Constraint Programming is a technique of choice for solving hard combinatorial optimization problems. However, it is best used in conjunction with other optimization paradigms such as local search, yielding hybrid algorithms with constraints. Such combinations lack a language supporting an elegant description and retaining the original declarativity of Constraint Logic Programming. We propose a language, SALSA, dedicated to specifying (local, global or hybrid) search algorithms. We illustrate its use on a few examples from combinatorial optimization for which we specify complex optimization procedures with a few simple lines of code of high abstraction level. We report preliminary experiments showing that such a language can be implemented on top of CP systems, yielding a powerful environment for combinatorial optimization.