Environment-Grounded Automated Prompt Optimization for LLM Game Agents

2026-06-16

AI Summary

This work addresses the high sensitivity of large language model (LLM) agents to prompting in interactive environments and the limitations of existing manual, task-specific prompt engineering. The authors propose a fine-tuning-free framework for automatic prompt optimization that decomposes agent behavior into a goal-conditioned describer and an action selector. A behavior analyzer attributes environmental rewards to specific prompt components, enabling targeted mutations. Integrating a multi-agent architecture, LLM-guided evolutionary search, and environment rollback validation, the method achieves substantial performance gains across all five BabyAI tasks in the BALROG benchmark. Notably, it attains a 72.5% success rate on the PutNext task—where RobustCoTAgent completely fails—demonstrating its robustness and effectiveness in complex reasoning scenarios.

Abstract

LLM agents in interactive environments are highly sensitive to their prompts, yet prompt engineering remains a manual, task-specific process. We introduce an automated prompt optimization framework for LLM agents that decomposes the observation-to-action pipeline into a goal-conditioned descriptor agent and an action selection agent, and iteratively refines each module's prompt through an LLM-driven evolutionary loop guided by environment returns. We propose a behavior analyzer to attribute episode outcomes to specific prompt components, and a mutator to propose targeted revisions to the prompt, before validating them through environment rollouts. We evaluate on all five BabyAI tasks in the BALROG benchmark, comparing our pipeline against BALROG's RobustCoTAgent under both plain and guided prompt initializations. Optimization improves performance consistently across tasks and conditions, without requiring updates to the model weights. On PutNext, a multi-step coordination task where the RobustCoTAgent achieves 0% success, our framework reaches up to 72.5% success rate using the same underlying LLM with optimized prompts. These results suggest that a multi-agent framework, combined with automatic prompt optimization, enhances LLMs without the need for fine-tuning or extensive human supervision.

Problem

• prompt engineering
• LLM agents
• interactive environments
• automated optimization
• environment grounding

Innovation

• automated prompt optimization
• LLM agents
• evolutionary prompt refinement
• multi-agent framework
• environment-grounded learning