ClawLoop：基于Harness原生训练循环的自我改进智能体
原标题：ClawLoop: Self-Improving Agents via Harness-Native Training Loop
AI 摘要
美团与北京大学的研究者提出 ClawLoop，一个无需外部教师模型、可自我改进的智能体训练框架。它把任务合成、训练（SFT/RL）和诊断反馈组成闭环，让智能体在 harness 环境中自主生成任务、收集轨迹并诊断弱点，再用于下一轮训练。以 Qwen3.6-27B 为循环模型，在 SkillsBench、Claw-Eval-Live 和 ClawGym-Bench 上取得持续提升，且合成任务可迁移到更小的 Qwen3.5-4B 和 9B 模型并带来增益。
正文节选
ClawLoop: Self-Improving Agents via Harness-Native Training Loops Abstract The rapid advancement of Large Language Models (LLMs) has driven their deployment toward general-purpose harness environments, where LLM-based agents operate in sandbox workspaces and interact directly with file systems and shells. While numerous post-training pipelines exist for tool-use agents, their counterparts for harness-based agents remain scarce and are predominantly one-shot and teacher-dependent: a stronger exte