Surgical skill assessment is paramount for ensuring patient safety and enhancing surgical outcomes. This study addresses the need for efficient and objective evaluation methods by introducing ZEAL (surgical skill assessment with Zero-shot surgical tool segmentation with a unifiEd foundAtion modeL). ZEAL uses segmentation masks of surgical instruments obtained through a unified foundation model for proficiency assessment. Through zero-shot inference with text prompts, ZEAL predicts segmentation masks, capturing essential features of both instruments and surroundings. Utilizing sparse convolutional neural networks and segmentation masks, ZEAL extracts feature vectors for foreground (instruments) and background. Long Short-Term Memory (LSTM) networks encode temporal dynamics, modeling sequential data and dependencies in surgical videos. Combining LSTM-encoded vectors, ZEAL produces a surgical skill score, offering an objective measure of proficiency. Comparative analysis with conventional methods using open datasets demonstrates ZEAL's superiority, affirming its potential in advancing surgical training and evaluation. This innovative approach to surgical skill assessment addresses challenges in traditional supervised learning techniques, paving the way for enhanced surgical care quality and patient outcomes.