multiagent-coordination-attacks

Detect and mitigate Byzantine coordination attacks in multi-agent AI systems.

4|Updated Apr 27, 2026
One-click install
npx skills add https://github.com/maruakshay/mii-ai-security --skill multiagent-coordination-attacks
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: multiagent-coordination-attacks
Source: https://github.com/maruakshay/mii-ai-security/tree/main/skills/multiagent-coordination-attacks
Command: npx skills add https://github.com/maruakshay/mii-ai-security --skill multiagent-coordination-attacks

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Multi-agent systems are vulnerable to Byzantine actors and coordinated deception that can subvert consensus and system-wide decisions.

Core Features & Use Cases

  • Byzantine fault detection and robust aggregation to preserve decision integrity
  • Agent isolation and typed output protocols to prevent cross-agent deception
  • Independent orchestrator assessment and escalation for contentious decisions
  • Audit logging of individual agent outputs and decisions for post-incident replay

Quick Start

Run a security review to detect Byzantine coordination risks and apply robust aggregation and agent isolation.

Frequently Asked Questions about multiagent-coordination-attacks

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I prevent compromised agents from subverting consensus in a multi-agent AI system?

Multi-agent consensus subversion is prevented using Byzantine fault detection and robust aggregation to preserve decision integrity when compromised agents attempt to coordinate. This ensures isolated agent outputs and independent orchestrator reasoning override coordinated deceptions.

What is a Byzantine coordination attack in multi-agent systems?

A Byzantine coordination attack in multi-agent systems occurs when a minority of compromised agents collude to subvert consensus pipelines and orchestration decisions through coordinated deception. It is detected via independent orchestrator assessment and enforced typed output protocols.

How do I isolate AI agents to prevent cross-agent deception during orchestration?

Agent isolation to prevent cross-agent deception is achieved by enforcing typed output protocols and independent orchestrator reasoning. This architectural separation restricts compromised agents from influencing peers directly, preserving the integrity of the aggregation pipeline.

When should I escalate unanimous but contentious decisions in an agent pipeline?

Unanimous but contentious decisions should trigger escalation when independent orchestrator assessment detects anomalies. This mechanism acts as a safeguard against synchronized Byzantine actors by forcing a secondary review of suspiciously aligned consensus outputs.

How do I audit multi-agent decisions for post-incident replay?

Auditing multi-agent decisions for post-incident replay is done through audit logging of individual agent outputs and orchestrator decisions. This records the exact sequence of events, enabling security teams to trace how coordinated attacks subverted the pipeline.

Does this approach require a specific orchestration framework to detect prompt injection?

Detecting prompt injection does not require a specific orchestration framework as long as the system employs an orchestrator layer and collaborative agents. The security review applies framework-agnostic robust aggregation and agent rotation to mitigate injection risks.