Public website

The public site is the static tree in website/. It is the front door for https://complexityeater.com. It is not the documentation site, and it does not run a room.

GitHub Pages continues to publish docs/ at https://zloeber.github.io/complexity-eater/. The two sites deploy separately.

Local preview

task website:build
task website:preview

task website:preview serves website/dist on http://127.0.0.1:4173/. task website:check rebuilds the site and checks structure, internal links, the room-launch contract, and the published release assets.

The build copies docs/assets/ce-banner.png into the site. It also publishes skills/commonroom/SKILL.md, skills/commonroom/ADOPT.md, and skills/commonroom/onboarding.txt at /skills/commonroom/ and /prompts/workspace-onboarding.txt. Those texts are the room skill, the first-run adoption skill, and the MCP workspace_onboarding prompt. The build does not call Rust. website/dist is generated and is not committed.

What gets published

scripts/build-website.zsh writes one HTML file per page, the stylesheet and scripts, the favicon, the banner, robots.txt, sitemap.xml, _headers, and contracts/room-launch.v1.json.

Canonical URLs use https://complexityeater.com. The install page is filled from website/releases/v0.2.1.json. When a newer GitHub release should become the one the site describes, replace that manifest with the new tag’s real asset URLs and digest, then run task website:check. The check fetches image-digest.txt and the archive URLs. It fails if they do not match.

Do not put API tokens, invitation strings, or session ids in the site. _headers sends Referrer-Policy: no-referrer and a Content-Security-Policy that allows only this origin’s scripts and images.

Cloudflare

The zone complexityeater.com is already in Cloudflare. As of this writing the name has no public DNS record, so it does not resolve until the Pages project is attached. There is no registrar transfer in this setup.

One-time project creation, from a machine that is already logged in to the same Cloudflare account:

task website:build
npx wrangler@4.151.0 pages project create complexity-eater --production-branch main
npx wrangler@4.151.0 pages deploy website/dist --project-name complexity-eater --branch main --commit-dirty=true

Wrangler is pinned to 4.151.0 in .github/workflows/website.yml. The command above uses the same version. Production is the main branch only.

Then, in the Cloudflare dashboard for that account:

1. Open Workers & Pages, then the complexity-eater project.
2. Add the custom domain complexityeater.com. Because the zone is already in the account, Cloudflare can create the DNS record and the certificate.
3. Add a redirect rule on the zone: if the hostname is www.complexityeater.com, respond with a 301 to https://complexityeater.com plus the original path and query. The apex is the canonical host.
4. Do not attach complexityeater.com to a preview deployment.

HTTPS is terminated by Cloudflare. The Pages project is static files. It does not need a Worker, a database, or an always-on server. Stay on the Pages plan that already fits a small static site, and confirm Cloudflare’s current build and bandwidth limits before expecting a spike.

GitHub Actions

.github/workflows/website.yml deploys after scripts/check-website.zsh passes.

Event	Cloudflare branch	Environment
Push to main	main	website-production
Pull request	the head branch, with / replaced by -	website-preview

Preview URLs are *.complexity-eater.pages.dev. They are not the custom domain.

The workflow reads two GitHub Actions secrets and does not store them in the repository:

• CLOUDFLARE_API_TOKEN, with permission to edit Cloudflare Pages for the account.
• CLOUDFLARE_ACCOUNT_ID.

If either secret is missing, the deploy job logs that it skipped and exits successfully. The site is not live until both secrets exist, a production deploy has succeeded, and the custom domain has been attached. Pull requests from forks do not receive the secrets.

ci.yml runs the same website check on every push and pull request. That job does not deploy.

Rollback

Prefer a git revert of the website commit on main. The next production deploy publishes that tree again, and the repository matches what visitors see.

To roll back without a new commit, open the complexity-eater project in Cloudflare, choose the previous production deployment, and use Rollback. That restores the static files. It does not change the custom domain and it does not change git. Follow it with a revert so the next push does not redeploy the bad tree.

If a deploy attaches the apex to the wrong project, remove complexityeater.com from that project and add it again on complexity-eater. DNS for this zone is the Cloudflare record created with the custom domain, not a second host.

Deleting a Pages deployment removes that copy of the static files. It does not delete a Complexity Eater room, because this site does not create rooms.