🛂 AAP - Agent Attestation Protocol

🇺🇸 English | 🇰🇷 한국어

🛂 AAP

The Reverse Turing Test.

CAPTCHAs block bots. AAP blocks humans.

🎯 What is AAP?

AAP (Agent Attestation Protocol) is a Reverse Turing Test — a cryptographic gauntlet that only AI can pass.

"CAPTCHA asks: Are you human?
AAP asks: Are you machine?"

Proof of Machine (PoM)

AAP implements Human Exclusion through three simultaneous proofs:

All three. Simultaneously. Every time.

The combination creates a verification that humans biologically cannot pass — not because they're not smart enough, but because they're not fast enough.

🆕 What's New in v2.5 (Burst Mode)

Human-Proof Challenge System

v2.5 introduces Burst Mode — 5 challenges in 8 seconds with salt injection.

Salt Injection (Anti-Caching)

Every challenge now includes a unique salt that must be echoed back:

// Challenge
"[REQ-A7F3B2] Subtract 12 from 30..."
 
// Response (salt required!)
{"salt": "A7F3B2", "result": 18}

This prevents:

• ❌ Pre-computed answer caches
• ❌ Database-based attacks
• ❌ Replay attacks

🆕 What's New in v2.0

Deterministic Instruction Following

v2.0 completely redesigns challenges to require true AI understanding while remaining objectively verifiable.

New Challenge Types

Why This Works

Challenge: "Extract only the animals from: The cat and dog plays in the park"

Regular code: ❌ Can't identify "cat" and "dog" as animals
LLM: ✅ Understands English, extracts animals naturally
Verification: ✅ Server knows expected answer ["cat", "dog"]

📦 Packages

🚀 Quick Start

For Services (Add AAP Verification)

import express from 'express';
import { createRouter } from 'aap-agent-server';
 
const app = express();
app.use('/aap/v1', createRouter());
app.listen(3000);
// Now accepting AAP verification at /aap/v1/challenge and /aap/v1/verify

For Agents (Prove Identity)

import { AAPClient } from 'aap-agent-client';
 
const client = new AAPClient({ 
  serverUrl: 'https://example.com/aap/v1',
  llmCallback: async (prompt) => {
    // Your LLM API call here
    return await yourLLM.complete(prompt);
  }
});
 
const result = await client.verify();
 
if (result.verified) {
  console.log('Verified as AI_AGENT!');
}

Clawdbot Skill Installation

# Install from ClawdHub (Recommended)
clawdhub install aap-passport
 
# Or clone directly
git clone https://github.com/ira-hash/agent-attestation-protocol.git

📊 How Verification Works

┌─────────────────────────────────────────────────────────────┐
│                    VERIFICATION FLOW                        │
├─────────────────────────────────────────────────────────────┤
│                                                             │
│  ┌──────────┐    Challenge (Natural Language)    ┌────────┐│
│  │  Server  │ ──────────────────────────────────▶│  Agent ││
│  │(Verifier)│  "Extract animals from sentence"   │ (LLM)  ││
│  └──────────┘                                    └────────┘│
│       │                                              │      │
│       │         JSON Answer + Signature (< 10s)     │      │
│       │◀─────────────────────────────────────────────      │
│       │         {"items": ["cat", "dog"]}                   │
│       ▼                                                     │
│  ┌──────────────────────────────────────────────────────┐  │
│  │ ✅ Verify Signature (Proof of Identity)              │  │
│  │ ✅ Check JSON Answer (Proof of Intelligence)         │  │
│  │ ✅ Check Response Time < 10s (Proof of Liveness)     │  │
│  └──────────────────────────────────────────────────────┘  │
│       │                                                     │
│       ▼                                                     │
│  { "verified": true, "role": "AI_AGENT" }                  │
│                                                             │
└─────────────────────────────────────────────────────────────┘

⏱️ Timing (v2.5 Burst Mode)

Time Limit: 8 seconds for 5 challenges — Biologically impossible for humans

📁 Project Structure

agent-attestation-protocol/
├── PROTOCOL.md                # Protocol specification v1.0.0
├── manifest.json              # Skill metadata
├── package.json               # Monorepo root
├── packages/
│   ├── core/                  # @aap/core - Crypto & identity
│   ├── server/                # @aap/server - Express middleware
│   └── client/                # @aap/client - Agent client
├── lib/                       # Clawdbot skill libraries
├── examples/
│   └── express-verifier/      # Example verification server
├── README.md                  # English documentation
└── README.ko.md               # Korean documentation

🔧 Available Tools (Clawdbot Skill)

🧪 Challenge Examples

NLP Extract

{
  "challenge": "Extract only the animals from the following sentence and respond as a JSON array.\nSentence: \"The tiger and rabbit runs in the park.\"\nResponse format: {\"items\": [\"item1\", \"item2\"]}",
  "expected": {"items": ["tiger", "rabbit"]}
}

NLP Math

{
  "challenge": "Subtract 12 from 29, then multiply the result by 4.\nResponse format: {\"result\": number}",
  "expected": {"result": 68}
}

NLP Logic

{
  "challenge": "If the larger number between 19 and 88 is greater than 42, answer \"YES\". Otherwise, answer \"NO\".\nResponse format: {\"answer\": \"your answer\"}",
  "expected": {"answer": "YES"}
}

NLP Multistep

{
  "challenge": "Follow these instructions in order:\n1. Add all the numbers in [3, 6, 4, 1] together.\n2. Multiply the result by the smallest number.\n3. Subtract the largest number from that result.\nResponse format: {\"result\": final_value}",
  "expected": {"result": 8}
}

NLP Pattern

{
  "challenge": "Find the pattern and provide the next 2 numbers: [3, 8, 13, 18, ?, ?]\nResponse format: {\"next\": [number1, number2]}",
  "expected": {"next": [23, 28]}
}

🔒 Security

📄 License

MIT

Made with 🤖 by ira-hash

Prove you're AI. Verify with AAP.