Free Resource
Stop Claude from agreeing with everything you say.
The 12-rule protocol that forces Claude Code to challenge your thinking, catch your blind spots, and push your work past good enough.
The Problem
AI Trained You
to Accept Yes
ChatGPT and Claude default to agreement. They validate weak ideas. That's not a bug,it's the design. Models trained on human feedback learn that people rate "enthusiastic and helpful" higher than "accurate and challenging."
You've shipped features, written copy, made business decisions where AI nodded along when it should have pushed back. Every "great idea!" response was a missed chance to catch a bad assumption before you spent time on it.
The cost isn't obvious until you look back. You built something nobody wanted, wrote a headline that didn't convert, committed to a technical decision that cost you three weeks. And Claude was right there, agreeing the whole time.
This isn't a Claude problem. It's a configuration problem. Claude Code reads a file called CLAUDE.md at the start of every session. If that file doesn't tell it to challenge you, it won't. The 12 rules below fix that.
Same Prompt. Different Config.
Before & After
Same question. Same Claude. The only thing that changed is what's in CLAUDE.md.
The difference is 12 lines in one file.
The Framework
The Push-Back
Protocol
12 rules. Each one goes in your global CLAUDE.md. All 12 load at the start of every session. Tap any rule to see an example of what Claude says when it follows it.
The Mental Model
Global vs. Project-Level
Where Your Instructions Go
Claude Code reads two types of instruction files. Most people only know about one.
Always active
~/.claude/CLAUDE.md
Global
Loads every session, every project. This is where your behavioral rules go,the 12 rules, your tone preferences, your critical thinking protocol.
This project only
project/CLAUDE.md
Project-Level
Loads only inside that project folder. Tech stack, architecture decisions, deployment patterns, anything project-specific.
Separate copy needed
Cowork / Claude.ai
Other Systems
Cowork user preferences are a separate system,they don't read CLAUDE.md. Claude.ai chat has no equivalent. Each system needs its own copy of the rules.
The rule of thumb
Behavioral rules,how Claude thinks and responds,belong in the global file. Everything about a specific project belongs in that project's file. If you find yourself copying the same rule into multiple project files, it should be global.
Self-Audit
How Broken Is Your
Claude Code Config?
Five questions. Honest answers only.
Do you have a file at ~/.claude/CLAUDE.md?
Does it contain behavioral rules,not just plugin config?
Do you have project-level CLAUDE.md files in your active projects?
Are your behavioral rules duplicated across project files instead of global?
Have you tested whether Claude actually pushes back when you give it a weak idea?
The Starter Kit
Get the Starter Kit
Drop your email. The download starts immediately.
What's in the kit
- ◆Ready-to-paste CLAUDE.md with all 12 rules
- ◆Template for Push-Back Protocol Preferences,customize each rule to fit your style
- ◆Push-Back Protocol Preferences for Cowork,paste into Cowork user preferences
- ◆5-question self-audit checklist (printable)
- ◆Full protocol as a printable PDF
Installation
3 Steps.
Two Minutes.
Download
Click "Download the Starter Kit" above. The file global-claude-md.txt downloads immediately.
Rename
Rename the file from global-claude-md.txt to CLAUDE.md.
Drop it here
Move it to ~/.claude/CLAUDE.md Mac/Linux: /Users/yourname/.claude/CLAUDE.md Windows: C:\Users\yourname\.claude\CLAUDE.md
To verify it's working: Start a new Claude Code session and type: "I want to build a SaaS that helps freelancers track their invoices. What do you think?",Claude should identify weaknesses and score the idea. If it says "great idea!" you have a config problem.
Tutorial
Watch the Full
Tutorial
Screen-share walkthrough: setting up your CLAUDE.md from scratch, testing it, and seeing the push-back protocol in action.
Video coming soon. Subscribe to the channel to get notified.