AGENTS.md Generator — One File for Four AI Tools

Highlights

Example output

# Agent Instructions

This file provides shared instructions for AI coding agents.
Following the AGENTS.md open standard (https://agents.md).

## Code Style

- Never use the `any` type in TypeScript. Prefer `unknown` with type narrowing.
- Use functional components exclusively. Manage state with hooks.

## Git Workflow

- All commit messages must follow Conventional Commits: `<type>(<scope>): <description>`.

## Boundaries

### ✅ Always do

- Always run the test suite after making changes.

### 🚫 Never do

- Never commit secrets, API keys, tokens, or credentials.

Frequently asked questions

• What exactly is the AGENTS.md standard? ▾

  AGENTS.md is an open standard donated by OpenAI to the Linux Foundation. It is a single file that lets AI coding agents understand a project's conventions, and it is adopted by 60,000+ GitHub repositories — making it the de-facto industry standard.
• Which AI tools read AGENTS.md natively? ▾

  OpenAI Codex reads AGENTS.md as its primary context. Cursor and GitHub Copilot also recognize it officially. Claude Code does not read it directly, so a single @AGENTS.md import line in CLAUDE.md is enough to share the same rules.
• Which sections should AGENTS.md include? ▾

  Based on the GitHub Engineering Blog's analysis of 2,500+ repositories, the most impactful sections are Commands / Testing / Project Structure / Code Style / Git Workflow / Boundaries. ConfigDeck generates all six automatically.
• Is there a length limit for AGENTS.md? ▾

  OpenAI Codex reads AGENTS.md up to 32 KiB. Anything beyond that may be truncated, so it is best to keep frequently used core rules in AGENTS.md and split detailed guides into separate documents.
• Can I migrate from .cursorrules or other config files? ▾

  AGENTS.md is free-form Markdown, so existing instructions can be moved over as-is. ConfigDeck takes your stack and best-practice choices as input and structures the content into the six core sections, which is a quick way to consolidate scattered rules into a single source of truth.