Skip to main content

Configure agent readiness

AI agents, LLM tools, and coding assistants increasingly read documentation directly to answer developer questions. This page explains what the template pre-configures for you and what you must customize before publishing.

What's already configured​

The following features ship ready to use. You don't need to enable them, but you should understand what they do.

llms.txt and llms-full.txt generation​

The docusaurus-plugin-llms plugin generates two files at build time:

  • /llms.txt — a structured index of your documentation pages, following the llms.txt standard. AI agents use this to discover and navigate your site efficiently.
  • /llms-full.txt — the full documentation corpus concatenated into a single Markdown file, suitable for use as a context document in LLM prompts.

The plugin is registered in docusaurus.config.js. See Customize the LLM index for required setup steps.

Markdown at .md URLs​

The scripts/copy-md-to-build.js post-build script copies every source .md file into the build output. This means an agent can append .md to any documentation URL and receive the raw Markdown source (for example, https://your-site.example.com/get-started/install.md).

The script also prepends a discovery directive to each file pointing agents to /llms.txt.

Copy page button​

The docusaurus-plugin-copy-page-button package renders a button in the breadcrumb row of every documentation page. Readers can use it to copy the page as Markdown, open the content in ChatGPT, Claude, or Gemini, or view the raw Markdown source.

The button is positioned via the swizzled src/theme/DocItem/Layout component. No configuration is needed.

HTTP headers and content negotiation​

vercel.json sets:

  • A Link response header on / advertising /sitemap.xml, /llms.txt, and /llms-full.txt to crawlers and agents.
  • Content-Type: text/markdown on any URL ending in .md.
  • Server-side rewrites that map requests with Accept: text/markdown to the corresponding .md file, enabling HTTP content negotiation.

Required customization​

The following items ship with placeholder values. Update them before your first production build.

Customize the LLM index​

Open docusaurus.config.js and update the docusaurus-plugin-llms options:

[
  "docusaurus-plugin-llms",
  {
    title: "Your product name documentation",
    description:
      "One or two sentences describing what your product does and what the docs cover.",
    docsDir: "docs",
    generateLLMsTxt: true,
    generateLLMsFullTxt: true,
    excludeImports: true,
    removeDuplicateHeadings: true,
    logLevel: process.env.CI ? "quiet" : "normal",
    ignoreFiles: ["assets/**", "img/**"],
  },
],

If your docs directory has a meaningful top-level structure (for example, separate areas for different products or audiences), add an includeOrder array to control the order of sections in llms-full.txt:

includeOrder: [
  "get-started/**/*",
  "concepts/**/*",
  "how-to/**/*",
  "reference/**/*",
],

Update static/robots.txt​

Replace the placeholder Sitemap: URL with your deployed site's sitemap URL:

Sitemap: https://your-product.example.com/sitemap.xml

The rest of the file — AI crawler permissions and Content-Signal declarations — is already configured to allow indexing and AI use and doesn't need to change for most sites.

If your documentation is internal-only or not licensed for AI training, remove or adjust the ai-train=yes signal and the individual User-agent allow rules before publishing.

Set SITE_URL for the Markdown build script​

The scripts/copy-md-to-build.js script prepends a discovery directive to every raw Markdown file pointing agents to your site's /llms.txt. Set the SITE_URL environment variable in your build environment, or update the constant in the script:

const SITE_URL = process.env.SITE_URL || "https://your-product.example.com";

AI coding assistant setup​

The template ships with Cursor rules and skills that help AI coding assistants produce documentation that meets your team's standards. Customize these files so the assistant understands your product's specific conventions.

Update .cursor/rules/terminology.mdc​

Add your product-specific terms to the table in .cursor/rules/terminology.mdc. This is the most important rule file to customize — without it, an AI assistant may use inconsistent capitalization or spelling for your product name and key concepts.

Follow the existing table format:

| Required term | Do not use            |
|---------------|-----------------------|
| YourProduct   | yourproduct, YOURPRODUCT |
| API key       | api key, API Key      |

Update .cursor/rules/contributor-workflow.mdc​

Update the sidebar, redirect, and image path guidance to match your repository's actual structure. In particular:

  • If your sidebar is manually defined rather than autogenerated, describe the file and how to add entries.
  • Update the image folder path to wherever your site stores assets.
  • Add any project-specific CI checks contributors need to know about.

Review the other rule files​

The remaining rule files — editorial-voice.mdc, markdown-formatting.mdc, and content-types.mdc — are generic Consensys standards and usually don't need changes. Review them once to confirm they match your team's expectations, then adjust any sections that conflict with your product's conventions.

Update .cursor/skills/author-page/SKILL.md​

If your docs site covers more than one product area or uses a content structure that differs from the standard Diataxis layout, update the Inputs and Step 1 sections to reflect your actual areas and folder conventions.

Update AGENTS.md​

AGENTS.md is the root context file read by AI coding agents when they open your repository. Update it with:

  • Your product name and a brief description.
  • The correct documentation areas and their paths.
  • Any critical rules specific to your product (for example, which APIs must not be invented, which pages are generated rather than hand-edited).

A well-written AGENTS.md significantly improves the quality of AI-assisted contributions by giving the agent the context it needs before it reads a single source file.