AgentGrad：多代理人系統之干預引導式提示優化

摘要

基於大型語言模型（LLM）的多智能體系統（MAS）透過採用專門化的多個智能體達到強大效能，然而其效能取決於每個智能體的提示設計。針對 MAS 提示優化，利用自然語言回饋引導提示更新的文字梯度方法已成為主流範式。本文指出既有文字梯度方法在兩個階段中的限制：梯度提取與梯度聚合。在梯度提取方面，先前研究選定一個目標提示，卻未驗證修改該提示是否能解決該失敗，並且在缺乏對相應智能體中間輸出的智能體層級監督下推導梯度。在梯度聚合方面，個別梯度被隨機分組與串接，經常混合不相關的失敗模式，並產生無法泛化的提示。為了解決這些限制，我們提出 AgentGrad，一個基於序列干預與語意文字梯度抽象的多智能體系統提示優化框架。對於每個失敗，序列干預一次修改一個智能體的行為，以找出修改後能解決該失敗的目標智能體。接著，目標智能體修改後的輸出可作為智能體層級監督，用於提取細粒度梯度。語意文字梯度抽象會將語意相似的梯度聚類，以防止混合不相關的失敗模式，並將每個聚類抽象化為一個能捕捉共同修正模式的泛化梯度。實驗結果顯示，AgentGrad 在五個 MAS 基準上達到最先進的效能，且相較於次快的基線方法，平均可將實際優化時間縮短 2.5 倍。

English

Large language model (LLM)-based multi-agent systems (MAS) achieve strong performance by employing specialized multiple agents, yet their performance depends on the prompt design of each agent. For MAS prompt optimization, textual gradient methods that guide prompt updates using natural-language feedback have emerged as a leading paradigm. In this paper, we identify limitations in two stages of existing textual gradient approaches: gradient extraction and gradient aggregation. In gradient extraction, previous works select a target prompt without verifying whether modifying it resolves the failure, and derive gradients without agent-level supervision over the corresponding agent's intermediate output. In gradient aggregation, individual gradients are randomly grouped and concatenated, often mixing unrelated failure modes and producing prompts that fail to generalize. To address these limitations, we propose AgentGrad, a prompt optimization framework for multi-agent systems based on sequential intervention and semantic textual gradient abstraction. For each failure, sequential intervention modifies the behavior of one agent at a time to identify the target agent whose modification resolves the failure. The modified output of the target agent then serves as agent-level supervision for extracting a fine-grained gradient. Semantic textual gradient abstraction clusters semantically similar gradients to prevent mixing unrelated failure modes, and abstracts each cluster into a generalized gradient that captures the shared corrective pattern. Experimental results show that AgentGrad achieves state-of-the-art performance across five MAS benchmarks and reduces wall-clock optimization time by 2.5times on average compared to the next-fastest baseline.