llms-txt-audit

llms-txt-audit is a small Python CLI and GitHub Action for validating /llms.txt files before they ship.

It checks whether your file follows the shape described by the official llms.txt proposal: a Markdown file with a project title, short summary, curated resource sections, and links to the pages an LLM should read for context.

Need to create the file first? Use the AltRepo llms.txt Generator & Validator to draft a spec-shaped file, then use this CLI to protect it in CI.

Why this exists

A useful llms.txt file is curated. It should not be a blind sitemap dump, and it should not contain robots.txt permission rules.

This tool helps teams catch common mistakes:

• missing project title
• missing blockquote summary
• no H2 resource sections
• no Markdown resource links
• robots.txt directives placed in llms.txt
• private-looking URLs such as /admin, /account, /checkout, or /wp-admin
• duplicate links
• links with no explanation
• relative links when absolute URLs are required

Install

For local development from this repository:

python -m pip install -e .

After installation, the command is available as:

llms-txt-audit --help

You can also run it without installing by setting PYTHONPATH:

PYTHONPATH=src python -m llms_txt_audit.cli public/llms.txt

Basic usage

Audit a local file:

llms-txt-audit public/llms.txt

Audit a directory that contains llms.txt:

llms-txt-audit public/

Audit a hosted file:

llms-txt-audit https://example.com/llms.txt

Discover /llms.txt from a site root:

llms-txt-audit https://example.com --discover

Print JSON for scripts or dashboards:

llms-txt-audit public/llms.txt --json

Use strict CI behavior:

llms-txt-audit public/llms.txt --strict --fail-on-warning

Example output

llms.txt audit: public/llms.txt

PASS  H1 title found. line 1
PASS  Blockquote summary found. line 3
PASS  3 H2 section(s) found.
PASS  4 Markdown resource link(s) found.
PASS  Optional section found.

Score: 100/100
Sections: 3  Links: 4  Optional: yes

GitHub Action

Use the action from inside this repository:

name: Validate llms.txt

on:
  pull_request:
  push:
    branches: [main]

jobs:
  llms-txt:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: ./llms-txt-audit
        with:
          target: public/llms.txt
          strict: "true"
          fail-on-warning: "true"

When this project is published as a standalone GitHub Action, replace ./llms-txt-audit with the public action reference.

The action runs directly from src, so it does not need to build or install the package during the workflow.

Action inputs

Input	Default	Purpose
target	public/llms.txt	Local path, directory, hosted /llms.txt URL, or site URL when discover is true.
discover	false	Treat target as a site root and audit https://site.com/llms.txt.
strict	false	Make recommended structure checks stricter.
require-optional	false	Warn when ## Optional is missing.
fail-on-warning	false	Exit non-zero if warnings are found.
no-relative-links	false	Warn when resource URLs are relative.

Exit codes

Code	Meaning
0	Audit completed without errors.
1	Audit found errors, or warnings when --fail-on-warning is enabled.
2	The target could not be read.

What this checks

Required structure

A strong llms.txt file should start like this:

# Project Name

> One short summary explaining what this site or project is.

Optional notes that help an LLM understand how to use the linked resources.

## Core
- [Overview](https://example.com/index.html.md): Project overview

## Docs
- [Quick start](https://example.com/docs/start.html.md): First setup path

## Optional
- [Sitemap](https://example.com/sitemap.xml): Complete canonical URL list

The audit checks for:

• one H1 title
• a blockquote summary
• H2 sections
• Markdown list links
• an optional ## Optional section

Link hygiene

The audit also flags:

• private-looking URLs
• duplicate URLs
• missing link descriptions
• unsupported URL schemes
• relative URLs when --no-relative-links is enabled

robots.txt confusion

llms.txt is for context, not crawler permissions. If the file contains lines like this:

User-agent: *
Disallow: /admin

…the audit warns you to move those rules to robots.txt.

What this does not do

This tool does not crawl your whole website or auto-generate a final curated file from every URL. That is intentional. A good llms.txt file should be selected by humans or documentation owners.

Run tests

The test suite uses Python's standard library only:

PYTHONPATH=src python -m unittest discover -s tests -v

On Windows PowerShell:

$env:PYTHONPATH='src'
python -m unittest discover -s tests -v