Awesome Claude Skills: Best Agent Skills and Extensions for 2026

Automated GitHub trackers indexed 24,403 repositories containing Claude Code skill and plugin definitions by late June 2026, according to the quemsah/awesome-claude-plugins project. Six months earlier, the count sat closer to 2,000. That 12x growth happened because Anthropic landed on a format simple enough that anyone could contribute: a SKILL.md file with YAML frontmatter and markdown instructions, dropped into a folder. Claude loads the skill only when it detects relevance to the current task, so installing dozens of them costs nothing in context until one fires.

The format follows the Agent Skills open standard at agentskills.io, meaning the same SKILL.md file works across Claude Code, Gemini CLI, Cursor, Aider, and Windsurf. Community directories like Awesome Claude at awesomeclaude.ai now catalog 169 skills across 13 categories. But with thousands more scattered across GitHub, the ecosystem needs a curated guide.

This roundup covers the skills worth installing first, organized by what they do. Every skill named here was verified as a real, installable project at the time of writing.

These skills change how Claude writes and reviews code. They enforce structure, prevent common LLM failure patterns, and add review workflows that catch problems before they reach your main branch.

Superpowers

By Jesse Vincent. Forces a plan-first, test-driven workflow: brainstorm, design, plan, write tests, implement, review. It runs implementation in isolated subagents using git worktrees, so failed attempts stay quarantined from your working tree. With 40,900+ GitHub stars and compatibility across nine coding agents, Superpowers is the most recommended single skill across every 2026 roundup.

Install: /plugin marketplace add obra/superpowers-marketplace

Karpathy Behavioral Skill

Andrej Karpathy published observations about LLM coding failure patterns in January 2026, and within weeks someone encoded them into a SKILL.md file. The skill enforces four rules: verify assumptions before acting, use the minimal solution, prevent unrelated changes from creeping in, and validate before execution. It hit 144,000+ GitHub stars, making it the fastest-growing Claude Code skill of the year. Zero runtime dependencies.

Install: npx skills add https://github.com/forrestchang/andrej-karpathy-skills --skill karpathy-guidelines

Systematic Debugging

Part of the Superpowers collection. Replaces the "change things and hope" approach with hypothesis-driven investigation. You form a hypothesis, test it in isolation, verify the root cause, then fix. Developers who adopt the methodology report saving roughly 2.5 hours per week on debugging tasks.

Install: npx skills add obra/superpowers --skill systematic-debugging

Code-Reviewer

By sanyuan0704. Runs structured code review against SOLID principles, architecture patterns, and security baselines. Tags every finding by severity (P0 through P3) so you can triage instead of reading a flat list. Pairs well with Superpowers for a complete write-then-review loop.

Frontend Design

The most-installed design skill in the ecosystem with 277,000+ downloads as of early 2026. It guides Claude to produce distinctive, production-grade interfaces instead of the generic AI design aesthetic that plagues most generated UIs. Covers typography, color systems, motion design, spatial composition, and depth effects. Works with React, Vue, Svelte, and plain HTML.

Install: npx skills add anthropics/skills --skill frontend-design

UI/UX Pro Max

Takes a different angle from Frontend Design. Where Frontend Design focuses on aesthetics, this skill emphasizes design reasoning. It ships with 50+ design styles, 161 color palettes, 57 font pairings, and 25 chart types. Every recommendation comes with an explanation for why that choice works, which makes it easier to modify the output without losing visual coherence.

Design Auditor By Ashutos1997. Audits existing designs against 17 professional standards. Useful when you need to review work that already exists rather than generate from scratch. Think of it as a design linter: it flags issues without rewriting your components.

Security skills matter because Claude generates code fast enough that manual review can't keep up. These skills apply security checks during generation, not after.

Trail of Bits Security Skills

From one of the most respected security research firms in the industry. The collection includes CodeQL and Semgrep integration for static analysis, plus guided code auditing workflows based on the firm's real-world audit methodology. If you install one security skill, make it this one.

VibeSec

Helps Claude write secure code during generation rather than catching vulnerabilities after the fact. Targets injection attacks, broken access control, and insecure deserialization before they land in your codebase.

OWASP Security

By agamm. Implements OWASP Top 10:2025 and ASVS 5.0 standards as actionable checklists that Claude applies during code review.

For research and data work, three skills stand out:

Claude Scientific Skills

By K-Dense-AI. Contains 125+ skills spanning bioinformatics, cheminformatics, clinical research, and computational materials science. This is the largest single-domain skill collection in the ecosystem, and it turns Claude into a capable lab notebook companion for researchers working across disciplines.

Paper Search

By ykdojo. Connects Claude to OpenAlex's database of 250+ million academic works. When you need to ground technical writing with peer-reviewed citations, this skill saves significant time compared to manual searches.

CSV Data Summarizer

Runs statistical analysis on CSV files including distributions, correlations, and anomaly detection. It handles the kind of exploratory data work that would otherwise mean switching to a Jupyter notebook or writing one-off pandas scripts.

Give your Claude skills a persistent workspace

A shared workspace with version history, semantic search, and MCP-native access for your agent's file operations. Upload skill outputs, index them automatically, and let teammates review in the browser. Start with a 14-day free trial.

Start 14-Day Trial

Composio

Connects Claude with 1,000+ external apps through normalized tool interfaces. Handles OAuth lifecycle management, structured schemas, and API credential isolation. If you need Claude to interact with Jira, Linear, Slack, or GitHub without building custom integrations, start here.

Install: git clone https://github.com/ComposioHQ/awesome-claude-skills.git

Agent-Browser

Headless browser automation using ref-based element targeting instead of brittle XPath or CSS selectors. With 14,000+ GitHub stars, it reflects how often agents need to navigate web interfaces for testing, scraping, or workflow automation. Supports parallel sessions and cloud browser providers for isolated testing environments.

Supermemory

A memory and context engine that extracts facts from conversations, tracks how they change over time, and resolves contradictions. Combines RAG with persistent memory across sessions. Syncs with Google Drive, Gmail, and Notion so context stays current without manual updates.

File Organizer

By Composio. Sorts generated files into consistent directory structures. When agent workflows produce dozens of output files per run, this skill keeps the output navigable for both the agent and the humans who review the work.

These automation skills generate real files: reports, analyses, organized directories. Those outputs need to persist across sessions and stay accessible to your team. Local disk works for individual use, but team workflows benefit from a shared workspace where outputs are versioned and searchable. Fast.io provides this layer. Enable Intelligence on a workspace and uploaded files are automatically indexed for semantic search and RAG chat. Agents can read and write through the Fast.io MCP server, while teammates review outputs in the browser without touching the command line. See the agent storage guide for a setup walkthrough.

Four directories cover most of the ecosystem:

1. Awesome Claude at awesomeclaude.ai/awesome-claude-skills catalogs 169 curated skills across 13 categories with descriptions and install links. Updated regularly and the closest thing to an official community index.

2. travisvn/awesome-claude-skills on GitHub has 13,800+ stars. The original community curated list, organized by official and community contributions.

3. ComposioHQ/awesome-claude-skills on GitHub focuses on integration-heavy skills and includes Composio's own collection alongside community picks.

4. Claude Skills Hub at claudeskills.info lists 658+ skills with community ratings, making it easier to filter by quality rather than browsing alphabetically.

You can also search GitHub directly for SKILL.md to find skills that haven't been listed in any directory yet.

Three Installation Methods

Plugin browser. Type /plugin in Claude Code, search for the skill, and install with a single confirmation. The fastest option when the skill is listed in the marketplace.

npx install. Run npx skills add <repo> --skill <name> to pull a skill directly from any GitHub repository. Works for skills that aren't in the plugin marketplace.

Manual drop. Download the SKILL.md file and place it in ~/.claude/skills/<skill-name>/SKILL.md for personal use, or .claude/skills/<skill-name>/SKILL.md inside your project for project-scoped access. Run /skills after restarting Claude Code to confirm it loaded.

What Goes in a SKILL.md File

Each skill is a folder containing at minimum a SKILL.md file. The file has two parts: YAML frontmatter between --- markers declares the skill's name and description, and markdown content below provides the instructions Claude follows. You can optionally include scripts, templates, and reference files in the same directory. Claude reads only the frontmatter at session start (roughly 100 tokens per skill) and loads the full body when it decides the skill is relevant to the current task.

---
name: my-deployment
description: Deploy the app to staging with rollback safety checks.
---

## Steps
1. Run the test suite
2. Build the production bundle
3. Push to the staging branch
4. Verify the health check endpoint responds

That file, placed in .claude/skills/my-deployment/SKILL.md, creates the /my-deployment slash command. The entire spec fits on an index card.