Agents Md Guide — AI skill for Claude Code

AGENTS.md Guide

A model-agnostic guide for writing concise, project-specific instructions in `AGENTS.md`, `CLAUDE.md`, and compatible agent entry files.

This public, open-source project does not introduce a new instruction format or agent harness. It defines a small set of authoring rules and a human-validated initialization workflow that any capable coding agent can apply after inspecting a repository.

Quick start

From the repository you want to initialize or update, give your preferred coding agent access to this guide:

Initialize this repository using
https://github.com/farhadeidi/agents-md-guide/blob/main/README.md as the sole
methodology. Read it in full and follow its required references. Ask at most one
question or request one approval per message, and keep the repository read-only
until I approve the exact patch.

The agent should inspect before proposing and produce the smallest useful result. Depending on the repository, that may be:

• a concise root AGENTS.md;
• a compatible entry such as CLAUDE.md pointing to AGENTS.md;
• a careful update to useful existing guidance;
• explicitly approved optional patterns adopted as project policy; or
• no change when additional instructions would provide no value.

The result must be adapted to the repository, not copied from this README or a universal template.

Purpose

Create the smallest set of durable project context and instructions that helps an agent make better decisions in the repository in front of it.

The result must be specific to that repository. It must not be a generic template filled with inferred stack details or broadly applicable engineering advice.

A valid result may be no change, or no instruction file at all, when the repository has no durable guidance worth loading on every task.

A useful result should let an unfamiliar agent quickly answer:

• What is this project, and what does it enable for its users or consumers?
• Which product qualities, domain rules, and