Read This First
Stop running Claude as one model answering once
Most people use about ten percent of what Claude can do: one model, one answer, take it or leave it. The other ninety percent is an Agent Graph. You split Claude into a small team that hands work down a line, and the move nobody makes is putting one agent on that team whose only job is to attack the work: hunt every weak spot, every hole, every lazy shortcut, and throw it back to be rebuilt until it survives.
So what reaches you is not Claude's first answer. It is the answer that already got torn apart by Claude and lived. You build the graph once and it hardens everything you feed it after that. Everything below is copy and paste, in order. It is plain prompting, so it works in Claude Code, the desktop app, or the API, on any plan.
Part 1 of 5Name the five agents
A graph is just agents with clear jobs and an order to pass work in. Keep it small. Four builders that do the real work, and one attacker that tries to break it. Define them once and reuse them for every job.
- The Planner. turns your request into a short spec and a checklist of what done looks like.
- The Builder. does the actual work against the spec.
- The Attacker. the whole trick. Its only job is to break the work, not to help.
- The Fixer. takes the attacker's findings and rebuilds the weak parts.
- The Judge. decides if it survived the attack or goes back for another round.
Free, for your email
Get the full setup steps + the copy-paste setup prompt
The rest of this guide (15 more sections) opens right here. The setup prompt, one paste into Claude that sets it all up, comes by email only. Use your real address.
Unlocks: Paste This Once to Set Up the Graph · Plan and build the first version · Paste This with your Task · Task: [describe What you Want in One or Two Lines] · result against it · +10 more
50,000+ follow where these guides come from. No spam, reply “unsubscribe” any time.