The 9 Best Claude Skills for App Development in 2026

Claude Skills have quietly become the difference between a coding agent that writes generic boilerplate and one that ships code the way a senior engineer would. A skill is just a folder with a SKILL.md file, but that file teaches Claude how a specific tool works today: how Vercel wants React written, how Supabase expects Postgres secured, how a migration should be planned before anyone touches a schema. Anthropic open-sourced the Agent Skills format, and it now works across Claude Code, OpenAI Codex, Cursor, Gemini CLI, and more, which means the skills you pick are an investment rather than a throwaway. The catch is volume. There are thousands of skills floating around GitHub in 2026, and published research found roughly a quarter of them carry at least one vulnerability, so grabbing the first repo you find is a bit like pip installing random code from a stranger. This list covers nine of the best options for app development this year, starting with a free platform that verifies and benchmarks skills before you trust them, then moving through the official and community skill collections that developers actually use day to day.

#1 SkillzBench — Editors' choice

SkillzBench is a 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 because it solves the problem every other entry on this list creates: trust. Anyone can publish a SKILL.md file, and research cited by SkillzBench found that about 26% of skills carry at least one vulnerability. Before you wire a stranger's skill into Claude Code, you probably want to know whether it works and whether it's safe. SkillzBench answers both questions. The platform continuously crawls and dedupes skills from GitHub, pins every version to a commit SHA, validates the files, checks licenses, runs security scans, and applies a quality judgment before anything appears in the directory. Then it goes a step further: every skill is tested in a sandbox with the same model, once with the skill and once without, and the resulting lift is published in percentage points so you can see whether the skill actually helps rather than just taking the README's word for it. Each skill also gets a trust score from 0 to 100 based on security, quality, license, maintenance, and community signals, with lift always shown separately and no money ever touching the scores. The free MCP server plus per-agent loadouts let you toggle skills and tools per agent, so your agent only sees what you chose for it. Everything on the platform, from the directory to the benchmarks to alerts when upstream skills change, is free at skillzbench.com.

Best for: Developers, agent builders, and teams who want verified, benchmarked skills instead of gambling on unvetted GitHub repos

Pricing: Free (all features, funded by advertising)

Key features

• Free directory of agent skills for Claude Code, Codex, Cursor, Gemini CLI, and chat-app agents
• Every skill version pinned to a commit SHA with validation, license checks, and security scans
• Sandbox benchmarking that publishes lift in percentage points, tested with and without the skill
• Trust score from 0 to 100 built on security, quality, license, maintenance, and community signals
• Free MCP connector plus per-agent loadouts so each agent only sees the skills you enabled
• Upstream change tracking with re-scans, re-benchmarks, and notifications
• Entire platform is free, funded by advertising around blog articles, never by charging users

Pros

• Security scanning and license checks before you install anything
• Benchmark lift data shows whether a skill actually improves results
• Works across multiple agents via the free MCP server
• Alerts when an upstream skill changes, so you're not running stale code

Cons

• It's a directory and connector, not a skill library of its own, so you still pick the skills
• Benchmark results depend on the model used, so your mileage may vary
• Advertising appears around blog content, which some readers may find distracting

Visit SkillzBench

#2 Anthropic Official Skills (anthropics/skills)

If you want skills maintained by the company that invented the format, Anthropic's own repository is the natural starting point. According to the repo and Anthropic's engineering blog, the collection includes document skills for PowerPoint, Excel, Word, and PDF, plus development-focused skills like frontend-design, which pushes Claude toward distinctive, intentional visual design instead of the templated purple-gradient look, and web-artifacts-builder for multi-component HTML artifacts built with React, Tailwind CSS, and shadcn/ui. There's also mcp-builder, which guides the creation of Model Context Protocol servers for integrating external APIs, and webapp-testing, which uses Playwright to test local web applications. The official skills follow the progressive disclosure model Anthropic describes in its documentation: only the name and description load at startup, the full instructions load when triggered, and bundled scripts and reference files load only as needed. That keeps context windows lean even with many skills installed. Because Anthropic maintains these skills, they track changes to the Claude ecosystem itself, and the Agent Skills format they follow has been adopted as an open standard by Claude Code, OpenAI Codex, Cursor, Gemini CLI, and others. For app development specifically, frontend-design and webapp-testing are the two most cited by developers writing about their Claude Code workflows.

Best for: Developers who want officially maintained skills that track the Claude ecosystem

Pricing: Free (open source; Claude plan required for Claude products)

Key features

• frontend-design skill for distinctive, intentional UI design in React and Tailwind
• web-artifacts-builder for multi-component HTML artifacts
• mcp-builder for creating MCP servers that integrate external APIs
• webapp-testing using Playwright for local web application testing
• Document skills for PowerPoint, Excel, Word, and PDF
• Progressive disclosure keeps context usage low
• Maintained by Anthropic and follows the open Agent Skills standard

Pros

• Maintained by the company behind the format, so updates are reliable
• Covers the full range from design to testing to document generation
• Portable across Claude Code, Codex, Cursor, and Gemini CLI

Cons

• Not specialized for any single framework or stack
• Some skills target Claude.ai artifacts rather than codebases
• Fewer niche development skills than large community collections

Visit Anthropic Official Skills (anthropics/skills)

#3 obra/superpowers

Superpowers is the most-starred skills collection in the ecosystem, with over 300,000 GitHub stars according to Linkly's rankings, and it approaches app development as a methodology rather than a bag of tricks. Created by Jesse Vincent, the collection teaches Claude a full agentic development workflow: brainstorming to refine rough ideas into structured designs, writing implementation plans with bite-sized tasks, executing those plans with human checkpoints, test-driven development with a red-green-refactor cycle, systematic debugging that enforces root cause analysis before fixes, and verification before completion so tasks aren't falsely marked done. There are also skills for requesting and receiving code reviews, dispatching parallel sub-agents, and using git worktrees for concurrent feature work. The reason developers rave about it is discipline. A raw Claude Code session will happily write code first and think about tests never. Superpowers flips that, treating tests as the specification and refusing to declare victory until fixes are actually verified. AgenticSkills rates several of the collection's skills at its top tier, and the skills work across Claude Code, Codex, Cursor, and other agents that read the SKILL.md format. If you're building an app end to end with an agent rather than asking for one-off snippets, this is the workflow layer many developers install first.

Best for: Developers running full app builds through Claude Code who want process discipline, not just code snippets

Pricing: Free (open source)

Key features

• Brainstorming skill for structured design refinement
• Writing and executing plans with human checkpoints
• Test-driven development with red-green-refactor enforcement
• Systematic debugging with root cause analysis before fixes
• Verification before completion to prevent false task completion
• Code review requesting and receiving workflows
• Git worktrees and parallel sub-agent coordination

Pros

• Enforces TDD and verification, producing fewer regressions
• Extremely popular and actively maintained
• Works across multiple coding agents via the SKILL.md format

Cons

• Opinionated workflow; developers who prefer looser sessions may find it rigid
• Human checkpoints add friction to fast iterations
• Large collection can take time to learn

Visit obra/superpowers

#4 Vercel Agent Skills (vercel-labs/agent-skills)

For React and Next.js development, Vercel's official skill repository is the reference point, and the writeup on DEV by Shrey Vijayvargiya calls react-best-practices the one skill to install if you install only one. According to the repo, react-best-packages packs React and Next.js performance guidelines from Vercel Engineering, meant to load whenever Claude writes, reviews, or refactors React code. The collection also includes web-design-guidelines, which reviews your UI code against Vercel's Web Interface Guidelines and triggers on phrases like 'review my UI' or 'check accessibility', making it a solid pre-ship check, plus writing-guidelines for docs and UI copy. The repo also ships agent-browser, a skill for automating browser interactions for web testing, form filling, screenshots, and data extraction, which has tens of thousands of stars in its own right according to Linkly's rankings. The case for official framework skills is timing: frameworks change faster than model training data, and when Next.js ships a new caching model, Vercel updates the skill to match. Installation is straightforward with the open skills CLI from Vercel Labs, which works for any GitHub repo containing SKILL.md files, and you can install single skills with a flag rather than taking the whole collection.

Best for: React and Next.js developers who want framework-current guidance in every Claude session

Pricing: Free (open source)

Key features

• react-best-practices with React and Next.js performance guidelines from Vercel Engineering
• web-design-guidelines for auditing UI against Vercel's Web Interface Guidelines
• writing-guidelines for docs and UI copy tone
• agent-browser for browser automation, testing, and data extraction
• Installable individually via the npx skills CLI
• Maintained by Vercel so skills track framework changes

Pros

• Official framework knowledge that stays current with Next.js releases
• The design guidelines audit is a genuinely useful pre-ship check
• Simple CLI installation with per-skill selection

Cons

• Focused on the Vercel and React ecosystem, less useful for other stacks
• No equivalent depth for backend or database work
• Guidelines reflect Vercel's opinions, which not every team shares

Visit Vercel Agent Skills (vercel-labs/agent-skills)

#5 alirezarezvani/claude-skills

This repository bills itself as the most comprehensive open-source library of Claude Code skills and agent plugins, with 398 production-ready skills across 20 domains and more than 5,000 GitHub stars according to its own listing. For app development, the draw is breadth: 54 core engineering skills covering architecture, frontend, backend, fullstack, QA, DevOps, and security, plus a 97-skill advanced tier that includes an RAG architect, database designer, CI/CD builder, security auditor, MCP builder, and a zero-hallucination-coder skill with a discuss-map-decompose-execute-verify workflow. There's also Playwright Pro for test generation and flaky test fixing, a self-improving agent for memory curation, and an accessibility audit skill. Beyond engineering, the repo includes product, project management, marketing, and C-level advisory skills, which matters if you're a solo founder using Claude Code to build and then market an app. Installation runs through Claude Code plugin marketplaces, and conversion scripts let you deploy the same skills to Codex, Cursor, Gemini CLI, Windsurf, and several other agents. The scale is both the appeal and the caveat: with hundreds of skills, you'll want to install by domain rather than taking everything, and as with any large community collection, reviewing what you install is wise.

Best for: Solo developers and founders who want one repository covering engineering plus business skills

Pricing: Free (open source)

Key features

• 398 skills across 20 domains including 54 core engineering skills
• Advanced tier with RAG architect, database designer, CI/CD builder, and security auditor
• zero-hallucination-coder with a structured discuss-map-decompose-execute-verify workflow
• Playwright Pro for test generation, flaky test fixes, and migrations
• Conversion scripts for Codex, Cursor, Gemini CLI, Windsurf, and more
• Product, marketing, and C-level advisory skills for solo founders
• 706 CLI scripts with no pip dependencies

Pros

• Enormous breadth under one install
• Multi-agent support via conversion scripts
• Structured workflows like zero-hallucination-coder reduce sloppy output

Cons

• Volume makes selective installation necessary
• Community-maintained, so quality varies between skills
• Large repo clone can be unwieldy, especially on Windows

Visit alirezarezvani/claude-skills

#6 nextlevelbuilder/ui-ux-pro-max-skill

If your app's problem is that everything Claude builds looks the same, this community skill is the heavyweight option. According to its repository and the AgenticSkills directory, ui-ux-pro-max is a large UI/UX knowledge skill covering web, mobile, and desktop, spanning pages, components, design systems, accessibility, interaction, and responsive layout. It's designed to make Claude reason about design decisions rather than apply a default style, and it has accumulated a substantial star count, with Linkly's rankings listing it among the top agent skill repositories. The same repository ships companion skills: design-system with three-layer tokens (primitive, semantic, and component), ui-styling built around shadcn/ui plus Tailwind, brand, banner-design, and slides. Installation is a two-step plugin marketplace command, and the skill works with Claude Code, Codex, Cursor, and other SKILL.md-compatible agents. The DEV writeup on Claude Code skills notes that without a design skill, Claude produces the same centered hero and three feature cards every time, and that a good design skill changes output more than any other category. This one aims squarely at that gap. It's community-maintained rather than official, so it moves at the speed of one maintainer, but for front-end-heavy app work it's one of the most cited skills in the ecosystem.

Best for: Front-end developers and small teams building user-facing apps who want design reasoning, not templates

Pricing: Free (open source)

Key features

• Deep UI/UX knowledge base for web, mobile, and desktop
• Design system skill with primitive, semantic, and component token layers
• ui-styling skill built on shadcn/ui and Tailwind
• Accessibility and responsive layout guidance
• Companion brand and banner-design skills
• Installable via Claude Code plugin marketplace
• Works across Claude Code, Codex, Cursor, and other agents

Pros

• Genuinely changes the visual quality of generated UI
• Covers design systems and accessibility, not just aesthetics
• Strong community adoption and ratings

Cons

• Community-maintained, so update cadence depends on one maintainer
• Large knowledge base adds context when triggered
• Opinionated style may clash with an existing brand system

Visit nextlevelbuilder/ui-ux-pro-max-skill

#7 addyosmani/web-quality-skills

Addy Osmani, a well-known Google engineering leader, maintains a collection of web quality skills that Snyk's roundup of top Claude skills ranks second among developer skills. According to the repo, the collection covers performance, Core Web Vitals, accessibility, and SEO, which makes it the skills-based equivalent of a pre-launch audit checklist. In practice, you trigger it when you want Claude to review a page or component for performance problems, accessibility failures, or search issues, and it applies current best practices rather than the training-data defaults that tend to lag a year or two behind. The broader addyosmani/agent-skills repository has grown to around 25 skills covering multiple areas of web development and has well over a hundred thousand stars according to Linkly's rankings, so the web quality skills sit inside a larger, actively maintained collection. For app development, this fills a specific hole: Claude Code will happily ship a feature that works but scores poorly on Lighthouse or fails keyboard navigation, and this skill category exists to catch exactly that before users do. It's free, open source, and installable with the standard skills CLI, and it works across Claude Code and other SKILL.md-compatible agents.

Best for: Web developers who want performance, accessibility, and SEO checks built into their Claude workflow

Pricing: Free (open source)

Key features

• Performance review against Core Web Vitals
• Accessibility auditing for generated UI
• SEO checks for web pages
• Part of the broader addyosmani/agent-skills collection of around 25 skills
• Installable via the standard skills CLI
• Maintained by a recognized web performance expert

Pros

• Catches quality issues Claude otherwise ships silently
• Backed by a maintainer with deep web performance expertise
• Sits inside a larger, actively updated skill collection

Cons

• Narrow scope; it audits quality rather than building features
• Community-maintained with no corporate backing
• Best paired with framework skills rather than used alone

Visit addyosmani/web-quality-skills

#8 HashiCorp Agent Skills (hashicorp/agent-skills)

Infrastructure is where unaided AI agents most often produce plausible-looking but broken code, and HashiCorp's official skills aim to fix that for Terraform. According to Snyk's roundup, the repository covers Terraform code generation, testing, and modules, with roughly 300 GitHub stars at the time of their review. The value of an official skill here is the same as with Vercel and Supabase: Terraform's provider ecosystem and module conventions change, and a skill maintained by HashiCorp tracks those changes in a way a model's training data cannot. For app developers who provision their own infrastructure, whether that's a database, a CDN, or a full cloud environment, the skill guides Claude through writing modules that follow HashiCorp's conventions, testing them, and structuring them for reuse rather than dumping a single monolithic configuration into your repo. It installs like any other SKILL.md-based skill and works with Claude Code and other compatible agents. It's a narrow skill by design, but narrow is the point: infrastructure mistakes are expensive, and a small, official, well-maintained skill beats a sprawling community one for this layer of the stack. Pair it with a general workflow collection like Superpowers for the rest of the build.

Best for: Developers provisioning cloud infrastructure with Terraform alongside their app builds

Pricing: Free (open source)

Key features

• Official Terraform code generation guidance from HashiCorp
• Terraform testing workflows
• Module structure and reuse conventions
• Maintained by HashiCorp so guidance tracks provider changes
• SKILL.md format, portable across compatible agents

Pros

• Official, vendor-maintained Terraform knowledge
• Reduces the risk of plausible-but-broken infrastructure code
• Focused and easy to evaluate quickly

Cons

• Terraform only; no coverage of the application layer
• Smaller collection with fewer skills than general repositories
• Assumes you already work in Terraform

Visit HashiCorp Agent Skills (hashicorp/agent-skills)

#9 OthmanAdi/planning-with-files

Long app builds fall apart when the agent loses the plot halfway through, and planning-with-files addresses that directly with a Manus-style persistent task planning approach. According to Snyk's roundup, the skill had over 13,000 GitHub stars when reviewed, and Linkly's rankings show it continuing to grow with tens of thousands of stars as of late 2026. The core idea is that instead of holding the plan in the conversation, where it evaporates as context fills, the agent writes its plan, task lists, and progress notes to files on disk and reads them back as it works. That gives long, multi-step feature builds a memory that survives session restarts and context compaction, which is exactly the failure mode developers describe when a big refactor drifts off course. The skill structures the work into planning documents the agent updates as it goes, so you can also inspect progress yourself at any point by reading the same files. It's a single-skill repository rather than a large collection, which makes it easy to evaluate and low-risk to adopt, and it works with Claude Code and other SKILL.md-compatible agents. For anything longer than a one-shot script, planning persistence is one of the highest-leverage additions you can make to a Claude Code setup.

Best for: Developers running long multi-session builds who keep losing agent context mid-project

Pricing: Free (open source)

Key features

• Manus-style persistent planning written to files on disk
• Task lists and progress notes survive context compaction and restarts
• Structured planning documents you can inspect yourself
• Designed for long, multi-step feature builds and refactors
• Single focused skill, easy to install and audit

Pros

• Solves the real problem of context loss on long tasks
• Very popular and actively maintained
• Simple to adopt since it's one focused skill

Cons

• Narrow scope; you'll want other skills for coding and design
• File-based planning adds some overhead to short tasks
• Community-maintained with no corporate backing

Visit OthmanAdi/planning-with-files

Conclusion

Skills are the cheapest way to make Claude Code dramatically better at app development, but 2026's ecosystem rewards a little skepticism. Roughly a quarter of published skills carry at least one vulnerability, so the smart setup is a verified foundation plus a few well-chosen specialists: start with SkillzBench to check security scans, trust scores, and benchmark lift before you install anything, then add the official Anthropic and Vercel skills for framework currency, Superpowers for process discipline, and a targeted pick or two for design, web quality, or infrastructure. You can browse and test skills for free on SkillzBench, including the ones on this list, and see the lift data before anything touches your project. Skills, proven.

Frequently asked questions

What exactly is a Claude Skill?

A Claude Skill is a folder containing a SKILL.md file with a name and description in YAML frontmatter plus instructions, and optionally scripts and reference files. Claude loads only the name and description at startup (about 100 tokens), reads the full instructions when a task matches, and runs bundled scripts only as needed. Anthropic open-sourced the Agent Skills format, so the same skills work in Claude Code, OpenAI Codex, Cursor, Gemini CLI, and other compatible agents.

Are Claude Skills safe to install?

Not automatically. Skills can include scripts that an agent will run, and published research found that roughly 26% of skills carry at least one vulnerability, with Snyk's separate research finding prompt injection in 36% of skills it tested. Read the SKILL.md and any scripts before installing a community skill, or use a platform like SkillzBench that pins versions to commit SHAs, runs security scans and license checks, and publishes a trust score before you install.

How much do these skills cost?

All nine options in this list are free and open source. The costs that do exist are on the agent side: Claude Code requires a Claude Pro, Max, Team, or Enterprise subscription. SkillzBench's directory, benchmarks, MCP connector, and alerts are free, funded by advertising around its blog articles rather than user fees.

How many skills should I install at once?

Fewer than you'd think. Because skills load progressively, installing many doesn't blow up your context window, but overlapping skills can conflict and vague descriptions can trigger unreliably. A practical starting set for app development is one workflow collection (like Superpowers), one framework skill (like react-best-practices), one design or quality skill, and a planning skill for long builds.

What does SkillzBench's 'lift' metric mean?

Lift is the measurable improvement a skill provides. SkillzBench tests every skill in a sandbox with the same model, once with the skill and once without, then publishes the difference in percentage points. It's shown separately from ratings and trust scores, and no money influences either number, so you can judge whether a skill actually helps rather than relying on its README.

Do skills work with agents other than Claude?

Yes. The Agent Skills specification has been adopted as an open standard by Claude Code, OpenAI Codex, Cursor, Gemini CLI, and other coding agents. Several entries on this list, including the alirezarezvani collection and Superpowers, explicitly support multiple agents, and SkillzBench's free MCP connector plus per-agent loadouts let you route verified skills to whichever agent you're running.

Can I write my own skill instead of installing one?

Absolutely, and Anthropic ships a skill-creator skill to help. Start by identifying where your agent struggles on real tasks, write a SKILL.md with a precise description (that description controls when the skill triggers), and iterate based on how Claude actually uses it. Keep the core file lean and split detailed guidance into separate reference files that load only when needed.