llmstxt

name	llmstxt
description	A proposal to standardise on using an /llms.txt file to provide information to help LLMs use a website at inference time. You MUST load this skill when reading websites, llms.txt, or when creating/updating that file.
license	MIT

llmstxt

WHEN TO USE

• When using the webfetch tool to read documentation from a website.
• When creating, updating, or maintaining an /llms.txt file for a project.
• When retrieving context about a website to help LLMs understand its structure.

WHEN NOT TO USE

• When reading local files that do not correspond to web documentation.
• When the task does not involve fetching external documentation or creating LLM context files.

Core Process

1. Check for llms.txt before reading docs: When you need to read documentation from a new website, always attempt to fetch https://<domain>/llms.txt first. This file often contains a curated overview of the documentation and links to LLM-friendly markdown files.
2. Look for expanded context files: Some sites provide /llms-ctx.txt (expanded context without optional URLs) or /llms-ctx-full.txt (full expanded context). Fetching these can provide all the documentation in a single context-friendly file.
3. Check for .md versions of pages: If reading a specific page https://<domain>/path/page.html, try fetching https://<domain>/path/page.html.md (or index.html.md for directories) as it might provide a clean markdown version.
4. Creating or updating llms.txt: When asked to create an llms.txt file for a site:
   • Ensure it's located in the root path /llms.txt (or a subpath if specified).
   • Use an H1 with the project name (required).
   • Follow with a blockquote > containing a short summary.
   • Include zero or more markdown sections with detailed information.
   • Include zero or more H2 (##) sections containing "file lists" of URLs.
   • In file lists, use standard markdown links [name](url) optionally followed by : and notes.
   • Use a ## Optional section for URLs that can be skipped if a shorter context is needed.

Best Practices

• Use concise, clear language in llms.txt files.
• When linking to resources, include brief, informative descriptions.
• Prefer markdown formats as they are easily understood by language models.

What to Avoid

• Do not use llms.txt as a full sitemap (use sitemap.xml for search engines).
• Avoid putting large unstructured data directly in /llms.txt; link to detailed .md files instead.

Common Pitfalls

• Forgetting to check llms.txt first: Jumping straight to searching individual HTML pages can lead to parsing complex HTML and missing curated LLM context. Always fetch /llms.txt or /llms-ctx.txt first.
• Incorrect file structure: Failing to include the required H1 or blockquote summary when generating the file.
• Overlooking .md extensions: Wasting tokens on raw HTML when appending .md to the URL could yield clean markdown.

References

• llms.txt Official Specification (links to docs)
• llms.txt proposal: The proposal for llms.txt Includes Background, Proposal, Format, Example, Existing standards, Directories, Integrations.
• llmstxt.org/llms.txt Includes links to docs.
• llms.txt sites directory A list of all llms.txt file locations across the web.
• AnswerDotAI/llms-txt repo Includes README.md with Background, Proposal, Format, Existing standards, Directories, Integrations and Next steps. Repository also includes llms_txt/.py Python mini scripts and nbs/.ipynb notebooks.