Hey Patrick

Reading material for the thing you said you wanted to build. Short pages. One thing to try on each.

You told Jeff you want an agent that runs in the background of your life, and that you want to get there your own way, at your own speed. Good. This is not a setup guide for Jeff’s machine. It is a map with a few marked trails, and you pick the trail.

It assumes you use Claude Code every day, because you do. Where a page mentions something you already know, it says so in a clause and moves on. The words go to what you are probably not using yet.

Every page fits on one screen and ends with something you can paste and a way to check it worked. Nothing needs to be read in order. The cheat sheet is the one page meant to be kept open.

Three doors

Level up at work The part fluent users skip. Delegation mechanics, verification as a gate, the prompt-cache reason a context switch cost you 33 cents, and what shipped this summer.

An agent that runs while you sleep What “always on” means, three ways to get there ranked by effort, and what your new UPS bought you.

The ship around the agents Four things that make agents about ten times more useful, with opinions. Tailscale in fifteen minutes.

Your Enterprise-D

You have been talking about a dashboard shaped like the Enterprise-D with agents mapped to the crew. Keep that. It is a better plan than most people’s because it forces three real questions early. Who is on the bridge. Who runs the night shift. What ship do they run on. The three doors above are those three questions, and Name your crew is the page for the fun part.

One thing to paste today

The thing you have had to type at Claude more than once is “Dude, in plain English, tell me exactly what you did and how to verify it in the browser.” That is a rule, and rules go in CLAUDE.md. Paste this into any session:

Add this section to my user-level CLAUDE.md at ~/.claude/CLAUDE.md, then show me the file:

# How to report back
- When you finish a task, tell me in plain English what you changed (which files) and exactly how I can verify it, as numbered steps I can follow in the browser or terminal.
- No closing caveats. No "one thing to note". No list of options unless I asked for options.
- Match the length of the answer to the length of the question.

How you know it worked. New session, /context, and ~/.claude/CLAUDE.md is listed under memory files. Then give it a small task and read the report.

And one setting. You said Opus had become verbose and robotic. There is now a built-in output style for exactly that. /config, Output style, Concise. Claude leads with the result and skips the narration, and still answers in full when you ask for detail. Needs Claude Code 2.1.237 or newer.

Who wrote this

Jeff asked his crew to put this together after your text thread on August 29. He read it before you did. You asked for the unslop skill to be run over it, and it was, every page.