Give the docs to an AI coding agent

Point Claude Code, Cursor or any coding agent at the handbook - llms.txt, every page as one text file, one page as Markdown, the docs search and the OpenAPI document.

Developers and AI

The docs site publishes the handbook in forms a coding agent can read: an index at llms.txt, every page as plain text in one file, a single page as Markdown, a search that answers JSON, and the developer API as an OpenAPI 3.1 document. This recipe picks the right one for the job and hands it to the agent.

Before you start

  • An AI coding agent that can read a web address or take pasted text: Claude Code, Cursor, Copilot or another.

  • Nothing else: every docs address below is public and needs no sign-in or key.

What the docs site publishes for agents

https://docs.blockoutgames.com/llms.txt

A short index in the llms.txt format: what the handbook covers, and links to the full text, the sitemap, the OpenAPI document, the API reference and each section.

https://docs.blockoutgames.com/llms-full.txt

Every page as plain text in one file, in reading order: each page's title, its address after Source:, its one-line description, then its text, pages separated by ---.

Copy page

A button on every docs page that copies the page as Markdown. Its menu offers Copy as Markdown and Open llms.txt.

https://docs.blockoutgames.com/api/search?q=

The docs search as JSON: the best-matching parts of pages for a query of 2 to 100 characters, best first, each with its page's slug, title and description, the part's heading (id and text, or null for a page's opening), an excerpt, the part's whole text and a score.

https://docs.blockoutgames.com/openapi.json

Every route, parameter and answer of /developer/v1 as OpenAPI 3.1, readable from any origin.

Steps

  1. Pick the size of context

    For one task, give the agent the one page it needs (Copy page). For a broad question about the studio, the panel or the Template, give it llms-full.txt. To call the developer API, give it openapi.json.

  2. Copy one page

    On the docs page, choose Copy page and paste the Markdown into the agent's chat.

  3. Or give it the addresses

    Paste the addresses and say what they are for, as in the prompt below. An agent that can fetch pages reads them itself.

  4. Let it search instead of reading everything

    An agent that can make requests can ask /api/search?q= for the parts that match its question, and fetch a whole page only when it needs one.

  5. Ask it to cite

    Ask the agent to name the page it used, by the Source: address in llms-full.txt or the slug in a search result, so you can check it.

A prompt for a coding agent
Read https://docs.blockoutgames.com/llms.txt first. For the Blockout developer API,
use https://docs.blockoutgames.com/openapi.json as the source of truth for routes,
parameters and answers. For anything else, search
https://docs.blockoutgames.com/api/search?q=<words> and read the page it names.
Cite the page address for every rule you follow.

Search the docs from a script

Shell
curl -s "https://docs.blockoutgames.com/api/search?q=rate+limit+429"

A query shorter than 2 or longer than 100 characters answers 400 with validation_failed. Answers may be cached for 5 minutes.

Generate a client from the OpenAPI document

Shell
curl -s https://docs.blockoutgames.com/openapi.json -o blockout-openapi.json

Feed the file to your OpenAPI tool, or to the agent with "write a typed client for these routes". The document describes the routes a game's key opens; the API reference shows the same, readable, with Try it.

What you get

An agent that answers from the handbook as published rather than from memory, cites the page it used, and writes API calls against the real OpenAPI document.

Updated