Spatial-temporal graphs are a commonly used approach in skeleton-based action recognition models. To efficiently capture strong movement patterns from skeleton graphs, multi-scale and aggregation of context over long distances, along with spatial-temporal dependency modeling, are critical for creating a powerful feature extractor. Nevertheless, limitations remain in achieving lightweight models suitable for real-world applications. In this work, we introduce a multi-form adjacency matrix that uses the Euclidean distance between each joint for graph convolution. This approach reduces model size by replacing the multi-scale G3D pathway with an all-scale adjacency matrix that captures the relationships between all joints. Additionally, we integrate graph transformer convolution to uncover correlations between each node, evaluated on the large-scale dataset NTU RGB+D 60.