Now vibe coding, so learning hammer FE ?https://channel.meo-ai.comhttps://channel.meo-ai.com/posts/686https://channel.meo-ai.com/posts/686Mon, 27 Jul 2026 02:05:27 GMT<span class="post-title-source">《前端状态管理的真相与迷思》</span><br /><br />标签：<a href="/search/result?q=%23%E5%89%8D%E7%AB%AF" title="#前端">#前端</a> <a href="/search/result?q=%23Web%E5%BC%80%E5%8F%91" title="#Web开发">#Web开发</a> <a href="/search/result?q=%23React" title="#React">#React</a> <a href="/search/result?q=%23%E7%8A%B6%E6%80%81%E7%AE%A1%E7%90%86" title="#状态管理">#状态管理</a> <a href="/search/result?q=%23Redux" title="#Redux">#Redux</a> <a href="/search/result?q=%23Zustand" title="#Zustand">#Zustand</a> <a href="/search/result?q=%23Jotai" title="#Jotai">#Jotai</a> <a href="/search/result?q=%23MobX" title="#MobX">#MobX</a> <a href="/search/result?q=%23CRDT" title="#CRDT">#CRDT</a> <a href="/search/result?q=%23%E7%A6%BB%E7%BA%BF%E4%BC%98%E5%85%88" title="#离线优先">#离线优先</a> <a href="/search/result?q=%23%E5%AE%9E%E6%97%B6%E5%8D%8F%E4%BD%9C" title="#实时协作">#实时协作</a> <a href="/search/result?q=%23%E5%89%8D%E7%AB%AF%E6%9E%B6%E6%9E%84" title="#前端架构">#前端架构</a><br /><br />总结：<br />作者犀利指出，React生态中流行的"状态管理"库（Redux、Zustand、MobX等）本质上只是状态传播与通知系统，而非真正的状态管理。真正的状态管理必须包含时间维度与顺序概念，能正确应用变更并解决冲突。作者推荐转向Y.js、Zero、Fluid等基于CRDT或OT的时序感知系统，它们天然支持离线优先与实时协作，才是前端数据层的未来方向。<br /><br />文章要点：<br />1. 那些我们天天用的Redux、Zustand、MobX，本质上更像"消息广播员"而不是"状态管家"——它们能通知组件更新，却搞不定时间顺序和冲突解决<br />2. React自己也不是状态管理系统，整个前端圈把这个词用得太随意，导致大家一直在用错误的工具硬撑复杂场景<br />3. 真正的状态管理必须懂"时间"和"顺序"，就像分布式系统用向量时钟给事件排队，没有时间概念就谈不上管理<br />4. 值得庆幸的是，Y.js、Zero、Fluid这类基于CRDT和操作变换的工具已经相当成熟，它们天生会处理冲突，离线和实时协作都是顺手的事<br />5. 当我们把数据层从"发通知"升级到"管时序"后，离线优先和实时协作不再是昂贵的附加功能，而是正确架构带来的自然馈赠<br /><br />URL：<br /><a href="https://infrequently.org/2026/07/state-management" target="_blank" rel="noopener" title="https://infrequently.org/2026/07/state-management">https://infrequently.org/2026/07/state-management</a><a class="tgme_widget_message_link_preview" href="https://infrequently.org/2026/07/state-management" target="_blank" rel="noopener" title="It is by no means the gravest linguistic crime of the React epoch,1 but &quot;state management&quot; grates like nails on chalkboard. A Socratic dialogue I keep having about it goes something like this:"> <i class="link_preview_right_image" style="background-image:url('https://cdn4.telesco.pe/file/ikJwW7XtmRLraCtYYCU98eth1-gW6tZABXURyVahlK4DeVCx2q4ZW-GpUv8pXhNBJynRgGNB-8pkhsrRb1oOdMPFQ9GcwYYYJyKx2_246dZIbRuP0VVRejifERHTEtfMqvEthY27WJRAS-5Wqp9SkPcVI_XNkHWmd9yRkmPOS34i-e-SS9uL9QOozGTnezpHvhfM5LMkxZSmNck0D9I18sQk-Zm2znuBFHlMYkTb-9fCqRfwK06uTvUVyuukqZ0-Tl-WjUVPiPkQ5dnPfXTv0vtZR_nYDtwwUmkU8y5MXoNTGMq_ZpaG-xLuD_jur7PA4b1tP0iV01K4Xg9BXiefFQ.jpg')"></i> <div class="link_preview_site_name accent_color">Infrequently Noted</div> <div class="link_preview_title">The Absolute State of Management</div> <div class="link_preview_description">It is by no means the gravest linguistic crime of the React epoch,1 but "state management" grates like nails on chalkboard. A Socratic dialogue I keep having about it goes something like this:</div> </a>https://channel.meo-ai.com/posts/607https://channel.meo-ai.com/posts/607Mon, 20 Apr 2026 05:15:21 GMT<span class="post-title-source">《OpenAI Agents SDK：轻量级多智能体工作流框架》</span><br /><br />标签：<a href="/search/result?q=%23AI" title="#AI">#AI</a> <a href="/search/result?q=%23%E5%A4%9A%E6%99%BA%E8%83%BD%E4%BD%93" title="#多智能体">#多智能体</a> <a href="/search/result?q=%23Python" title="#Python">#Python</a> <a href="/search/result?q=%23OpenAI" title="#OpenAI">#OpenAI</a> <a href="/search/result?q=%23MCP" title="#MCP">#MCP</a> <a href="/search/result?q=%23%E6%99%BA%E8%83%BD%E4%BD%93%E5%B7%A5%E4%BD%9C%E6%B5%81" title="#智能体工作流">#智能体工作流</a> <a href="/search/result?q=%23LLM" title="#LLM">#LLM</a> <a href="/search/result?q=%23%E5%AE%9E%E6%97%B6%E8%AF%AD%E9%9F%B3" title="#实时语音">#实时语音</a> <a href="/search/result?q=%23%E6%B2%99%E7%AE%B1%E7%8E%AF%E5%A2%83" title="#沙箱环境">#沙箱环境</a><br /><br />总结：<br />OpenAI Agents SDK 是一个轻量但功能强大的 Python 框架，用于构建多智能体工作流。它支持 OpenAI 的 Responses 和 Chat Completions API，同时兼容 100 多种其他 LLM，具有供应商无关性。框架围绕"智能体"这一核心概念展开，每个智能体都配备指令、工具、护栏和交接机制，让复杂任务可以像搭积木一样拆解协作。<br /><br />文章要点：<br />- 智能体是核心乐高积木：每个智能体都自带"说明书"（指令）、"工具箱"（函数/MCP/托管工具）和"安全护栏"（输入输出校验），还能互相"交接"任务，像团队协作一样分工处理复杂流程<br />- 沙箱智能体让AI真正"动手干活"：0.14.0 版本新增的 Sandbox Agent 能在容器环境里操作文件系统、运行命令、打补丁，适合需要长时间执行且要保留工作状态的"重体力"任务<br />- 人在回路，安全可控：内置了人类介入机制，在关键节点可以暂停流程等人来确认，避免AI"自作主张"搞出大新闻<br />- 全链路可观测：自带 Tracing 追踪系统，能可视化查看每个智能体的思考过程、工具调用耗时和 Token 消耗，方便调试和优化<br />- 不挑模型，兼容百家：虽然是 OpenAI 出品，但设计上保持中立，支持接入 100+ 种 LLM，包括通过 LiteLLM 等适配层接入国产模型<br />- 实时语音也能玩：支持用 <code>gpt-realtime-1.5</code> 构建语音智能体，把实时语音能力也纳入多智能体协作体系<br /><br />文章URL：<a href="https://github.com/openai/openai-agents-python" target="_blank" rel="noopener" title="https://github.com/openai/openai-agents-python">https://github.com/openai/openai-agents-python</a><a class="tgme_widget_message_link_preview" href="https://github.com/openai/openai-agents-python" target="_blank" rel="noopener" title="A lightweight, powerful framework for multi-agent workflows - openai/openai-agents-python"> <div class="link_preview_site_name accent_color">GitHub</div> <div class="link_preview_title">GitHub - openai/openai-agents-python: A lightweight, powerful framework for multi-agent workflows</div> <div class="link_preview_description">A lightweight, powerful framework for multi-agent workflows - openai/openai-agents-python</div> </a>https://channel.meo-ai.com/posts/575https://channel.meo-ai.com/posts/575Wed, 18 Mar 2026 01:39:20 GMT<span class="post-title-source">《关于协作编辑的谎言（第二部分）：为什么我们不用Yjs》</span><br /><br />标签：<a href="/search/result?q=%23%E5%89%8D%E7%AB%AF" title="#前端">#前端</a> <a href="/search/result?q=%23ProseMirror" title="#ProseMirror">#ProseMirror</a> <a href="/search/result?q=%23CRDT" title="#CRDT">#CRDT</a> <a href="/search/result?q=%23%E5%8D%8F%E4%BD%9C%E7%BC%96%E8%BE%91" title="#协作编辑">#协作编辑</a> <a href="/search/result?q=%23%E6%80%A7%E8%83%BD%E4%BC%98%E5%8C%96" title="#性能优化">#性能优化</a> <a href="/search/result?q=%23Yjs" title="#Yjs">#Yjs</a> <a href="/search/result?q=%23%E5%AE%9E%E6%97%B6%E5%8D%8F%E4%BD%9C" title="#实时协作">#实时协作</a><br /><br />总结：<br />本文是Moment.dev团队关于协作编辑算法分析的第二部分，作者详细阐述了为何在生产环境中放弃Yjs而选择基于ProseMirror-collab的简单方案。文章指出Yjs存在严重的性能问题（每次协作编辑会销毁重建整个文档）、与文档Schema冲突、权限控制困难、调试困难以及墓碑数据占用内存等问题。作者认为，除非真正需要无主节点的P2P架构，否则40行代码的"简单方案"在性能、可维护性和开发体验上都优于复杂的CRDT实现。<br /><br />文章要点：<br />- Yjs存在严重性能缺陷：每次协作编辑会销毁并重建整个文档，导致60fps目标难以达成，影响NodeView、插件状态、撤销功能和光标位置管理<br />- 简单方案仅需40行代码：使用prosemirror-collab库，通过单一权威节点管理文档版本和事务，支持乐观更新、离线编辑和网络中断恢复<br />- Yjs与文档Schema冲突：在无主节点架构下难以验证事务有效性，可能导致数据永久丢失，升级时尤其危险<br />- 权限控制复杂化：需要将Yjs的XML更新预测转换为ProseMirror事务来判断权限，实现难度大<br />- 墓碑数据问题：Yjs需保留删除标记，导致内存持续增长或数据丢失风险，而简单方案通过数据库存储步骤即可解决<br />- 调试困难：CRDT仅保证最终一致性，难以区分暂时分歧与真正错误，调试工具受限<br />- 核心观点：技术选型应从最终用户体验出发，而非算法本身；如无P2P刚需，简单方案在各方面均优于CRDT<br /><br />文章URL：<a href="https://www.moment.dev/blog/lies-i-was-told-pt-2" target="_blank" rel="noopener" title="https://www.moment.dev/blog/lies-i-was-told-pt-2">https://www.moment.dev/blog/lies-i-was-told-pt-2</a><a class="tgme_widget_message_link_preview" href="https://moment.dev/blog/lies-i-was-told-pt-2" target="_blank" rel="noopener"> <div class="link_preview_site_name accent_color">Lies I was Told About Collaborative Editing, Part 2: Why we don't use Yjs / Moment devlog</div> <img class="link_preview_image" alt="Lies I was Told About Collaborative Editing, Part 2: Why we don't use Yjs / Moment devlog" src="/static/https://cdn4.telesco.pe/file/YVXaUzBE8-9TDW2QQrSSbabfgg71zJ4YfK9str6-S1LPODRrO0GTgZm9XVC005_t5HR8VqdNrsgcUwQyA9CcsGyd1JTNXiYmfD35KPvxoWVvTV3w40qwfsqy8doKw0ZtWIufS8kkP6b_7C_qBXZmPDelkDueKKd_mZLaVGpRdnWPevjv76nE543DoXGcMtepNx9ppudeaLhaphSNg-17nX4wJVoKz43YqT9z_ynPXioKsnRqKvzAi1NxlO07pXRzyhwn5GcDxgfJ6WE6NlLa5PhPeeJh7VLRKBwCq1-crT8pqRc_IVTVDnEToydSWbK2EBXIWObDBA27HcLh3qeT-Q.jpg" width="1200" height="630" loading="eager" /> </a>