How to use the llms.txt Analysis and Generation agent skill

What is an agent skill and what does this one do?

An agent skill is a folder containing a SKILL.md file of instructions that an agent such as Claude Code loads when a task matches it. This skill covers the llms.txt standard, a proposed plain-text Markdown file placed at a site's root. The file summarises the site and points AI systems to its most important pages, and the source describes it as analogous to robots.txt, except it says what is useful to understand rather than what not to access.

Once the skill is installed, the agent can fetch a site's existing llms.txt and llms-full.txt, check them against the format rules, and report problems such as missing sections, relative URLs or broken links. It can also crawl a site and draft a new llms.txt from scratch.

• Validate: fetch /llms.txt and /llms-full.txt, record the HTTP status and check the structure.
• Assess: score completeness, accuracy and usefulness of the file.
• Generate: crawl the site and draft a new llms.txt.

How do you install the geo-llmstxt skill?

Download the ZIP for the skill from the directory page and unzip it. The skill folder contains SKILL.md. Place the folder in one of these locations:

• For all your projects: unzip into ~/.claude/skills/
• For one project only: unzip into the project's .claude/skills/ folder
• In the Claude apps: upload the skill under Settings → Capabilities → Skills

How do you trigger the skill with a natural request?

You do not need a special command. Describe the task in plain language and name the domain or site. The agent matches the request to the skill's description. Clear requests work best, such as stating whether you want a check of an existing file or a new file written.

• Audit: "Check the llms.txt on example.com and tell me what is wrong with it."
• Generate: "Crawl example.com and draft an llms.txt file for me."
• Compare: "Does example.com have an llms-full.txt, and is it consistent with its llms.txt?"

What does the agent need to run it?

The skill expects Claude Code or a compatible agent with these tools: Read, Grep, Glob, Bash, WebFetch and Write. It also needs network access so it can fetch the target website. The skill does not run scripts, but it does use the network.

• Read, Grep, Glob, Bash, WebFetch and Write tools available to the agent
• Network access to the target website

What should you check before publishing a generated file?

The source states that the file must sit at the root of the domain, for example https://example.com/llms.txt. Generated content should be reviewed by you, because the skill's format rules call for factual descriptions and accurate Key Facts, and the agent can only report what it found on the pages it fetched.

Use the format rules as a checklist: an H1 title first, a blockquote description under 200 characters, at least one H2 section, absolute https URLs, and a description after each colon in every page entry.

• Title as the first line, using a single # heading
• Description in a blockquote under 200 characters
• At least one ## section, with 10-30 page entries in total
• Full absolute URLs, not relative paths
• Key Facts and Contact sections with information you can verify

What are the safety notes for scanning sites?

The skill fetches pages from the web, so only use it on sites you own or have permission to scan. Crawling someone else's site without permission is outside what this skill is meant for.

The skill's stated benefits are claims in its text and are not verified by the files it produces. Publishing an llms.txt does not guarantee that AI systems will read it, cite your pages or rank them. The source also notes that the file cannot make AI systems read it, so it is only useful on a domain where you control the root and can publish a file.

• Only scan sites you own or are permitted to scan.
• Treat the output as a draft and verify every fact and URL.
• Do not expect guaranteed citations or rankings.

What tips improve the results?

Give the agent a precise goal and the domain. Ask for the llms-full.txt check as well, because the source describes it as a deeper companion file that AI systems may load after llms.txt. Keep descriptions specific, since the format asks for 10-30 words per page entry and specific details help differentiate pages.

• Name the exact domain and whether you want an audit, a draft or both.
• Ask the agent to note redirects and any 403 responses, which it should flag as misconfigurations.
• Keep llms.txt concise, around 50-150 lines, and put deeper detail in llms-full.txt.