The introduction of MOCHA, a new optimization method for agent skills, significantly enhances performance by addressing multi-objective constraints that existing prompt optimizers overlook. This advancement leads to notable improvements in task correctness and the discovery of more optimal skill variants across various applications.