Primer
How agent skills work
"Skill" is a deliberately broad word in this directory. It covers any artifact that extends what an AI agent can do — read a PDF, query a database, write a slide deck, file an expense report, draft a contract. The artifacts ship in different formats depending on which platform you're using. This page is the short tour.
The platforms, in plain English
Claude Skills
Anthropic's native format. A Claude Skill is a folder with a SKILL.md and optional helper code; Claude reads the SKILL.md, decides when to use the skill, and runs the code in its sandbox. Used inside Claude Code, Claude Desktop, and claude.ai.
MCP Servers
The Model Context Protocol is an open standard. An MCP server is a process — usually Node.js, Python, or Go — that exposes tools to any MCP-aware client. The same MCP server works in Claude Desktop, Cursor, Windsurf, VS Code, and many other clients.
Cursor Rules
A .cursorrules file in your project root, or a folder of .mdc rule files, that steers Cursor's AI assistant. Lighter than a Claude Skill — no execution sandbox, just instruction text — but extremely effective at constraining a coding assistant to your team's conventions.
Windsurf Rules
The Codeium / Windsurf equivalent of Cursor Rules. Same idea, slightly different file format and IDE.
Custom GPTs and OpenAI Apps
GPTs published in the OpenAI GPT Store. Closed-source, hosted by OpenAI, accessed at chatgpt.com/g/g-… URLs. Useful for end-users who don't want to install anything, less useful when you need to inspect or customise the implementation.
n8n Templates
Importable workflow JSON for the n8n automation platform. Each template is a multi-step pipeline (Gmail → OpenAI → Sheets, for example) that you can run on n8n.io or self-host. Strong for ops automations that span SaaS apps.
Zapier Templates and Make Scenarios
Pre-built automations for the two largest no-code workflow platforms. Templates are starting points — you connect your accounts, customise the steps, and run. Distinct from n8n in that the platform is hosted-only.
LangChain and LlamaIndex Packs
Code-level building blocks for developers who are writing their own agents. A LangChain or LlamaIndex pack ships as an npm or PyPI package; it composes with your application logic, doesn't run on its own.
GitHub Copilot Extensions
Extensions that plug into Copilot Chat inside VS Code or github.com. Each extension exposes its own tools (often via MCP under the hood) so Copilot can do platform-specific things — manage GitHub Actions, query a database, edit Figma.
Agent and System Prompts
Curated prompt collections — sometimes a single SYSTEM_PROMPT.md, sometimes a multi-thousand-entry library. Used as either drop-in instructions for an agent or as research material for prompt engineers.
The lines are blurring
A growing number of skills ship simultaneously as a Claude Skill, an MCP server, a Cursor rule, and an npm package — same logic, different packaging. The skill score we publish is the same regardless of which form factor you install; the install steps differ per platform. Where a skill ships across multiple platforms, every form is listed on the same canonical page and you can copy the install command that matches your stack.
Variants and forks are also collapsed to a canonical entry. If twenty people fork the same MCP server with cosmetic changes, you see one well-described listing, not twenty near-duplicates fighting for attention.
Browsing
The directory is browsable on five axes:
- Capability — what the skill does (writing, code, research, documents, data, automation, communication, creative).
- Role — who you are (sales, marketing, engineering, legal, finance, HR, ops).
- Industry — the vertical you work in (legal, healthcare, fintech, e-commerce, real estate, government).
- Platform — the format (Claude Skill, MCP, Cursor rule, GPT, n8n template, etc.).
- Discover — curated collections (trending, top rated, recently updated, editor's picks).
Every skill is tagged across all five axes simultaneously, so the same skill appears in many browse paths without ever getting its own duplicate page.
What we don't index (yet)
- Closed-source enterprise skills with no public install path.
- Skills published only inside private Slack or Discord communities.
- Generic example code from tutorials and SDK docs that doesn't ship as an installable artifact.
- Hardware-only products and consumer devices.
If you think we should cover a platform we don't list, the submission form takes both individual skills and entire platform suggestions.