Developer shares low-tech Markdown and tagging system to give AI agents repo context

A software developer has published a lightweight repository convention designed to help AI coding agents navigate codebases without relying on large prompt files or vendor-specific indexing tools. The approach uses two main components: per-directory index.md files that guide agents through the codebase top-down, and inline @tag: tokens that link related code and documentation across different folders. Tags are registered in a central docs/tags.md file to prevent naming drift and keep the tag set reviewable. The convention is compatible with tools like Codex, Cursor, and Claude Code, as well as plain command-line search using grep. The author acknowledges that stale documentation is the primary risk, and recommends treating doc and tag updates as mandatory parts of any code change.