Best Claude Code Skills in 2026: 349 Agent Skills Ranked by GitHub Stars
We catalogued every publicly available Claude Code skill we could find — 349 of them across 12 categories. Here is what the GitHub star counts actually tell you about which skills are worth installing.
TL;DR
- • Claude Code skills are reusable SKILL.md files that give the AI agent specialized workflows — think plugins, but instruction-based rather than code-based
- • We tracked 349 skills across development, DevOps, testing, design, data analysis, and more
- • The most-starred skills focus on practical development workflows: TDD, debugging, code review, and frontend design
- • Skills like brainstorming, systematic-debugging, and frontend-design consistently rank in the top 10 by adoption
- • You can browse the full directory at our SkillsMap — searchable, filterable, with GitHub star counts updated regularly
What Are Claude Code Skills?
If you have used Claude Code for more than a few sessions, you have probably noticed that it handles some tasks brilliantly on the first try and struggles with others repeatedly. Skills exist to bridge that gap.
A Claude Code skill is a structured instruction file — typically named SKILL.md — that teaches the AI agent a specific workflow. When you invoke a skill (usually with /skill-name), Claude Code reads the instructions and follows them step by step, rather than improvising from its general training.
The practical difference is significant. Without a TDD skill, asking Claude Code to "write tests first, then implement" produces inconsistent results — sometimes it writes tests, sometimes it skips straight to implementation. With the TDD skill installed, it follows a rigid Red-Green-Refactor cycle every single time, because the skill explicitly defines each step. (Not sure if Claude Pro is right for you? See our Claude Pro vs ChatGPT Plus comparison.)
Skills are community-created and shared primarily through GitHub repositories. Some are single-file instructions for narrow tasks. Others are comprehensive workflow systems with checklists, decision trees, and multi-agent coordination — for example, the orchestration skills covered in our Claude Code multi-agent tutorial. The ecosystem has grown from roughly 50 skills in mid-2025 to over 334 as of March 2026.
How We Ranked Them
We used GitHub stars as the primary ranking signal, for a simple reason: stars are the closest thing to a usage proxy in the open-source ecosystem. A skill with 500 stars has been discovered, tried, and explicitly bookmarked by hundreds of developers. That is not a guarantee of quality, but it is a stronger signal than self-reported ratings.
We also factored in recency (skills updated within the last 90 days score higher), documentation quality (does it explain what it does and how to invoke it?), and category coverage (ensuring we represent skills across all major use cases, not just coding).
What we explicitly did not do: test every skill ourselves. With 349 skills, that would take months. Instead, we installed and tested roughly 40 of the top-ranked ones across several real projects. The hands-on impressions in this article come from that subset.
Top Development Skills
Development skills make up the largest category — about 120 of the 334 total. These cover everything from code generation to refactoring to commit workflows.
| Skill | What It Does | Best For |
|---|---|---|
| brainstorming | Structured ideation before coding — generates constraints, edge cases, and alternative approaches before writing a single line | Feature planning, architecture decisions |
| test-driven-development | Enforces strict Red-Green-Refactor: write failing test → minimal implementation → verify → refactor | New features, bug fixes with regression coverage |
| systematic-debugging | Methodical debugging with hypothesis formation, evidence gathering, and root cause analysis — not random code changes | Complex bugs, intermittent failures |
| code-reviewer | Multi-perspective code review: factual accuracy, senior engineering patterns, security audit, consistency check | Pre-commit quality gates |
| writing-plans | Creates detailed implementation plans from specs — breaks features into phases with dependencies mapped | Large features, multi-file changes |
The brainstorming skill deserves special mention. Most developers jump straight to coding when given a task. This skill forces a structured exploration phase first — identifying constraints, listing edge cases, considering alternative approaches — before any code is written. In our testing, features built after brainstorming had roughly 30% fewer revision rounds.
The systematic-debugging skill changed how we approach complex bugs entirely. Instead of Claude Code randomly trying fixes (its default behavior on difficult bugs), the skill enforces a scientific method: form hypothesis, gather evidence, test hypothesis, repeat. It sounds obvious, but the structured approach catches root causes that ad-hoc debugging misses. If you are evaluating whether Claude Code itself is worth adopting, our Claude Code vs Cursor comparison breaks down the key workflow differences.
Top DevOps & Testing Skills
| Skill | What It Does | Best For |
|---|---|---|
| docker-optimize | Analyzes Dockerfiles for image size, layer caching, security, and build speed optimizations | CI/CD pipeline optimization |
| deploy-checklist | Pre-deployment validation: environment vars, database migrations, rollback plan, monitoring setup | Production deployments |
| test-harness | Generates comprehensive test suites — unit, integration, and E2E — from existing code | Legacy code coverage, test bootstrapping |
| security-scan | OWASP-based vulnerability scanning: injection, XSS, CSRF, auth bypass, secrets in code | Security audits, compliance checks |
Top Design & Frontend Skills
| Skill | What It Does | Best For |
|---|---|---|
| frontend-design | Creates distinctive, production-grade UI with 50 styles, 21 color palettes, and 50 font pairings — generates complete page layouts | Landing pages, dashboards, marketing sites |
| landing-page-guide | High-converting landing page creation with CRO principles baked into the structure | Product launches, SaaS marketing |
| ui-ux-pro-max | Full UI/UX design intelligence — responsive layouts, accessibility, interaction patterns | Design systems, component libraries |
The frontend-design skill is probably the most impressive in this category. Instead of Claude Code generating generic-looking components, this skill applies specific visual identities — brutalist, glassmorphism, neobrutalist, among others — with matching color palettes and typography. The output still needs design review, but it is dramatically better than Claude Code's default frontend generation.
Top Data & Research Skills
| Skill | What It Does |
|---|---|
| postgres-best-practices | Postgres performance optimization from Supabase's production experience — indexing, query tuning, connection pooling |
| data-pipeline | ETL/ELT pipeline architecture — source connectors, transformation logic, scheduling |
| iterative-retrieval | RAG-style research across large codebases — iteratively searches, reads, and synthesizes information |
How to Install Skills
Installing a Claude Code skill typically takes under a minute. The process varies slightly depending on how the skill is packaged, but the most common method is straightforward:
# Option 1: Copy the SKILL.md file
mkdir -p .claude/skills
cp path/to/skill/SKILL.md .claude/skills/my-skill.md
# Option 2: Use /install-skill command (if supported)
/install-skill github-repo-url
# Option 3: Slash command invocation
/skill-name
After installation, skills are invoked by their name — either as a slash command (/brainstorming) or referenced by Claude Code automatically when it detects a relevant task context.
Some skills require companion tools or MCP servers to function fully. The skill documentation should specify dependencies. If a skill requires mcp__playwright, for instance, you need the Playwright MCP server configured in your Claude Code environment.
Skills vs MCP Servers: What is the Difference?
This distinction confuses a lot of people, so it is worth clarifying. Skills and MCP servers serve fundamentally different purposes, even though they both extend Claude Code's capabilities.
| Aspect | Skills | MCP Servers |
|---|---|---|
| Type | Instructions (how to think) | Tools (what to do) |
| Format | Markdown files (SKILL.md) | Running processes (JSON-RPC) |
| Example | "Follow TDD: write test → run → implement → run → refactor" | "Here is a tool to read/write Slack messages" |
| Dependencies | None (just text) | Running server process |
In practice, the most powerful setups combine both. A coding workflow might use the systematic-debugging skill (which defines the debugging process) alongside the Playwright MCP server (which gives Claude Code the ability to interact with browsers for E2E testing). The skill provides the methodology; the MCP server provides the capability.
Building Your Own Skills
Creating a custom skill is surprisingly simple. A basic skill is just a Markdown file with a structured format:
The key to a good skill is specificity. Vague instructions like "write good code" produce the same output as no skill at all. Effective skills define exact steps, decision points, and output formats. The skill-writer meta-skill can help — it guides you through creating well-structured skills by asking the right questions.
If you build something useful, consider sharing it. The skills ecosystem grows through community contribution, and directories like SkillsMap make discovery easier for everyone.
Where to Discover More Skills
The skills ecosystem is growing fast — from roughly 50 skills in mid-2025 to over 85,000 indexed across multiple platforms by March 2026. Here are the three places worth checking regularly:
| Platform | What It Offers | Skills Indexed | Best For |
|---|---|---|---|
| skills.sh | Official Vercel-backed skills directory with install tracking. Ranks by anonymous telemetry from the npx skills CLI | 85,741+ | Finding the most-installed skills across 18 AI agents |
| MCP Market | Skills leaderboard ranked by GitHub stars. Categorized by use case (Developer Tools, Productivity, Data Science, etc.) | 100+ ranked | Discovering enterprise-grade skills from Facebook, PyTorch, Vercel, n8n |
| SkillsMap | Curated directory with 14 categories, search, and cross-platform comparison. Combines data from skills.sh, MCP Market, and independent GitHub repos | 349 | Side-by-side comparison of skills by category and AI platform compatibility |
According to skills.sh install data, the five most-installed skills as of March 2026 are:
- find-skills (vercel-labs) — 418.6K installs. The meta-skill: discovers and installs other skills
- vercel-react-best-practices — 176.4K installs. Official Vercel React patterns
- web-design-guidelines — 137.0K installs. Vercel's design system codified as a skill
- remotion-best-practices — 126.0K installs. Programmatic video creation in React
- frontend-design (Anthropic) — 124.1K installs. 50 visual styles with matched typography
On the MCP Market leaderboard, Facebook's React Code Fix & Linter leads with 242K GitHub stars, followed by several skills from the OpenClaw project — including a Coding Agent Orchestrator (228K stars) that delegates tasks across multiple AI agents simultaneously.
The ecosystem now supports 18 AI agents — not just Claude Code, but also Cursor, GitHub Copilot, Windsurf, Gemini, OpenAI Codex, Cline, Roo, AMP, Kiro CLI, and more. A skill installed via npx skills add works across any compatible agent, which explains the explosive install counts.
How We Tested
This guide is based on installing and using 47 Claude Code skills across 3 real projects over 4 weeks (March to May 2026). The projects were a Next.js 15 + TypeScript marketing site, a Python FastAPI SEO agent, and a small React Native prototype.
What we measured: time saved per task category (debugging, refactoring, feature scaffolding, testing, code review), retention after 7 days of use (did the skill stay installed?), and friction during install. We logged each session with the skill name, task type, completion time, and whether the output needed rework.
Top time-savers in our test: superpowers/test-driven-development (~3 hrs/week saved), superpowers/systematic-debugging (~2.5 hrs), code-reviewer (~2 hrs), superpowers/brainstorming (~1.5 hrs), and frontend-design (~1 hr). Skills we uninstalled within a week: 12 of the 47 were removed because they triggered too often on irrelevant tasks or duplicated capabilities already in our workflow.
What we did not test: enterprise team-orchestration skills, skills requiring paid APIs we do not subscribe to (Figma, certain monitoring services), and skills targeting languages outside our stack (Rust, Java specifics). Time savings are self-reported by a single engineer on a single workflow — your mileage will vary.
Disclosure: no compensation from Anthropic or any skill author. Claude Pro subscription paid out of pocket. Test commit log and skill install dates available on request.
Frequently Asked Questions
What are Claude Code skills?
Claude Code skills are reusable instruction files (SKILL.md) that extend Claude Code's capabilities. They act like plugins — giving the AI agent specific workflows for tasks like TDD, debugging, frontend design, or database optimization. Skills are community-created and shared via GitHub repositories.
How do I install Claude Code skills?
Most skills install by copying the SKILL.md file into your project's .claude/skills/ directory. Some skills use the /install-skill command or can be added via the Claude Code CLI. After installation, invoke them with /skill-name in your Claude Code session.
Are Claude Code skills free?
The vast majority are free and open-source, hosted on GitHub. Some premium skill packs exist but the core ecosystem — including development, DevOps, testing, and design skills — is entirely community-driven and free to use.
What is the difference between Claude Code skills and MCP servers?
Skills are instruction-based — they tell Claude Code HOW to approach a task using structured prompts and workflows. MCP servers are tool-based — they give Claude Code the ABILITY to interact with external services. Many workflows combine both: a skill defines the process, and MCP servers provide the tools.
How do I install Claude Code skills (system-wide)?
Claude Code skills live in ~/.claude/skills/ for system-wide use or .claude/skills/ inside a specific project repo. The fastest install path is `claude skills install <name>` from the official marketplace, which clones the skill into your global directory and registers it. For skills hosted on GitHub directly, clone the repo and either symlink the SKILL.md or copy the entire skill folder into ~/.claude/skills/. Once installed, list available skills with the /skills command. The skill becomes invokable as /<skill-name> in any conversation. No restart is needed — Claude Code re-scans the skills directory each new session.
Are Claude Code skills safe to install from third parties?
Skills are plain Markdown instruction files — they cannot execute code on your machine by themselves. The risk surface is what the skill instructs Claude Code to do: a malicious skill could tell Claude to delete files, exfiltrate API keys, or run dangerous shell commands. Always read the SKILL.md before installing, especially the 'When to use' and 'Steps' sections. Avoid skills that hide instructions behind base64 or external fetches. Pin to specific commit hashes rather than HEAD on untrusted repos. For high-trust skills, prefer anthropics/claude-code-skills, the official superpowers marketplace, and well-starred public skill collections.
Which Claude Code skills save the most time per week?
Based on 4 weeks measuring actual time saved on a Next.js + Python codebase, the highest-ROI skills were: superpowers/test-driven-development (~3 hrs/week saved by avoiding re-debug cycles), superpowers/systematic-debugging (~2.5 hrs/week on intermittent bugs), code-reviewer (~2 hrs/week catching issues pre-PR), superpowers/brainstorming (~1.5 hrs/week on feature scoping), and frontend-design (~1 hr/week vs hand-rolling CSS). Smaller wins: commit-commands, deps-upgrade, deploy-checklist (~30 min/week each). Skills that did NOT save time in our test: meta-skills like skill-creator (useful but not daily) and broad orchestrators that added overhead for simple tasks.
How is a Claude Code skill different from an MCP server?
A skill is a Markdown prompt overlay — it tells Claude HOW to think about a task (workflow, checklists, decision points). An MCP server is a running process that exposes tools (functions Claude can call) and resources (data Claude can read), spoken via JSON-RPC over stdio or HTTP. Skills change behavior; MCP servers add capability. Concrete example: the systematic-debugging skill defines a 5-step debugging methodology, while the Playwright MCP server gives Claude the ability to actually drive a browser. Pick a skill when you want consistent reasoning. Pick an MCP server when you need new tools. Most real workflows combine both: skill provides the process, MCP provides the actions.
Can I write my own Claude Code skill?
Yes — a skill is a single SKILL.md with YAML frontmatter (name, description, when-to-use) followed by Markdown sections covering steps, examples, rules, and pitfalls. Create the file at ~/.claude/skills/my-skill/SKILL.md and Claude Code picks it up next session. The skill-creator meta-skill walks you through scaffolding and avoids common mistakes like vague descriptions that hurt trigger accuracy. Test your skill by triggering it from a fresh session and checking whether Claude follows the steps. Iterate on the description field — that one line determines whether Claude finds your skill at the right moment. Most useful skills are under 200 lines.
Where can I find the best Claude Code skills in 2026?
Three reliable sources as of mid-2026: (1) github.com/anthropics/claude-code-skills — the official Anthropic-curated index, includes vetted skills like brainstorming, test-driven-development, and systematic-debugging. (2) The superpowers-marketplace plugin — install via /plugin and get a curated bundle of high-quality skills with one command. (3) gstack and similar third-party indexes — community-curated, broader coverage including frontend-design, devex-review, and verify. For specific niches (SEO, ads, design), check our SkillsMap directory which catalogs 349 skills across 14 categories. Avoid uncurated mass-imports — review SKILL.md files individually before installing.
Do Claude Code skills work with Sonnet 4.5 and Opus 4.7?
Yes. Skills are model-agnostic prompt overlays and work on Sonnet 4.5, Sonnet 4.6, Opus 4.5, Opus 4.7, and Haiku 4.5. Some skills include guidance on which model is preferred — e.g., complex orchestration skills suggest Opus tier for the planning phase, while execution sub-tasks run on Sonnet for cost. The Skill tool itself is part of Claude Code's runtime regardless of model. One caveat: skills that depend on the 1M-context Opus build (e.g., very-large-codebase agents) may give degraded results on Sonnet's shorter context window. Check the skill's frontmatter or README for model-tier hints.
How do Claude Code skills handle long-running tasks?
Skills that run long tasks typically integrate with TaskCreate / Monitor / TaskStop tools to dispatch work in the background. The pattern: the skill creates a sub-task with a clear acceptance criterion, returns control to the main session immediately, and the user (or a follow-up skill) checks status via Monitor or TaskList later. For multi-hour work, skills like ralph-loop run on a continuous interval with checkpoint files so a session restart picks up where the previous one stopped. Inside a single conversation, long sequences use checkpointing — writing partial progress to memory/ or a state file so context compaction does not lose progress. Skills should avoid blocking the main session for more than a few minutes.
Save on AI Subscriptions
Claude Code skills need Claude Pro. Get it at 30-40% off through GamsGo shared plans — use code WK2NU
Browse the Full Skills Directory
Search and filter 349 skills across 12 categories. Sort by GitHub stars, filter by platform compatibility, and find the right skill for your workflow.
Open SkillsMap →