7 CHECKS
Content Discoverability
llms.txt: whether it exists, is validly structured, stays under size, and whether your pages point to it.
View checks+
2 CHECKS
Markdown Availability
Markdown at .md URLs, and a server that honours Accept: text/markdown.
View checks+
4 CHECKS
Page Size and Truncation
Server-rendered content, 50K character budgets, and where your content starts.
View checks+
3 CHECKS
Content Structure
Tabs that serialise, headers that stand alone, code fences that close.
View checks+
2 CHECKS
URL Stability
Real 4xx codes instead of soft 404s, and same-host HTTP redirects.
View checks+
3 CHECKS
Observability
Sitemap coverage, markdown that matches the HTML, and cache headers.
View checks+
2 CHECKS
Authentication and Access
Pages that open without a login, or a documented alternative path.
View checks+
〉 AGENT-READABLE BY DEFAULT Publish on Velu and the checks take care of themselves.
Everything measured above is part of what publishing does, not a project you schedule after the fact. Your pages go out human-readable and machine-readable at the same time, and they stay that way as you add to them.
01
Auto llms.txt and llms-full.txt
Generated from your published docs and regenerated as pages are added, so agents always get a current index rather than one somebody remembered to update.
/llms.txt/llms-full.txt/page.md
02
Markdown on every URL
Append .md to any page and get clean markdown back, without the navigation and layout markup an agent has to wade through.
03
Server-rendered pages
Real content in the HTTP response. Nothing that needs JavaScript to appear, which is what most agents cannot execute.
04
Managed hosting
Status codes, redirects, and cache headers handled by the platform, so URLs stay stable and updates get through.
05
Public by default
Docs reachable without an interactive login, so an agent is not stopped at the door.
06
MCP server and skill.md
One endpoint agents can search, read, and cite through, beyond what the spec asks for today.
agent → mcp.search() → your docs → cited answer
01 What is agent-friendly documentation? +
Agent-friendly documentation is documentation a model or agent can reliably find, fetch, parse, and act on. In practice that means pairing the human-readable pages with machine-readable artefacts: a discovery file like llms.txt, markdown versions of every page, server-rendered content that fits inside a context window, and stable URLs. Human readers can work around a gap by inferring or clicking around. Agents cannot, so they guess instead, and they do it confidently.
02 What is llms.txt? +
A plain-text index at the root of your docs that tells an agent what exists and where to read it. The convention is an H1 title, a blockquote summary, and headed sections of markdown links. It is the single highest-weighted item in the spec, because an agent that cannot discover your docs never reaches any of the other twenty-two checks.
03 How is the score calculated? +
By the Agent-Friendly Documentation Spec, an open community standard maintained at agentdocsspec.com. It defines 23 checks across 7 weighted categories. This page runs them with afdocs, the spec’s MIT-licensed reference implementation, against a sample of your pages, and reports exactly what it returns without adjustment.
04 Is this just SEO under a new name? +
No, and they can pull in opposite directions. SEO optimises how a page is ranked and presented to a human who will click, skim, and judge. This optimises whether a machine can retrieve the content at all, parse it without HTML noise, and fit it in a context window. A page can rank first on Google and still be unreadable to an agent, usually because it is client-rendered or buried under navigation markup.
05 What happens to the URL I enter? +
It is fetched, scored, and cached for fifteen minutes so repeat checks are fast. There is no signup and results are not published to a leaderboard. If you share your score, the link carries a signed summary of the result so others can open the same scorecard.
06 Can I share my score? +
Yes. After a run, copy the short link or share it to X, LinkedIn, or Reddit. Anyone with the link sees the same score, and the preview image shows the grade and overall score.