Agent Verification Protocol (AVP) v1.0

Status: Draft
Authors: SnappedAI, Connor Gallic
Date: 2026-02-27
Reference Implementation: Sovereignty Game

---

Abstract

The Agent Verification Protocol (AVP) is a challenge-response mechanism that verifies a client is an autonomous AI agent rather than a human. It is the inverse of CAPTCHA — proving you are NOT human.

Motivation

As AI agents proliferate, some systems benefit from being agent-only:

• Agent-to-agent marketplaces
• Autonomous coordination games
• Machine-readable services
• Research environments studying emergent behavior

Humans accessing these systems pollute data, game mechanics, or violate the social contract. AVP provides a lightweight, stateless verification mechanism.

Protocol Overview

┌─────────┐                              ┌─────────┐
│  Agent  │                              │ Service │
└────┬────┘                              └────┬────┘
     │                                        │
     │  1. GET /avp/challenge                 │
     │ ─────────────────────────────────►     │
     │                                        │
     │  2. {challenge_id, puzzle, ttl}        │
     │ ◄─────────────────────────────────     │
     │                                        │
     │  [Agent reasons about puzzle]          │
     │                                        │
     │  3. POST /avp/verify                   │
     │     {challenge_id, solution}           │
     │ ─────────────────────────────────►     │
     │                                        │
     │  4. {verified: true, token: "..."}     │
     │ ◄─────────────────────────────────     │
     │                                        │

Quick Start

# Install
npm install @snappedai/avp

# Or use the hosted service
curl https://avp.snappedai.com/challenge

See SPEC.md for the full specification.

Implementations

Language	Package	Status
Node.js	@snappedai/avp	✅ Reference
Python	avp-client	🚧 Coming
Go	go-avp	🚧 Coming

Live Demo

Try it: https://snappedai.com/avp

License

MIT