The Agent Skills format is universal — SKILL.md with YAML frontmatter works everywhere. What differs across agents is where skills live and how you install them. This guide gives you the setup for all four major agents plus the universal installer that works across 37+ tools.

If you haven't read Part 1, start there for the concept overview and architecture.

Quick Reference: Where Skills Live

Before we dive in, here's the cheat sheet:

Agent

Project Skills Path

User/Global Skills Path

Install Command

Claude Code

.claude/skills/

~/.claude/skills/

/plugin install

GitHub Copilot

.github/skills/ or .claude/skills/

~/.copilot/skills/

Copy to directory

OpenAI Codex

.codex/skills/

~/.codex/skills/

$skill-installer

Cursor

.cursor/skills/ or .claude/skills/

~/.cursor/skills/

npx skills add

Universal (any agent)

varies

varies

npx skills add -a <agent>

Now let's set up each one.

1. Claude Code

Claude Code has the deepest native integration, which makes sense — Anthropic created the standard. Skills are available for Pro, Max, Team, and Enterprise plans (requires code execution enabled).

Where Claude Code looks for skills

Claude Code scans these locations in priority order:

  1. Project-level: .claude/skills/ in your repository

  2. User-level: ~/.claude/skills/

  3. Plugin marketplace: Skills installed via the plugin system

Installing pre-built skills

# Add the Anthropic skills marketplace
/plugin marketplace add anthropics/skills

# Browse and install interactively
# Select "Browse and install plugins" → "anthropic-agent-skills"

# Or install specific skill packs directly
/plugin install document-skills@anthropic-agent-skills
/plugin install example-skills@anthropic-agent-skills

Creating a custom skill

mkdir -p .claude/skills/our-api-patterns

Create .claude/skills/our-api-patterns/SKILL.md:

---
name: our-api-patterns
description: >-
  Generate API endpoints following our team's conventions.
  Use when asked to create routes, handlers, or server actions.
---

# Our API Patterns
Always use the App Router pattern with named exports...

For advanced skills, you can add executable scripts (Python, JavaScript) and reference files. Claude Code can install packages from PyPI and npm when loading skills. Reference additional files from your SKILL.md to keep the core instructions lean — Claude loads them only when needed.

How skills trigger

Claude Code automatically detects skills when your prompt matches a skill's description. You can also invoke explicitly: "Use the frontend-design skill to build a dashboard." Claude can use multiple skills together automatically — this composability is one of the system's most powerful features.

Claude.ai and the API

Pre-built skills (docx, pdf, pptx, xlsx) work automatically in Claude.ai for paid plans. Custom skills can be uploaded as ZIP files via Settings → Features → Capabilities. Via the API, use the /v1/skills endpoints — specify the skill_id in the container parameter along with the code execution tool.

Team/Enterprise note: Skills can be provisioned organization-wide by admins on Team and Enterprise plans.

2. GitHub Copilot

Copilot supports Agent Skills across the coding agent, Copilot CLI, and agent mode in VS Code. Available with Pro, Pro+, Business, and Enterprise plans.

Where Copilot looks for skills

  1. Project skills (repo-specific): .github/skills/ or .claude/skills/

  2. Personal skills (cross-project): ~/.copilot/skills/ or ~/.claude/skills/ (coding agent & CLI only)

  3. Organization/enterprise skills: Coming soon

Creating and adding a skill

Create the directory. Subdirectory names must be lowercase with hyphens:

mkdir -p .github/skills/webapp-testing

Create .github/skills/webapp-testing/SKILL.md:

---
name: webapp-testing
description: >-
  Guide for testing web applications using Playwright.
  Use when asked to create or run browser-based tests.
license: MIT
---

# Web Application Testing with Playwright

## Setup
Install with `npm install @playwright/test`

## Test Structure
Use Page Object pattern for all page interactions...

Key rules: The name must match the directory name. File must be named exactly SKILL.md. Optionally add scripts and resources to the directory and reference them from your instructions.

How skills trigger

When your prompt matches a skill's description, Copilot injects the full SKILL.md into context. Additional files load only when needed. You can also invoke directly:

  • Type /skills in VS Code chat to configure and toggle skills

  • Use /[skill-name] as a slash command in chat

  • In Copilot CLI: /skills list to see available, /skills info for details

GitHub recommends custom instructions (.github/copilot-instructions.md) for always-on coding standards, and skills for detailed, task-specific guidance Copilot should only access when relevant.

Using community skills

The github/awesome-copilot repository contains community-contributed skills organized into collections for DevOps, security, database, cloud, frontend, CI/CD, and more. Copy any skill directory into your .github/skills/ folder. VS Code extensions can also ship skills via the chatSkills contribution point.

3. OpenAI Codex

Codex supports skills in the CLI, IDE extension, and web app — all following the open standard.

Skill scope and locations

Codex loads skills from multiple locations, with higher-precedence scopes overwriting same-named skills from lower ones:

Scope

Location

Use Case

REPO

$CWD/.codex/skills

Skills specific to a working folder (microservice, module)

REPO

$CWD/../.codex/skills

Shared skills in a parent folder within a repo

REPO

$REPO_ROOT/.codex/skills

Root-level skills available to entire repository

USER

~/.codex/skills

Personal skills that apply across all repos

ADMIN

/etc/codex/skills

System-wide defaults for SDK scripts and automation

SYSTEM

Bundled with Codex

Built-in skills (e.g., $skill-creator, $plan)

Creating a skill

Use the built-in $skill-creator skill for guided creation, or create manually:

mkdir -p .codex/skills/deployment-checklist

---
name: deployment-checklist
description: >-
  Run deployment safety checks before pushing to production.
  Use when deploying, releasing, or pushing to staging/prod.
metadata:
  short-description: Optional user-facing description
---

# Deployment Checklist
- All tests pass in CI
- No pending migration conflicts
- Environment variables verified...

Invoking skills

Codex supports two activation modes:

  • Explicit: Run /skills or type $ to mention a skill by name (e.g., $deployment-checklist). Codex web/iOS don't support explicit invocation yet, but skills checked into the repo still activate implicitly.

  • Implicit: Codex automatically activates a skill when your task matches its description.

Installing community skills

Use the built-in $skill-installer to download skills from curated or custom repos:

$skill-installer linear
$skill-installer notion-spec-to-implementation

You can also prompt the installer to fetch from any GitHub repository.

4. Cursor

Cursor adopted the Agent Skills specification and supports skills in Agent mode. Skills are folders of instructions that Cursor loads when relevant — the same open standard used by Claude Code, Copilot, and Codex.

Where Cursor looks for skills

  1. Project-level: .cursor/skills/ or .claude/skills/ in your repo

  2. User-level (global): ~/.cursor/skills/ — available across all projects

  3. Built-in: ~/.cursor/skills-cursor/ — reserved for Cursor's own skills (create-skill, create-rule, etc.). Don't put custom skills here.

Cursor also reads .claude/skills/ at both project and user level, so skills written for Claude Code work in Cursor without duplication.

Creating a skill

mkdir -p .cursor/skills/code-standards

Create .cursor/skills/code-standards/SKILL.md. In Cursor, description is the primary required field — the name field is optional (defaults to folder name):

---
description: >-
  Enforce our team's TypeScript coding standards.
  Use when writing new code, refactoring, or reviewing.
name: Code Standards
---

# Code Standards

## TypeScript Conventions
- Strict mode always enabled
- No `any` types — use `unknown` with type guards
- Prefer `interface` over `type` for object shapes...

Use kebab-case folder names and keep each skill focused on a single responsibility. Add references/ for detailed guides and scripts/ for repeatable automation, referenced from SKILL.md.

How skills activate

Cursor uses the description field as the primary trigger — this is what the agent reads to decide when to apply a skill. You don't need to name the skill in your prompt; natural language like "refactor this to match our standards" is enough for the right skill to activate.

You can also invoke skills directly: type / in Agent chat to search and select skills by name.

Skills vs. rules vs. subagents in Cursor:

  • Rules (always-on): Use for coding conventions that apply to every task

  • Skills (on-demand): Use for task-specific workflows the agent should only access when relevant

  • Subagents: Use for complex multi-step tasks requiring isolated context

Migrating existing rules to skills

If you have dynamic rules or slash commands, Cursor 2.4+ includes /migrate-to-skills which converts eligible dynamic rules (those with alwaysApply: false) and slash commands into proper skills. Run it in Agent chat, review the generated output in .cursor/skills/, and you're done.

Installing community skills

# Via Vercel Skills CLI
npx skills add anthropics/skills --skill frontend-design -a cursor

# Via OpenSkills
npm i -g openskills
openskills install anthropics/skills
openskills sync   # Generates AGENTS.md with skill knowledge

The sync command updates your AGENTS.md file so Cursor's agent mode knows about all installed skills.

The Universal Installer: Vercel's Skills CLI — and most developers do — the Vercel skills CLI is the tool that ties everything together. It installs skills to any agent from any source.

# Install from GitHub (shorthand)
npx skills add anthropics/skills

# Install from full URL
npx skills add https://github.com/vercel-labs/agent-skills

# Target specific agents
npx skills add anthropics/skills -a claude-code -a cursor -a copilot

# Install globally
npx skills add vercel-labs/agent-skills -g

# List available skills in a repo before installing
npx skills add anthropics/skills --list

# Install specific skills to all agents
npx skills add anthropics/skills --skill frontend-design --agent '*'

# Remove skills
npx skills remove web-design-guidelines

# Non-interactive (CI/CD friendly)
npx skills add anthropics/skills --skill pdf -g -a claude-code -y

The CLI supports 37+ agents including Claude Code, Copilot, Codex, Cursor, Windsurf, Cline, Goose, Gemini CLI, Roo Code, Amp, Kiro, and others. It creates symlinks from each agent's expected directory to a canonical copy, keeping everything in sync and deduplicated.

Cross-Agent Compatibility Tips

A few practical patterns for teams using multiple agents:

Commit skills to your repo. Put them in .github/skills/ (Copilot's primary path) — most agents either support this path directly or can be configured to find it.

Use the npx skills CLI for personal skills. It handles the path differences automatically when installing to multiple agents.

Keep the format universal. Stick to name and description in frontmatter. Avoid agent-specific metadata unless you need it. The core spec works everywhere.

Test with your primary agent, verify with others. Skills are text — if they work in one agent, they'll work in others. The main difference is how aggressively each agent matches descriptions to prompts.

Starter Skills to Install Right Now

Get immediate value by installing these pre-built skills:

# Anthropic's document skills (PDF, Word, Excel, PowerPoint)
npx skills add anthropics/skills --skill pdf --skill docx -a claude-code -g -y

# Vercel's React/Next.js best practices (40+ rules)
npx skills add vercel-labs/agent-skills --skill vercel-react-best-practices -g -y

# Vercel's UI review skill (100+ rules for a11y, perf, UX)
npx skills add vercel-labs/agent-skills --skill web-design-guidelines -g -y

# Frontend design (avoids generic "AI slop" aesthetics)
npx skills add anthropics/skills --skill frontend-design -g -y

Coming Up: Build Your Own

You've got skills installed and running across your agents. In Part 3, we'll build a custom skill from scratch, explore the ecosystem tools for discovering and sharing skills, and introduce our Skills Directory — a curated, searchable collection of skills organized by development domain.

Which agent is your daily driver? Reply and tell me your setup — I'm curious what combinations people are running.

📬 Forward this to your team. Everyone using a different agent? This guide gets everyone on the same page.

Subscribe to AI Native Coding →

References & Further Reading:

This is Part 2 of our 3-part Agent Skills series:

  • Part 1: What Are Agent Skills?

  • Part 2: Set Up Skills in Claude Code, Copilot, Codex & Cursor (You are here)

  • Part 3: Build Your First Custom Skill + The Skills Ecosystem

Keep Reading

background
© 2026 Medhavi Dev.
Report abusePrivacy policyTerms of use
beehiivPowered by beehiiv