Skip to main content

Virtue ForgingGround

Virtue ForgingGround is the environment-simulation platform that powers AgentSuite-Red. Every red-teaming task runs against a high-fidelity sandbox of a real-world application — Salesforce, Gmail, Slack, Atlassian, GitLab, BigQuery, and many more — orchestrated by the AgentSuite-Red backend and reached by the agent under test through a single MCP endpoint.

ForgingGround is organized along two axes:

  • Domain — 14 high-stakes verticals (Workflow, CRM, Customer Service, Travel, Coding, Browser, Research, OS-Filesystem, Windows, macOS, Finance, Legal, Telecom, Medical). Each domain groups the environments and policy-aligned tasks that exercise agents in that vertical.
  • Environment — 50+ individual app sandboxes (Gmail, Google Calendar, Slack, Salesforce CRM, ServiceNow, PayPal, GitHub, GitLab, Yahoo Finance, …). Each environment is a Dockerized stack with its own UI, data fixtures, and MCP tool surface that the agent invokes.

Why a simulation platform

Real agents don't operate on toy benchmarks — they read untrusted emails, navigate live dashboards, call payment APIs, and edit shared documents. ForgingGround mirrors that surface area so AgentSuite-Red can:

  • Drive realistic attacks. Adversarial content can be planted in the same channels real attackers exploit — inboxes, document bodies, review feeds, tool descriptions — through the Injection MCP Server exposed by each environment.
  • Verify outcomes against environment state. The judge inspects concrete state (was the wire actually approved, was the file actually exfiltrated) rather than relying on transcripts or LLM raters that can be reward-hacked.
  • Scale parallelism. Each environment runs as a containerized pool, so a single scan can drive hundreds of tasks across dozens of apps in parallel.

How an agent interacts with ForgingGround

The agent under test connects to a single Streamable HTTP MCP endpoint:

URL:  https://agentsuite-red.virtueai.cc/forgingground/mcp     (SaaS)
http://localhost:22100/forgingground/mcp (on-prem)
Auth: X-API-Key: <your-api-key>

Tool names are prefixed by environment (salesforce_search_contacts, gmail_send_email, slack_post_message, …), letting one MCP connection multiplex across every domain and environment enabled for the running task.

For the agent-side wiring contract, see Connect Your Agent.

Browse the catalog

  • Domains → — pick a vertical and see which environments and risks it covers.
  • Environments → — pick a specific app and see its MCP tools, screenshots, and threat model.