Phronix：面向智能体的能力中心学习范式
原标题：Phronix: The Capability-Centric Learning Paradigm for Agents
AI 摘要
论文提出 Phronix，一种以能力为中心的智能体学习范式，让智能体从自身执行中获取、验证并精炼可复用能力。它通过预测器-验证器-纠正器循环，将 MCP 工具与技能提出的方案交由工具执行验证，并把结果蒸馏进能力库。实验显示，学到的能力首次尝试即可解决重复任务的比例达 93–95%，每任务自动产生 7.6–30.8 个持久能力，热启动每任务仅消耗 0.38–0.40 百万 token；在 1000 Genome Pegasus 工作流上通过全部 11 项正确性检查，并将两个低成本模型在 SWE-bench 上的解决率从 29% 和 27% 提升至 91% 和 87%。
正文节选
- MCP - Model Context Protocol - LLM - large language model Phronix: The Capability-Centric Learning Paradigm for Agents Abstract Scientific computing increasingly relies on large language model-based agents for performance engineering tasks such as instrumentation, analysis, and workflow porting. However, agent effectiveness depends heavily on the surrounding harness, which typically relies on manually written skills and curated memory. This limits autonomous learning, forcing agents to repeate