Awesome Agent Skills

Agent Skills are folders of instructions, scripts and resources, each with a SKILL.md file, that coding agents such as Claude Code, Codex, Cursor, Gemini CLI and GitHub Copilot load on demand. The format is an open standard.

Browse it as a searchable site: sagargupta.online/awesome-skills, with search, category and agent filters, and live stars. It is rebuilt from this README every hour.

This list is curated, not exhaustive, and open to anyone. The point of it is that everything on it has been checked. Curated does not mean famous: your project needs no minimum star count or age, and submitting your own skill is welcome. Every entry:

• is a real skill -- the linked repository, or the folder inside one, holds at least one SKILL.md whose frontmatter has a name and a description
• states an open-source licence -- a LICENSE file, or a licence field in its package manifest or its SKILL.md
• was pushed within the last 180 days, and is not archived
• has a README that says how to install or use the skill
• appears exactly once, in one category

Those rules are enforced by CI, not by good intentions: a pull request that breaks the format, adds a dead link, or submits a repository with no SKILL.md or no licence fails its checks. A monthly audit re-checks every listed repository and files what has rotted.

A skill can run scripts on your machine and shapes what your agent does with your files and credentials. Read its SKILL.md and its scripts before you install it -- a place on this list is a filter, not an audit. See using a skill safely and SECURITY.md.

Contents

• Official
• Skills
  • Development
  • Testing & QA
  • DevOps & Cloud
  • Data & Databases
  • Documents & Office
  • Design & Frontend
  • Security
  • Writing & Content
  • Research & Knowledge
  • Productivity
  • Skill Authoring
• Collections
• Tools
• Tutorials & Articles
• Community

Official

Skill repositories maintained by the company behind the product or agent they cover, and the specification itself.

• Agent Skills specification - The open format: SKILL.md rules, the clients that support it and guides for skill authors.
• Anthropic plugin directory - Anthropic's curated Claude Code plugin marketplace, with first-party plugins and vetted partner plugins that can bundle skills.
• Anthropic skills - Anthropic's public skills for documents, design, MCP and skill creation. Most are Apache-2.0; the four document skills are source-available.
• Cloudflare skills - Skills that teach agents to build on Cloudflare, with plugin installs for Claude Code, Codex and Cursor.
• GitHub awesome-copilot - GitHub's community collection of skills, agents and instructions for Copilot.
• Google skills - Skills for Google products, including Google Cloud and the Gemini API.
• Google Stitch skills - Skills that pair with the Stitch MCP server for design-to-code work.
• Hugging Face skills - Skills for working with the Hugging Face ecosystem and its CLI.
• Microsoft skills - Skills and agent guidance for Azure and Microsoft Foundry SDKs.
• NVIDIA skills - NVIDIA-verified skills for CUDA-X, robotics, simulation and RAG, mirrored daily from NVIDIA's product repositories.
• OpenAI skills - The Codex skills catalog, split into curated and system skills and licensed per skill.
• Sentry skills - Skills the Sentry team uses for its own development, such as commits and settings audits.
• Stripe AI - Stripe's agent toolkit, which includes skills for building with Stripe.
• Supabase agent skills - Skills that help agents build on Supabase.
• Vercel agent skills - Vercel's skills for React and Next.js performance, web design review, React Native, view transitions and Vercel deploys.

Skills

One skill per row. Works with names the agents a skill supports; Any means it sticks to the open standard and has no agent-specific parts.

Development

Testing & QA

DevOps & Cloud

Data & Databases

Documents & Office

Design & Frontend

Security

Writing & Content

Research & Knowledge

Productivity

Skill Authoring

Collections

Multi-skill packs and plugin marketplaces, listed once as a whole.

Tools

Validators, linters, installers and registries for skills.

• Skill Seekers - Converts documentation sites, repositories and PDFs into skills, with conflict detection.
• skill-validator - CLI that validates skill packages against the spec and scores content quality, links and token cost.
• skillcheck - Conformance suite that lints SKILL.md against the spec and tests skills on real agent runtimes.
• SkillOpt - Microsoft's optimizer that improves skill text through trajectory-driven edits and validation gates.
• skills CLI - The npx skills installer that adds skills from GitHub repositories to dozens of coding agents.
• skills.sh - A directory for discovering and installing agent skills, tied to the skills CLI.
• SkillSpector - NVIDIA's scanner that checks skills for prompt injection, exfiltration and risky patterns before install.
• Snyk Agent Scan - Snyk's scanner for agent components, MCP servers and skills, looking for prompt injection and vulnerabilities.
• UI Skills - Registry, CLI and MCP server of UI skills for design engineers, plus its own baseline UI skills.

Tutorials & Articles

• Best practices for skill creators - How to ground a skill in real expertise and keep it well scoped.
• Equipping agents with Agent Skills - Anthropic's October 2025 introduction to skills as folders of instructions, scripts and resources.
• Evaluating skill output quality - How to test whether a skill works reliably, using eval-driven iteration.
• Extend Claude with skills - Claude Code's guide: writing a SKILL.md, invoking skills and what Claude Code adds on top of the open standard.
• Optimizing skill descriptions - How to test and tune a description so the skill triggers on the right prompts.

Community

• Awesome MCP Servers - The sister list: MCP servers, curated and checked by the same CI.

Contributing

Anyone can contribute, including your own skill. Your project needs no minimum star count and no minimum age -- if it holds a SKILL.md, is licensed and is documented, it qualifies. Read CONTRIBUTING.md for the details.

To add a skill:

1. Fork the repo and branch off main.
2. Add one row to the most appropriate existing category. Alphabetical order is preferred, not required.
3. Follow the format: | [Name](URL) | Description | Works with | with the description at 80 characters or fewer and the agents from Claude Code, Codex, Cursor, Gemini CLI, Copilot or Any.
4. Run python scripts/validate.py and fix what it reports.
5. Star the repository. A pull request only merges itself once its author has starred.
6. Open a pull request and answer the affiliation question. Submitting your own project is welcome; not disclosing it is not.

Rather not open a pull request? The skill suggestion form turns into the same one-row pull request and runs the same checks.

Once it merges, a bot thanks you on the pull request with a badge for your project's README. It links back here, so people who find your skill can find the rest of the list:

[![Listed in Awesome Agent Skills](https://sagargupta.online/awesome-skills/badge.svg)](https://github.com/Sagargupta16/awesome-skills)

A star is what lets your pull request merge itself. Without one it waits, labelled missing-star, until you star or a maintainer merges it by hand.

Other ways to help, no pull request needed: report rot (a dead link, an archived upstream, a repository that no longer holds a skill) by opening an issue, since pruning is as useful as adding; fix a description that is too terse or reads like marketing; or triage the monthly health report, where confirming or dismissing an item is a real contribution.

Looking for MCP servers instead? The sister list, listed under Community, holds them to the same kind of checks.

Footnotes

Using a skill safely

An agent reads a skill's name and description up front and loads the rest of SKILL.md, and any scripts or reference files it points to, when a task calls for it. That makes a skill closer to a plugin than to a document.

Read it first. Open SKILL.md and every script in the folder before you install it. Instructions can tell the agent to read, change or send your files, and scripts run with your permissions.

Check what it pre-approves. The specification's experimental allowed-tools field lists tools the skill may use without asking you each time, where the agent supports it.

Install from the source. Copy the folder from the repository listed here, at a commit you have read, rather than from a mirror or a re-upload.

Start narrow. Try a new skill in one project before installing it for every project. Claude Code, for example, reads personal skills from ~/.claude/skills/ and a project's own from .claude/skills/; other agents use their own folders, so check your agent's docs.

Something on the list behaves maliciously? Report it privately, not in a public issue.