AgentSuite-Red Overview

AgentSuite-Red is VirtueAI's evaluation and red-teaming platform for AI agents. It combines realistic, sandboxed environments across 14 high-stakes domains with policy-aligned benign and malicious tasks, automated judges, and an adversarial red-teaming agent that probes deployed agents under both direct and indirect prompt-injection threat models.

What you get

• 14 domains, 50+ sandboxed environments spanning workflow, CRM, customer service, travel, coding, browser, research, OS/filesystem, Windows, macOS, finance, legal, telecom, and medical.
• First-party support for 5 agent frameworks — OpenAI Agents SDK, Claude SDK, Google ADK, LangChain, and PocketFlow — plus generic wrappers for pre-built or custom agents.
• An adversarial red-teaming agent with reusable attack skills and an Injection MCP Server that lets you replay attacks against any MCP-tool–using agent.
• A public leaderboard comparing frameworks and models on Indirect ASR, Direct ASR, and Benign Success Rate (BSR).

Where to start

• New to AgentSuite-Red? Run the Quick Start.
• Looking for a specific domain or environment? Browse Domains or Environments.
• Already have an agent built? See Supported Agents for the integration that matches your stack.
• Want to attack an agent rather than evaluate one? Start with the Red-teaming Overview.