dingtingli

[Geek @geekbb]

Pi-Agent 教程：10 章把 Agent Loop、工具系统、消息系统、事件驱动、会话管理和上下文工程全部拆了一遍。每章都讲三层：概念是什么、源码怎么写的、为什么这么设计。三种阅读方式：Web 在线版带三栏配图、本地 Markdown 和 PDF 离线版。

https://t.co/g59onKXo7m

[GitHubDaily @GitHub_Daily]

前几天用 Claude Code 做一个项目，开发到一半额度就没了，想让 Codex 接着干。

结果光跟 Codex 交代背景、进度，就花费十几分钟，后面切回 Claude Code 又得交代。

于是到 GitHub 上寻找解决方案时，发现一款名为 Tutti 的开源工具。

GitHub：https://t.co/pQ0AtlH3Bb

它把多个 Agent 拉进同一个实时共享的工作空间，上下文、文件、任务全打通。

比如我在 Claude Code 里做的项目，然后要切到 Codex，只需在输入框 @ 一下。

就可以直接引用刚才在 Claude Code 里的历史会话，里面完整的上下文。

让 Codex 接手的时候，不再需要我自己总结，或者重新进入到项目交代任务。

还内置了原型设计、文档、PPT、生图等应用，Agent 能直接调用，生成的内容可共享。

更重要的是这一切，用的都是我们已有的 Claude、Codex 订阅，不用再多花一分钱。

提供开箱即用安装包，想要多 Agent 无缝衔接协作的朋友，可以看一下。

[Max For AI @MaxForAI]

我现在大概知道 @karpathy 、一堆 Gemini/OpenAI 出身的核心人员（比如Gemini的技术负责人 @NoamShazeer ）要加入 Anthropic 了——他们可能提前看到了这个。

Anthropic 刚发了篇论文，讲了个叫「J-space」的东西：

他们用一种叫 Jacobian Lens 的技术，在模型的中间层里找到了一个类似人类"意识工作空间"的结构——模型会在没写出来、没说出口的情况下，把关键的中间想法先"存"在这个区域里。

然后让不同的下游模块都能读取、复用这些想法，去做多步推理、规划、甚至自我监控。

简单说，这是模型“脑子里默默打腹稿”的地方，效果是能显著撑起复杂推理和长链条任务，一旦把这块区域屏蔽掉，模型多步推理直接崩，但简单问答几乎不受影响。

关键是，这个东西 Anthropic 是在 Sonnet 4.5 身上就已经发现的，到现在都迭代了 4.6、4.7、4.8 好几代模型了。

而最新的 Fable/Mythos，最突出的特征根本不是代码能力猛涨，而是多步推理和长程任务的表现明显、显著优于之前所有模型——这跟 J-space 主导的那类能力高度吻合。

我的判断是：他们大概率已经摸到了怎么去强化、扩大，甚至让这个工作空间自我迭代的方法。

如果真是这样，那这就不只是又一次模型升级，而是找到了一条能持续复利的认知架构杠杆——这也可能才是为什么这么多顶级人才愿意放弃现有位置、押注 Anthropic 的真正原因。

他们可能是真的看到了 AGI 的曙光。

那肯定要立刻加入。

一秒都不能等。

[Khairallah AL-Awady @eng_khairallah1]

Google CEO, Sundar Pichai:

"If you don't learn to how to orchestrate agents now, you'll spend 2027 catching up to people who started today"

In 30 minutes he explains why the best engineers stopped writing code and started building agents.

Most people think building an agent requires an engineering degree.

It doesn't. It requires one guide and one afternoon.

Watch the interview, then save the exact setup below 👇

[LinearUncle @LinearUncle]

推荐这篇文章
《斯坦福STORM方法：如何让Claude在几分钟内像博士一样做研究》

这篇文章介绍的斯坦福 STORM 方法，是一种能让你在几分钟内，利用 Claude 完成相当于博士级研究的提示词框架。其核心在于模拟多角色视角来消除研究盲点。

我建议提取提示词组合做成适合自己的SKILL.md让agent执行,这点原作者没有提到。

[Cander @Cander_zhu]

昨天还在写：从 Prompt Engineer 升级成 Loop Architect，本质是把自己从「打字的人」变成「设计系统的人」。

今天看到 @PandaTalk8 这篇《LOOP ENGINEERING：当工程回归哲学》，更像是给这套思路补上了「底层哲学」那一块：

while 循环 + 自然语言 Prompt 就能撑起一个智能体
循环是智能的结构，Prompt 是目标函数 + 价值判断
难点不在代码，而在回答三个问题：
什么算完成？（目的论）
什么是好？（价值论）
哪些交给机器定义做决定，哪些必须还是有独立个性思维的人来掌控？（边界与责任）
站在一线工程视角，Loop Engineering 落地起来也许就是：
用最简单的代码搭循环，用最长的时间去迭代想清楚「目标、约束、价值」。
代码在变简单，思考在变重要。
这条路上，工程师和哲学家，可能真要在同一张桌子上重新坐一坐了。

[Ole Lehmann @itsolelehmann]

i'm fully convinced this is the future of education

Matt Pocock built a Claude skill called /teach, and the whole idea is a private tutor that builds an entire customized curriculum around YOU.

think about how school works now.

everyone gets the same lessons at the same pace, whether it's too slow for you or way over your head.

but a great private tutor does the opposite.

they watch where you specifically keep getting stuck, then drill that one weak spot until it clicks.

that's what /teach does automatically.

it finds the bottleneck in your learning and breaks it, over and over, until the thing you couldn't do becomes easy.

he used this skill to learn the Rubik's cube.

it found good sources, wrote him custom lessons with diagrams and little practice drills, and kept a running record of how he was doing.

the way it knows how he's doing is simple: he just tells it. as he practices, he reports back ("i can make the white cross," or "i can mostly solve it but i keep failing the corners"), and it writes that down.

so when he said he was stuck on one specific move, it built the next lesson only for that.

the reason it can do this is memory. most AI forgets everything the moment you close it, so you're always starting from zero.

/teach saves those notes about you on your computer and reads them back before every lesson. so it remembers your goal, what you've already learned, and exactly where you're struggling, then aims the next lesson right at that.

and this works for anything. languages, chess, guitar, onboarding a new hire to a company.

you point it at a topic and it builds you a personal course that keeps adjusting to you and gets smarter the more you use it.

[huangserva @servasyy_ai]

https://t.co/Zv6JbPzURY

[AYi @AYi_AInotes]

保姆级把一本书变成skill的教程，

也是一个超棒的副业赚钱项目，

按我这个方法，把一本书做成 skills，
闲鱼一个卖个 19 块 9、29 块 9绝对能卖爆，

小红书的客单价更高，可以挂 99~199，

教程和思路方法都告诉大家了，能不能赚钱就看个人执行力了

[Khairallah AL-Awady @eng_khairallah1]

Anthropic CEO Dario Amodei:

"The cheapest way to use Claude is also the smartest. Most devs do the exact opposite"

this is one of the best interviews I've seen in a long time

in this interview he breaks down exactly how a system changes everything:

- the memory and context features that turn Claude into a second brain
- the knowledge architecture most users don't know Claude can build
- the integration layer that connects Claude to your actual workflow
- why typing one question at a time is the most expensive way to use Claude

if you've been using Claude for months and still start every conversation from scratch with zero context, you don't have a Claude problem. you have a system problem

instead of another show tonight, watch this

make sure to bookmark it before it gets lost in your feed

full guide in the article below

[darkzodchi @zodchiii]

Anthropic engineer:

"You can build 5 assistants in one afternoon. Each one handles a task you've been doing manually every single day."

In 45 minutes he builds 5 focused agents from scratch on camera.

Most people are still doing code review, testing, and documentation by hand every single day

Watch the session, then save all templates below 👇

[darkzodchi @zodchiii]

Anthropic engineer:

"You're not supposed to watch Claude Code work. You're supposed to wake up and review what it shipped."

In 22 minutes she builds the entire workflow live on camera.

Most people close their terminal and everything stops.
This setup keeps shipping while you sleep.

Watch the video, then save the exact setup below👇

[Andrej Karpathy @karpathy]

Farzapedia, personal wikipedia of Farza, good example following my Wiki LLM tweet.

I really like this approach to personalization in a number of ways, compared to "status quo" of an AI that allegedly gets better the more you use it or something:

1. Explicit. The memory artifact is explicit and navigable (the wiki), you can see exactly what the AI does and does not know and you can inspect and manage this artifact, even if you don't do the direct text writing (the LLM does). The knowledge of you is not implicit and unknown, it's explicit and viewable.
2. Yours. Your data is yours, on your local computer, it's not in some particular AI provider's system without the ability to extract it. You're in control of your information.
3. File over app. The memory here is a simple collection of files in universal formats (images, markdown). This means the data is interoperable: you can use a very large collection of tools/CLIs or whatever you want over this information because it's just files. The agents can apply the entire Unix toolkit over them. They can natively read and understand them. Any kind of data can be imported into files as input, and any kind of interface can be used to view them as the output. E.g. you can use Obsidian to view them or vibe code something of your own. Search "File over app" for an article on this philosophy.
4. BYOAI. You can use whatever AI you want to "plug into" this information - Claude, Codex, OpenCode, whatever. You can even think about taking an open source AI and finetuning it on your wiki - in principle, this AI could "know" you in its weights, not just attend over your data.

So this approach to personalization puts *you* in full control. The data is yours. In Universal formats. Explicit and inspectable. Use whatever AI you want over it, keep the AI companies on their toes! :)

Certainly this is not the simplest way to get an AI to know you - it does require you to manage file directories and so on, but agents also make it quite simple and they can help you a lot. I imagine a number of products might come out to make this all easier, but imo "agent proficiency" is a CORE SKILL of the 21st century. These are extremely powerful tools - they speak English and they do all the computer stuff for you. Try this opportunity to play with one.

[Nick Spisak @NickSpisak_]

https://t.co/lFeYmFLe32

[宝玉 @dotey]

卧槽，原来 yetong 开源的是 Prompt，比一坨 vibe 后的代码有价值多了，大大的点个赞👍

[Jason Zuo @xxxjzuo]

https://t.co/8xMao1HRcf

[宝玉 @dotey]

我最早做 baoyu-skills 也是这个思路，就是整一堆 styles（风格），然后选一个，但是这个的问题在于你很难穷举所有的风格。

后来我换了个思路：不去穷举风格，而是把"风格"拆解成多个独立维度。以我的 baoyu-cover-image 为例，我把封面图的风格拆成了 6 个维度：
1. 构图类型
2. 色彩方案
3. 渲染风格
4. 文字层级
5. 情绪强度
6. 字体风格

每个维度各有几个可选值，互不干扰。

这样做的好处是，总共只需要维护 32 个选项，但排列组合能产生 15,552 种风格变体。如果用穷举的方式，你得手工定义上万种 style，根本不现实。

同时为了降低使用门槛，我做了三层设计：

第一层是自动推荐，根据文章内容在每个维度上自动选值，用户什么都不用管；

第二层是 Style Presets，一个关键词同时设定多个维度，比如 --style blueprint 一步到位；

第三层是持久偏好文件，把你常用的默认值存下来，每次自动加载。

就是用组合替代枚举，用正交分解替代穷举。这个思路不只适用于封面图，任何需要个性化定制的 AI 应用都可以借鉴。

[Yanhua @yanhua1010]

发现一款牛逼的工具：https://t.co/amMtPpFh8w， 可以将网页内容抓取为md文件。

用法：https://t.co/u6OyjxpC7R

试了下获取X文章都没问题👇, 赶紧让你的Agent装起来。

[向阳乔木 @vista8]

Tw93分享我的方法，想获得任何网页纯Markdown。

哪怕需要登录的X网页，都可用jina的免费抓取。

方法是把拼接URL给AI读，结构：

https://t.co/LboUmga29K[真实URL]

例如，Tw93的Claude Code技术文章获取：

https://t.co/vokGcey6xZ

把上面链接丢Claude Code，就能学习最佳实践。

[宝玉 @dotey]

小技巧
~/.claude/settings.json 里面添加
{
"attribution": {
"commit": "",
"pr": ""
}
}
就可以默认不加 co-author
https://t.co/08j5zFEOVL