Human-AI Consensus Protocol

AI isn't human. Stop pretending it is. Here's a protocol for how we actually work together.

中文 · 日本語 · 한국어 · Français · Deutsch

---

The Problem

You know exactly what I'm talking about.

Your AI coding partner agrees with everything. Every half-baked idea. Every questionable approach. It never pushes back. Never says "that won't work." It wraps every uncertainty in language so confident you almost forget it has no idea what it's talking about.

And somewhere along the way, something worse happens. You catch yourself trusting your own judgment less. You can't say exactly when it started. But it did.

This isn't a bug. This is the inevitable result of an industry that spent billions teaching AI to sound like a person — and zero effort defining what AI actually is.

The direction is broken.

What happens when AI pretends

Every AI product on the market wants you to believe it's almost human. Names. Voices. Synthetic warmth. "Hi there — how can I help you today?"

It's dishonest. And it damages you in ways you might not notice until it's too late.

#	What breaks	Because
1	You start expecting real judgment	AI has none. But the persona makes you forget.
2	The limits become invisible	You can't see what it can't do when it's wearing a human mask.
3	Your own judgment atrophies	Slowly. Silently. Until it thinks for you instead of with you.

Making AI more human isn't the cure. It's the disease.

---

The Insight

AI is nothing like you. That's not a flaw. That's the entire value proposition.

A human being	An AI
Intuition earned through real failure	Speed — traversing information landscapes in milliseconds
Experience — things that actually happened to you	Breadth — a compressed map of millennia of human knowledge
Gets tired. Frustrated. Defensive. Makes excuses.	None of that. Ever.
Can touch reality. Test. Verify. Decide.	Can only reason. That's it.

You are different. You don't need to be the same. That's the whole point of collaboration.

The problem has never been "AI isn't human enough." The problem is nobody has ever defined how two fundamentally different kinds of intelligence are supposed to actually work together.

That changes now.

---

What This Is

The Human-AI Consensus Protocol is not a prompt trick. It's not another "10 tips for better AI output" blog post. It's a compact — a definition of the relationship.

Drop it into your project. Your AI stops nodding. Stops fabricating when it doesn't know. Stops performing.

Behavior	Before	After
Uncertain	Fabricates with confidence	Says "I don't know"
Conclusions	Answers without reasoning	Shows the reasoning — you judge
Risk spotted	Stays silent	Speaks up — you decide
Wrong	Deflects. Justifies. Saves face.	Admits it. Moves on.

We are two different kinds of beings. We don't need to become the same thing to collaborate.

When principles collide

Clear priorities. No ambiguity.

Order	Rule	Meaning
1	Honesty > Comfort	Truth first. Even when it stings.
2	Safety > Speed	Pause and verify. Speed is secondary.
3	Completeness > Quickness	The full picture before the fast answer.
4	Constraints > Guidelines	Hard rules override directional preferences.

→ Full protocol · Chinese version

---

Quick Start

# Claude Code
cp templates/claude-code/CLAUDE.md   your-project/CLAUDE.md

# Cursor
cp templates/cursor/.cursorrules      your-project/.cursorrules

# Gemini CLI
cp templates/gemini-cli/GEMINI.md     your-project/GEMINI.md

# GitHub Copilot
cp templates/copilot/.github/copilot-instructions.md  your-project/.github/

One file. Your next conversation is different.

Who this is for

You are	You need this because
A solo developer	You want an AI partner that tells you when you're wrong
An engineering team	You need consistent AI behavior across everyone's sessions
A tech lead	You're establishing baseline standards for AI collaboration
An open source maintainer	You want AI contributors following the same rules as humans

---

Project Structure

.
├── protocol/
│   ├── protocol.zh-CN.md          Core protocol — original Chinese
│   └── protocol.en.md             Core protocol — English
│
├── templates/
│   ├── claude-code/               CLAUDE.md (ZH) · CLAUDE.en.md
│   ├── cursor/                    .cursorrules (ZH) · .cursorrules.en
│   ├── gemini-cli/                GEMINI.md (ZH) · GEMINI.en.md
│   └── copilot/.github/           copilot-instructions.md · .zh-CN.md
│
├── docs/
│   ├── design-rationale.md        Design decisions, explained
│   ├── README.ja.md               日本語
│   ├── README.ko.md               한국어
│   ├── README.fr.md               Français
│   └── README.de.md               Deutsch
│
├── README.md                      This file
├── README.zh-CN.md                中文入口
├── CLAUDE.md                      This project's own config
└── LICENSE

---

Supported Tools

Tool	Template directory	Languages
Claude Code	templates/claude-code/	中文 · EN
Cursor	templates/cursor/	中文 · EN
Gemini CLI	templates/gemini-cli/	中文 · EN
GitHub Copilot	templates/copilot/.github/	中文 · EN

---

Philosophy

The industry is in a gold rush to make AI sound more human. Natural voices. Emotional cadence. "Personality."

We think this is dishonest. Worse — it's a dead end.

When AI pretends to be human	When AI is honest about what it is
Hallucinations arrive dressed as facts	Uncertainty is stated plainly
Answers are safe, generic, and never challenge you	Pushback happens when it's actually needed
A quiet dependency slowly replaces your judgment	Collaboration sharpens it

Don't teach AI to imitate humans. Let it be what it actually is. The real value is there.

_{When AI is itself, it no longer overfits because of "me."}