- Published on
Claude Code announces support for AGENTS.md, the 'AI universal manual'
- Authors

- Name
- DeepHorus
On September 19, Thariq Shihipar (online handle @trq212), an engineer on Anthropic's Claude Code team, posted on X (formerly Twitter) announcing that the 2.1.277 version of Claude Code released that day officially supports AGENTS.md files. This update is regarded by the industry as an important step toward standardization of AI programming tools. AGENTS.md, also known as the 'universal AI instruction manual,' aims to provide shared project-level rules for different AI tools.
Claude Code is an AI programming tool launched by Anthropic. Previously, it relied on CLAUDE.md files to define exclusive configurations such as workflows, MCP, sub-agents, or commands. These configurations are deeply tied to specific tools, and developers often need to rewrite them when switching tools. AGENTS.md, on the other hand, is designed as a universal rules file, with the goal of being understood and followed by all AI tools. This provides a more universal foundation for project configuration.
According to Shihipar's tweet, starting from version 2.1.277, if there is no CLAUDE.md in the project folder, Claude will automatically check for and use AGENTS.md. Users can also manually switch this behavior in the /config settings to decide which configuration file takes precedence. This design ensures backward compatibility while providing an entry point for universal configuration.
CLAUDE.md carries Claude Code-specific workflow, MCP, sub-agent, or command configurations, while AGENTS.md aims to let different tools share project-level basic rules, which can be seen as universal rules that all AI should follow. This distinction makes AGENTS.md more suitable as a cross-platform project-level 'instruction manual.' For teams using different AI programming tools, this universality is particularly important.
For ordinary users and developers, AGENTS.md can significantly reduce tool-switching costs and configuration costs. With Claude Code's support this time, no matter which AI programming tool a user uses, a project can have a unified 'instruction manual for AI.' When switching tools, taking over a project, or participating in open-source collaboration, there is no need to repeatedly configure for different tools. This helps improve development efficiency and reduce repetitive work.
As AI programming tools become increasingly diverse, universal configuration files are expected to become an industry convention. Anthropic's early support for AGENTS.md may prompt more tools to follow suit, thereby forming a more open AI development collaboration ecosystem. In the future, project-level AI rules may become as common as README, becoming a basic component of developer collaboration. This change deserves continued attention.