Quantitative assessment for post-stroke spasticity remains a significant challenge due to the encountered variable resistance during passive stretching, which can lead to the widely used modified Ashworth scale (MAS) for spasticity assessment depending heavily on rehabilitation physicians. To address these challenges, a high-force-output triboelectric soft pneumatic actuator (TENG-SPA) inspired by a lobster tail is developed. The bioinspired TENG-SPA can generate approximately 20 N at 0.1 MPa, providing sufficient stretching force for spastic fingers. The anti-interference, durability, and electrical output characteristics of the TENG-SPA under varying conditions—such as different air pressures, bending frequencies, and simulated spastic finger stretching—are explored, demonstrating TENG-SPA's ability to sense resistance during the stretching process. Furthermore, a TENG-SPA-enabled hand rehabilitation robot system integrated with the convolutional neural network (CNN) is further developed, which is tested in a clinical trial involving 15 stroke patients. The results have demonstrated that a classification accuracy for the levels of finger spasticity reaches 93.3% and the MAS scores predicted by the CNN regression model exhibit a strong linear relationship with the actual MAS (R2 = 0.8451, p < 0.01). This study presents promising potential applications in digital rehabilitation medicine, human–machine interaction, biomedicine, and related fields.