FlowEvo：透過工作流程與可執行技能的共同演化實現自我進化的代理系統

摘要

大型語言模型代理可透過在推論時建構工作流程來適應複雜任務，但在單一執行回合中發現的程序通常會在執行後被捨棄。現有的技能庫提供可重複使用的可執行常式，但通常是在離線狀態下組裝，且不會從代理自身的工作流程中成長。我們提出 FlowEvo——一個免訓練框架，讓工作流程與技能在推論時共同演化。FlowEvo 將成功的工作流程編譯為可呼叫的技能，存放在持久化的技能庫中，並透過直接執行或作為建構新工作流程的上下文來使用檢索到的技能。它也會追蹤每個技能的下游效用，並抑制導致負遷移的技能。在共享的 GPT-4o-mini 骨幹模型下，FlowEvo 在 ALFWorld、HumanEval、MBPP、GSM8K 與 MATH-500 的完整標準分割上，於 8 個基準方法中達到最高準確率。在 ALFWorld 上，它達到 85.6%，比最強基準高出 26.4 個百分點，且使用的 token 數量約為其三分之一。在涵蓋 7B 至 671B 參數的 10 個基礎模型上，FlowEvo 在 50 個模型-資料集比較中有 49 個優於 ExpeL。程式碼已公開於 https://github.com/DEFENSE-SEU/FlowEvo。

English

Large language model agents can adapt to complex tasks by constructing workflows at inference time, but procedures discovered in one episode are usually discarded after execution. Existing skill libraries provide reusable executable routines, but are typically assembled offline and do not grow from the agent's own workflows. We introduce FlowEvo, a training-free framework in which workflows and skills co-evolve at inference time. FlowEvo compiles successful workflows into callable skills, stores them in a persistent bank, and uses retrieved skills either through direct execution or as context for constructing new workflows. It also tracks each skill's downstream utility and suppresses skills that cause negative transfer. Using a shared GPT-4o-mini backbone, FlowEvo achieves the highest accuracy among 8 baselines on the full standard splits of ALFWorld, HumanEval, MBPP, GSM8K, and MATH-500. On ALFWorld, it reaches 85.6%, 26.4 points above the strongest baseline, while using roughly one third as many tokens. Across 10 base models spanning 7B to 671B parameters, FlowEvo outperforms ExpeL in 49 of 50 model-dataset comparisons. Code is available at https://github.com/DEFENSE-SEU/FlowEvo.