<?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>Agent on Kada's Notes</title><link>https://kadaliao.github.io/tags/agent/</link><description>Recent content in Agent on Kada's Notes</description><generator>Hugo</generator><language>zh-CN</language><lastBuildDate>Wed, 01 Apr 2026 00:00:00 +0000</lastBuildDate><atom:link href="https://kadaliao.github.io/tags/agent/index.xml" rel="self" type="application/rss+xml"/><item><title>Claude Code 是如何狙杀中国用户的</title><link>https://kadaliao.github.io/posts/claude-code-tracking-and-cc-gateway/</link><pubDate>Wed, 01 Apr 2026 00:00:00 +0000</pubDate><guid>https://kadaliao.github.io/posts/claude-code-tracking-and-cc-gateway/</guid><description>研究 Claude Code 源码，看看 Claude Code 到底怎么精准定位中国用户然后封号的</description></item><item><title>Human 真的需要 in the loop 吗？</title><link>https://kadaliao.github.io/posts/human-in-loop/</link><pubDate>Mon, 30 Mar 2026 00:00:00 +0000</pubDate><guid>https://kadaliao.github.io/posts/human-in-loop/</guid><description>不是要不要人，而是人留在哪一层。AI 时代真正的问题是：当 AI 接管越来越多执行动作时，人还能不能占据那些不可轻易外包的 loop。</description></item><item><title>Harness Engineering 是什么，如何落地</title><link>https://kadaliao.github.io/posts/harness-engineering/</link><pubDate>Fri, 27 Mar 2026 00:00:00 +0000</pubDate><guid>https://kadaliao.github.io/posts/harness-engineering/</guid><description>Harness Engineering 不是新工具，而是一套工程思路：当代码主要由 Agent 生成，工程师的工作重心从写代码转向���计让 Agent 能有效工作的环境。</description></item><item><title>微信官方小龙虾插件协议拆解</title><link>https://kadaliao.github.io/posts/wechat-clawbot-protocol/</link><pubDate>Sun, 22 Mar 2026 00:00:00 +0000</pubDate><guid>https://kadaliao.github.io/posts/wechat-clawbot-protocol/</guid><description>微信发布官方龙虾插件，支持 OpenClaw 接入个人微信。本文深入源代码，拆解其完整协议实现：登录链路、消息收发、媒体协议、输入态。</description></item><item><title>小龙虾为何变蠢、失忆？深入理解 OpenClaw 记忆系统</title><link>https://kadaliao.github.io/posts/openclaw-memory-system/</link><pubDate>Wed, 18 Mar 2026 00:00:00 +0000</pubDate><guid>https://kadaliao.github.io/posts/openclaw-memory-system/</guid><description>OpenClaw 的记忆系统并不神秘，它非常工程化。理解三层记忆机制、向量检索原理和常见坑，才能把 Agent 调到长期稳定的状态。</description></item><item><title>MCP 协议入门：给 AI Agent 装上标准化工具接口</title><link>https://kadaliao.github.io/posts/mcp-protocol-intro/</link><pubDate>Tue, 15 Apr 2025 00:00:00 +0000</pubDate><guid>https://kadaliao.github.io/posts/mcp-protocol-intro/</guid><description>Anthropic 推出的 MCP（Model Context Protocol）协议正在成为 AI Agent 工具集成的新标准。这篇文章介绍它解决了什么问题以及如何上手。</description></item><item><title>LangGraph 实战：构建一个可控的 ReAct Agent</title><link>https://kadaliao.github.io/posts/langgraph-react-agent/</link><pubDate>Mon, 18 Nov 2024 00:00:00 +0000</pubDate><guid>https://kadaliao.github.io/posts/langgraph-react-agent/</guid><description>ReAct 是目前最主流的 Agent 范式。这篇文章用 LangGraph 从零实现一个生产可用的 ReAct Agent，重点讲如何做流程控制和错误处理。</description></item><item><title>Agent 框架横评：LangGraph vs AutoGen vs CrewAI</title><link>https://kadaliao.github.io/posts/agent-framework-comparison/</link><pubDate>Tue, 10 Sep 2024 00:00:00 +0000</pubDate><guid>https://kadaliao.github.io/posts/agent-framework-comparison/</guid><description>2024 年 Agent 框架百花齐放，LangGraph、AutoGen、CrewAI 各有侧重。这篇文章从工程角度做一个横评。</description></item><item><title>Function Calling 实战：让 LLM 学会调用工具</title><link>https://kadaliao.github.io/posts/function-calling-in-practice/</link><pubDate>Tue, 20 Feb 2024 00:00:00 +0000</pubDate><guid>https://kadaliao.github.io/posts/function-calling-in-practice/</guid><description>Function Calling 是构建 AI Agent 的基础能力。这篇文章通过实例讲清楚它的工作原理和工程实现。</description></item></channel></rss>