Game-theoretic LLM: Agent Workflow for Negotiation Games