Résumé : In this report we present knowledge-based techniques for automating the use of a modular set of programs (or program supervision in short). Our goal is to provide tools which are both general and flexible. First we describe how program supervision knowledge can be modeled and we propose a knowledge description language, YAKL, to easily express this knowledge. Then we present two dedicated program supervision engines which are extensions of the OCAPI program supervision engine. The PEGASE engine provides mechanisms for managing sophisticated repair strategies. The second one PHENIX is a first attempt to introduce more dynamicity in the planning phase of PEGASE. Examples are shown through an image processing application, the morphological description of galaxies.