Agents.md践行Unix哲学，开发者应该跟进

Agentic AI Foundation推行了AGENTS.md这一开放Markdown格式，用于记录项目上下文和为AI代理提供指导1。该格式建立在Unix"一切皆文件"的设计哲学之上，让代理能够通过文件、流和进程访问配置，从而实现工具的灵活组合1。

有开发者通过自己的dotfiles项目（fmind/dot）展示了如何在实践中应用这一理念1。该项目采用显式的代理指导、验证命令和skill目录来支持代理与人类的协作工作流1。关键指令包括编辑chezmoi源文件而非已安装配置、检查Git差异、保留不相关编辑，以及运行验证检查1。dot CLI工具由Typer和Pydantic构建，提供类型化命令用于系统维护和代理协调1。

该开发者强调，在接纳他人的dotfiles配置前应充分理解其中的工具和设置，"如果你不理解工具和配置，你就不是主人，而是木偶"1。其工作流采用Ghostty、Fish、Zellij、LazyGit、Neovim等工具，并遵循"一个项目一个会话、一个任务一个标签、一个程序一个窗格"的组织方式1。

A new open Markdown format called AGENTS.md is gaining traction as a standardized way to provide AI agents with clear project guidance and context 1. Developed by the Agentic AI Foundation, the format leverages Unix philosophy—particularly the principle that everything is a file—to enable agents to understand and modify development environments through composable tools and accessible configuration files 1.

The approach emphasizes that agents should interact with source configurations rather than deployed settings, verify changes through Git diffs, and preserve unrelated edits 1. A practical example of this methodology is demonstrated through the dot CLI tool, built with Typer and Pydantic, which offers typed commands for system maintenance and agent coordination 1. The author stresses a critical principle: "never use someone else's dotfiles as-is. If you do not understand the tools and configuration, you are not the master. You are the puppet." 1

In practice, the workflow integrates tools like Ghostty, Fish, Zellij, LazyGit, and Neovim, organized around a structured principle: one project per session, one task per tag, and one program per pane 1.