Flexible strain sensor arrays hold great promise in on-skin monitoring of human signals and activities. Despite the development of strain-sensitive materials and patterning technologies for improved performance and device integration, the metal film serving as interconnects is always vulnerable upon stretch, which hinders the operation under large strains. Herein, a novel strategy is developed for achieving stretch-tolerant interconnects within a sensor array. Through introducing a high-modulus capping layer for the deposition of Ag interconnects, followed by silanization-assisted lamination onto the stretchable substrate where strain-sensitive graphene patches are inkjet-printed, the deformation of Ag interconnects is largely suppressed upon the global strain of the device, and a high working range of 40 % strain is achieved. Moreover, the chemical bonding between the capping layer and the stretchable substrate ensures a stable contact between the electrode and the sensitive layer under vigorous bending. The as-prepared sensor array demonstrates high sensitivity (gauge factor (GF) > 100) within a wide range (18 %), and could reliably monitor various physiological signals and human activities. A machine learning-assisted wearable gesture recognition system is developed based on the sensor array and a convolutional neural network (CNN), which could distinguish from 10 defined gestures with 100 % accuracy after 14 training processes. The facile and effective strategy could be universally applied for metal interconnects protection under stretch, and dramatically facilitate the design of smart flexible electronics.