Earth Observation (EO) missions are among the most critical and rapidly growing applications of satellite constellations due to their usage in different domains including climate monitoring, disaster management, agricultural planning, and national security. These missions generally require high availability, timely target monitoring, and system-level resiliency and adaptability in the presence of failures or dynamic environmental conditions. However, current satellite constellation adaptation heavily depends on human operators to interpret telemetry data, identify the root causes of faults, and issue telecommands to perform recovery actions. In most cases, faulty satellites would enter safe mode which significantly affect the mission success rate and objectives. Moreover, most of the existing automated and autonomous approaches focus on adaptation and resilience at the single?satellite level. To overcome these limitations, mission-aware coordinated adaptation augmented by Deep Reinforcement Learning (DRL) is proposed to seamlessly integrate mission success with satellite health and resilience. To evaluate and validate our proposed mechanism, we conducted a preliminary analysis by using unresponsive/malfunctioning reaction wheel (RW) fault with two different reasons (causes) in a satellite cluster. The experiments show that our mechanism outperforms traditional methods in terms of mission success rate and fault compensation, and they demonstrate the importance and impact of root?cause analysis in decision making and recovery.