AI Agent Specifications & Standards

The AI coding ecosystem is converging on a set of open formats and protocols. Each standard solves a different problem — from teaching agents new skills to connecting them with external tools. Here’s what you need to know about each one.

SKILL.md

The Agent Skills Format

Modular, on-demand capabilities for AI coding agents. The open standard for teaching agents new tasks.

agentskills.io

AGENTS.md

Project Context for AI Agents

Project-level instructions that give AI agents context about your codebase, conventions, and build steps.

Linux Foundation

MCP

Model Context Protocol

A universal protocol for connecting AI models to external tools, data sources, and services.

Anthropic

llms.txt

Making Websites AI-Readable

A simple text file that helps AI models understand your website content in a structured, machine-friendly way.

Community Standard

.cursorrules

Custom Rules for Cursor

Project-level configuration that tells Cursor how to write code for your specific codebase and conventions.

Cursor / Anysphere

CLAUDE.md

Project Instructions for Claude

A markdown file that gives Claude Code persistent context about your project, style preferences, and workflows.

Anthropic

SOUL.md

Personal Identity for AI Agents

A markdown file that encodes your personality, voice, worldview, and opinions so AI agents can embody your identity.

Aaron Mars / OpenClaw

How these standards work together

Think of it in three layers: Rules (CLAUDE.md, .cursorrules) set project-wide constraints that apply to every task. Skills (SKILL.md) provide modular capabilities loaded on demand for specific tasks. Tools (MCP servers) connect agents to external services and data sources. And llms.txt makes your documentation discoverable by AI models.

Read the ecosystem overview →Browse skills →