PLAY2PROMPT: Zero-shot Tool Instruction Optimization for LLM Agents via Tool Play
This addresses the challenge of integrating specialized tools with LLMs in true zero-shot settings, offering a scalable solution for domain-specific applications.
The paper tackles the problem of zero-shot tool usage for LLM agents with minimal or noisy documentation by proposing PLAY2PROMPT, an automated framework that systematically explores tool behaviors through trial-and-error to refine documentation and generate usage examples without labeled data. Experiments show it significantly improves zero-shot tool performance across open and closed models.
Large language models (LLMs) are increasingly integrated with specialized external tools, yet many tasks demand zero-shot tool usage with minimal or noisy documentation. Existing solutions rely on manual rewriting or labeled data for validation, making them inapplicable in true zero-shot settings. To address these challenges, we propose PLAY2PROMPT, an automated framework that systematically "plays" with each tool to explore its input-output behaviors. Through this iterative trial-and-error process, PLAY2PROMPT refines tool documentation and generates usage examples without any labeled data. These examples not only guide LLM inference but also serve as validation to further enhance tool utilization. Extensive experiments on real-world tasks demonstrate that PLAY2PROMPT significantly improves zero-shot tool performance across both open and closed models, offering a scalable and effective solution for domain-specific tool integration.