Learning Communication Skills in Multi-task Multi-agent Deep Reinforcement Learning