Agents

Claude Code Adds AGENTS.md Support via New Mods

Claude Code now supports AGENTS.md files starting in version 2.1.277, leveraging a new customization system that lets developers tailor how the AI assistant interacts with their codebases.

Simon Willison1 day agoAgents
Illustration generated for this story

Anthropic has updated its developer tool, Claude Code, to support AGENTS.md configuration files starting in version 2.1.277. According to a statement from Thariq Shihipar, the tool will now automatically look for and execute instructions inside an AGENTS.md file if a standard CLAUDE.md file is not present in the project folder. This update marks a shift toward more flexible, agent-based configurations for developers using the command-line assistant.

The new AGENTS.md integration is built on top of an upcoming feature called Claude Code mods. These mods represent a new framework designed to customize the underlying Claude Code harness. While the AGENTS.md support is delivered as a built-in mod, developers will soon have the ability to build their own custom versions of project instructions to suit their specific workflow requirements.

For software engineers and development teams, this update provides a more structured and standardized way to define how AI agents interact with a local repository. Instead of relying solely on generic prompts or the default CLAUDE.md configuration, practitioners can use AGENTS.md to establish precise boundaries, task definitions, and behavioral guidelines tailored to agentic workflows. The source code for this built-in mod is publicly accessible, allowing developers to inspect how the harness processes these instructions.

This release signals a broader push toward modular customization in AI development tools. As engineering teams increasingly rely on autonomous agents to handle complex codebase modifications, the ability to configure these systems through a flexible modding framework will be essential. The upcoming release of custom mods will allow teams to share specialized instructions and automate repetitive setup tasks across diverse programming environments.

This is our own summary of reporting by Simon Willison

More in Agents