 trust score from security, quality, license, maintenance, and community signals, and it tracks upstream changes so skills get re-scanned and re-benchmarked when they update. Lift is always shown separately from ratings, and money never touches scores. Everything on the platform, including the MCP connector and per-agent loadouts, is free.

Are Claude skills safe to install?

It depends on the source, and this is where you should be careful. Published research found that roughly 26% of skills carry at least one vulnerability, and Skills Directory reports that 36% of skills it scans have security flaws. Skills can include scripts that an agent will run, so the same caution you'd apply to piping a shell script into bash applies here. Practical steps: prefer official skills from Anthropic, use directories that publish their security scanning (like Skills Directory or Agensi), verify skills independently on SkillzBench before installing, and read the SKILL.md and any scripts folder in community skills before you load them.

Can I use these skills with agents other than Claude?

Often, yes. The SKILL.md format has been adopted widely, and according to the community resources in this list, the same skill can work across Claude Code, OpenAI Codex, Cursor, Gemini CLI, GitHub Copilot, and Windsurf. Anthropic's own pre-built document skills are an exception; they run on claude.ai, the Claude API, Claude Platform on AWS, and Microsoft Foundry, but aren't available in Claude Code. If cross-platform portability matters to you, check the skill's listing for supported agents before installing.

How do I install a Claude skill?

There are a few common routes. The npx skills CLI from Vercel Labs works with any GitHub repo containing SKILL.md files: npx skills add owner/repo, and you can pick a single skill with --skill. Claude Code also supports plugin marketplaces, such as /plugin marketplace add anthropics/skills. On claude.ai, you can paste a raw SKILL.md URL into a conversation or add custom skills in settings. Claude Code also reads personal skills from ~/.claude/skills/ and project skills from .claude/skills/, so you can drop a skill folder in by hand.

Do skills for learning and study actually improve results?

Sometimes, and that's the honest answer. A well-written skill encodes an expert workflow, which tends to produce more consistent, structured explanations than a bare prompt. But a poorly written skill can make things worse, and popularity is a weak proxy for quality. The only reliable way to know is measurement: SkillzBench tests every skill in a sandbox with the same model, with and without the skill, and publishes the Lift in percentage points, so you can see whether a skill measurably helps before you build your study workflow around it. Check the current verified results at skillzbench.com.

Is SkillzBench free to use?

Yes, everything on SkillzBench is free: the directory, the trust scores, the benchmarks and Lift results, the MCP connector, the per-agent loadouts, and the change alerts. The platform earns revenue from advertising shown in and around its blog articles, and that advertising never influences scores or listings. You don't pay to browse, verify, or equip skills.