MILO: Automated Harness Discovery via Orchestrated Multi-Agent Evolution

• 类型：arxiv
• 标识：2609.38349
• 链接：https://arxiv.org/abs/2609.38349
• 主分类：evaluation
• 形态：method
• TLDR：Modern agentic systems combine an AI model with a harness that controls execution and environmental interactions. Harness design strongly affects long-horizon performance, yet its combinatorial search space demands substantial human effort that must be repeated as models change. Existing automated methods explore this space narrowly, optimizing only components such as prompts or skills or becoming trapped by fixed, exploitative search strategies. We introduce MILO (Meta-evolutionary Island Orchestration), a framework that co-evolves agent harnesses and the strategy used to discover them. MILO
• 副分类：agent
• 待LLM分类：否
• 标题中文：MILO：通过编排式多 Agent 进化实现自动化 Harness 发现
• TLDR中文：现代 Agentic 系统将 AI 模型与控制执行及环境交互的 harness 相结合。Harness 设计显著影响长期性能，但其组合式搜索空间需要大量人工投入，且模型更迭时必须重新进行。现有自动化方法对该空间的探索范围狭窄，仅优化 prompt 或 skill 等单一组件，或受限于固定且易陷入局部利用的搜索策略。本文提出 MILO（Meta-evolutionary Island Orchestration），一个协同进化 Agent harness 与其发现策略的框架……
• 来源文件：
• /inbox/tom/_candidates/2026-10-02-agent-rag-longcontext-candidates.json