This article investigates the resilient output consensus of continuous-time cyber-physical systems, where cyber-attacks with arbitrary forms occur in the communication layer. To make the system possess higher security, we first propose a more general attack model, which is not subject to any constraints. Then, we design a dynamics controller whose internal state is transmitted in real-time from the communication layer to the physical layer through the information exchange with neighboring subsystems. Utilizing the notion of trusted agents, we propose a resilient output consensus algorithm, which does not require any prior knowledge about attacks. Also, we obtain the appropriate gain matrices by solving the linear output regulation equation. Furthermore, a sufficient condition on the connection relation of trusted agents is provided to prove that all output values can reach a resilient consensus. Finally, we exhibit two numerical simulations to confirm the validity of the theoretical results.