The paper proposes a Distributed Inter-Layer Topology Optimization (DITO) method for resource-limited Low Earth Orbit (LEO) Double-Layered Satellite Networks (DLSNs) composed of Remote-Sensing Layer Satellites (RLSs) and Communication Layer Satellites (CLSs). The DITO method aims to maximize task-aware observation benefits with limited onboard resources by optimizing inter-layer links based on time-space relationships between RLSs and CLSs. The method considers practical constraints such as link switching intervals and laser transceiver limitations, and introduces a new time-slot division method to linearize the nonlinear relationship between adjacent time-slots. The proposed model ensures the feasibility of inter-layer connections and supports onboard optimization for future space networks. Simulation results in four constellation systems demonstrate the effectiveness and efficiency of the DITO method, showing that it can achieve optimal solutions close to centralized global solutions with significantly reduced computational time.The paper proposes a Distributed Inter-Layer Topology Optimization (DITO) method for resource-limited Low Earth Orbit (LEO) Double-Layered Satellite Networks (DLSNs) composed of Remote-Sensing Layer Satellites (RLSs) and Communication Layer Satellites (CLSs). The DITO method aims to maximize task-aware observation benefits with limited onboard resources by optimizing inter-layer links based on time-space relationships between RLSs and CLSs. The method considers practical constraints such as link switching intervals and laser transceiver limitations, and introduces a new time-slot division method to linearize the nonlinear relationship between adjacent time-slots. The proposed model ensures the feasibility of inter-layer connections and supports onboard optimization for future space networks. Simulation results in four constellation systems demonstrate the effectiveness and efficiency of the DITO method, showing that it can achieve optimal solutions close to centralized global solutions with significantly reduced computational time.