# Agent files

Agent files (AGENTS.md / CLAUDE.md / SKILL.md)

**In short:** Publish an AGENTS.md of 200 bytes or more at your site root, with real instructions for agents using your product or API.

- Category: Ecosystem Integration · weight 11 of 100 within the category
- Ecosystem Integration is worth 20 in technical, 15 in generic, regulated and commerce (points of 100, by scoring profile)
- Audit ID E7 (as shown on reports) · rubric v2.1

> **Changed in v1.1.** E7 needs a real file. An AGENTS.md or CLAUDE.md under 200 bytes no longer earns credit.

## What it checks

Whether the site publishes agent instruction files: AGENTS.md, CLAUDE.md or SKILL.md.

## Why it matters

AGENTS.md is the convention coding agents read, inside a code repository, for instructions on how to work there. Publishing one at your site root is this rubric’s recommendation, not a documented agent behaviour: it puts the same kind of instructions for your product or API where an agent or developer can fetch them, alongside llms.txt.

## When it applies

Always, for any llms.txt that exists.

An audit that doesn’t apply, or that we couldn’t verify, is removed from your score and its weight shared among the Ecosystem Integration audits that do apply. It is never scored as a zero.

## How it’s scored

1.0 AGENTS.md · 0.6 CLAUDE.md · 0 absent. A file under 200 bytes counts as absent (v1.1). SKILL.md is captured but not yet credited.

> **Measurement.** A live scan fetches /AGENTS.md, /CLAUDE.md and /SKILL.md from the site root, the same as our research corpus, so a CLAUDE.md on its own scores 0.6. A response that is an HTML page, not a 200, or under 20 bytes is treated as absent.

## How to fix

1. Publish an AGENTS.md at the root with real instructions for agents using your product or API.
2. Make it substantive: a placeholder under 200 bytes earns nothing.
3. If you already have CLAUDE.md, publish the same content as AGENTS.md for full credit.

## Example

```
# AGENTS.md

## Using the Acme API

- Authenticate with a bearer token from https://acme.com/keys
- Use the sandbox base URL for tests: https://sandbox.api.acme.com/v1
- Stay under 100 requests a minute; on a 429, wait for the Retry-After header before retrying
```

## Frequently asked questions

### Which agent file earns credit: AGENTS.md, CLAUDE.md or SKILL.md?

`AGENTS.md` at your site root scores 1.0; `CLAUDE.md` alone scores 0.6 (a scan fetches `/AGENTS.md`, `/CLAUDE.md` and `/SKILL.md` from the root); `SKILL.md` is fetched but not credited. AGENTS.md is an open, tool-neutral convention for code repositories: coding agents read it from the project they are working in (Claude Code, for example, reads a project’s AGENTS.md when there is no CLAUDE.md). We know of no agent that documents fetching `/AGENTS.md` from a website; crediting a file at your site root is this rubric’s opinion of good practice. If you publish one file, make it AGENTS.md.

Sources: [agents.md — AGENTS.md, a simple, open format for guiding coding agents](https://agents.md/) · [Claude Code docs — How Claude remembers your project (CLAUDE.md and AGENTS.md)](https://code.claude.com/docs/en/memory) · [LLMSRank — Agent files audit](https://llmsrank.dev/docs/audits/agent-files.md)

### Why doesn’t my short AGENTS.md count?

A file under 200 bytes counts as absent: a placeholder earns nothing. The file must also answer HTTP 200 as text, so an HTML page served at `/AGENTS.md` (a soft 404) is rejected. Fill it with real instructions, such as how to authenticate, sandbox and production base URLs, and which docs to read first.

Sources: [LLMSRank — Agent files audit](https://llmsrank.dev/docs/audits/agent-files.md) · [agents.md — AGENTS.md, a simple, open format for guiding coding agents](https://agents.md/)

### What should a public AGENTS.md on a website contain, and how is it different from llms.txt?

AGENTS.md is plain Markdown with no required fields: the instructions an agent needs to work with your product or API, such as authentication, base URLs, key endpoints or SDKs, limits and conventions. llms.txt is a curated map of your content; AGENTS.md tells an agent how to act. Link each from the other so an agent that finds one finds both.

Sources: [agents.md — AGENTS.md, a simple, open format for guiding coding agents](https://agents.md/) · [llmstxt.org — The /llms.txt file, v2 (modified 2026-08-10)](https://llmstxt.org/)

### Is it worth publishing a SKILL.md?

Not for this audit yet: we capture `/SKILL.md` but give it no credit. SKILL.md belongs to Agent Skills, where each skill is a folder whose SKILL.md starts with YAML frontmatter (a required `name` and `description`) followed by instructions. Publish AGENTS.md first, and add skills if you want agents to load task-specific instructions on demand.

Sources: [Agent Skills — Specification (SKILL.md)](https://agentskills.io/specification) · [LLMSRank — Agent files audit](https://llmsrank.dev/docs/audits/agent-files.md)

### How many sites publish an AGENTS.md, in LLMSRank’s research?

Very few. In our research corpus of 187 sites (captured 2026-10-10, scored with rubric v2.1), 11 (5.9%) earned full marks with a real AGENTS.md, 93.0% scored zero, and the mean was 0.065. The count did not move while the corpus grew by 31 files, so the share fell. Good examples include zapier.com, apify.com, developers.deepgram.com, firecrawl.dev and docs.poolside.ai.

Sources: [LLMSRank — How we score (rubric v2.0)](https://llmsrank.dev/docs/method.md) · [LLMSRank — Agent files audit](https://llmsrank.dev/docs/audits/agent-files.md)

See also: [all audits](https://llmsrank.dev/docs/audits.md) · [how scoring works](https://llmsrank.dev/docs/method.md)