# Agents Md Generator

> Generate or update a robust AGENTS.md file (the "README for AI coding agents") for a project. This skill ensures your repository is optimized for AI tools (like Copilot, Cursor, and Windsurf) by providing structured context, commands, architecture diagrams, and coding standards. It automatically detects project metadata from pyproject.toml and generates a project tree view.

- Skill: `knuckles-team/agents-md-generator` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add knuckles-team/agents-md-generator`
- Raw SKILL.md: https://api.skillmd.com/api/skills/knuckles-team/agents-md-generator/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- License: MIT
- Author: Knuckles-Team (https://skillmd.com/u/knuckles-team)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/knuckles-team/agents-md-generator

---


# AGENTS.md Generator

Generates a robust `AGENTS.md` file designed specifically for AI coding agents. This file acts as a "README for machines," providing them with the exact project context, rules, and instructions they need to work effectively.

## Features
- **Project Metadata Extraction**: Automatically parses `pyproject.toml` for project name, description, and scripts.
- **Dynamic File Tree**: Generates a text-based tree view of the project structure, ignoring common noise directories (`.git`, `__pycache__`, etc.).
- **Mermaid Diagrams**: Includes standardized Architecture and Workflow diagrams to help agents understand the system's design.
- **Battle-Tested Template**: Uses a structure optimized for high agent performance, including "Always/Ask first/Never" boundaries.

## Usage

Run the generator script with the path to your project root:

```bash
python scripts/generate_agents_md.py <project-path>
```

### Options
- `<project-path>`: (Required) Path to the root directory of the project to document.

## AGENTS.md Standard

The generated `AGENTS.md` follows the emerging open standard for AI agent documentation:
1. **Tech Stack & Architecture**: High-level overview of technologies and component interactions.
2. **Commands**: Exact, file-scoped commands for building, testing, and linting.
3. **Project Structure**: Quick reference to key files and the dynamic tree view.
4. **Code Style & Conventions**: Specific "Always" rules with concrete Good/Bad code examples.
5. **Dos and Don'ts**: Specific behavioral guidelines.
6. **Safety & Boundaries**: Explicit "Always do", "Ask first", and "Never do" constraints.
7. **When Stuck**: Instructions on how to handle ambiguity (e.g., "propose a plan first").