Generate an llms.txt-style file from GitHub

What this workflow is for

Use this when your source material already exists but is scattered across pages, notes, docs, or exports.

Riflet helps collect and structure the material. You still decide what belongs in the public llms.txt file.

Why GitHub is a particular case

A repository is the most natural fit, because the README, the docs directory, and the API reference are already written for an outside reader. They form a ready-made spine for llms.txt with almost no rewriting.

Prefer the documentation over the source. llms.txt is a map, and a link to the docs entrypoint is more useful to an AI system than a listing of every module.

What to leave out of a GitHub bundle

llms.txt is published. Everything in it is readable by anyone and by every AI system that fetches it, so the exclusions matter more than the inclusions.

• Issue templates, CI configuration, and dependency manifests.
• Internal architecture decision records that assume team context.
• Anything in a private repository, even when the docs feel public.

Workflow

Add the repository source, include the files relevant to the task, and use filters to remove generated output, dependencies, and test fixtures.

• Collect the authoritative GitHub pages or files.
• Remove private, duplicate, stale, or low-value material.
• Export a concise Markdown bundle.
• Use the bundle to draft or refresh llms.txt and llms-full.txt.

What to include

Prioritize stable, public, canonical information: product overview, important URLs, docs entrypoints, use cases, and contact details.

Avoid turning llms.txt into a full website dump. It should help AI systems discover the best sources, not replace your site.