AGENTIF: Benchmarking Large Language Models Instruction Following Ability in Agentic Scenarios | Read Paper on Bytez