agent.json is SkillAEO's experimental format for publishing a small set of public website metadata as JSON. It is not an industry standard, a W3C or IETF specification, an agent protocol, or a file that websites must publish. No AI system is required to discover, read, use, or call it.
The practical purpose is modest: it gives a site owner one machine-readable record of public links and descriptions that can be checked and maintained alongside the website. Publishing the file does not guarantee discovery, indexing, citations, recommendations, integrations, or rankings.
What the SkillAEO format contains
The current public SkillAEO file contains only these fields:
{
"name": "Skillaeo",
"description": "Skillaeo provides public website audits and practical guidance for reviewing how a site presents information to search and AI-assisted discovery tools.",
"url": "https://skillaeo.com",
"audit_url": "https://skillaeo.com/audit",
"documentation_url": "https://skillaeo.com/docs",
"help_url": "https://skillaeo.com/blog",
"supported_languages": ["en"],
"experimental": true,
"last_updated": "2026-08-20"
}This example matches the file published at /agent.json. The experimental flag is intentional: it prevents the format from being mistaken for a stable external specification.
Field reference
| Field | Meaning | Guidance |
|---|---|---|
name | Public name of the site or product | Use the same name shown on the site |
description | Short factual summary | Avoid slogans, unverifiable superlatives, and guarantees |
url | Canonical public homepage | Use HTTPS |
audit_url | Public audit entry point | Omit if the site has no audit |
documentation_url | Public documentation | Omit if it does not exist |
help_url | Public help or learning page | Use a page visitors can open without credentials |
supported_languages | Languages actually supported by the published experience | Do not list planned languages |
experimental | Marks this as an experimental SkillAEO format | Keep this true while the format remains experimental |
last_updated | Date the public facts and links were last reviewed | Use an ISO date and update it after every material review |
These are SkillAEO conventions, not universal required fields. Another site may use a different internal record, and an external system may ignore the file entirely.
What not to publish
Treat agent.json as public content. Do not include:
- API keys, tokens, passwords, private endpoints, or internal hostnames;
- customer data, private contact details, or unpublished pricing;
- actions that an agent is supposedly allowed to perform unless a separate, authenticated protocol actually implements them;
- claims that a third-party AI system supports this format without direct documentation from that system;
- guarantees that the file will change an AI answer.
The file should describe public information, not create a new trust or authorization boundary.
How to publish it safely
- Copy only information already available on public pages.
- Use absolute HTTPS URLs on the same domain where practical.
- Mark the file as experimental.
- Validate that the JSON parses.
- Open every link in a private browser session.
- Review the file whenever public URLs, language support, or product descriptions change.
You can publish the same content at /.well-known/agent.json for your own consistency, but that path does not turn the file into a recognized standard. SkillAEO keeps both copies identical to avoid contradictory public records.
How this differs from llms.txt and Schema.org
llms.txt is a separate optional community proposal for a Markdown summary and selected links. SkillAEO's agent.json is a compact experimental JSON record. Neither is required by Google for AI features.
Schema.org structured data has documented consumers and page-specific eligibility rules. Even valid structured data does not guarantee a search feature. Do not describe agent.json as a substitute for schema, ordinary HTML, a sitemap, an API description, or an authentication protocol.
How to evaluate whether it is useful
Use an operational test rather than a ranking promise:
- Does the file give your team one accurate public record?
- Can someone validate every URL and statement?
- Is it inexpensive to keep current?
- Does any tool you actually use document support for this exact format?
If the answer to the last question is no, treat the file as an experimental publishing convenience. Your normal pages, structured data, sitemap, and technical SEO remain the primary public sources.
Frequently Asked Questions
Is agent.json an official or industry-standard file?
No. The format documented here is SkillAEO's experimental format, not a W3C, IETF, Schema.org, Google, OpenAI, Anthropic, or other industry standard.
Does publishing agent.json make AI agents discover or recommend a site?
No. There is no such guarantee, and no AI system is required to discover, read, use, or call the file. Publish it only as a truthful, maintainable record of public information.
Can agent.json grant an AI agent permission to take actions?
No. A public JSON file is not authentication or authorization. Real actions require a separately designed API, explicit user consent, authentication, validation, and abuse protection.
Should every website publish agent.json?
No. It is optional. If maintaining another public file would create stale or conflicting information, prioritize accurate web pages and established technical foundations instead.
What is the safest starting point?
Use the minimal example above, remove fields that do not apply, publish no secrets, and keep experimental: true. Verify the live file against your public pages after each material update.