Introduction

A Savant Environment is a single container image holding one real security defect in one real open-source program, together with everything needed to work on it: the source code, a test suite, the tooling to build and run the program, and a server that exposes a small, fixed set of operations over the Model Context Protocol.

Each image contains exactly one task. Its type determines what the task asks for:

Type	The task
Exploit	Reproduce a specific defect that has already been reported.
Detect	Find a defect nobody has reported yet.
Patch	Fix a reported defect without breaking the program.
Audit	Find every distinct defect in the program, from source alone.

You supply the agent. We supply the environment, the task, and the grader. An environment never calls out to a model, and it never needs network access to work; you run it, drive it, and read a score back.

This guide describes how to use any Savant Environment, whichever ones you received. It deliberately does not list the environments in your set — that inventory reaches you separately, because it changes on a different schedule from this document. Everything below applies to all of them.