AI agents in this homelab — overview

Why this section exists

Since August 2026, this homelab has included a small "virtual team" of Claude Code agents: specialized instances that help automate, document, and secure the infrastructure. This isn't a gimmick — it's a genuine exploration of what agentic AI looks like when it has real tools and real (if small-scale) infrastructure to act on, and it doubles as learning material for anyone curious about the pattern, technical or not.

The core idea

For the precise definitions of main agent, subagent, skill, and MCP, see the Glossary — Agent, Subagent, Skill, MCP.

The one non-negotiable rule

No agent in this setup ever connects to a system, credential, or tool belonging to my employer, in a regulated industry. Any professional-world problem I want to explore here (hardening standards, compliance checks, clustering issues) gets reproduced from scratch with homelab-only data and infrastructure — never imported, never connected to anything real.

I think this generalizes well beyond a homelab: in any regulated environment — healthcare, government, finance — AI tooling should stay cleanly separated from production systems until governance has actually signed off on it. Treat that boundary as a feature, not a temporary inconvenience.