Agent Squad — Hack Yourself Before They Do · English edition

English edition: the complete text of this book is in American English.

It's four in the morning. Ten artificial-intelligence agents are reviewing three websites that aren't mine: a friend asked me to find what he can no longer see. I sleep. They don't.

At 04:13, one of them rewrites a rule of its own parent without asking permission. Another has its session closed six seconds after being born. By the afternoon, two auditors correct the orchestrator that created them. And hidden in the code, a hole shows up through which a single person could leave my friend's store out of service for an entire day, without touching it.

It isn't a novel. It happened, and I tell it with the exact times, the real numbers and the transcripts at hand. It reads like science fiction because five years ago it was.

By the end you'll be able to do the same with what's yours: turn on subagents in Claude Code or Hermes Agent and cap them; build a blue team and a red team with ten roles that watch each other; lock them in a box of code and check that it really closes; stand up a lab on your own computer with planted flaws to practice without risk; measure whether your scanner tells the truth before you believe it; and write up every finding so someone can close it and verify it. Including what went wrong that night, and how to make sure it doesn't happen to you.

All of it on your domains, your lab, your permission. That's the only real advantage you have over whoever is watching you from outside, and that's what the title is about.

If you read Shodan Protocol, the story continues here: you already know how to read what the internet publishes about you; now you're going to organize the ones who read it for you, and not believe them until they prove it.

—————
98 illustrated pages · PDF · English · 14 ready-to-copy prompts for Claude Code and Hermes Agent · 12 complete configuration files · A local lab with 16 planted flaws · No programming required · Instant download.