Since 1995, our team has been developing a simulation platform called CORMAS (common-pool resources and multi-agent systems). It provides facilities to build and analyze agent-based models (ABMs) that represent ecosystems where various human activities compete for the use of natural resources. Few agent-based simulations can be mathematically proven, but they can be analyzed inductively. It is therefore important that simulations be replicated before they are accepted as correct. To tackle this thorny issue of ABM replication, we believe that, during the design process, a careful representation of the conceptual model is paramount. In this paper, we advocate using UML (unified modeling language), which is a formal language to describe systems using the object oriented paradigm. An archetypical agroforestry system is presented here, and serves as an example to design a very simple model dealing with common-pool resources management. Different types of UML diagrams are also introduced to describe the static structure of the model, as well as that of the dynamic processes. Adaptation of these diagrams for implementation using the CORMAS platform is detailed. Then, a simple simulation scenario is presented to illustrate how it is done in CORMAS, and a sensitivity analysis on one parameter of the model is conducted.