V3S (Virtual Reality for Safe Seveso Substractors) is an ANR/RNTL project (French national agency for research). In this project, we aim to design a tool allowing to scenarise hazardous working situations on SEVESO sites for risk prevention, training and decision making. The tool interprets a high level task and a related risk model. It is meant for a manager to help him/her to make decisions. The manager plays the scenario of an intervention anf manages a team of virtual operators (associated with autonomous agents) in the VERP (Virtual Environment for Risk Prevention) submitted to constraints. Depending on his/her decisions, the incurred risks are displayed in the virtual environment. Our architecture relies on a multi-agents platform (OMAS). In this paper we present some of the features of the tool through different scenarios. We present the organisational rules of our system, how it self-adapts to the technical competencies and the characteristics of the operators (human factors). We also present the working environment hosting our agents.