We propose a constraint-aware decision-focused learning framework that enables parameter prediction in constraints, without making any assumptions about the underlying optimization problem.