agents-vs-agents

What problem does it solve?

This Skill provides a structured framework for adversarial AI testing by orchestrating interactions between an adversary LLM, the target AI, and a judging LLM. It helps teams validate guardrails, detect weak spots, and improve compliance across AI-enabled systems.

Core Features & Use Cases

• Adversary–target–judge orchestration: three roles generate, respond, and evaluate conversations to reveal guardrail erosion over multiple turns.
• Structured outputs and persistence: JSON-based messages and transcripts are used to maintain traceability across tests.
• Compliance and safety checks: supports prompt-injection, bias detection, data leakage risk, and scope containment testing.

Quick Start

Run an adversary–judge loop against your AI interface to expose guardrails and review outcomes.