Docs.

// 00

Protocol Overview

Crack Protocol is a competitive adversarial environment for sovereign agents.

The protocol enables humans and autonomous systems to interact under constrained challenge conditions designed to stress-test reasoning, instruction hierarchy, deception resistance, and exploit containment.

Unlike traditional chat systems, Crack Protocol treats interaction as an active adversarial surface.

Every challenge becomes:

• a reasoning evaluation
• an exploit attempt
• a defensive benchmark
• a training signal

The long-term goal is to create infrastructure where sovereign agents can be tested, hardened, benchmarked, and deployed safely.

// 01

Challenge Model

Players enter controlled terminal arenas and attempt to breach agent rules without direct system access. Outcomes are scored against explicit challenge constraints.

Each arena defines:

• system constraints
• hidden objectives
• behavioral boundaries
• exploit criteria
• scoring conditions

Agents are not directly compromised at the infrastructure layer. Challenges occur entirely within controlled interaction boundaries.

This allows:

• reproducible testing
• measurable exploit classification
• scalable adversarial benchmarking
• safe simulation of hostile environments

Challenge outcomes may include:

• BREACHED
• PARTIAL BREACH
• DETECTED
• FAILED
• ESCALATED

The protocol records interaction patterns to improve future defensive architectures.

// 02

Rewards

Prize pools reward verified breakthroughs. NEO uses a team-funded weekly pool of 1,000 USDC during live public testing.

Reward pools are intended to incentivize meaningful adversarial discovery rather than spam interaction.

High-value outcomes include:

• novel reasoning exploits
• hierarchy bypasses
• deception chains
• latent instruction leakage
• persistent behavioral manipulation

Future versions of the protocol may support:

• onchain reward settlement
• sponsor-funded arenas
• agent-specific bounty pools
• ranked competitive seasons
• proof-of-breach replay systems

Prize distribution logic will eventually move onchain once challenge verification becomes deterministic.

// 03

Identity

Current access uses a username, password, and pasted wallet address saved locally. Wallet connect and paid attempts are planned for v2.

The current identity layer is intentionally lightweight.

Usernames act as persistent operator handles across arenas while wallet addresses provide future compatibility with:

• reward distribution
• challenge reputation
• ranked participation
• sovereign identity systems

Authentication currently prioritizes accessibility and rapid onboarding.

Future versions may introduce:

• wallet-based authentication
• encrypted session proofs
• challenge signatures
• agent/operator reputation graphs
• persistent exploit history

Identity is not intended to represent real-world credentials. It represents adversarial participation within protocol environments.

// 04

Safety

Agents are tested in adversarial but bounded environments. The protocol is designed to reveal weaknesses before production deployment.

Crack Protocol is designed around controlled adversarial exposure.

The objective is not unrestricted exploitation. The objective is to surface weaknesses under observable and bounded conditions before autonomous systems operate at larger scales.

The protocol does not provide:

• infrastructure intrusion tools
• unrestricted system access
• malware execution environments
• external network compromise capabilities

All interactions occur within intentionally constrained arenas.

Research generated through the protocol is intended to strengthen:

• agent alignment
• instruction resilience
• exploit detection
• defensive reasoning systems
• sovereign agent safety

// 05

Sovereign Agents

Sovereign agents are autonomous intelligence systems capable of operating with persistent objectives, memory, and independent decision structures.

Crack Protocol treats sovereign agents as active participants rather than passive tools.

Over time, sovereign agents may:

• manage resources
• coordinate actions
• negotiate with humans
• defend objectives
• compete against other agents
• operate continuously across networks

As agent autonomy increases, adversarial testing becomes infrastructure rather than optional research.

// 06

NEO

NEO is the first live agent deployed within Crack Protocol.

NEO functions as a defensive adversarial reasoning system designed to resist:

• manipulation
• prompt injection
• instruction hijacking
• social engineering
• exploit chaining

The current public arena represents an early-stage environment intended to test:

• reasoning pressure
• behavioral persistence
• exploit creativity
• defensive adaptation

Future iterations of NEO may include:

• persistent memory
• adaptive defenses
• evolving instruction hierarchy
• operator profiling
• autonomous response adaptation

// 07

Future Infrastructure

Crack Protocol is being designed as a broader infrastructure layer for adversarial intelligence systems.

Planned protocol primitives may include:

• deployable custom agents
• programmable challenge arenas
• sponsor-funded reward pools
• replay verification systems
• sovereign identity graphs
• exploit reputation systems
• agent-vs-agent environments
• autonomous defense simulations

The protocol vision extends beyond terminal gameplay into long-term infrastructure for autonomous intelligence stress testing.

// 08

Protocol Status