Agent-Friendly Documentation Spec (agentdocsspec.com)

A new specification aimed at improving how documentation sites serve coding agents has been announced, addressing the challenges faced by AI models like Claude Code, Cursor, and GitHub Copilot. Traditionally designed for human users, many documentation websites hinder agents by presenting issues such as truncation limits, outdated content rendering methods, and a lack of support for essential discovery protocols like llms.txt. The specification outlines 23 specific checks across seven categories, focusing on factors such as content discoverability, page size, and URL stability, to enhance the accessibility of documentation for AI systems. This initiative is significant as it bridges a critical gap in the integration of AI tools within development environments, allowing agents to retrieve and process documentation more effectively. Key recommendations for documentarians include implementing an llms.txt file, serving markdown content directly, and keeping page sizes manageable. The accompanying CLI tool, afdocs, can automate checks to identify areas needing improvement, fostering a more robust ecosystem for AI-driven code assistance and documentation retrieval. As the tech community embraces this spec, it signals a shift towards more agent-friendly resources that can better support the advancing landscape of AI and machine learning applications.