arXiv 论文提出不更新权重的 Agent Harness 进化方法
Beyond Weight Updates: Evolving Agent Harnesses for Domain Specialization
做垂直领域 Agent 的开发者该看看:不训练模型也能专业化。论文用 GEPA 只改 prompt、工具和技能库,就在 SWE-bench Verified 等三个基准上逼近 SFT/RL 的效果,算力开销小很多,还能和权重训练叠加,特别适合没 GPU 的团队。
arXiv 新论文《Beyond Weight Updates: Evolving Agent Harnesses for Domain Specialization》提出将 LLM Agent 的领域专业化从权重更新扩展到 harness 层面。作者把 system prompt、工具 schema、技能库和规划脚手架的进化建模为 RL 式外循环,用反思式进化优化器 GEPA 实例化,保持策略模型冻结。
生成的 harness 适配器无需 GPU 训练,可按垂直领域加载,并与基于权重的 RL 组合使用。在 Codev(一个 Claude Code 风格的编码 Agent)上,团队于 SWE-bench Verified、OfficeBench 和 Terminal-Bench 三个基准测试,发现权重更新带来最强的单路径收益,harness 进化以极小算力恢复大部分收益,两者结合效果最好。
论文还分析 SFT 专业化的一种失败模式:推理链变短、截断增加,而 harness 适配器在结构上可避免该问题,定位为权重更新的非破坏性补充。
Beyond Weight Updates: Evolving Agent Harnesses for Domain Specialization
Domain specialization of LLM agents is usually framed as a weight-update problem: collect trajectories, run reinforcement learning (RL) or supervised fine-tuning (SFT), and deploy a new checkpoint. We study a complementary axis: evolving the \textbf{agent harness}---the system prompt, tool schemas, skill library, and planning scaffolding---while keeping the policy frozen. We cast harness evolution as an RL-style outer loop and instantiate it with GEPA, a reflective evolutionary optimizer. The resulting harness adapters require no GPU training, are loadable per vertical domain, and compose with weight-based RL. Experiments with Codev, a Claude Code-style coding agent, on SWE-bench Verified, OfficeBench, and Terminal-Bench show a consistent pattern: weight updates give the strongest single-path gains, harness evolution recovers most of those gains at a small fraction of the compute, and combining both paths is best overall. We also analyze a failure mode of SFT-based specialization---shortened reasoning chains and increased truncation---and argue that harness adapters avoid it by construction. Our results position harness evolution as a practical, non-destructive complement to weight updates for vertical agents.
This report is categorized under AITOP's "Papers" desk. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include GEPA, SWE-bench Verified, Agent, SFT.
Source attribution: curated from arXiv cs.AI (@BoHao Chen). Cross-source corroboration: 2 independent sources have covered this development (including shao__meng), confirming industry-wide signal strength.
- shao__meng01:02Source