Open Agent is my attempt to make this more structured: - Containerized workspaces with systemd-nspawn (or host workspaces if you trust the agent) - A git repo for your library of skills/tools/rules that syncs automatically - A web dashboard to start missions, view thinking/tool calls, browse history and manage environments - An iOS client for simple monitoring

It's opinionated: the backend is deliberately thin (no agentic logic), and it supports both Claude Code and OpenCode as runtimes. Goal is to be able to easily switch between them if needed (you keep one unified library that is "translated" to the various formats).

Still rough around the edges. Would appreciate feedback on what's missing or what you'd do differently