Set up a second brain that shows its work
Seven steps: where the base lives, the script that lays it out, and the five things /onboard asks. Nothing on this page is sent anywhere, and nothing is written to disk: it hands you three things to paste yourself.
1Where it lives
Put it on a volume you can carry, mounted at a fixed path. The whole knowledge base becomes one file to copy or back up, and the fixed path is what lets scheduled jobs keep working after the base moves house. Encrypt it if the notes are personal.
2The skeleton
Folders, version control, one commit, the prompts installed as slash commands, three tools, and a hook that greets each new session with what is waiting. It runs only in an empty folder, and the fresh volume from step 1 counts as one. You can read it first: it is about two hundred lines and reaches for nothing beyond curl, git and the everyday mkdir, touch, rm, grep, wc, tr and dirname.
4Your gates
The one decision nobody can make for you: which actions must never happen without your explicit yes. Nothing is ticked for you. Most people should tick the first four, and leave one out only if they truly would not stop for it; tick any of the rest, and add your own. Editing anything under 4_Archive/ is not on the list because it is not a choice: the rulebook refuses it outright.
5Your areas
What you are responsible for on an ongoing basis. Each becomes a folder under 2_Areas/, and the rulebook will propose new ones rather than create them on the way past. Two to start from; add as many as you need, a few words each.
7Write the rulebook
Start Claude Code in that folder and run /onboard. It asks the same five things, one at a time, before it writes anything. Answer as it goes, or hand it everything from this page up front with this:
Handed everything up front, it writes CLAUDE.md from these answers, so read them before you paste. What it writes is committed and in git, and /onboard run again shows you the rulebook and asks before replacing it.