<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>Skill优化 on MessageDaily</title><link>https://inkeast.github.io/MessageDaily/tags/skill%E4%BC%98%E5%8C%96/</link><description>Recent content in Skill优化 on MessageDaily</description><generator>Hugo</generator><language>zh-cn</language><lastBuildDate>Sat, 31 Oct 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://inkeast.github.io/MessageDaily/tags/skill%E4%BC%98%E5%8C%96/index.xml" rel="self" type="application/rss+xml"/><item><title>把不确定性编译出去，还是把污染拦在门外：SkillCompiler 与 VEvo-Harness 双精读</title><link>https://inkeast.github.io/MessageDaily/posts/2026-10-31-harness-self-evolution-duo/</link><pubDate>Sat, 31 Oct 2026 00:00:00 +0000</pubDate><guid>https://inkeast.github.io/MessageDaily/posts/2026-10-31-harness-self-evolution-duo/</guid><description>自演化 agent 的核心矛盾是演化收益与库污染、执行不确定性之间的权衡。本文合并精读两篇 2026 年 10 月新论文：SkillCompiler 用操作级选择性编译把良定义操作变成确定性程序，LiveMath EIR@3 从 32% 降到 11%；VEvo-Harness 用反事实双臂重放做因果门控，有害入库率从 33-40% 压到约 2%。两条路线分别代表编译消不确定性与因果守护防污染，本文逐一拆解其问题定义、解法、实验证据与适用边界。</description></item></channel></rss>