Autonomous valet parking systems face significant challenges in generating optimal trajectories in complex environments, particularly in dead-end configurations where traditional planning methods often fail. This paper presents SA-RS (Skeleton-Assisted Reeds-Shepp), a hierarchical path planning framework that integrates skeletal extraction with Reeds-Shepp curves. The proposed approach optimizes parking entry poses through reverse trajectory analysis, generates efficient global paths via skeletal guidance, and refines trajectories to ensure kinematic feasibility. Simulation results show that SA-RS achieves comparable path optimality to the Hybrid A* benchmark while reducing computational complexity by an order of magnitude. In scenarios with severe spatial constraints, where conventional planners struggle with low success rates and exponential computational growth, SA-RS maintains consistent performance through predictive reverse motion planning. Quantitative evaluations across diverse parking configurations validates their robustness in generating collision-free, kinematically feasible trajectories, demonstrating its potential for real-world autonomous parking applications.