Claude Code Loops - Stop Prompting, Start Shipping

You are the bottleneck. You prompt, you wait, you read the diff, you prompt again. Your agent stops the second you stop typing.


Again.


Meanwhile, developers on Reddit are closing their laptops and waking up to merged pull requests. One of them left the house for four hours and came back to seven shipped PRs, unattended. The difference isn't talent. It's a loop that knows when it's done.


This book gives you a fleet of agents that ships real code by Sunday. You start your first loop in Chapter 1, on a real repository, in ten minutes. By Chapter 4 you are running three kinds of loop and know which one to reach for. By Chapter 6 a separate verifier signs off before anything reaches your branch, so what you wake up to is work you can merge instead of work you have to check. By Chapter 9 you know what a night of agent work costs before you start it. By Chapter 12 your fleet runs on a thirty-day schedule, wired so one loop's verified output becomes the next one's input.


This isn't another Claude Code tour. The number one Claude Code book on Amazon publishes its table of contents: eleven chapters, and not one names a loop. Its own four-star reader says so out loud. The loop shelf has ninety-six books, and its entire first page carries ten ratings, of which the leader's eight are all unverified. Across the top three listings, the entire vocabulary of cost, budget and tokens appears exactly once. You can't ask Claude to build the system that replaces you at the keyboard. You can't ask it to prove its own work honestly. You can't ask it what a night of agent work will cost before you spend it. This is the book those authors needed before they wrote theirs.


Every loop, hook, gate and receipt in the book ships in a free companion GitHub repo: one folder per chapter, and one command that installs the whole set into a project of your own.


Here's what you'll build:

1. A loop running on your own repository ten minutes from now.

2. Three kinds of loop, built and priced, so you reach for the right one instead of the loud one.

3. A verifier that signs off before code reaches your branch, which the working agent cannot talk its way past.

4. A recovery ladder that gets a stuck run moving again instead of stalling your morning.

5. A per-iteration receipt in tokens, minutes and dollars, so a night of agent work has a price you knew in advance.

6. A break-even model that puts one overnight run against your own hourly rate.

7. A fleet of loops on a thirty-day schedule, wired into a graph that feeds itself.


The loop shelf grew by four books yesterday. Ninety-six titles, six languages, and ten ratings across the whole first page. Nobody has won it yet. This one is 186 pages you build rather than read, every command tested on Linux, WSL and macOS, with a companion repo whose paths actually resolve. Scroll up and grab your copy.