HarnessEvolve：基于参考轨迹的可靠智能体自进化方法
Self-improving agents have a basic problem: when a long run fails, they often do…
Agent 自进化的核心痛点是归因难，这篇论文给出了类似软件调试的完整闭环方案，有具体数据和验证流程，值得 Agent 开发者深入研究。
Self-improving agents have a basic problem: when a long run fails, they often do not know which step actually caused it.
自我改进的智能体存在一个基本问题：当一次长流程运行失败时，它们通常不知道是哪一步导致了失败。
If an agent is going to improve itself, it needs more than failure feedback: HarnessEvolve treats agent self-improvement like software debugging: find where a failed run first went off track, fix the recurring cause, then reject any edit that breaks existing behavior.
如果智能体要改进自身，它需要的不仅仅是失败反馈：HarnessEvolve 将智能体的自我改进视为软件调试过程：找出失败运行首次偏离轨道的步骤，修复重复出现的原因，然后拒绝任何破坏现有行为的修改。
It clusters those errors into recurring patterns and can edit the whole agent harness: prompts, skills, tools, scripts, and execution logic.
它将这些错误聚类为重复出现的模式，并能编辑整个智能体框架（harness）：包括提示词、技能、工具、脚本和执行逻辑。
On CloudCoreNetwork-QA with Qwen3.6-27B, full HarnessEvolve reached 86.9% accuracy; removing reference trajectories dropped it to 57.8%.
在 CloudCoreNetwork-QA 数据集上使用 Qwen3.6-27B 模型时，完整的 HarnessEvolve 达到了 86.9% 的准确率；移除参考轨迹后，准确率降至 57.8%。
The full system also beat the strongest baseline there by 21.6 percentage points.
完整系统在该任务上也比最强的基线方法高出 21.6 个百分点。
Candidate edits then face gates for training-data leakage, prompt bloat, regressions on recent batches, and held-out validation.
候选修改随后需经过多项关卡测试，包括训练数据泄露检查、提示词膨胀检测、近期批次的回归测试以及保留验证集评估。
– arxiv. org/abs/2609.00829
– arxiv.org/abs/2609.00829
Title: "HarnessEvolve: Learning from Reference Trajectories for Reliable Agent Self-Evolution"
标题：《HarnessEvolve：基于参考轨迹学习以实现可靠的智能体自我进化》
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力