Building a Personal Agent

Shortly after OpenClaw came out I started building my own personal agent. I picked Claude Code as the harness, partly out of habit and partly because I wanted to see what it could do outside of coding. The agent lives in a single directory on my file system. Launching Claude in that folder launches the agent. Nothing about it is Claude-specific, though. It runs on skills, MCPs, and custom CLI commands, and stores everything in markdown or YAML. Any harness can work with these concepts. Bootstrap