A large language model text file gives an agent a compact starting point for understanding a website. It can identify the project, summarize its purpose, and link to the most useful documentation or machine-readable resources without requiring the agent to crawl and interpret every page.
The file should be hand-curated, current, and concise. It is not an access-control mechanism, a complete replacement for documentation, or a guarantee that every model will use it. Links should resolve to authoritative content, and changes should be tested from the perspective of several retrieval tools.
