Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

awesome-agents-md banner

awesome-agents-md

The Rosetta Stone for AI coding agents. Production-grade agent configs for every popular stack — translated for every major tool.

Awesome License: MIT PRs Welcome Stack Count Tools Covered


What is this?

Every AI coding agent reads a different config file:

Tool Reads
Claude Code CLAUDE.md
Cursor .cursorrules / .cursor/rules/*.mdc
Windsurf .windsurfrules
GitHub Copilot .github/copilot-instructions.md
Gemini CLI GEMINI.md
Codex (OpenAI) AGENTS.md
Universal AGENTS.md

These files do the same thing — tell the agent how to write code in your project. But each tool has its own format, its own affordances, and its own conventions.

awesome-agents-md collects production-grade configs for popular stacks, in every format. Pick your stack, copy the file, ship.

If you switch from Cursor to Claude Code, you don't lose your config — you find the equivalent here.

Why this matters

Without good agent configs:

  • Your AI writes inconsistent code that doesn't match your team's conventions
  • You re-explain the same project rules every session
  • Switching agents means starting over

With them:

  • Day-one productivity from any agent
  • One source of truth for your team's conventions
  • Onboarding new tools doesn't reset your setup

🚀 Quick start

  1. Find your stack in the table below.
  2. Copy the relevant config files into your project root.
  3. Edit the placeholders (project name, paths, opinionated choices).
  4. Commit and ship.
# Example: bootstrap a Next.js 15 project
curl -sL https://raw.githubusercontent.com/bradleybeatz1313/awesome-agents-md/main/stacks/nextjs-15/AGENTS.md   -o AGENTS.md
curl -sL https://raw.githubusercontent.com/bradleybeatz1313/awesome-agents-md/main/stacks/nextjs-15/CLAUDE.md   -o CLAUDE.md
curl -sL https://raw.githubusercontent.com/bradleybeatz1313/awesome-agents-md/main/stacks/nextjs-15/.cursorrules -o .cursorrules

Or symlink everything to a single canonical file — see the "write once" pattern.


📚 Stacks

15 stacks. AGENTS.md baseline for everything; multi-tool coverage for the most popular.

Stack AGENTS.md CLAUDE.md .cursorrules .windsurfrules copilot-instructions.md GEMINI.md
Frontend / full-stack
Next.js 15
SvelteKit 2 + Svelte 5
Nuxt 3
Astro 5
Backend / APIs
FastAPI (Python)
Django 5 (Python)
Express 5 (Node)
NestJS
Go (stdlib)
Rust + Axum
Mobile / desktop
React Native (Expo)
Tauri 2
Infra / data / monorepo
Turborepo monorepo
PyTorch research
Terraform

🙋 Don't see your stack? Open an issue or contribute one. Phoenix, SwiftUI, Solid, Hono, Tanstack Start, Spring Boot, Laravel, Rails, and more are all welcome.


📖 What's in a good config?

Every AGENTS.md in this repo follows the same structure — we wrote up the universal base so contributors can stay consistent:

  • What this project is — context for every later decision
  • Quick commands — the exact CLI invocations the agent needs
  • Project structure — where things live
  • Conventions — what we do, with examples
  • Patterns we avoid — what we don't do, also with examples
  • Common pitfalls — gotchas that have actually bitten the team
  • Out of scope — things to ask before doing

Length target: 80–250 lines. Long enough to be useful; short enough to stay in the agent's context.


🧠 The "write once, work everywhere" pattern

If you don't want to maintain six different files, symlink them:

# Linux / macOS
ln -sf AGENTS.md CLAUDE.md
ln -sf AGENTS.md GEMINI.md
ln -sf AGENTS.md .cursorrules
ln -sf AGENTS.md .windsurfrules
mkdir -p .github && ln -sf ../AGENTS.md .github/copilot-instructions.md

The trade-off: symlinks share content but lose tool-specific affordances (Claude Code's sub-agents, Cursor's glob targeting, etc.). For most teams the shared baseline is enough.

When you do need tool-specific files, this repo shows both patterns side-by-side — see Next.js 15 for the full multi-tool example.


📋 Tool reference

A quick map of which AI tool reads which file. Detailed comparison.

Tool File Notes
AGENTS.md spec AGENTS.md The emerging standard. Codex reads it. Most tools fall back to it.
Claude Code CLAUDE.md Project-level + ~/.claude/CLAUDE.md global + .claude/agents/*.md sub-agents
Cursor .cursorrules (legacy) / .cursor/rules/*.mdc (modern) Modern format supports per-glob targeting
Windsurf .windsurfrules + ~/.codeium/windsurf/memories/global_rules.md Workspace + global tiers
GitHub Copilot .github/copilot-instructions.md + .github/instructions/* Path-targeted instructions for monorepos
Gemini CLI GEMINI.md Reads from project root upward
Codex AGENTS.md First major tool to adopt the universal standard
Aider .aider.conf.yml + CONVENTIONS.md YAML config + free-form conventions
Cline .clinerules/*.md Multiple files in rules directory
Continue .continue/config.yaml + .continue/rules/*.md Modern Continue uses a rules directory

🤝 Contributing

PRs are very welcome. Adding a new stack takes 30–60 minutes; we'll review within a few days.

See CONTRIBUTING.md for:

  • The quality bar for an AGENTS.md
  • How to add a new stack directory
  • Style guide for writing rules
  • What we accept (new stacks, new tool variants, corrections) and what we don't (low-quality configs, marketing)

Quick-start contribution:

  1. Fork the repo
  2. Create stacks/your-stack-name/ and add AGENTS.md + README.md
  3. Update the table in this README
  4. Open a PR

❓ FAQ

Q: Is AGENTS.md an actual standard or just a convention? It's an emerging convention with a public spec. Codex was the first major tool to adopt it; others increasingly read it as a fallback. Putting one in your repo costs nothing and benefits any tool that supports it.

Q: Why not just use .cursorrules everywhere? Cursor's format is tied to Cursor. Claude Code looks for CLAUDE.md, GitHub Copilot for .github/copilot-instructions.md. If your team uses more than one tool — or might in the future — AGENTS.md is the universal baseline.

Q: How long should an agent config be? 80–250 lines. Long enough to encode real conventions; short enough to stay in the agent's context window without crowding out the actual code. If yours is longer, you're probably writing the framework manual.

Q: Why these specific stacks? The 15 stacks here cover what most teams ship in 2026: Next.js for full-stack web, FastAPI/Django/Express/NestJS for APIs, Go/Rust for systems, Expo/Tauri for client apps, PyTorch/Terraform for specialized work. Missing your stack? Open an issue or contribute.

Q: Are these configs battle-tested? The conventions in each config reflect widely-adopted patterns from the relevant ecosystem (HackSoft for Django, Stripe-style for Express, the AGENTS.md spec for everything). They're starting points, not gospel — adapt to your team.

Q: Can I use these in commercial projects? Yes — MIT licensed. Copy, modify, redistribute. Attribution is appreciated but not required.

Q: How do I keep my configs in sync across tools? Either symlink (single source of truth, lose tool-specific features) or maintain AGENTS.md as the canonical version and copy/extend into tool-specific files. The Next.js 15 stack shows both patterns.


🌟 Star history

If you find this useful, a star helps others find it.

Star History Chart


📜 License

MIT © awesome-agents-md contributors


Related


Made by developers tired of writing the same config six times.

About

The Rosetta Stone for AI coding agents — production configs for every popular stack

Topics

Resources

Contributing

Stars

2 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors