llms-txt-kit

· strattlabs.com

Generate and validate spec-compliant llms.txt and llms-full.txt files.

llms.txt is a simple, growing convention for making a site legible to LLMs and AI agents: a single Markdown file at your site root that indexes your most important pages, plus an optional llms-full.txt that inlines the full content. This kit gives you a small, dependency-free CLI and library to generate those files from your existing Markdown content and to validate files you already have.

• Zero runtime dependencies. Pure TypeScript; nothing to audit downstream.
• Generator — crawl a content directory and emit llms.txt + llms-full.txt.
• Validator/linter — check an existing file against the spec with precise, rule-coded diagnostics.
• Library API — parse, format, validate, generate for programmatic use.

Status: v0.1 — early but production-shaped. The API surface in src/index.ts is the stable contract.

Install

pnpm add -D @strattlabs/llms-txt-kit
# or: npm i -D @strattlabs/llms-txt-kit / yarn add -D @strattlabs/llms-txt-kit

Requires Node.js ≥ 18.18.

CLI

Generate

llms-txt generate \
  --content ./docs \
  --base-url https://example.com \
  --title "Example Docs" \
  --summary "Everything you need to integrate with Example." \
  --out .

Writes llms.txt and llms-full.txt to --out (default: current directory). Use --stdout to print llms.txt instead of writing files.

Per-file metadata is read from optional YAML frontmatter; sensible fallbacks are derived from the document body and file path when it is absent:

---
title: Quickstart
description: Make your first API call in five minutes.
section: Guides
order: 1
draft: false
---

# Quickstart

...

| Frontmatter key | Effect | Fallback | | --------------- | -------------------------------------- | -------------------------------------------------- | | title | Link text in the index | first # H1, else humanized filename | | description | Trailing notes on the link | first prose line (truncated) | | section | ## Section the link is grouped under | top-level folder name, else Docs | | order | Sort order within a section | alphabetical by title | | url | Override the derived link URL | base-url + path (ext stripped, index → folder) | | draft | true excludes the file | included |

A section literally named Optional is emitted last, matching the spec's special-case semantics (consumers may skip it when they need a shorter context).

Validate

llms-txt validate ./llms.txt
llms-txt validate ./llms.txt --strict          # warnings fail too
llms-txt validate ./llms.txt --no-absolute-urls # don't warn on relative URLs

Exit code is 1 when there are errors (or warnings under --strict), else 0 — suitable for CI.

Library

import { generate, validate, parse, format } from '@strattlabs/llms-txt-kit';

const { llmsTxt, llmsFullTxt } = await generate({
  contentDir: './docs',
  baseUrl: 'https://example.com',
  title: 'Example Docs',
});

const result = validate(llmsTxt);
if (!result.valid) {
  for (const d of result.diagnostics) {
    console.error(`${d.rule} ${d.severity}: ${d.message}`);
  }
}

// Lossless AST round-trips: parse(format(doc)) deep-equals doc.
const doc = parse(llmsTxt);
const text = format(doc);

Validation rules

| Rule | Severity | Meaning | | ----- | -------- | --------------------------------------------------- | | LT001 | error | Missing required H1 title | | LT002 | error | More than one H1 | | LT003 | warning | Section has no links | | LT004 | error | Malformed list item (not - [title](url)) | | LT005 | error | Link missing a title or URL | | LT006 | warning | Duplicate URL | | LT007 | error | Content before the H1 title | | LT008 | warning | Heading deeper than H2 (index should be flat) | | LT009 | warning | No summary blockquote | | LT010 | warning | No sections at all | | LT011 | warning | Relative URL (absolute recommended for portability) | | LT012 | warning | Link declared before any section heading |

Validation is structural: it checks that a file conforms to the llms.txt format. It does not fetch URLs or judge content quality.

Why MIT?

This kit operates entirely on a public standard and formats public content into a public file format. We want the widest possible adoption — embedding it in other tools, docs pipelines, and frameworks — with zero friction. MIT is the lowest-friction, most-recognized permissive license for that goal. (Apache-2.0's explicit patent grant adds little here: there is no novel, patentable method in a text formatter, and the extra ceremony would only slow adoption.)

Examples

See examples/:

• examples/content/ — a small Markdown tree you can run the generator against.
• examples/llms-invalid.txt — a deliberately broken file that trips most validation rules; try llms-txt validate examples/llms-invalid.txt.

Development

pnpm install
pnpm test          # vitest
pnpm typecheck
pnpm lint
pnpm build         # emit dist/

License

MIT © 2026 Stratt Labs

Part of the Stratt Labs toolkit for the agentic web

Small, standards-friendly tools from Stratt Labs:

• agent-readiness-manifest — an open declaration spec for agent discovery.
• schema-for-agents — JSON-LD recipes for agent-legible commerce and content.
• webmcp-starter — a minimal WebMCP reference page.