Crate vex_adversarial 

VEX Adversarial

Red/Blue agent pairing for adversarial verification.

Every “Blue” agent can have a “Red” shadow that challenges its conclusions, reducing hallucinations and improving reliability.

Key Types

• ShadowAgent — Red agent that challenges Blue’s outputs
• Debate — Multi-round debate between agents
• Consensus — Voting protocols for agreement

Quick Start

use vex_adversarial::{ShadowAgent, ShadowConfig, Consensus, ConsensusProtocol, Vote};
use vex_core::{Agent, AgentConfig};

// Create a Blue agent
let blue = Agent::new(AgentConfig::default());

// Create its Red shadow
let shadow = ShadowAgent::new(&blue, ShadowConfig::default());

// Detect issues in a claim
let issues = shadow.detect_issues("This always works 100% of the time.");
// Returns: ["Over-generalization: Verify if universal claims 'always'/'never' hold true for all edge cases.", "Statistics: Determine if percentages are sourced or if they are illustrative placeholders."]

Consensus Voting

use vex_adversarial::{Consensus, ConsensusProtocol, Vote};

let mut consensus = Consensus::new(ConsensusProtocol::SuperMajority);
consensus.add_vote(Vote::new("agent1", true, 0.9));
consensus.add_vote(Vote::new("agent2", true, 0.8));
consensus.add_vote(Vote::new("agent3", false, 0.7));
consensus.evaluate();

assert!(!consensus.reached); // 2/3 is not > 2/3

Re-exports

pub use consensus::Consensus;

pub use consensus::ConsensusProtocol;

pub use consensus::SuperMajorityConfig;

pub use consensus::Vote;

pub use debate::Debate;

pub use debate::DebateRound;

pub use reflection::ReflectionAgent;

pub use reflection::ReflectionConfig;

pub use reflection::ReflectionResult;

pub use shadow::ShadowAgent;

pub use shadow::ShadowConfig;

Modules

consensus

Consensus protocols for multi-agent agreement

debate

Debate protocol between Blue and Red agents

reflection

Reflection agent for self-improvement

shadow

Shadow agent spawning and management