<?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>Engineering on 智识脉搏</title>
    <link>https://cuigh.com/zh/tags/engineering/</link>
    <description>Recent content in Engineering on 智识脉搏</description>
    <generator>Hugo</generator>
    <language>zh-CN</language>
    <lastBuildDate>Mon, 13 Jul 2026 09:51:14 +0800</lastBuildDate>
    <atom:link href="https://cuigh.com/zh/tags/engineering/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>把生产 AI 代理从 Claude Opus 4.8 切到 GPT-5.6 Sol：表面账之外的三个工程坑</title>
      <link>https://cuigh.com/zh/posts/gpt-5-6-sol-production-migration-2026/</link>
      <pubDate>Mon, 13 Jul 2026 09:51:14 +0800</pubDate>
      <guid>https://cuigh.com/zh/posts/gpt-5-6-sol-production-migration-2026/</guid>
      <description>Ploy 把核心生产代理从 Claude Opus 4.8 迁到 GPT-5.6 Sol，公开账是快 2.2 倍、便宜 27%、视觉分 0.970 vs 0.936。本文不讲这个，讲藏在账背后的三个真工程坑：工具参数越权（25 个字段全填默认值，52% 文件读空）、prompt 缓存断崖（partial-prefix 被砍）、reasoning replay 用 server-side item reference 翻车。任何想把生产代理迁到 GPT-5.6 的 builder，都必须先按这三轴重写堆栈。</description>
    </item>
  </channel>
</rss>
