Reconfigurable Intelligent Surface (RIS) is a technology that optimizes wireless communication performance by intelligently controlling signal reflections. In RIS-assisted communication systems, deep learning-based methods require a large number of parameters and high computational complexity to improve channel state information (CSI) feedback performance. In this paper, we propose a lightweight autoencoder network, referred to as LwCSI-Net, for implementing a low-complexity feedback design in RIS-assisted multiple-input single-output (MISO) communication systems. Specifically, we design a novel lightweight feedback scheme that integrates multi-layer convolutional networks (CNNs) with attention mechanisms. The network employs one-dimensional convolution, which reduces the number of parameters to be learned by sliding the convolutional kernel in one direction, while maintaining efficient feature extraction capabilities. Additionally, the attention mechanism enables the model to focus on important features, reducing redundant computations and further lowering complexity. Experimental results show that, for compression ratios (CR) of 16, 32, and 64, LwCSI-Net significantly improves CSI reconstruction performance while maintaining fewer parameters and lower computational complexity, achieving an average complexity reduction of 28.58% compared to state-of-the-art (SOTA) CSI feedback autoencoder architectures.