RRSI：智能体 harness 的正则化递归自我改进
原标题：Paper page - RRSI: Regularized Recursive Self-Improvement of Agent Harnesses
AI 摘要
Google Research 提出 RRSI（正则化递归自我改进）方法，用于自动优化 LLM 智能体的 harness（提示词、控制流、工具、记忆与上下文管理）。该方法通过时间退火预算约束候选编辑、并用 critic 与 pruner 筛选提案，以缓解递归进化对训练任务的过拟合。在涵盖编码、智能体工作区与工程设计等八项基准上，RRSI 在进化所用划分上最高提升 14.1 分，在五个分布外基准上最高提升 4.7 分，且策略 token 消耗减少 30%。代码与项目页面已公开。
正文节选
RRSI: Regularized Recursive Self-Improvement of Agent Harnesses Abstract An LLM agent's capability is largely magnified by its harness, namely the prompts, control flow, tooling, memory, and context management surrounding the frozen backbone model. Recent methods increasingly automate this process by iteratively proposing and selecting component-wise edits of an agent harness, practically establishing a form of recursive self-improvement (RSI) at the agent-system level. However, such recursive e