This paper studies a novel distributed fault estimation framework for multi-agent systems under directed topology, subject to time-varying multiplicative and additive faults. Both actuator and sensor faults are simultaneously addressed by introducing an augmented system. A two-step design process is presented aimed at joint estimation of faults, system state, and exogenous disturbance, which involves an lth-order proportional-integral observer and a constrained least square estimator. Utilizing the relative output of neighbor information enhances the accuracy of fault and state estimation. Output sharing is realized by a dynamic event-triggered communication protocol, which effectively saves network resources. The design conditions of the observer are formulated as an optimization problem subject to linear matrix inequalities, ensuring guaranteed H-infinity performance with regard to not only estimation error but also event error. Simulation results validate the effectiveness and feasibility of the proposed method.