Loading skills catalog...

🧠 Agent Skills Directory

Informed discoverability for AI agent skills

Find the RIGHT skills for your project β€’ Quality scored β€’ Maintenance tracked β€’ Security validated

πŸ’‘ The Problem: There are over 30K+ skills. Which ones should you use in your project? Our Solution: Quality scores, LGTM validation, and security scanning to help you make informed decisions. Browse here, install with skills.sh. You don't know which? Use our scanner to discover the best for your project.
- Total Skills
- Providers
- Categories
- Showing
Loading skills...

🌟 Have a Skill Repository?

We aggregate skills from trusted sources. Submit your repository to be included in the directory.

βž• Request New Source

πŸ“¦ Curated Skill Bundles

Pre-configured collections of skills grouped by use case. Install an entire stack with one command!

- bundles available
Loading bundles...

🌐 Ecosystem Exports

Pre-filtered skill catalogs optimized for different AI ecosystems. Use these endpoints directly in your tools, MCP servers, or integrations.

πŸ€–

Claude Skills

-

High-quality skills (score β‰₯ 50) compatible with Claude Code

cdn.jsdelivr.net/gh/dmgrok/agent_skills_directory@main/exports/claude-skills.json
πŸ™

Copilot Skills

-

High-quality skills (score β‰₯ 50) compatible with GitHub Copilot

cdn.jsdelivr.net/gh/dmgrok/agent_skills_directory@main/exports/copilot-skills.json
πŸ”—

MCP-Compatible

-

Skills with MCP integration support

cdn.jsdelivr.net/gh/dmgrok/agent_skills_directory@main/exports/mcp-compatible.json
⭐

Premium Skills

-

Top-quality skills with score β‰₯ 70

cdn.jsdelivr.net/gh/dmgrok/agent_skills_directory@main/exports/premium-skills.json
🟒

Active Skills

-

Recently maintained skills (updated within 6 months)

cdn.jsdelivr.net/gh/dmgrok/agent_skills_directory@main/exports/active-skills.json

🏷️ Badges for Skill Authors

Add these badges to your repository README to show your skill is listed in the Agent Skills Directory.

πŸ“Œ "Listed on" Badge

Show that your skill is part of the curated directory:

Listed on Agent Skills Directory
[![Listed on Agent Skills Directory](https://img.shields.io/badge/Listed_on-Agent_Skills_Directory-6366f1?style=flat)](https://dmgrok.github.io/agent_skills_directory/)

πŸ“Š Dynamic Catalog Badges

Live-updating badges powered by shields.io endpoints:

Skills count Providers count Quality
![Skills](https://img.shields.io/endpoint?url=https://cdn.jsdelivr.net/gh/dmgrok/agent_skills_directory@main/exports/badge-skills.json)
![Providers](https://img.shields.io/endpoint?url=https://cdn.jsdelivr.net/gh/dmgrok/agent_skills_directory@main/exports/badge-providers.json)
![Quality](https://img.shields.io/endpoint?url=https://cdn.jsdelivr.net/gh/dmgrok/agent_skills_directory@main/exports/badge-quality.json)

⭐ Quality Score Badge

Show your skill's quality score (replace {score} and {color}):

Quality Score 85 Quality Score 65 Quality Score 45

πŸ”Œ Validate Your Skill

Every skill submitted to the directory goes through a comprehensive automated validation pipeline. Use our reusable GitHub Action to validate locally, or submit and our pipeline handles the rest.

πŸ›‘οΈ Validation Pipeline β€” All Checks We Run

1
πŸ“„ YAML Frontmatter Validation
Parses the --- frontmatter block in SKILL.md. Validates structure, required fields, and correct types. A malformed YAML block or missing required field is an immediate hard failure.
name present description present category present tags is array version present globs optional valid YAML syntax no duplicate keys
Required
2
🏷️ Naming & Format Conventions
Validates skill name format and version string. Reserved generic names are flagged to keep the catalog meaningful.
lowercase only alphanumeric + hyphens 2–50 chars no leading/trailing hyphen no reserved names test, demo, sample, app… semver 2.0 version description 10–200 chars ≀10 keywords known runtime known capabilities LICENSE present
Required
3
πŸ“ Content Quality Analysis
Analyzes the instruction body for completeness. Detects placeholder / template text that was never filled in, and checks for recommended structural sections and code examples.
β‰₯100 chars total body β‰₯50 words no TODO no FIXME no TBD no Lorem ipsum no "placeholder" text no "add your content here" Usage / How to use section Examples section fenced code blocks <5000 words (not too long)
Quality
4
πŸ”’ Secrets Detection β€” 30+ Patterns
Two-layer scan. Primary: detect-secrets (Yelp) with entropy-based detection and 20+ provider-specific plugins. Fallback: built-in regex patterns for 30+ known secret formats. Also runs gitleaks v8 against the full repository tree.
OpenAI sk-… / sk-proj-… Anthropic sk-ant-api… GitHub PAT ghp_… GitHub OAuth gho_… GitHub fine-grained github_pat_… GitHub App ghu_/ghs_ AWS Access Key AKIA… AWS Secret Key (entropy) Google API AIza… Google OAuth ya29.… Slack bot xoxb-… Slack webhook URL Stripe live sk_live_… Stripe test sk_test_… Discord bot token Discord webhook URL npm token npm_… PyPI token pypi-AgE… Twilio API SK… SendGrid SG.… Private key PEM header generic api_key=… generic secret=… generic token=… Base64 high-entropy strings Hex high-entropy strings JWT tokens gitleaks full-repo scan
Required
5
πŸ›‘οΈ Injection & Malicious Pattern Detection
Scans skill instructions for patterns that could cause an AI agent to execute dangerous operations. Covers shell injection, code injection, RCE, obfuscation, and data exfiltration vectors.
destructive rm -rf / targeting system dirs pipe-to-shell curl … | bash pipe-to-shell wget … | sh pipe-to-python curl … | python3 code injection eval(input(…)) code injection exec(request.…) dynamic import __import__(argv) shell injection subprocess(shell=True) SQL injection via string concat obfuscation base64.decode β†’ exec prompt injection attempts jailbreak instructions
Lakera Guard AI is also optionally used to detect prompt-injection in skill content.
Required
6
⭐ LGTM Quality Scoring (0–100)
Automated quality assessment using the LGTM Agent Skills action. Minimum score: 70/100 to pass. Sub-scores are weighted and visible in the PR comment.
Spec Compliance Security KPI Content Quality Testing coverage Originality score <70 β†’ fail score 70–84 β†’ manual review score β‰₯85 β†’ auto-approve
Automated
7
πŸ”„ Duplicate & Similarity Check
Checks the full catalog for exact name clashes and near-duplicates. Uses rapidfuzz (C-optimized) β†’ python-Levenshtein β†’ pure-Python fallback for the similarity calculation. Top-3 similar names are surfaced in the report.
exact name match β†’ block Levenshtein similarity >80% β†’ warn top-3 similar names reported mirror / fork detection normalized similarity ratio (0–1)
Quality
8
πŸ§ͺ Provider Integration Test
Live end-to-end test against the actual repository. Confirms the provider config resolves correctly, files are reachable, and at least one valid SKILL.md can be parsed before the PR is created.
GitHub API tree fetch recursive tree traversal SKILL.md reachable via raw URL frontmatter parseable skills_path_prefix valid branch exists (main/master) rate-limit check (needs GITHUB_TOKEN)
Automated

πŸ”§ Run Validation in Your Repo

# .github/workflows/validate.yml
name: Validate Skill
on: [push, pull_request]
jobs:
  validate:
    uses: dmgrok/agent_skills_directory/.github/workflows/validate-skill.yml@main
    with:
      skill-path: '.'

This validates frontmatter schema, required fields, content length, and scans for leaked secrets. On success, it suggests a badge to add to your README.

πŸ“‘ Distribution Channels

The Agent Skills Directory is available through multiple distribution channels for maximum reach.

🌐 CDN (jsDelivr)

Direct JSON access via CDN. Zero setup, instant integration.

cdn.jsdelivr.net/gh/dmgrok/agent_skills_directory@main/catalog.json

⚑ CLI Tool

npm-like package manager. Search, install, manage skills.

brew install dmgrok/tap/skills

πŸ€– MCP Server

Mother Skills MCP for auto-detection and installation.

github.com/dmgrok/mcp_mother_skills

🐍 PyPI

Python package installation for CLI access.

pip install agent-skills-directory

πŸ“¦ GitHub Releases

Versioned releases with vYYYY.MM.DD tags.

github.com/dmgrok/agent_skills_directory/releases

πŸ” Web Browser

This site! Browse, search, filter, and install skills.

dmgrok.github.io/agent_skills_directory

⚑ Skills CLI

The package manager for AI agent skills. Search, install, and manage skills from your terminalβ€”just like npm for packages.

macOS Installation

⚑ Quick Install Script

curl -fsSL https://raw.githubusercontent.com/dmgrok/agent_skills_directory/main/install.sh | sh

πŸ“₯ Direct Download

Download for macOS (ARM64 & Intel)

Linux Installation

🍺 Homebrew (if installed)

brew install dmgrok/tap/skillsdir

πŸ“₯ Direct Download

Download for Linux (x64)

Windows Installation

Install from Source (Python Required)

🐍 pip install (editable)

git clone https://github.com/dmgrok/agent_skills_directory.git
cd agent_skills_directory && pip install -e .

πŸ› οΈ Command Reference

skillsdir search <query>

Discovery

Search for skills by name, description, or tags

$ skillsdir search "web scraping"
Found 5 skill(s):

  anthropic/web-researcher
  Advanced web research and data extraction
  #web #research #scraping

  github/browser-use
  Browser automation for web interaction
  #browser #automation #web

skillsdir info <skill-id>

Discovery

View detailed information about a skill

$ skillsdir info anthropic/pdf

pdf
────────────────────────────────────────
  id:          anthropic/pdf
  provider:    anthropics
  category:    documents
  description: Process and analyze PDF documents
  tags:        pdf, documents, extraction
  repo:        https://github.com/anthropics/skills

  Not installed. Run: skillsdir install anthropic/pdf

skillsdir install <skill-id>

Install

Install a skill to ~/.skills/installed/ (global) or project directory

$ skillsdir install github/web-design-reviewer
β†’ Installing github/web-design-reviewer...
βœ“ Installed github/web-design-reviewer β†’ ~/.skills/installed/github/web-design-reviewer

# Install a specific version
$ skillsdir install anthropic/pdf@1.2.0

# Force reinstall
$ skillsdir install anthropic/pdf --force

# Global install (default)
$ skillsdir install anthropic/pdf
β†’ Installing anthropics/pdf [global]...
βœ“ Installed anthropics/pdf β†’ ~/.skills/installed/anthropics/pdf

# Project install (auto-detects agent)
$ skillsdir install anthropic/pdf --project
β†’ Installing anthropics/pdf [project (copilot)]...
βœ“ Installed anthropics/pdf β†’ .github/skills/pdf
β†’ Tip: Reference this skill in your .github/copilot-instructions.md

# Project install with specific agent
$ skillsdir install anthropic/pdf -p --agent claude
β†’ Installing anthropics/pdf [project (claude)]...
βœ“ Installed anthropics/pdf β†’ .claude/skills/pdf
β†’ Tip: Reference this skill in your CLAUDE.md

# Other agents
$ skillsdir install anthropic/pdf -p --agent copilot  # .github/skills/
$ skillsdir install anthropic/pdf -p --agent codex    # .codex/skills/
$ skillsdir install anthropic/pdf -p --agent cursor   # .cursor/skills/

skillsdir detect

Discovery

Show detected agent and skill paths for current directory

$ skillsdir detect

Agent Detection
────────────────────────────────────────
  Detected agent:  GitHub Copilot (copilot)
  Project path:    /path/to/your/project

Installation Paths
────────────────────────────────────────
  Project paths (--project):
    βœ“ .github/skills
      .claude/skills

  Personal path (--global):
      ~/.copilot/skills

  Instructions file:
    βœ“ .github/copilot-instructions.md

skillsdir list

Manage

List all installed skills grouped by location

$ skillsdir list

Installed skills (3):

  [global]
    anthropic/pdf@1.0.0
    Process and analyze PDF documents

  [project (copilot)]
    github/web-design-reviewer@0.0.0
    Visual inspection of websites

  [project (claude)]
    vercel/nextjs-app-router@1.2.0
    Next.js App Router development

# Output as JSON
$ skillsdir list --json

skillsdir init

Publish

Create a new skill.json manifest interactively

$ skillsdir init

Creating skill.json

skill name (my-skill): web-scraper
version (1.0.0): 
description: Scrape and extract data from websites
author: Your Name
license (MIT): 
runtime (universal): mcp
keywords (comma-separated): web, scraping, extraction

βœ“ Created skill.json
βœ“ Created SKILL.md

skillsdir uninstall <skill-id>

Manage

Remove an installed skill

$ skillsdir uninstall anthropic/pdf
Remove anthropic/pdf from ~/.skills/installed/anthropic/pdf? [y/N] y
βœ“ Uninstalled anthropic/pdf

skillsdir update

Manage

Check for and apply updates to installed skills

$ skillsdir update

Updates available:
  anthropic/pdf
  vercel/nextjs-app-router

Update 2 skill(s)? [y/N] y
βœ“ Updated anthropic/pdf
βœ“ Updated vercel/nextjs-app-router

skillsdir config

Manage

View or modify CLI configuration

$ skillsdir config list
{
  "registry": "https://cdn.jsdelivr.net/.../catalog.json",
  "cache_ttl": 3600,
  "auto_update": true
}

$ skillsdir config set cache_ttl 7200
βœ“ Set cache_ttl = 7200

πŸ“€ Publishing Commands

skillsdir login

Publish

Authenticate with GitHub for publishing skills

$ skillsdir login

GitHub Authentication
────────────────────────────────────────
β†’ Opening browser to generate a personal access token...
  URL: https://github.com/settings/tokens/new?...

Paste your token: ghp_xxxxxxxxxxxx
βœ“ Authenticated as: your-username
βœ“ Token saved to ~/.skills/config.json

skillsdir validate

Publish

Validate a skill directory before publishing

$ skillsdir validate ./my-skill

Validating Skill
────────────────────────────────────────
  Directory: /path/to/my-skill
β†’ Fetching catalog for duplicate check...

βœ… Valid

Checks passed:
  βœ“ skill.json schema valid
  βœ“ SKILL.md has content
  βœ“ No secrets detected
  βœ“ No malicious patterns
  βœ“ No duplicate name

skillsdir publish

Publish

Publish a skill to GitHub and optionally submit to the directory

$ skillsdir publish

Publishing Skill
────────────────────────────────────────
β†’ Running validation checks...
βœ“ Valid skill: web-scraper@1.0.0
βœ“ Authenticated as: your-username
β†’ Skill ID: your-username/web-scraper
β†’ Pushing to GitHub...
βœ“ Created release: v1.0.0

βœ“ Published your-username/web-scraper@1.0.0

  To add to the official directory:
    skillsdir publish --submit

# Submit to official directory
$ skillsdir publish --submit
β†’ Creating submission issue...
βœ“ Submission created!
  https://github.com/dmgrok/agent_skills_directory/issues/123

# Preview without changes
$ skillsdir publish --dry-run

🎯 What are Agent Skills?

Agent Skills are modular, reusable instruction packages that extend AI coding agents with specialized capabilities. Unlike monolithic instruction files, skills are composable, version-controlled, and automatically matched to your project context.

❌ Traditional Approach

  • CLAUDE.md - Single file, grows unwieldy
  • .github/copilot-instructions.md - Agent-specific
  • .cursorrules - Vendor lock-in
  • AGENTS.md - Manual maintenance
πŸ”΄ One giant file per project πŸ”΄ No reusability across projects πŸ”΄ Manual updates required πŸ”΄ Agent-specific formats

βœ… Agent Skills Approach

  • SKILL.md - Focused, single-purpose
  • Works with Claude, Copilot, Codex
  • Auto-matched to tech stack
  • Community-maintained registry
🟒 Modular & composable 🟒 Share across projects 🟒 Auto-updates from registry 🟒 Agent-agnostic standard

πŸ“ Skill Structure

my-skill/
β”œβ”€β”€ SKILL.md        # Instructions + metadata (required)
β”œβ”€β”€ scripts/        # Executable automation (optional)
β”œβ”€β”€ references/     # Documentation files (optional)
└── assets/         # Templates, schemas (optional)

πŸ› οΈ How to Build a Good Skill

  • Start with a clear goal and a single responsibility per skill.
  • Keep instructions concise; prefer step lists over prose.
  • Include context guards (prereqs, tech stack) to avoid misuse.
  • Scope to file types using globs (e.g., *.tsx, **/*.test.ts) so the skill only applies where relevant.
  • Document inputs/outputs and expected files touched.
  • Add short examples or test commands that validate success.
  • Reference trusted sources (docs, standards) when relevant.
  • Check for safety: no secrets, destructive commands, or ambiguous actions.
  • Use tags and category from the catalog schema for easy discovery.

πŸ“– See the Agent Skills specification for full guidelines.

🎯 Scoping Skills to File Types

Skills can be limited to specific files, paths, or contexts using the globs field or context guards:

MethodExampleUse Case
globs (frontmatter) ["*.tsx", "*.jsx"] Only apply to React component files
globs (path patterns) ["src/components/**/*"] Only apply within specific directories
globs (test files) ["**/*.test.ts", "**/*.spec.ts"] Only apply to test files
Context guards (markdown) "Skip when editing config files" Soft guidance for the agent

πŸ“ Sample SKILL.md Template

---
name: my-awesome-skill
description: Brief description of what this skill does
category: development
tags: [typescript, testing, react]
globs: ["*.tsx", "*.jsx", "src/components/**/*"]
version: 1.0.0
---

# My Awesome Skill

## Context
- **Applies to:** `*.tsx`, `*.jsx` files in React projects
- **Skip when:** Editing config files, markdown, or non-component code
- **Requires:** Node.js 18+, React 18+

## Instructions

1. **Step One** β€” Clear action the agent should take
2. **Step Two** β€” Next action with expected outcome
3. **Step Three** β€” Final step or validation

## Examples

```bash
# Command to verify the skill worked
npm test
```

## References
- [Official Docs](https://example.com/docs)

πŸš€ How to Use Skills

Choose the method that works best for your workflow

πŸ€– Mother Skills MCP Server Recommended

Auto-detects your tech stack and installs matching skills automatically.

1

Configure MCP Server

Add to .vscode/mcp.json (Copilot) or .mcp.json (Claude):

{
  "mcpServers": {
    "mother-skills": {
      "command": "node",
      "args": ["/path/to/mcp-mother-skills/dist/index.js"],
      "env": { "MOTHER_PROJECT_PATH": "." }
    }
  }
}
2

Sync Skills

In your AI agent conversation:

sync_skills

Mother detects React, TypeScript, Docker, etc. and installs relevant skills.

3

Install Specific Skills or Bundles

install_skill("pdf")
install_bundle("web-development")
πŸ“¦ Get Mother Skills MCP

πŸ“š Reference Documentation

Additional details about agents, file structure, and manifest format

Skills are installed to different directories based on the detected agent:

AgentProject PathPersonal PathInstructions File
Claude .claude/skills/ ~/.claude/skills/ CLAUDE.md
Copilot .github/skills/ ~/.copilot/skills/ .github/copilot-instructions.md
Codex .codex/skills/ ~/.codex/skills/ AGENTS.md
Cursor .cursor/skills/ ~/.cursor/skills/ .cursorrules
# Global skills (shared across projects)
~/.skills/
β”œβ”€β”€ config.json          # CLI configuration
β”œβ”€β”€ installed/           # Global skills
β”‚   └── anthropic/pdf/
└── cache/               # Catalog cache

# Project skills (per-agent)
your-project/
β”œβ”€β”€ .claude/skills/      # Claude skills
β”‚   └── pdf/
β”œβ”€β”€ .github/skills/      # Copilot skills
β”‚   └── web-design-reviewer/
β”œβ”€β”€ .codex/skills/       # Codex skills
└── .cursor/skills/      # Cursor skills

Define your skill's metadata, dependencies, and capabilities:

{
  "name": "my-skill",
  "version": "1.0.0",
  "description": "A skill that does amazing things",
  "author": "Your Name",
  "license": "MIT",
  "runtime": "mcp",
  "keywords": ["automation", "productivity"],
  "entry": "SKILL.md",
  "dependencies": {
    "anthropic/browser-use": "^1.0.0"
  },
  "capabilities": ["web-browsing", "file-system"],
  "inputs": [
    { "name": "url", "type": "string", "required": true }
  ],
  "outputs": [
    { "name": "result", "type": "object" }
  ]
}

πŸ“– See the full skill manifest schema.

πŸŽ›οΈ Filtering via URL

Share filtered views by adding query params to the docs URL:

?provider=anthropics
?provider=vercel
?category=development
?search=notion       # or ?q=...
?tags=git,api        # comma-separated
?id=anthropics/pdf   # open modal directly

Combine params as needed, e.g. ?provider=openai&category=documents&search=pdf or ?provider=vercel.