Transformer-based Working Memory for Multiagent Reinforcement Learning with Action Parsing Y aodong Y ang