A continuous evaluation framework for enterprise AI agent skills is proposed, combining outcome-level and process-level checks to detect behavioral drift in skills due to changing tool APIs, models, and specifications.