Tag
The paper presents ACES, a framework for continuous evaluation of AI agent skills through live trials, measuring Skill Lift to quantify added value, and demonstrating its effectiveness on enterprise repositories compared to scan-only gates.