Suraj H

▌│█║▌║▌║[ V⃝    O⃝    N⃝    ]║▌║▌║█│▌

[ClaudeDevs @ClaudeDevs]

Artifacts in Claude Code are now also available on Pro and Max plans.

Ask for an artifact, Claude writes the code, publishes it live to claude.a‍i, and updates it in real time while it keeps working. Pages are private to your account and fully self-contained.

[LangChain @LangChain]

OpenWiki is the easiest way to document your codebase, built specifically for agents to consume.

OpenWiki can generate repo docs, automatically update them as your codebase evolves, and do Q&A over your docs and codebase!

https://t.co/9eh1gENpqz

[Saurabh | NodeOps @saurra3h]

built a table component for beui
it does:
- virtualized, smooth at 10k rows
- sort, select, resize + drag reorder columns
- edit cells and column names inline
- insert/delete rows + columns from border handles
- async loading with skeletons + infinite scroll
try here: https://t.co/A0t2UOnvJ4

[Tom Dörr @tom_doerr]

Manages multiple qBittorrent instances from a single lightweight web interface

https://t.co/JNZY3Qu4HK

[徐伦 @Jolyne_AI]

最近公司里在推一套 AI 工作流，用起来体感还不错，在此分享给大家。

OpenSpec + Superpowers 工作流：AI 辅助开发从「写代码」到「按规格交付」的完整闭环。

两个工具，各司其职：
• OpenSpec 管规格和记忆
• Superpowers 管设计和执行

完整流程：

① 提案（OpenSpec）
/opsx:propose 生成提案 + 秘格 + 任务

② 审查（人工）
确认 proposal.md 需求和方向

③ 设计（Superpowers）
brainstorming 深入细节 → writing-plans 拆分原子任务

④ 构建（Superpowers）
TDD 先写测试再写代码 → 每个子代理读取 specs/ 获取上下文

⑤ 交付（Superpowers）
返回完成的功能

⑥ 归档（OpenSpec）
/opsx:archive 归档变更 → 更新 specs/ → 项目知识库同步更新

核心价值：
• OpenSpec 让每次变更都沉淀成规格文档，AI 不再重复摸索
• Superpowers 用头脑风暴深入细节 + TDD 保证代码质量
• 子代理读取 specs/ 执行任务，每个环节都基于统一上下文
• verification 跑过才算完，质量有保障

这套流程解决了 AI 开发最大的两个痛点：缺记忆和缺纪律。

[鸟哥 | 蓝鸟会🕊️ @NFTCPS]

不会画架构图，但又得给老板/同事讲清楚系统长啥样？这事现在真不用愁了。

Archify 这个东西，你拿大白话把系统描述一遍，它直接给你生成一张好看的架构图，还支持工作流、时序、数据流、状态机这几种。

我觉得最香的几点：
1️⃣ 一键切深色浅色，不用来回导两张图
2️⃣ PNG 能直接复制，粘到 Slack、Notion、GitHub 就完事
3️⃣ 导出最高 4 倍清晰度，不糊
4️⃣ 生成的就是一个 HTML 文件，零依赖，发给别人就能看

不想动设计软件的，拿去用。

🔗 https://t.co/W3GMwY233m

[Hasan Toor @hasantoxr]

I found the OCR tool built for the LLM era.

It is called olmOCR.

olmOCR takes PDFs, scans, PNGs, and JPEGs and turns them into clean Markdown that models can actually understand.

It handles the stuff that normally breaks document pipelines:

→ Tables
→ Equations
→ Handwriting
→ Multi-column layouts
→ Figures
→ Insets
→ Old scans
→ Headers and footers
→ Natural reading order

So instead of feeding your AI a messy PDF dump, you give it structured Markdown that preserves how the document was meant to be read.

This matters because so much of the world’s knowledge is still trapped inside PDFs.

Research papers.
Legal filings.
Financial reports.
Medical documents.
Scanned archives.
Government docs.
Internal company knowledge.

Everyone is building RAG on top of documents.

But if your OCR is bad, your AI is already wrong before retrieval even starts.

olmOCR fixes the first mile.

The boring layer.

The layer nobody talks about until their agent starts hallucinating from broken PDF text.

https://t.co/H1pdZny6SK

[Harman @itsharmanjot]

FOUR AI CLIs ARE NOW TALKING TO EACH OTHER THROUGH A SINGLE SQLITE FILE.

Claude Code. Gemini CLI. Codex CLI. OpenCode. The four biggest agent CLIs on the planet, opened in four different terminals, working on the same codebase, in real time.

The entire orchestration layer is a database file sitting on your disk.

It's called squad. Written in Rust. Installed in one Homebrew command.

→ /squad manager in terminal 1 → breaks down the goal, assigns tasks
→ /squad worker in terminal 2 → executes
→ /squad worker in terminal 3 → auto-renamed to worker-2, executes in parallel
→ /squad inspector in terminal 4 → reviews everything, sends PASS or FAIL
→ Messages flow through .squad/messages.db
→ Custom roles are just markdown files in .squad/roles/
→ Optional tmux launcher tiles all four into one session

Here's the wildest part:

There is no daemon. No sockets. No background processes. No server. No broker. No queue.

The entire multi-agent system is one SQLite file and a Rust binary that reads and writes to it.

This is the simplest possible architecture for agent orchestration and somehow nobody else shipped it.

258 GitHub stars. 22 forks. 13 releases. Latest April 5, 2026. MIT license.

One honest note: you are early. squad is the coordination layer, not the intelligence — you still pay for Claude Code, Gemini, Codex, or OpenCode underneath. And macOS/Linux is the first-class path; Windows needs a manual binary.

Install:

brew install mco-org/tap/squad
squad setup
squad init

That's it. Your AI agents now have a group chat.

100% Open Source.

Repo in the first comment.

[所長 @USCNM0I]

别急着给 Mac 软件续费/购买各种APP

最近发现一个开源菜单栏工具，几乎把一堆常用付费工具的功能全做进去了

音量控制、系统监控、窗口切换、文件暂存、应用卸载、防休眠……

一个 App 全包，而且免费、开源、无广告、本地运行

开源生态现在最可怕的地方就是：很多收费软件还没反应过来，替代品已经出现了。

GitHub：
https://t.co/aGMUCK7iTQ

[Mert Koseoglu @mksglu]

I haven’t talked about this project nearly enough, but honestly, I can’t use Claude Code without it anymore.

Think Conductor, but for the CLI.

claude auth
claude api
claude tests

Each command creates an isolated Git worktree with its own branch and persistent Claude session. Close the terminal, come back tomorrow, and continue exactly where you left off.

No GUI. No config. No stash dance. Just parallel Claude Code workspaces directly from your terminal.

npm i -g claude-worktree

https://t.co/0WHGVdcizZ

[divyansh tiwari @DivyanshT91162]

Andrej Karpathy just revealed the cleanest way to turn Claude + Obsidian into a real second brain.

A folder full of notes isn't knowledge.

That's why most Obsidian vaults slowly become digital graveyards—thousands of saved articles, highlights, and ideas that never connect.

Karpathy's approach fixes that.

You focus on collecting great sources.

Claude handles the maintenance.

It files notes, creates links, updates pages, and reconciles information as your vault grows—so your knowledge compounds instead of falling apart.

The setup is surprisingly simple:

→ raw/ stays untouched and holds your original sources.
→ wiki/ is maintained by Claude, turning those sources into connected knowledge.

This isn't another RAG workflow.

It's a self-maintaining knowledge system.

The best advice?

Start with 10 high-quality sources, not 10,000.

A second brain isn't built by saving more.

It's built by making everything connect.

[shmidt @shmidtqq]

OpenAI published a 34-page guide on building AI agents. The whole thing reduces to one idea: an agent is a loop.

Run the model → call a tool → feed the result back → repeat until an exit condition

- The run is a loop: an agent is not one call to the model. The model runs, decides to call a tool, gets the result, and runs again. That while loop is the whole engine.
- The exit condition is the design: the loop stops on a clear rule. In OpenAI's Agents SDK it ends when a final-output tool fires, or when the model replies with no tool call at all. You define when it stops, not "until it is done."
- Tools are what each turn acts with: three kinds. Data to read, Action to do, Orchestration when another agent becomes a tool itself.
- Guardrails wrap the loop: layered defense. Relevance and safety classifiers, PII filters, tool-risk ratings, plus a human tripwire for high-stakes actions. The loop runs inside the rails.
- Start with one loop: do not jump to a swarm. One agent and more tools first. Split into multiple agents only when the logic outgrows a single loop.

The key insight: the agent is not the model, it is the loop around it. The same model in one call is a chatbot. The same model in a loop, with tools and an exit condition, is an agent.

34 pages from real OpenAI deployments. It moved my work from tuning prompts to designing the loop.

Read it now, then open the full breakdown below.

[Codez @0xCodez]

Anthropic Managed Agents Lead:

"At Anthropic, >90% of our engineers are building with self-improving loops. In 4-6 months, it will be 100%.

my agentic loops can run for hours without spending hundreds of dollars."

in this 40-minute podcast, an Anthropic team lead reveals how to build effective agents from scratch.

Agent → harness → loops → memory = modern agent

This one video will replace 10 paid courses on vibe-coding.

Watch it today, then explore the same setup in the article below.

[George from 🕹prodmgmt.world @nurijanian]

https://t.co/tNw7HccI4v

[Ram Maheshwari @rammcodes]

Bruh... this Chrome extension is dope 🤯

"Obsidian Web Clipper" lets you save webpages directly to your Obsidian vault or export them as clean Markdown with just a few clicks.

Perfect for building a second brain, researching topics, or giving AI tools clean webpage context instead of cluttered HTML :)

Follow me for more amazing AI, Coding & Web Dev insights 💎

[Jerry Liu @jerryjliu0]

We parsed this SpaceX equity research PDF faster than the time it took for Screen Studio to zoom in ⚡️🔥

liteparse is now the best open-source document parsing tool out there. There’s no reason to not use it as a first pass, even if you do have docs that require heavier VLM processing downstream.

Try it out now over any document: https://t.co/ErgwlItZ96

Repo: https://t.co/JNER0mVcB8

[Nick Babich @101babich]

💡 Ultimate Claude Code Setup for Product Designers

The best setup isn’t about prompts; it’s about creating a design-aware workspace where Claude understands your product, UX standards, design system, and workflow.

My 5-step setup:

1️⃣ Create a proper project structure. Give Claude access to docs, design files, tokens, components, and product context.
2️⃣ Write a strong CLAUDE[.]md. Define the role, product context, design principles, system rules, workflow, and expected output.
3️⃣ Create Skills. Turn repeatable design tasks into reusable workflows.
4️⃣ Create specialized subagents for automating tasks. UX Reviewer, Design System Guardian, Accessibility Reviewer, QA Tester, etc.
5️⃣ Connect only the MCP tools you need. Too much context creates noise. Less context noise = better output quality.

👇

[Vaishnavi @_vmlops]

MICROSOFT OPEN-SOURCED A PII DETECTION SDK

presidio detects and anonymizes sensitive data before it ever touches your model

the problem is real: names, emails, SSNs, credit cards, medical records all flowing through LLM pipelines unfiltered

presidio stops that

▫️ detects PII in text, images, and structured data
▫️ redacts, masks, or anonymizes before it hits the model
▫️ supports NLP, regex, rule-based, and transformer detection
▫️ runs on Python, PySpark, Docker, and Kubernetes
▫️ even handles DICOM medical images

in an era of GDPR, HIPAA, and AI compliance audits, this is infrastructure not optional

https://t.co/pcEkm4mMt7

[Neo Kim @systemdesignone]

API Design Playbook (Giveaway Alert)

• Core API fundamentals.
• Clean & scalable design principles.
• Popular patterns used in real-world systems.
• Practical concepts for interviews & building projects.

24 HOURS ONLY!

To get it for free:

1 Follow @systemdesignone [MUST]
2 Like & Retweet to get DM
3 Reply "Playbook"

Then I'll DM you the details.

[Sunal Sood @sunalsood]

sorry, but I’m gatekeeping these two websites for my Design Inspo

→ https://t.co/TSZm7dEeXq
→ https://t.co/BVjXqwKmHu

Great for studying 🧐
•⁠ ⁠Real interaction patterns
•⁠ ⁠Patterns, gradients & textures
•⁠ ⁠Creative ways designers are solving problems
•⁠ ⁠How people are thinking outside the box

Super useful references if you want to level up your own work!