<?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>Reliability on Hex4C59's Blog</title><link>http://hex4c59.cc/tags/reliability/</link><description>Recent content in Reliability on Hex4C59's Blog</description><generator>Hugo -- 0.150.0</generator><language>zh-cn</language><lastBuildDate>Mon, 23 Mar 2026 13:45:25 +0800</lastBuildDate><atom:link href="http://hex4c59.cc/tags/reliability/index.xml" rel="self" type="application/rss+xml"/><item><title>Reflection：Agent 如何审视自己的输出并纠正错误</title><link>http://hex4c59.cc/agent/reflection-agent/</link><pubDate>Mon, 23 Mar 2026 14:00:00 +0800</pubDate><guid>http://hex4c59.cc/agent/reflection-agent/</guid><description>解释 Reflection 为什么是 Agent 的质量保证层，以及如何用输出后审查、执行中自我纠错和多 Agent 交叉审查发现并修正隐性错误。</description></item><item><title>如何评测一个 Agent：不只是看它回答得好不好</title><link>http://hex4c59.cc/agent/agent-evaluation/</link><pubDate>Mon, 23 Mar 2026 10:00:00 +0800</pubDate><guid>http://hex4c59.cc/agent/agent-evaluation/</guid><description>解释 Agent 评测为什么不能只看最终答案，并给出覆盖任务完成率、过程合理性、错误恢复与稳定性的评测框架。</description></item></channel></rss>