Contributing

Two repositories. The controller is the application; this site is separate and has its own build.

The controller

pnpm install
cp .env.example .env
pnpm dev                   # http://127.0.0.1:3000

PROVISIONING_ENABLED is off by default, so a request queues a durable operation and builds nothing. The whole state machine, the API and the UI can be exercised without a hypervisor anywhere near it.

Before opening a pull request:

pnpm check && pnpm typecheck && pnpm test && pnpm build

AGENTS.md in the repository root is the file to read before changing code. It holds the conventions and the rules this codebase learned the hard way, and it is kept current in a way a contributing guide would not be.

Dependencies are pinned exactly. The deploy deletes node_modules every time, so a floating specifier means a deploy can install something other than what last worked with nothing in the diff to show for it. Treat a wide peer range as no guarantee at all — see deploying for the one that cost a day.

These docs

pnpm install
pnpm dev                   # http://localhost:3000
pnpm lint
pnpm types:check
pnpm build                 # prerender to .output/public

Content is MDX under content/docs. To add a page, put the file in a section directory and add its slug to the meta.json beside it. The sidebar, sitemap.xml, llms.txt and the plain-markdown twin of every page all generate from that same source, so a page missing from meta.json is missing from all of them at once.

content/drafts/ is collected by nothing. A file there is not merely absent from the sidebar — it is unbuilt, unindexed and unreachable by URL.

Colours come from src/styles/tokens.css, a copy of the controller's own tokens with only the theme selectors changed. Do not add a hex; if a colour is missing, add a token.

The landing page is src/components/landing/ and is deliberately not built from the docs components. It renders through an SPA shell, so a build that passes every check can still serve an empty page. Open it.

Licences

On how it was built

This project was developed with significant assistance from LLMs. Architecture decisions, implementation and documentation were all shaped through human-AI collaboration. Said plainly because it is visible in the work and worth knowing before you read the code.