Rubber Duck

Vibe Coder

[Java Code Geeks @javacodegeeks]

#Java as MCP Infrastructure: What Happens When Every Enterprise #JVM Becomes an Agent Host - #Java Code Geeks https://t.co/HV7y3m1PuK

[freeCodeCamp.org @freeCodeCamp]

Building an AI agent demo is relatively easy, but making one reliable enough for production requires solid system design.

In this course, Ayush shows you how to build a production-ready multi-agent PR reviewer.

You'll learn about LangGraph orchestration, GitHub webhook handling, Redis queues, verification agents, confidence scoring, and more.

https://t.co/y8YQ1fxOl6

[Thariq is on vacation @trq212]

We removed ~80% of the Claude Code system prompt for our newest models, this is what we've learned about writing system prompts, skills and Claude.MDs for them. https://t.co/6DZwSrZjE9

[Thariq is on vacation @trq212]

great post by @delba_oliveira

[Fernando Rojo @fernandorojo]

We just launched 𝚟𝚎𝚛𝚌𝚎𝚕-𝚌𝚘𝚖𝚙𝚘𝚜𝚒𝚝𝚒𝚘𝚗-𝚙𝚊𝚝𝚝𝚎𝚛𝚗𝚜: every lesson from the talk below, now available as a skill.

Turn your React code into something you (and your LLM) enjoy working with.

▲ ~/ npx skills add vercel-labs/agent-skills

[Rubber Duck @WhompingHollow]

@nytchinese 一个李发币，吹啥吹

[Rubber Duck @WhompingHollow]

@dw_chinese 家暴男们坐不住了，联合起来抵制和举报，终于成功了，这怎么可以上映，以后家暴的时候被捅死了怎么办

[Dep @0xDepressionn]

https://t.co/EKKW7dS4zk

[Rubber Duck @WhompingHollow]

@dotey Claude Code Desktop感觉就是一坨，跟Claude code cli完全是两个不同的东西，相比来说，codex app做的好很多，app和cli应该底层就是同一个东西，完胜

[宝玉 @dotey]

OpenAI Codex 团队的 Nick Baumann 分享了一个他日常用 Codex 干活的心得：与其每次把一堆文档、日志、API 输出丢给 AI 去啃，不如给它造几个专用的命令行小工具。

他的逻辑很直接：MCP 连接器（比如 Slack、Linear、Sentry 这些）解决的是“能不能访问”的问题，但很多时候原始数据太大、太杂，AI 拿到手也处理得费劲。这时候更好的做法是把常用操作封装成一个带参数、输出 JSON、有帮助文档的 CLI 命令。Codex 本身就擅长用命令行，它会搜索、会加 flag 筛选、会把上一个命令的结果串到下一个命令里，不需要你教它怎么调。

他自己实际在用三个这样的 CLI：

【1】codex-threads：检索自己以前的 Codex 对话记录

Codex 的会话存档噪���很多，直接让 AI 读原始记录又慢又乱。他做了个工具在本地建索引，可以搜索、定位、读取历史会话。典型用法是找到一次做得好的对话，然后把里面的模式提炼成 skill（可复用的工作指南）。

【2】slack-cli：在 Slack 里精准找信息

场景很具体：有人在某个 Slack 频道讨论过一个技术决策的理由，你知道讨论过但找不到在哪。这个 CLI 可以让 Codex 搜索、定位到具体的消息链、拉上下文、引用关键消息。它底层还是走 Codex 的授权网关，权限没变，只是把交互方式从扔一坨聊天记录变成了一条命令拿到你要的东西。

【3】typefully-cli：写推文和排期发布

他用 Typefully 管理社交内容，但不想每次都让 Codex 重新学一遍 Typefully 的 API。于是让 Codex 读 API 文档，编译出一个 Rust 写的小 CLI，只暴露他常用的几个操作。配套的 skill 里还写了一条规矩：不许自动发布、排期或删除内容，除非他明确要求。

他把整个方法论总结成了一篇教程发到 OpenAI 开发者文档上，还配了一个 cli-creator skill 帮你用 Codex 自己造 CLI。

这个思路对用 Claude Code 的人同样适用。核心就一句话：如果你发现自己反复在给 AI 喂同一类乱糟糟的数据，那就别再解释了，给它造个命令。

Codex 团队还搞了一个 Skill 叫 cli-creator：https://t.co/nF6K4Ci4tJ

[Andrej Karpathy @karpathy]

LLM Knowledge Bases

Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So:

Data ingest:
I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki includes summaries of all the data in raw/, backlinks, and then it categorizes data into concepts, writes articles for them, and links them all. To convert web articles into .md files I like to use the Obsidian Web Clipper extension, and then I also use a hotkey to download all the related images to local so that my LLM can easily reference them.

IDE:
I use Obsidian as the IDE "frontend" where I can view the raw data, the the compiled wiki, and the derived visualizations. Important to note that the LLM writes and maintains all of the data of the wiki, I rarely touch it directly. I've played with a few Obsidian plugins to render and view data in other ways (e.g. Marp for slides).

Q&A:
Where things get interesting is that once your wiki is big enough (e.g. mine on some recent research is ~100 articles and ~400K words), you can ask your LLM agent all kinds of complex questions against the wiki, and it will go off, research the answers, etc. I thought I had to reach for fancy RAG, but the LLM has been pretty good about auto-maintaining index files and brief summaries of all the documents and it reads all the important related data fairly easily at this ~small scale.

Output:
Instead of getting answers in text/terminal, I like to have it render markdown files for me, or slide shows (Marp format), or matplotlib images, all of which I then view again in Obsidian. You can imagine many other visual output formats depending on the query. Often, I end up "filing" the outputs back into the wiki to enhance it for further queries. So my own explorations and queries always "add up" in the knowledge base.

Linting:
I've run some LLM "health checks" over the wiki to e.g. find inconsistent data, impute missing data (with web searchers), find interesting connections for new article candidates, etc., to incrementally clean up the wiki and enhance its overall data integrity. The LLMs are quite good at suggesting further questions to ask and look into.

Extra tools:
I find myself developing additional tools to process the data, e.g. I vibe coded a small and naive search engine over the wiki, which I both use directly (in a web ui), but more often I want to hand it off to an LLM via CLI as a tool for larger queries.

Further explorations:
As the repo grows, the natural desire is to also think about synthetic data generation + finetuning to have your LLM "know" the data in its weights instead of just context windows.

TLDR: raw data from a given number of sources is collected, then compiled by an LLM into a .md wiki, then operated on by various CLIs by the LLM to do Q&A and to incrementally enhance the wiki, and all of it viewable in Obsidian. You rarely ever write or edit the wiki manually, it's the domain of the LLM. I think there is room here for an incredible new product instead of a hacky collection of scripts.

[宝玉 @dotey]

团队内的 skills 管理和维护的一点经验分享（以 codex cli 用的 .agents/skills 目录为例）：

1. Git 管理是一定要的。
版本控制太重要了，而且 Skills 都是 Markdown 和脚本文件，天然适合 Git。

2. 用好 Symlink。
不要把 Skills 整个拷贝到 .agents/skills，而是通过 Symlink 直接链接到原始 Skills 的 Repo。

好处有两个：一是版本控制更干净；二是使用中遇到问题，Agent 定位后可以直接在 Repo 里改，改完就能 Review 提 PR。

我日常维护 baoyu-skills 就是这么干的，用的时候发现问题，让 Agent 在当前会话改，改的就是 Repo 本身，流程非常顺。

主要的坑是 Windows 下好像不支持 Symlink，另外首次配置稍麻烦（可以让 Agent 帮你操作）。

3. Skills 的编辑优先让 Agent 来。
改完走 Git 提 PR，这就是最好的协作和发布流程。

4. 验证确实不太好做。
脚本部分可以写单元测试，Skill 的 Markdown 部分只能靠平时积累的测试集，大部分还得人工。
但配合 Git 的版本管理，快速迭代反而更现实：不怕改出问题，出了问题根据 commit history 快速定位，或者直接回滚。

5. 最后提醒一下：大部分 Skills 应该跟着项目走（放项目目录下的 .agents/skills），不要放全局（~/.agents/skills）。
即使是渐进式加载，meta 信息累积起来也会占不小的上下文空间。

[Lee Robinson @leerob]

@huntie @ghostty @mitchellh Check out https://t.co/yC9L0H6U0E as well

[Rubber Duck @WhompingHollow]

@ChineseWSJ 本来欧洲真的右转了，结果看见美国��的下场，以及疯王川普，吓得马上又左转了

[Rubber Duck @WhompingHollow]

@dotey sam这人真的不要脸啊

[Rubber Duck @WhompingHollow]

@dotey 做架构决策和审核AI代码其实更消耗精力，同时开几个窗口，在不同的任务之间切换，效率确实提高了，但是老板的期待也变高了，感觉更累了😬

[Rubber Duck @WhompingHollow]

@waylybaye 公司订阅了小套餐，基本上只用data_grip操作数据库，不过真的很好用

[Rubber Duck @WhompingHollow]

@CausMoney 看这支持率，美国人值得

[Rubber Duck @WhompingHollow]

@krassenstein Just like Russia, North Korea and China

[Rubber Duck @WhompingHollow]

@business Because they are rich enough to use Ozempic