Agent-safe infrastructure

Public · this site runs on it

Trigger Agents deploy my apps. They should never hold root on a server or the keys to my DNS.

Harden a fresh Ubuntu server in one script: key-only SSH on a random port, firewall, fail2ban, automatic updates and a verify mode.
Run one shared Caddy per server. Projects never open ports; they join a Docker network and Caddy routes to them.
Accept web traffic only from Cloudflare, so the server's real address is never in DNS.
Every script checks state first, can be re-run and takes --dry-run. All of them run as an unprivileged deploy user.

LimitDNS changes run from my workstation with a scoped token. The server never holds it.

RollbackA site-remove script takes a project off the proxy. A full server rebuild was tested in October 2026.

0: root or DNS credentials on the server
2: public repos

Stack Bash, Caddy, rootless Docker, Cloudflare API, Ubuntu 26.04

Code caddy-behind-cloudflare serverbeef

Why

Coding agents are good at deploying apps and bad at knowing when to stop. I wanted a setup where the worst an agent can do on a server is break the one project it was working on.

How

serverbeef hardens a new Ubuntu 26.04 server in one idempotent script. caddy-behind-cloudflare then installs a shared Caddy, adds sites and connects domains through Cloudflare, all over SSH as an unprivileged user. Each step checks the current state, can show what it would do and can be repeated without breaking anything.

How this site is deployed

wenyangho.com is a static Astro build served by nginx in a small container. The container joins the shared proxy network on the same server as my other projects, and Caddy routes to it behind Cloudflare. No ports open on the container and no secrets on the server.