AGENTS.md vs CLAUDE.md vs llms.txt

AGENTS.md is an open convention: a plain-Markdown file at your repo root that tells coding agents what they need to know to work in your project — build and test commands, code style, conventions, and gotchas that would clutter a human README. There's no required schema; it's just standard Markdown.

It started across tools like OpenAI Codex, Cursor, and Jules, and is now stewarded by the Agentic AI Foundation under the Linux Foundation. A wide range of agents read it, which is the point: write it once, and many tools use it.

CLAUDE.md is Claude Code's native memory file. It's loaded at the start of every session and holds the same kind of context — commands, conventions, architecture, standing instructions — but it's specific to Claude Code, with support for scoped files (org, user, project, local) and @-imports.

Importantly, Claude Code reads CLAUDE.md, not AGENTS.md, with no automatic fallback. The recommended pattern is a CLAUDE.md that imports AGENTS.md (via an @AGENTS.md import or a symlink), so both tools share one source of truth and you add Claude-specific notes below.

llms.txt is a different animal. Proposed by Jeremy Howard in September 2024, it's a Markdown file at your website root (/llms.txt) that gives LLMs a clean, curated map of your content instead of forcing them to parse messy HTML. The only required element is an H1 with the project name, followed by a summary and lists of links.

A related convention, llms-full.txt, concatenates the full text of those pages into one file for single-fetch ingestion — though that's a popularized extension, not part of the original spec. Adoption is real but contested: no major LLM provider has officially committed to consuming llms.txt, and some, like Google, have publicly dismissed it. Treat it as a low-cost bet, not a settled standard.