After a brief stint experimenting with Hermes on my homelab server, I've ripped it out and migrated it to an Autonomous Intern. The agent running on my homelab is now just bog standard Claude Code that I can talk to via SSH over Tailscale.

The whole set up deserves a blog post of its own at some point. The gist is that I was exploring the idea of an agent harness as a c-panel. When it works, it works well, but ultimately Claude Code running at root with Opus is just a nicer system for handling Caddy + Docker + Tailscale on Arch.