llms.txt Validator

llms.txt Validator — Catch Formatting Errors Before an AI Agent Does

One missing closing parenthesis in a markdown link is enough to break an llms.txt file. Not visibly — the file still opens fine in a text editor — but a parser reading it for an AI agent will either skip that line or choke on it entirely, and you won't know unless you check. That's the gap this tool fills: paste your file, get back the exact line where something doesn't match the spec.

What This Validator Checks

The llms.txt format, proposed by Jeremy Howard and the Answer.AI team, has a small number of structural rules. This tool checks each of them:

• The file starts with exactly one H1 heading (#) — this is the only required section
• A blockquote summary (>) sits near the top, right after the H1
• Section headings use H2 (##) — anything deeper isn't part of the spec
• Links inside H2 sections follow - [Name](URL): optional notes
• Every URL inside a link is actually a valid URL

Errors mean the file breaks the spec in a way that will likely cause a parser to misread it — a missing H1, a link with unmatched brackets, a URL that doesn't parse. Warnings are softer: an H3 buried inside a section, a bare URL instead of a full markdown link, an extra H2 section that holds prose or metadata instead of a link list. None of those are illegal markdown, and none of them match the spec exactly either.

Why Some "Errors" Are Actually Just Warnings

The original spec text is specific: a link list entry needs "a required markdown hyperlink [name](url)." Plenty of real llms.txt files in the wild skip that and just drop bare URLs under a heading instead. It's off-spec, but most context-building tools can still pull a URL out of a line whether or not it's wrapped in brackets, so flagging every one of those as a hard error produces a wall of red text for a file that mostly works. Same with extra sections like "Project" or "Guidance for AI Systems" — not part of the original two-section model, but common enough in practice that this tool treats them as a deviation worth noting, not a failure.

What still counts as a real error: no H1 at the top, more than one H1, a link with broken bracket or parenthesis syntax, or a URL that isn't a URL at all. Those are the things a parser is actually likely to trip over.

A Few Things Worth Knowing

This tool checks structure, not content. It won't tell you whether your project summary is clear, or whether you've linked the right pages — that part is still on you. A file can pass every check here and still be a weak llms.txt if the summary is vague or the wrong pages are linked.

It also runs entirely in your browser. Nothing you paste gets sent anywhere or stored, so testing a draft before it's public is safe.