Reference · Actionable

Agent manifest

Where you tell an agent what it may do, before it guesses.

What it is

/agents.md is a plain-language file stating what automated clients are allowed to do on your site, which endpoints exist, what the limits are, and who to contact.

It is the counterpart to robots.txt for the era where the visitor can act rather than only read.

Why an agent cares

Right now an agent that wants to behave well on your site has nothing to consult. It infers your policy from your rate limits, which means it discovers your rules by breaking them.

Few sites publish one yet, which makes it cheap and still distinctive.

How to fix it

# Agent policy for acme.com

## Permitted
- Reading any public page.
- Calling the public API at https://api.acme.com (see /openapi.json).
- Submitting the contact form at /contact.

## Not permitted
- Automated account creation.
- Bulk export of customer-facing listings.

## Limits
60 requests/minute per IP. Identify yourself in the User-Agent.

## Contact
agents@acme.com — tell us if a limit is blocking legitimate use.

How we test it

We fetch /agents.md, /.well-known/agents.json and /.well-known/ai-plugin.json, and reject soft-404s that return your HTML shell.

Related

Does your site pass this one?

Run the full scan and find out — along with the other 18.