kubical-ai/llms-txt-creator

llms-txt Creator

Creates spec-compliant llms.txt and llms-full.txt files for documentation projects.

Read references/llms-txt-spec.md for the full specification, format details, and directory submission information.

When to Use This Skill

• Creating a new llms.txt from scratch (item #1)
• Creating or auto-generating llms-full.txt (item #2)
• Fixing a malformed or incomplete llms.txt
• Adding llms.txt to an existing docs project

---

Workflow

Step 1: Understand the project

Explore the docs project to gather:

1. Project name and description — read package.json, README.md, or the site's H1/homepage
2. Content structure — list the content directory (src/content/, src/pages/, docs/) to understand sections and hierarchy
3. Site URL — find in astro.config.mjs, next.config.js, or package.json homepage field
4. Framework — Astro, Next.js, Docusaurus, etc. (determines where to place the file)

Step 2: Create llms.txt

Format requirements (see references/llms-txt-spec.md for full spec):

# [Project Name]

> [One-paragraph description of what the project is, who it's for, and what the docs cover. Should be 2-4 sentences.]

## [Section Name]

- [Link text](https://full-url/path): Brief 5-10 word description
- [Link text](https://full-url/path): Brief description

## [Section Name]
...

Curation guidelines:

• Include 20-40 links total — curate, don't dump everything
• Put the most important content first (Getting Started, Key Concepts)
• Group links into 3-6 logical H2 sections matching the site's top-level structure
• Always use absolute URLs (https://...)
• Always include : Description after each link

Where to place it:

• Astro/Next.js/Vite: public/llms.txt
• Docusaurus: static/llms.txt
• Other: site root's static assets directory

Validate before saving:

• H1 present ✓
• Blockquote present (line starting with >) ✓
• At least one H2 section ✓
• All URLs are absolute ✓
• Links point to real pages (spot-check 3-5)

Step 3: Create llms-full.txt (if requested)

Option A — Build script (recommended for maintained docs):

Create scripts/generate-llms-full.js (or .mjs):

import { glob } from 'glob';
import { readFile, writeFile } from 'fs/promises';
import matter from 'gray-matter';
import path from 'path';

const SITE_URL = 'https://YOUR_SITE_URL'; // update this
const CONTENT_DIR = 'src/content'; // update this

const files = (await glob(`${CONTENT_DIR}/**/*.{md,mdx}`)).sort();
const parts = [
  `# [Project Name] — Full Documentation`,
  `Generated: ${new Date().toISOString()}`,
  `Source: ${SITE_URL}`,
  `Pages: ${files.length}`,
  '',
];

for (const file of files) {
  const raw = await readFile(file, 'utf8');
  const { data: fm, content } = matter(raw);
  const title = fm.title || path.basename(file, path.extname(file));
  const slug = file
    .replace(CONTENT_DIR, '')
    .replace(/\.(md|mdx)$/, '')
    .replace(/\/index$/, '');
  parts.push('---');
  parts.push(`# ${title}`);
  parts.push(`URL: ${SITE_URL}${slug}`);
  parts.push('');
  parts.push(content.trim());
  parts.push('');
}

await writeFile('public/llms-full.txt', parts.join('\n'));
console.log(`✓ Generated llms-full.txt (${files.length} pages)`);

Add to package.json scripts:

{
  "scripts": {
    "build": "node scripts/generate-llms-full.mjs && [existing build command]"
  }
}

Option B — Static generation (for projects without a build pipeline):

Manually concatenate the 10-20 most important pages into a single file using the format above.

Step 4: Verify

For llms.txt:

• Open the file and read it — does it represent the project accurately?
• Check that 3-5 random links in the file would resolve if visited

For llms-full.txt:

• Check the file size (should be >10KB for a real docs project)
• Verify the header is correct (date, page count)
• Confirm a few page sections are present with correct titles

Step 5: Report

Tell the user:

1. Where the file(s) were created
2. How many links are in llms.txt (and which sections)
3. For llms-full.txt: page count and file size
4. Next step: run the build to verify the file is served correctly
5. Remind them about directory registration (item #25): submit to llmstxthub.com, llmstxtdirectory.org

---

Common Mistakes to Avoid

• No blockquote → llms.txt parsers may skip the file; always include > summary
• Relative URLs → use full https:// URLs always
• Too many links → curate to 20-40; more is not better
• Wrong placement → must be at /llms.txt, not /docs/llms.txt
• Manual llms-full.txt → auto-generate in build to keep it fresh