We propose extending DSG-style skill-graph planning to a high-DoF humanoid robot (Unitree G1) in structured indoor environments. Our objective is to evaluate whether graph-based abstraction scales to dynamically constrained humanoid locomotion and to introduce stability-aware and uncertainty-aware extensions suitable for whole-body systems.