<?xml version="1.0" encoding="UTF-8"?>
<rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:media="http://search.yahoo.com/mrss/" version="2.0"><channel><title>命令行小屋</title><link>https://blog.kennethcheng.cc/</link><atom:link href="https://blog.kennethcheng.cc/feed.xml" rel="self" type="application/rss+xml"/><description>命令行小屋</description><generator>Halo v2.25.4</generator><language>zh-cn</language><image><url>https://blog.kennethcheng.cc/upload/DOGE.png</url><title>命令行小屋</title><link>https://blog.kennethcheng.cc/</link></image><lastBuildDate>Tue, 4 Aug 2026 19:12:16 GMT</lastBuildDate><item><title><![CDATA[零代码造 Agent 的时代来了：LangSmith Fleet 完全上手指南]]></title><link>https://blog.kennethcheng.cc/archives/ling-dai-ma-zao-agent-de-shi-dai-lai-liao-langsmith-fleet-wan-quan-shang-shou-zhi-nan</link><description><![CDATA[<img src="https://blog.kennethcheng.cc/plugins/feed/assets/telemetry.gif?title=%E9%9B%B6%E4%BB%A3%E7%A0%81%E9%80%A0%20Agent%20%E7%9A%84%E6%97%B6%E4%BB%A3%E6%9D%A5%E4%BA%86%EF%BC%9ALangSmith%20Fleet%20%E5%AE%8C%E5%85%A8%E4%B8%8A%E6%89%8B%E6%8C%87%E5%8D%97&amp;url=/archives/ling-dai-ma-zao-agent-de-shi-dai-lai-liao-langsmith-fleet-wan-quan-shang-shou-zhi-nan" width="1" height="1" alt="" style="opacity:0;">让业务人员 5 分钟用自然语言打造会自我进化的生产级 AI 智能体 本文配套所有架构图均为 Mermaid 源码，可直接复制到任何支持 Mermaid 的 Markdown 编辑器中渲染。 📑 目录 为什么需要 LangSmith Fleet？ Fleet 到底是什么？ Fleet 的三大核心特性]]></description><guid isPermaLink="false">/archives/ling-dai-ma-zao-agent-de-shi-dai-lai-liao-langsmith-fleet-wan-quan-shang-shou-zhi-nan</guid><dc:creator>KennethCheng</dc:creator><enclosure url="https://blog.kennethcheng.cc/apis/api.storage.halo.run/v1alpha1/thumbnails/-/via-uri?uri=%2Fupload%2FChatGPT%2520Image%25202026%25E5%25B9%25B48%25E6%259C%25884%25E6%2597%25A5%252001_29_17.png&amp;size=m" type="image/jpeg" length="0"/><category>LangChain </category><category>Ai</category><pubDate>Mon, 3 Aug 2026 17:59:02 GMT</pubDate></item><item><title><![CDATA[别让你的 AI Agent 裸奔：Sandbox 沙箱隔离从入门到选型]]></title><link>https://blog.kennethcheng.cc/archives/bie-rang-ni-de-ai-agent-luo-ben-sandbox-sha-xiang-ge-chi-cong-ru-men-dao-xuan-xing</link><description><![CDATA[<img src="https://blog.kennethcheng.cc/plugins/feed/assets/telemetry.gif?title=%E5%88%AB%E8%AE%A9%E4%BD%A0%E7%9A%84%20AI%20Agent%20%E8%A3%B8%E5%A5%94%EF%BC%9ASandbox%20%E6%B2%99%E7%AE%B1%E9%9A%94%E7%A6%BB%E4%BB%8E%E5%85%A5%E9%97%A8%E5%88%B0%E9%80%89%E5%9E%8B&amp;url=/archives/bie-rang-ni-de-ai-agent-luo-ben-sandbox-sha-xiang-ge-chi-cong-ru-men-dao-xuan-xing" width="1" height="1" alt="" style="opacity:0;">你的 AI Agent 会写代码、跑命令、操作文件 —— 但你真的放心让它直接跑在你的电脑上吗？ 如果 Agent 误读了某个 prompt 来一句 rm -rf /，或者 curl 到了不安全网络…… 沙箱（Sandbox），就是给 Agent 套的"笼子"。 本文将讲清两种主流方案： 模式一：A]]></description><guid isPermaLink="false">/archives/bie-rang-ni-de-ai-agent-luo-ben-sandbox-sha-xiang-ge-chi-cong-ru-men-dao-xuan-xing</guid><dc:creator>KennethCheng</dc:creator><enclosure url="https://blog.kennethcheng.cc/apis/api.storage.halo.run/v1alpha1/thumbnails/-/via-uri?uri=%2Fupload%2FChatGPT%2520Image%25202026%25E5%25B9%25B48%25E6%259C%25883%25E6%2597%25A5%252023_40_55.png&amp;size=m" type="image/jpeg" length="0"/><category>LangChain </category><category>Ai</category><pubDate>Mon, 3 Aug 2026 15:41:15 GMT</pubDate></item><item><title><![CDATA[用 Remotion Skills + DeepAgents，让 AI 帮你"写"视频代码]]></title><link>https://blog.kennethcheng.cc/archives/yong-remotion-skills-deepagents-rang-ai-bang-ni-xie-shi-pin-dai-ma</link><description><![CDATA[<img src="https://blog.kennethcheng.cc/plugins/feed/assets/telemetry.gif?title=%E7%94%A8%20Remotion%20Skills%20%2B%20DeepAgents%EF%BC%8C%E8%AE%A9%20AI%20%E5%B8%AE%E4%BD%A0%22%E5%86%99%22%E8%A7%86%E9%A2%91%E4%BB%A3%E7%A0%81&amp;url=/archives/yong-remotion-skills-deepagents-rang-ai-bang-ni-xie-shi-pin-dai-ma" width="1" height="1" alt="" style="opacity:0;">一行自然语言描述 → 一个完整的 React 视频项目。这是程序化视频生成的新范式。 一、背景：为什么需要 Remotion Skills？ 传统视频制作流程是这样的： #bytemd-mermaid-1785761231546-190{font-family:"trebuchet ms",verd]]></description><guid isPermaLink="false">/archives/yong-remotion-skills-deepagents-rang-ai-bang-ni-xie-shi-pin-dai-ma</guid><dc:creator>KennethCheng</dc:creator><enclosure url="https://blog.kennethcheng.cc/apis/api.storage.halo.run/v1alpha1/thumbnails/-/via-uri?uri=%2Fupload%2FChatGPT%2520Image%25202026%25E5%25B9%25B48%25E6%259C%25883%25E6%2597%25A5%252019_57_07.png&amp;size=m" type="image/jpeg" length="775340"/><category>LangChain </category><category>Ai</category><pubDate>Mon, 3 Aug 2026 12:47:34 GMT</pubDate></item><item><title><![CDATA[告别失忆 Agent：LangChain Memory 双轨制完全指南]]></title><link>https://blog.kennethcheng.cc/archives/gao-bie-shi-yi-agent-langchain-memory-shuang-gui-zhi-wan-quan-zhi-nan</link><description><![CDATA[<img src="https://blog.kennethcheng.cc/plugins/feed/assets/telemetry.gif?title=%E5%91%8A%E5%88%AB%E5%A4%B1%E5%BF%86%20Agent%EF%BC%9ALangChain%20Memory%20%E5%8F%8C%E8%BD%A8%E5%88%B6%E5%AE%8C%E5%85%A8%E6%8C%87%E5%8D%97&amp;url=/archives/gao-bie-shi-yi-agent-langchain-memory-shuang-gui-zhi-wan-quan-zhi-nan" width="1" height="1" alt="" style="opacity:0;">从 Checkpoint 到 Store，让 Agent 真正"记住"你 🎯 写在最前 想象这样一个场景： 用户：「你好，我叫李雷，是个男生。」 Agent：「你好李雷！」 （第二天） 用户：「我是谁？」 Agent：「抱歉，我不知道。」 这就是典型的失忆 Agent——它只能看到当前对话窗口里的]]></description><guid isPermaLink="false">/archives/gao-bie-shi-yi-agent-langchain-memory-shuang-gui-zhi-wan-quan-zhi-nan</guid><dc:creator>KennethCheng</dc:creator><enclosure url="https://blog.kennethcheng.cc/apis/api.storage.halo.run/v1alpha1/thumbnails/-/via-uri?uri=%2Fupload%2FChatGPT%2520Image%25202026%25E5%25B9%25B48%25E6%259C%25883%25E6%2597%25A5%252011_37_44.png&amp;size=m" type="image/jpeg" length="0"/><category>LangChain </category><category>Ai</category><pubDate>Mon, 3 Aug 2026 03:37:56 GMT</pubDate></item><item><title><![CDATA[上下文工程完全指南：用「写、选、压、隔」四把手术刀，根治 LLM 的"上下文崩溃"]]></title><link>https://blog.kennethcheng.cc/archives/shang-xia-wen-gong-cheng-wan-quan-zhi-nan-yong-xie-xuan-ya-ge-si-ba-shou-shu-dao-gen-zhi-llm-de-shang-xia-wen-beng-kui</link><description><![CDATA[<img src="https://blog.kennethcheng.cc/plugins/feed/assets/telemetry.gif?title=%E4%B8%8A%E4%B8%8B%E6%96%87%E5%B7%A5%E7%A8%8B%E5%AE%8C%E5%85%A8%E6%8C%87%E5%8D%97%EF%BC%9A%E7%94%A8%E3%80%8C%E5%86%99%E3%80%81%E9%80%89%E3%80%81%E5%8E%8B%E3%80%81%E9%9A%94%E3%80%8D%E5%9B%9B%E6%8A%8A%E6%89%8B%E6%9C%AF%E5%88%80%EF%BC%8C%E6%A0%B9%E6%B2%BB%20LLM%20%E7%9A%84%22%E4%B8%8A%E4%B8%8B%E6%96%87%E5%B4%A9%E6%BA%83%22&amp;url=/archives/shang-xia-wen-gong-cheng-wan-quan-zhi-nan-yong-xie-xuan-ya-ge-si-ba-shou-shu-dao-gen-zhi-llm-de-shang-xia-wen-beng-kui" width="1" height="1" alt="" style="opacity:0;">为什么你的 Agent 聊着聊着就"发疯"？为什么它会忘记几分钟前说过的话，又为什么它会"言之凿凿"地胡说八道？ 答案藏在 上下文 里。 构建一个能跑通 demo 的 Agent 容易，构建一个能稳定生产的 Agent 难。难就难在：上下文窗口是有限的，而世界是无限的。一旦你往里塞的内容失控，模型就]]></description><guid isPermaLink="false">/archives/shang-xia-wen-gong-cheng-wan-quan-zhi-nan-yong-xie-xuan-ya-ge-si-ba-shou-shu-dao-gen-zhi-llm-de-shang-xia-wen-beng-kui</guid><dc:creator>KennethCheng</dc:creator><enclosure url="https://blog.kennethcheng.cc/apis/api.storage.halo.run/v1alpha1/thumbnails/-/via-uri?uri=%2Fupload%2FChatGPT%2520Image%25202026%25E5%25B9%25B48%25E6%259C%25883%25E6%2597%25A5%252011_05_52.png&amp;size=m" type="image/jpeg" length="0"/><category>LangChain </category><category>Ai</category><pubDate>Mon, 3 Aug 2026 03:06:36 GMT</pubDate></item><item><title><![CDATA[告别上下文爆炸：LangChain Agent Skills 完全实战指南]]></title><link>https://blog.kennethcheng.cc/archives/gao-bie-shang-xia-wen-bao-zha-langchain-agent-skills-wan-quan-shi-zhan-zhi-nan</link><description><![CDATA[<img src="https://blog.kennethcheng.cc/plugins/feed/assets/telemetry.gif?title=%E5%91%8A%E5%88%AB%E4%B8%8A%E4%B8%8B%E6%96%87%E7%88%86%E7%82%B8%EF%BC%9ALangChain%20Agent%20Skills%20%E5%AE%8C%E5%85%A8%E5%AE%9E%E6%88%98%E6%8C%87%E5%8D%97&amp;url=/archives/gao-bie-shang-xia-wen-bao-zha-langchain-agent-skills-wan-quan-shi-zhan-zhi-nan" width="1" height="1" alt="" style="opacity:0;">当你的 AI Agent 需要接入十几个工具，每个工具的"使用说明"都塞进 prompt 里时，模型会发出这样的哀嚎："我上下文已经爆炸了，怎么干活啊？？" 这篇文章会带你深入理解 Agent Skills 这套由 Anthropic 提出的开放标准，并通过 LangChain DeepAgent]]></description><guid isPermaLink="false">/archives/gao-bie-shang-xia-wen-bao-zha-langchain-agent-skills-wan-quan-shi-zhan-zhi-nan</guid><dc:creator>KennethCheng</dc:creator><enclosure url="https://blog.kennethcheng.cc/apis/api.storage.halo.run/v1alpha1/thumbnails/-/via-uri?uri=%2Fupload%2FGemini_Generated_Image_dewmazdewmazdewm.png&amp;size=m" type="image/jpeg" length="0"/><category>LangChain </category><category>Ai</category><pubDate>Mon, 3 Aug 2026 02:30:05 GMT</pubDate></item><item><title><![CDATA[一文搞懂 LangSmith：从可观测到 Studio 调试，让 LLM Agent 告别黑盒]]></title><link>https://blog.kennethcheng.cc/archives/yi-wen-gao-dong-langsmith-cong-ke-guan-ce-dao-studio-diao-shi-rang-llm-agent-gao-bie-hei-he</link><description><![CDATA[<img src="https://blog.kennethcheng.cc/plugins/feed/assets/telemetry.gif?title=%E4%B8%80%E6%96%87%E6%90%9E%E6%87%82%20LangSmith%EF%BC%9A%E4%BB%8E%E5%8F%AF%E8%A7%82%E6%B5%8B%E5%88%B0%20Studio%20%E8%B0%83%E8%AF%95%EF%BC%8C%E8%AE%A9%20LLM%20Agent%20%E5%91%8A%E5%88%AB%E9%BB%91%E7%9B%92&amp;url=/archives/yi-wen-gao-dong-langsmith-cong-ke-guan-ce-dao-studio-diao-shi-rang-llm-agent-gao-bie-hei-he" width="1" height="1" alt="" style="opacity:0;">作为 LangChain 团队三大核心产品之一，LangSmith 提供观察、调试和部署 LLM 应用的统一平台。本文带你从入门到 Studio 实战，彻底搞懂 LangSmith 的使用方式。 一、初识 LangSmith 当我们成功构建一个 Agent 后，常常会被这些问题困扰： 这个请求到底经]]></description><guid isPermaLink="false">/archives/yi-wen-gao-dong-langsmith-cong-ke-guan-ce-dao-studio-diao-shi-rang-llm-agent-gao-bie-hei-he</guid><dc:creator>KennethCheng</dc:creator><enclosure url="https://blog.kennethcheng.cc/apis/api.storage.halo.run/v1alpha1/thumbnails/-/via-uri?uri=%2Fupload%2FChatGPT%2520Image%25202026%25E5%25B9%25B48%25E6%259C%25883%25E6%2597%25A5%252000_35_17.png&amp;size=m" type="image/jpeg" length="0"/><category>LangChain </category><category>Ai</category><pubDate>Sun, 2 Aug 2026 16:36:03 GMT</pubDate></item><item><title><![CDATA[LangChain V1.1 & V1.2 双版本深度解读：看 Profile 抽象如何重塑中间件生态]]></title><link>https://blog.kennethcheng.cc/archives/langchain-v1.1-v1.2-shuang-ban-ben-shen-du-jie-du-kan-profile-chou-xiang-ru-he-chong-su-zhong-jian-jian-sheng-tai</link><description><![CDATA[<img src="https://blog.kennethcheng.cc/plugins/feed/assets/telemetry.gif?title=LangChain%20V1.1%20%26%20V1.2%20%E5%8F%8C%E7%89%88%E6%9C%AC%E6%B7%B1%E5%BA%A6%E8%A7%A3%E8%AF%BB%EF%BC%9A%E7%9C%8B%20Profile%20%E6%8A%BD%E8%B1%A1%E5%A6%82%E4%BD%95%E9%87%8D%E5%A1%91%E4%B8%AD%E9%97%B4%E4%BB%B6%E7%94%9F%E6%80%81&amp;url=/archives/langchain-v1.1-v1.2-shuang-ban-ben-shen-du-jie-du-kan-profile-chou-xiang-ru-he-chong-su-zhong-jian-jian-sheng-tai" width="1" height="1" alt="" style="opacity:0;">📅 2025 年 11 月 25 日 V1.1 发布，2025 年 12 月 15 日 V1.2 紧随其后。两个月两次发版，LangChain 团队究竟在下怎样一盘棋？ LangChain 在 V1.0 正式版发布后的短短两个月内，密集推出了 V1.1 和 V1.2 两个版本。与其说是"功能更新"]]></description><guid isPermaLink="false">/archives/langchain-v1.1-v1.2-shuang-ban-ben-shen-du-jie-du-kan-profile-chou-xiang-ru-he-chong-su-zhong-jian-jian-sheng-tai</guid><dc:creator>KennethCheng</dc:creator><enclosure url="https://blog.kennethcheng.cc/apis/api.storage.halo.run/v1alpha1/thumbnails/-/via-uri?uri=%2Fupload%2FChatGPT%2520Image%25202026%25E5%25B9%25B48%25E6%259C%25882%25E6%2597%25A5%252023_40_18.png&amp;size=m" type="image/jpeg" length="730127"/><category>LangChain </category><category>Ai</category><pubDate>Sun, 2 Aug 2026 15:38:58 GMT</pubDate></item><item><title><![CDATA[从上下文隔离到任务协作：SubAgent 的概念、创建与应用实践]]></title><link>https://blog.kennethcheng.cc/archives/cong-shang-xia-wen-ge-chi-dao-ren-wu-xie-zuo-subagent-de-gai-nian-chuang-jian-yu-ying-yong-shi-jian</link><description><![CDATA[<img src="https://blog.kennethcheng.cc/plugins/feed/assets/telemetry.gif?title=%E4%BB%8E%E4%B8%8A%E4%B8%8B%E6%96%87%E9%9A%94%E7%A6%BB%E5%88%B0%E4%BB%BB%E5%8A%A1%E5%8D%8F%E4%BD%9C%EF%BC%9ASubAgent%20%E7%9A%84%E6%A6%82%E5%BF%B5%E3%80%81%E5%88%9B%E5%BB%BA%E4%B8%8E%E5%BA%94%E7%94%A8%E5%AE%9E%E8%B7%B5&amp;url=/archives/cong-shang-xia-wen-ge-chi-dao-ren-wu-xie-zuo-subagent-de-gai-nian-chuang-jian-yu-ying-yong-shi-jian" width="1" height="1" alt="" style="opacity:0;">从上下文隔离到任务协作：SubAgent 的概念、创建与应用实践 在构建复杂 Agent 应用时，主代理往往需要同时处理任务规划、工具调用、信息检索、数据分析和结果生成等工作。 如果所有任务都由一个 Agent 直接完成，容易出现以下问题： 上下文越来越长，导致上下文膨胀 大量工具调用记录干扰主任务]]></description><guid isPermaLink="false">/archives/cong-shang-xia-wen-ge-chi-dao-ren-wu-xie-zuo-subagent-de-gai-nian-chuang-jian-yu-ying-yong-shi-jian</guid><dc:creator>KennethCheng</dc:creator><enclosure url="https://blog.kennethcheng.cc/apis/api.storage.halo.run/v1alpha1/thumbnails/-/via-uri?uri=%2Fupload%2FChatGPT%2520Image%25202026%25E5%25B9%25B48%25E6%259C%25882%25E6%2597%25A5%252023_37_06.png&amp;size=m" type="image/jpeg" length="742845"/><category>LangChain </category><category>Ai</category><pubDate>Sun, 2 Aug 2026 14:43:20 GMT</pubDate></item><item><title><![CDATA[从浅到深：Deep Agents 核心能力与架构全解析]]></title><link>https://blog.kennethcheng.cc/archives/cong-qian-dao-shen-deep-agents-he-xin-neng-li-yu-jia-gou-quan-jie-xi</link><description><![CDATA[<img src="https://blog.kennethcheng.cc/plugins/feed/assets/telemetry.gif?title=%E4%BB%8E%E6%B5%85%E5%88%B0%E6%B7%B1%EF%BC%9ADeep%20Agents%20%E6%A0%B8%E5%BF%83%E8%83%BD%E5%8A%9B%E4%B8%8E%E6%9E%B6%E6%9E%84%E5%85%A8%E8%A7%A3%E6%9E%90&amp;url=/archives/cong-qian-dao-shen-deep-agents-he-xin-neng-li-yu-jia-gou-quan-jie-xi" width="1" height="1" alt="" style="opacity:0;">从浅到深：Deep Agents 核心能力与架构全解析 一文掌握 LLM 智能体从"简单工具循环"到"复杂任务规划"的进化之路 🎯 引子：为什么简单 Agent 不够用？ 在 LLM 应用开发中，最常见的 Agent 模式就是： LLM + 工具循环调用 — 用户给指令，模型判断要不要调工具，调完]]></description><guid isPermaLink="false">/archives/cong-qian-dao-shen-deep-agents-he-xin-neng-li-yu-jia-gou-quan-jie-xi</guid><dc:creator>KennethCheng</dc:creator><enclosure url="https://blog.kennethcheng.cc/apis/api.storage.halo.run/v1alpha1/thumbnails/-/via-uri?uri=%2Fupload%2FGemini_Generated_Image_hdc108hdc108hdc1.png&amp;size=m" type="image/jpeg" length="0"/><category>LangChain </category><category>Ai</category><pubDate>Sun, 2 Aug 2026 12:19:41 GMT</pubDate></item><item><title><![CDATA[File System 中间件完全指南：四种后端让你的 Agent 学会“读写文件”]]></title><link>https://blog.kennethcheng.cc/archives/file-system-zhong-jian-jian-wan-quan-zhi-nan-si-zhong-hou-duan-rang-ni-de-agent-xue-hui-du-xie-wen-jian</link><description><![CDATA[<img src="https://blog.kennethcheng.cc/plugins/feed/assets/telemetry.gif?title=File%20System%20%E4%B8%AD%E9%97%B4%E4%BB%B6%E5%AE%8C%E5%85%A8%E6%8C%87%E5%8D%97%EF%BC%9A%E5%9B%9B%E7%A7%8D%E5%90%8E%E7%AB%AF%E8%AE%A9%E4%BD%A0%E7%9A%84%20Agent%20%E5%AD%A6%E4%BC%9A%E2%80%9C%E8%AF%BB%E5%86%99%E6%96%87%E4%BB%B6%E2%80%9D&amp;url=/archives/file-system-zhong-jian-jian-wan-quan-zhi-nan-si-zhong-hou-duan-rang-ni-de-agent-xue-hui-du-xie-wen-jian" width="1" height="1" alt="" style="opacity:0;">在构建复杂 Agent 时，上下文（Context）管理 是最关键的问题之一。当工具调用结果非常庞大时——例如网络搜索或 RAG 返回的大量信息——上下文窗口会被迅速填满，导致 Agent 无法持续运行。 File System 中间件 正是为了解决这一问题而设计的：它把文件系统作为 Agent 的]]></description><guid isPermaLink="false">/archives/file-system-zhong-jian-jian-wan-quan-zhi-nan-si-zhong-hou-duan-rang-ni-de-agent-xue-hui-du-xie-wen-jian</guid><dc:creator>KennethCheng</dc:creator><enclosure url="https://blog.kennethcheng.cc/apis/api.storage.halo.run/v1alpha1/thumbnails/-/via-uri?uri=%2Fupload%2FGemini_Generated_Image_iyvd4miyvd4miyvd.png&amp;size=m" type="image/jpeg" length="0"/><category>LangChain </category><category>Ai</category><pubDate>Sun, 2 Aug 2026 04:21:21 GMT</pubDate></item><item><title><![CDATA[Agent Engineering 一文通：4 个核心认知 + 10 条实战洞察]]></title><link>https://blog.kennethcheng.cc/archives/agent-engineering-yi-wen-tong-4-ge-he-xin-ren-zhi-10-tiao-shi-zhan-dong-cha</link><description><![CDATA[<img src="https://blog.kennethcheng.cc/plugins/feed/assets/telemetry.gif?title=Agent%20Engineering%20%E4%B8%80%E6%96%87%E9%80%9A%EF%BC%9A4%20%E4%B8%AA%E6%A0%B8%E5%BF%83%E8%AE%A4%E7%9F%A5%20%2B%2010%20%E6%9D%A1%E5%AE%9E%E6%88%98%E6%B4%9E%E5%AF%9F&amp;url=/archives/agent-engineering-yi-wen-tong-4-ge-he-xin-ren-zhi-10-tiao-shi-zhan-dong-cha" width="1" height="1" alt="" style="opacity:0;">当大模型开始"思考"，Demo 和 Production 之间隔着一道叫做 Agent Engineering 的鸿沟。 前言 你或许已经听过无数次"AI Agent"这个词。但一个尴尬的现实是：90% 的 Agent Demo 永远停在 demo。 它们能在本地惊艳地跑通，可在真实用户面前，要么"]]></description><guid isPermaLink="false">/archives/agent-engineering-yi-wen-tong-4-ge-he-xin-ren-zhi-10-tiao-shi-zhan-dong-cha</guid><dc:creator>KennethCheng</dc:creator><enclosure url="https://blog.kennethcheng.cc/apis/api.storage.halo.run/v1alpha1/thumbnails/-/via-uri?uri=%2Fupload%2FGemini_Generated_Image_cds408cds408cds4.png&amp;size=m" type="image/jpeg" length="0"/><category>LangChain </category><category>Ai</category><pubDate>Sun, 2 Aug 2026 03:56:12 GMT</pubDate></item><item><title><![CDATA[Agent 中间件双雄实战：Tool Selector 与 To-do List 完全指南]]></title><link>https://blog.kennethcheng.cc/archives/agent-zhong-jian-jian-shuang-xiong-shi-zhan-tool-selector-yu-to-do-list-wan-quan-zhi-nan</link><description><![CDATA[<img src="https://blog.kennethcheng.cc/plugins/feed/assets/telemetry.gif?title=Agent%20%E4%B8%AD%E9%97%B4%E4%BB%B6%E5%8F%8C%E9%9B%84%E5%AE%9E%E6%88%98%EF%BC%9ATool%20Selector%20%E4%B8%8E%20To-do%20List%20%E5%AE%8C%E5%85%A8%E6%8C%87%E5%8D%97&amp;url=/archives/agent-zhong-jian-jian-shuang-xiong-shi-zhan-tool-selector-yu-to-do-list-wan-quan-zhi-nan" width="1" height="1" alt="" style="opacity:0;">让 LLM 工具调用既精准又清晰——两个中间件，搞定 Agent 的两大痛点。 📖 写在前面 随着 LangChain Agent 越来越普及，开发者会很快撞上两堵"墙"： 工具太多：给 Agent 挂了 20 个工具，模型在调用时一脸懵。 任务太长：跨多步的复杂任务跑下来，用户只看到一行最终回复]]></description><guid isPermaLink="false">/archives/agent-zhong-jian-jian-shuang-xiong-shi-zhan-tool-selector-yu-to-do-list-wan-quan-zhi-nan</guid><dc:creator>KennethCheng</dc:creator><enclosure url="https://blog.kennethcheng.cc/apis/api.storage.halo.run/v1alpha1/thumbnails/-/via-uri?uri=%2Fupload%2FGemini_Generated_Image_e6dlzwe6dlzwe6dl.png&amp;size=m" type="image/jpeg" length="0"/><category>LangChain </category><category>Ai</category><pubDate>Sat, 1 Aug 2026 14:11:53 GMT</pubDate></item><item><title><![CDATA[AI Agent 的「记忆压缩术」：LangChain Summarization 中间件完全指南]]></title><link>https://blog.kennethcheng.cc/archives/ai-agent-de-ji-yi-ya-suo-shu-langchain-summarization-zhong-jian-jian-wan-quan-zhi-nan</link><description><![CDATA[<img src="https://blog.kennethcheng.cc/plugins/feed/assets/telemetry.gif?title=AI%20Agent%20%E7%9A%84%E3%80%8C%E8%AE%B0%E5%BF%86%E5%8E%8B%E7%BC%A9%E6%9C%AF%E3%80%8D%EF%BC%9ALangChain%20Summarization%20%E4%B8%AD%E9%97%B4%E4%BB%B6%E5%AE%8C%E5%85%A8%E6%8C%87%E5%8D%97&amp;url=/archives/ai-agent-de-ji-yi-ya-suo-shu-langchain-summarization-zhong-jian-jian-wan-quan-zhi-nan" width="1" height="1" alt="" style="opacity:0;">当 Agent 对话越来越长，模型开始"记不住"早先聊过什么——这时候，Summarization 中间件就是你的救命稻草。 本文将带你从概念、创建、触发流程、规则制定到参数调优，彻底掌握 LangChain Summarization 中间件。 📑 目录 什么是 Summarization 中间]]></description><guid isPermaLink="false">/archives/ai-agent-de-ji-yi-ya-suo-shu-langchain-summarization-zhong-jian-jian-wan-quan-zhi-nan</guid><dc:creator>KennethCheng</dc:creator><enclosure url="https://blog.kennethcheng.cc/apis/api.storage.halo.run/v1alpha1/thumbnails/-/via-uri?uri=%2Fupload%2FGemini_Generated_Image_hiz471hiz471hiz4.png&amp;size=m" type="image/jpeg" length="0"/><category>LangChain </category><category>Ai</category><pubDate>Sat, 1 Aug 2026 12:24:12 GMT</pubDate></item><item><title><![CDATA[Human-in-the-Loop 实战指南:在 LangChain 中为 Agent 装上"决策刹车"]]></title><link>https://blog.kennethcheng.cc/archives/human-in-the-loop-shi-zhan-zhi-nan-zai-langchain-zhong-wei-agent-zhuang-shang-jue-ce-cha-che</link><description><![CDATA[<img src="https://blog.kennethcheng.cc/plugins/feed/assets/telemetry.gif?title=Human-in-the-Loop%20%E5%AE%9E%E6%88%98%E6%8C%87%E5%8D%97%3A%E5%9C%A8%20LangChain%20%E4%B8%AD%E4%B8%BA%20Agent%20%E8%A3%85%E4%B8%8A%22%E5%86%B3%E7%AD%96%E5%88%B9%E8%BD%A6%22&amp;url=/archives/human-in-the-loop-shi-zhan-zhi-nan-zai-langchain-zhong-wei-agent-zhuang-shang-jue-ce-cha-che" width="1" height="1" alt="" style="opacity:0;">随着 LLM Agent 越来越"敢动手",我们开始需要一种机制 —— 让 AI 在做关键动作前停下来,让人看一眼,再由人来决定放行、修改还是拒绝。 这就是 Human-in-the-Loop (HITL):AI 不再是"自动执行",而是"提建议 → 被审核 → 必要时修正"。 本文带你从概念到 L]]></description><guid isPermaLink="false">/archives/human-in-the-loop-shi-zhan-zhi-nan-zai-langchain-zhong-wei-agent-zhuang-shang-jue-ce-cha-che</guid><dc:creator>KennethCheng</dc:creator><enclosure url="https://blog.kennethcheng.cc/apis/api.storage.halo.run/v1alpha1/thumbnails/-/via-uri?uri=%2Fupload%2FGemini_Generated_Image_mtq3x0mtq3x0mtq3.png&amp;size=m" type="image/jpeg" length="0"/><category>LangChain </category><category>Ai</category><pubDate>Fri, 31 Jul 2026 04:38:20 GMT</pubDate></item><item><title><![CDATA[LangChain Agent 中间件完全指南：从拦截器设计到自定义钩子实战]]></title><link>https://blog.kennethcheng.cc/archives/langchain-agent-zhong-jian-jian-wan-quan-zhi-nan-cong-lan-jie-qi-she-ji-dao-zi-ding-yi-gou-zi-shi-zhan</link><description><![CDATA[<img src="https://blog.kennethcheng.cc/plugins/feed/assets/telemetry.gif?title=LangChain%20Agent%20%E4%B8%AD%E9%97%B4%E4%BB%B6%E5%AE%8C%E5%85%A8%E6%8C%87%E5%8D%97%EF%BC%9A%E4%BB%8E%E6%8B%A6%E6%88%AA%E5%99%A8%E8%AE%BE%E8%AE%A1%E5%88%B0%E8%87%AA%E5%AE%9A%E4%B9%89%E9%92%A9%E5%AD%90%E5%AE%9E%E6%88%98&amp;url=/archives/langchain-agent-zhong-jian-jian-wan-quan-zhi-nan-cong-lan-jie-qi-she-ji-dao-zi-ding-yi-gou-zi-shi-zhan" width="1" height="1" alt="" style="opacity:0;">读懂中间件机制，让你的 Agent 从"能跑"走向"可控、可观测、可生产"。 写在前面 如果你已经用 LangChain 搭过 Agent，大概率会遇到这些痛点： 想给 Agent 加上日志，却不知道在哪一层埋点 想做调用次数限流，防止 token 成本失控 想在关键工具调用前让人工把关，但不想重写]]></description><guid isPermaLink="false">/archives/langchain-agent-zhong-jian-jian-wan-quan-zhi-nan-cong-lan-jie-qi-she-ji-dao-zi-ding-yi-gou-zi-shi-zhan</guid><dc:creator>KennethCheng</dc:creator><enclosure url="https://blog.kennethcheng.cc/apis/api.storage.halo.run/v1alpha1/thumbnails/-/via-uri?uri=%2Fupload%2FGemini_Generated_Image_mtz3fmmtz3fmmtz3.png&amp;size=m" type="image/jpeg" length="0"/><category>LangChain </category><category>Ai</category><pubDate>Thu, 30 Jul 2026 10:58:40 GMT</pubDate></item><item><title><![CDATA[一文搞懂 LangChain Message：从 4 种消息类型到 Agent 状态流转]]></title><link>https://blog.kennethcheng.cc/archives/yi-wen-gao-dong-langchain-message-cong-4-zhong-xiao-xi-lei-xing-dao-agent-zhuang-tai-liu-zhuan</link><description><![CDATA[<img src="https://blog.kennethcheng.cc/plugins/feed/assets/telemetry.gif?title=%E4%B8%80%E6%96%87%E6%90%9E%E6%87%82%20LangChain%20Message%EF%BC%9A%E4%BB%8E%204%20%E7%A7%8D%E6%B6%88%E6%81%AF%E7%B1%BB%E5%9E%8B%E5%88%B0%20Agent%20%E7%8A%B6%E6%80%81%E6%B5%81%E8%BD%AC&amp;url=/archives/yi-wen-gao-dong-langchain-message-cong-4-zhong-xiao-xi-lei-xing-dao-agent-zhuang-tai-liu-zhuan" width="1" height="1" alt="" style="opacity:0;">如果你正在学习 LangChain 或基于 LangChain 搭建 Agent，那么「Message（消息）」是绕不开的第一道坎。 它既是模型的输入输出，也是对话的上下文单元，更是 Agent 状态机的"燃料"。 本文将带你系统掌握 Message 的全部要点：4 种消息类型 → 工具调用流程 →]]></description><guid isPermaLink="false">/archives/yi-wen-gao-dong-langchain-message-cong-4-zhong-xiao-xi-lei-xing-dao-agent-zhuang-tai-liu-zhuan</guid><dc:creator>KennethCheng</dc:creator><enclosure url="https://blog.kennethcheng.cc/apis/api.storage.halo.run/v1alpha1/thumbnails/-/via-uri?uri=%2Fupload%2FGemini_Generated_Image_t066kwt066kwt066.png&amp;size=m" type="image/jpeg" length="0"/><category>LangChain </category><category>Ai</category><pubDate>Thu, 30 Jul 2026 10:20:11 GMT</pubDate></item><item><title><![CDATA[一文读懂 LangChain Tools：从 @tool 装饰器到 StructuredTool，让大模型真正'动手']]></title><link>https://blog.kennethcheng.cc/archives/yi-wen-du-dong-langchain-tools-cong-tool-zhuang-shi-qi-dao-structuredtool-rang-da-mo-xing-zhen-zheng-dong-shou</link><description><![CDATA[<img src="https://blog.kennethcheng.cc/plugins/feed/assets/telemetry.gif?title=%E4%B8%80%E6%96%87%E8%AF%BB%E6%87%82%20LangChain%20Tools%EF%BC%9A%E4%BB%8E%20%40tool%20%E8%A3%85%E9%A5%B0%E5%99%A8%E5%88%B0%20StructuredTool%EF%BC%8C%E8%AE%A9%E5%A4%A7%E6%A8%A1%E5%9E%8B%E7%9C%9F%E6%AD%A3%27%E5%8A%A8%E6%89%8B%27&amp;url=/archives/yi-wen-du-dong-langchain-tools-cong-tool-zhuang-shi-qi-dao-structuredtool-rang-da-mo-xing-zhen-zheng-dong-shou" width="1" height="1" alt="" style="opacity:0;">写在前面 大语言模型（LLM）很强，但"只会说话"是它最大的短板。Tools（工具） 就是 LangChain 给出的解法——通过标准化的接口让模型能够调用外部能力，从而真正"动手"。 本文从一张概念图出发，系统讲清楚： @tool 装饰器如何把一个普通函数变成工具 模型发出 tool_calls]]></description><guid isPermaLink="false">/archives/yi-wen-du-dong-langchain-tools-cong-tool-zhuang-shi-qi-dao-structuredtool-rang-da-mo-xing-zhen-zheng-dong-shou</guid><dc:creator>KennethCheng</dc:creator><enclosure url="https://blog.kennethcheng.cc/apis/api.storage.halo.run/v1alpha1/thumbnails/-/via-uri?uri=%2Fupload%2FGemini_Generated_Image_v6kqvcv6kqvcv6kq.png&amp;size=m" type="image/jpeg" length="0"/><category>LangChain </category><category>Ai</category><pubDate>Thu, 30 Jul 2026 10:07:02 GMT</pubDate></item><item><title><![CDATA[LangChain Agent 完全指南：从 Tools 到 ReAct 循环的工程实践]]></title><link>https://blog.kennethcheng.cc/archives/langchain-agent-wan-quan-zhi-nan-cong-tools-dao-react-xun-huan-de-gong-cheng-shi-jian</link><description><![CDATA[<img src="https://blog.kennethcheng.cc/plugins/feed/assets/telemetry.gif?title=LangChain%20Agent%20%E5%AE%8C%E5%85%A8%E6%8C%87%E5%8D%97%EF%BC%9A%E4%BB%8E%20Tools%20%E5%88%B0%20ReAct%20%E5%BE%AA%E7%8E%AF%E7%9A%84%E5%B7%A5%E7%A8%8B%E5%AE%9E%E8%B7%B5&amp;url=/archives/langchain-agent-wan-quan-zhi-nan-cong-tools-dao-react-xun-huan-de-gong-cheng-shi-jian" width="1" height="1" alt="" style="opacity:0;">智能体 (Agent) = 大语言模型 + 工具 + 系统提示，让 LLM 拥有"手脚"和"行为准则"，能自主规划、调用工具、迭代求解复杂任务。 本文基于 LangChain v1.0，系统梳理 Agent 的核心概念、Tools 定义、Agent 构建、系统提示、结构化输出与多种调用方式，配套可直]]></description><guid isPermaLink="false">/archives/langchain-agent-wan-quan-zhi-nan-cong-tools-dao-react-xun-huan-de-gong-cheng-shi-jian</guid><dc:creator>KennethCheng</dc:creator><enclosure url="https://blog.kennethcheng.cc/apis/api.storage.halo.run/v1alpha1/thumbnails/-/via-uri?uri=%2Fupload%2FChatGPT%2520Image%25202026%25E5%25B9%25B47%25E6%259C%258829%25E6%2597%25A5%252000_06_37.png&amp;size=m" type="image/jpeg" length="0"/><category>LangChain </category><category>Ai</category><pubDate>Tue, 28 Jul 2026 16:07:29 GMT</pubDate></item><item><title><![CDATA[故障排查记录：阿里云 fnOS 网络不可达 (Network is Unreachable)]]></title><link>https://blog.kennethcheng.cc/archives/gu-zhang-pai-cha-ji-lu-a-li-yun-fnos-wang-luo-bu-ke-da-network-is-unreachable</link><description><![CDATA[<img src="https://blog.kennethcheng.cc/plugins/feed/assets/telemetry.gif?title=%E6%95%85%E9%9A%9C%E6%8E%92%E6%9F%A5%E8%AE%B0%E5%BD%95%EF%BC%9A%E9%98%BF%E9%87%8C%E4%BA%91%20fnOS%20%E7%BD%91%E7%BB%9C%E4%B8%8D%E5%8F%AF%E8%BE%BE%20%28Network%20is%20Unreachable%29&amp;url=/archives/gu-zhang-pai-cha-ji-lu-a-li-yun-fnos-wang-luo-bu-ke-da-network-is-unreachable" width="1" height="1" alt="" style="opacity:0;">💡 故障排查记录：阿里云 fnOS 网络不可达 (Network is Unreachable) 本文档记录了阿里云 fnOS 系统出现 network is unreachable 故障的完整排查与修复过程。 故障现象： 服务器突然断网，无法 ping 通外网。 1. 发现问题：查看当前路由表]]></description><guid isPermaLink="false">/archives/gu-zhang-pai-cha-ji-lu-a-li-yun-fnos-wang-luo-bu-ke-da-network-is-unreachable</guid><dc:creator>KennethCheng</dc:creator><enclosure url="https://blog.kennethcheng.cc/apis/api.storage.halo.run/v1alpha1/thumbnails/-/via-uri?uri=%2Fupload%2FGemini_Generated_Image_wuk4kswuk4kswuk4.png&amp;size=m" type="image/jpeg" length="0"/><category>fnOS</category><pubDate>Mon, 4 May 2026 09:29:07 GMT</pubDate></item></channel></rss>