<?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>Learning on On My Way</title>
    <link>https://tuanzi-wow.cn/categories/learning/</link>
    <description>Recent content in Learning on On My Way</description>
    <generator>Hugo</generator>
    <language>zh-CN</language>
    <atom:link href="https://tuanzi-wow.cn/categories/learning/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>学习顺序</title>
      <link>https://tuanzi-wow.cn/learning-order/</link>
      <pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate>
      <guid>https://tuanzi-wow.cn/learning-order/</guid>
      <description>学习顺序相关学习笔记与实践要点。</description>
      <content:encoded><![CDATA[<p>1.Prompt Engineering</p>
<ol>
<li>CoT</li>
<li>few-shot</li>
<li>zero-shot</li>
</ol>
<p>2.Context Engineering:</p>
<ol>
<li>
<p>写入策略</p>
</li>
<li>
<p>检索:主要是RAG相关
建立索引:</p>
<ol>
<li>文档解析(转化为纯文本)</li>
<li>chunking</li>
<li>Embedding</li>
<li>存入向量数据库</li>
</ol>
<p>​            生成:前部分类似于上面,
多路召回
重排序
GraphRAG和LightRAG</p>
</li>
<li>
<p>压缩策略
借鉴Claude Code的五层压缩机制</p>
</li>
<li>
<p>隔离策略</p>
</li>
</ol>
<p>3.Agent</p>
<ol>
<li>Function calling/Tool calling/Tool use</li>
<li>记忆系统</li>
<li>几种工作范式:React,Plan-and-Execute,Reflection</li>
<li>MCP: Host,Client,Server</li>
<li>Skills</li>
<li>A2A</li>
</ol>
<p>4.Harness:</p>
<ol>
<li>上下文精细化</li>
<li>工具系统的可控调用</li>
<li>任务执行的全局编排</li>
<li>记忆与状态的分层管理</li>
<li>独立的评估和观测系统</li>
<li>约束校验和失败恢复机制</li>
</ol>
<p>5.Loop Engineering:</p>
<ol>
<li>自动化(Trigger)</li>
<li>worktree</li>
<li>Skill</li>
<li>Connector</li>
<li>Skill</li>
<li>Sub-agent</li>
</ol>
<p>6.框架</p>
<ol>
<li>LangChain</li>
<li>LangGraph</li>
<li>AutoGen</li>
<li>等等</li>
</ol>
]]></content:encoded>
    </item>
  </channel>
</rss>
