<?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>Re on AioGeoLab | GEO优化实验室</title><link>https://aiogeolab.com/tags/re/</link><description>Recent content in Re on AioGeoLab | GEO优化实验室</description><generator>Hugo -- 0.151.0</generator><language>zh</language><copyright>2025 TardiAI</copyright><lastBuildDate>Sun, 17 May 2026 11:00:12 +0800</lastBuildDate><atom:link href="https://aiogeolab.com/tags/re/index.xml" rel="self" type="application/rss+xml"/><item><title>Agent做完了，但它不知道自己做对了没有丨反馈环：执行链里那个被默认跳过的环节</title><link>https://aiogeolab.com/posts/tgje-121-reflection-loop-how-agents-check-their-own-work/</link><pubDate>Sun, 17 May 2026 00:00:00 +0800</pubDate><guid>https://aiogeolab.com/posts/tgje-121-reflection-loop-how-agents-check-their-own-work/</guid><description>Agent执行完任务，不代表任务做对了——它没有内置的&amp;#34;回头看&amp;#34;机制，不会自己发现错误。本文拆解反馈环（Reflection）的底层逻辑，呈现从自检型、双Agent型到迭代循环型的三种设计形态，并以OpenClaw与Hermes的真实实现为例，说明反馈环不是固定的解，而是需要根据目标做出的设计选择。</description></item></channel></rss>