skills.sh: The Open Directory That Turns AI Agents Into Specialists

AI coding agents are everywhere. Claude Code, Cursor, GitHub Copilot, Codex, Windsurf, Gemini CLI - the list grows weekly. They all share the same limitation: out of the box, they know how to write code, but they do not know how your team writes code. They lack the procedural knowledge - the step-by-step workflows, conventions, and guardrails - that turn a generic assistant into a specialist. skills.sh is the open ecosystem that fixes this.

What skills.sh Is

skills.sh is an open directory and package manager for agent skills - reusable markdown files that give AI coding agents procedural knowledge. Think of it as npm for agent behavior. A skill tells an agent not just what to do, but how to do it: which steps to follow, what to verify, which patterns to apply, and when to stop.

Installation is a single command:

`npx skills add <owner/repo>`

That is it. The skill gets installed into your project, and every supported agent picks it up automatically. No configuration, no plugins, no API keys.

Why Agent Skills Matter

Without skills, every conversation with an AI agent starts from zero. You explain your testing strategy, your commit conventions, your deployment process - every single time. Skills encode that knowledge once and make it available permanently. The agent stops guessing and starts following your playbook.

This is not a theoretical benefit. The numbers on skills.sh tell the story: over 90,000 total installs across the ecosystem, with the top skills each seeing tens of thousands of downloads. Developers are actively adopting this pattern because it delivers measurable improvements in agent output quality.

19 Agents, One Skill Format

The most compelling aspect of skills.sh is its breadth. A single skill works across 19 different AI agents:

• Claude Code - Anthropic's CLI agent
• Cursor - AI-native code editor
• GitHub Copilot - Microsoft's coding assistant
• Codex - OpenAI's agent
• Gemini CLI - Google's terminal agent
• Windsurf - Codeium's AI IDE
• AMP - Sourcegraph's coding agent
• Cline, Roo, Kilo, Goose, Trae, OpenCode - and more

Write a skill once, use it everywhere. This cross-agent compatibility is what makes the ecosystem viable - you are not betting on a single vendor.

The Skills Leaderboard: What Developers Actually Need

The skills.sh leaderboard ranks skills by install count, with trending and hot categories. The top skills reveal what developers struggle with most when working with AI agents:

Patterns in the Top Skills

Three categories dominate:

• Browser automation - agent-browser and browser-use are the #1 and #3 skills. Agents that can navigate the web, fill forms, and extract data are in massive demand.
• Workflow discipline - writing-plans, verification-before-completion, and finishing-a-development-branch are all from the obra/superpowers repository. They enforce structure: plan before you code, verify before you claim done, clean up when you finish. These skills exist because agents tend to skip these steps without explicit guidance.
• Meta-skills - skill-creator from Anthropic themselves teaches agents how to create new skills. The ecosystem is bootstrapping itself.

Key Repositories Worth Knowing

A few repositories have emerged as the foundational pillars of the ecosystem:

anthropics/skills

Anthropic's official skill repository. Contains skill-creator (117,800+ installs), the meta-skill that teaches Claude Code how to build new skills. If you are using Claude Code, this is where you start.

obra/superpowers

The most prolific repository in the ecosystem, with multiple top-10 skills. Focuses on developer workflow: planning, verification, branch management, test-driven development, and code review. Treats the agent like a junior developer who needs process guardrails - and the results speak for themselves.

vercel-labs/agent-browser

The #1 most-installed skill at 142,800+ installs. Gives agents the ability to interact with websites - navigating pages, filling forms, clicking buttons, extracting data, and testing web applications. Built by Vercel Labs.

coreyhaines31/marketingskills

Proves that skills are not just for developers. This repository includes pricing-strategy, site-architecture, and other marketing-focused skills. With 31,500+ installs on pricing-strategy alone, there is clear demand for business and marketing expertise encoded as agent skills.

How Skills Work Under the Hood

A skill is a markdown file with YAML frontmatter. It gets installed into your project directory (typically `.claude/skills/` for Claude Code or equivalent for other agents), and the agent loads it when relevant. The skill contains:

• Trigger conditions - when to activate (e.g., 'when writing tests', 'before committing', 'when debugging')
• Step-by-step instructions - the procedure the agent should follow
• Guardrails - what to verify, what to avoid, when to stop
• Context - background knowledge the agent needs to make good decisions

Because skills are plain text files stored in your repository, they are version-controlled, reviewable, and shareable. No vendor lock-in, no SaaS dependency, no runtime cost.

Getting Started in 5 Minutes

Here is how to start using skills.sh today:

• Browse the leaderboard at skills.sh to find skills relevant to your workflow
• Install with a single command: `npx skills add obra/superpowers` installs the entire superpowers collection
• Start your agent - the skills are picked up automatically. No configuration needed.
• Create your own - install `npx skills add anthropics/skills` to get the skill-creator, then ask your agent to create custom skills for your team's workflows

The feedback loop is immediate. Install a skill, run your agent, and observe the difference in output quality. The verification-before-completion skill alone can eliminate the most common agent failure mode: claiming work is done when it is not.

What This Means for Development Teams

The skills ecosystem is significant because it solves the knowledge transfer problem that every team using AI agents faces. Today, the senior developer who knows how to structure a PR, run a proper code review, or debug a production incident carries that knowledge in their head. When they use an AI agent, they manually transfer that knowledge through prompts - every time.

Skills formalize that transfer. A team lead can encode their review checklist, deployment procedure, or debugging methodology as a skill, install it across the team's projects, and every agent used by every team member follows the same process. It is institutional knowledge at machine speed.

The Broader Trend

skills.sh is part of a larger shift in how developers interact with AI. We are moving from prompting (telling the agent what to do each time) to programming (encoding behavior that persists). Skills are to AI agents what configuration files are to software - declarative, portable, and composable.

The ecosystem is still early. With 91,000+ total installs and growing, the library of available skills is expanding daily. The most impactful skills have not been written yet - they will come from teams encoding their unique domain expertise into reusable, shareable formats.

Build Smarter Agent Workflows

At webvise, we use agent skills extensively in our own development workflow - from automated code review to deployment verification. If you are looking to integrate AI agents into your team's processes or build custom automation workflows, get in touch. We help teams move from ad-hoc prompting to structured, repeatable AI-assisted development.