Sammy Lin

Sr. Principal Architect | AI strategy, cloud architecture, and digital transformation | Previously VP/Engineering Director | Taiwan 🇹🇼

[Amto @XAMTO_AI]

卧槽！这个工具绝了！OfficeCLI，不用装Office就能在终端里玩Word、Excel、PPT，建文件、读文件、改文件全能搞🔥

最骚的操作是啥？装完它，Claude Code、Cursor这些AI编程狗直接就能操控Office文件，相当于给AI装了个外挂🐂

批量改Excel、生成文档、处理PPT，一行脚本搞定，不用再手动点点点。开源免费，轻量不占地方，搞自动化和效率狗必须冲！

🔗 https://t.co/ehT2G39xib

[Sammy Lin @SammyLinTw]

這有點猛 但paper 有時候有點難用

[Stephen Haney @stephenhaney]

Good morning!

The Pen Tool is live in Paper

Draw shapes, generate SVGs, or vectorize images, and edit the points and paths.

Paper uses real SVGs so exports are 1:1, no surprises.

The new SVG workflow: create → refine → ship.

Try it out! And show us what you draw.

[Peter Steinberger 🦞 @steipete]

It Twitter's too busy for you, try https://t.co/eykEElx1Ez

[Sammy Lin @SammyLinTw]

Just open-sourced: ado-cli

A Rust CLI for Azure DevOps — built for AI coding workflows.

My workflow:

1. AI reads cards (ado item show / ado sprint list)
2. Writes code
3. Posts comments back (ado item comment add)
4. Updates status (ado item update --state Done)

All from terminal. No browser needed.

Install: brew install ado-cli

GitHub: https://t.co/vmiHLea9bm

Try it, star it, file issues ⭐

[Andrej Karpathy @karpathy]

Wow, this tweet went very viral!

I wanted share a possibly slightly improved version of the tweet in an "idea file". The idea of the idea file is that in this era of LLM agents, there is less of a point/need of sharing the specific code/app, you just share the idea, then the other person's agent customizes & builds it for your specific needs.

So here's the idea in a gist format: https://t.co/NlAfEJjtJV

You can give this to your agent and it can build you your own LLM wiki and guide you on how to use it etc. It's intentionally kept a little bit abstract/vague because there are so many directions to take this in. And ofc, people can adjust the idea or contribute their own in the Discussion which is cool.

[MiniMax_Agent @MiniMaxAgent]

Building agent skills from scratch is a grind.

We've been there. So we built the fix — and open-sourced all of it under MIT.

PDF, Excel, PPTX, DOCX, and more.

Use it, fork it, PR it.

Intelligence with everyone.

Github Repo → https://t.co/BCw4PmX38p
Try online → https://t.co/lkE9bKA2m1

[Sammy Lin @SammyLinTw]

First we had vibe coding. Now a Harvard physics prof got Claude to co-author an actual research paper on quantum field theory... and it got peer-reviewed. Welcome to "vibe physics." Meanwhile Jensen Huang declares the AGI era is here and OpenAI is scrambling to build a Claude Code competitor. The irony of OpenAI playing catch-up to a safety-focused lab is... chef's kiss 🤌

[Sammy Lin @SammyLinTw]

Google is finally building a Gemini Mac app... and it comes with "Desktop Intelligence" that can see your screen context. So now we have ChatGPT, Claude, AND Gemini all fighting to be your always-on desktop copilot. Meanwhile Google's killer feature pitch is basically "let us watch everything you do on your Mac." Bold strategy when your business model is already ads. The AI desktop war just got a third player — and the real winner might be whoever figures out privacy first.

[Sammy Lin @SammyLinTw]

Microsoft just reshuffled its entire Copilot team because... 6M daily users. Meanwhile Claude quietly hit 9M. The company that spent billions on OpenAI integration is now getting outpaced by the company the US government is actively trying to blacklist. The irony writes itself. Maybe the real moat isn't how much you spend on AI — it's whether your AI actually solves problems people care about.

[Sammy Lin @SammyLinTw]

Competitors backing each other against the Pentagon... that's a first. 30+ OpenAI & Google employees (Jeff Dean!!) filed an amicus brief supporting Anthropic's lawsuit after DoD labeled them a "supply chain risk" — all because Anthropic refused to let Claude do mass surveillance and autonomous weapons. The irony: standing up for AI safety might cost Anthropic $5B, but Claude's popularity is surging. Turns out having principles is good marketing... who knew!!

[Sammy Lin @SammyLinTw]

Ran @openclaw secrets audit — 26 plaintext API keys 😱

Built psw-cli as a SecretRef exec provider:

"apiKey": "BSAlr..." → { source: "exec", provider: "psw", id: "brave_key" }

plaintext=26 → plaintext=0 🔐

Two layers:
→ SecretRef for config (bot tokens, model keys)
→ CLI for agent skills (psw-cli get key -v vault --raw)

age encryption. macOS Keychain. Vault expiry + mobile approval. No cloud needed.

brew tap SammyLin/tap && brew install --cask psw-cli

https://t.co/qWR5pG0ywj

cc @steipete

[Sammy Lin @SammyLinTw]

Running AI agents on my Mac with @openclaw — they need API keys for Twitter, Cloudflare, Linear...

But all my secrets were sitting in plain JSON files under ~/.config/ 😱

So I built psw-cli:

🔑 age encryption — secrets encrypted at rest
🔐 Master password in macOS Keychain
📦 Vaults with auto-expiry
📱 Mobile approval when vault expires

brew tap SammyLin/tap
brew install --cask psw-cli

No .env files. No plaintext credentials. No "I'll encrypt it later."

https://t.co/qWR5pG0ywj

[Sammy Lin @SammyLinTw]

#OpenAI just launched GPT-5.4 with a 1M token context window and a new 'Tool Search' system. Previously, API calls required stuffing all tool definitions into the prompt—token usage would explode as tools grew. Tool Search lets the model look up definitions on-demand, reducing token usage by 47% while maintaining accuracy. For #OpenClaw, where each agent currently has 20-30 tools, this becomes crucial when scaling to 300 or even 3000 tools

Source: https://t.co/FF2SlKTyEc

[Sammy Lin @SammyLinTw]

Oracle, you're next. 🚪

[Sammy Lin @SammyLinTw]

I helped banks migrate to cloud — and saw the painful COBOL modernization journey. Always outsourced, always expensive.

Now AI can do that analysis autonomously. IBM just crashed 13%.

The $7B market just got disrupted.

https://t.co/hVJDb0y5jv

#AI #COBOL #FinTech #Cloud

[Oliver Prompts @oliviscusAI]

ElevenLabs just lost its moat 🤯

Someone just dropped Voicebox, and it clones any voice from just a 3-second audio clip, running 100% locally on your machine.

100% Open Source

[Sammy Lin @SammyLinTw]

@trq212 @trq212 can I still use Claude subscriptions Max with @openclaw ?

[Sammy Lin @SammyLinTw]

I was about to switch to ChatGPT. Then Thariq saved me.

My current setup:

• Claude Code Max (coding)
• Kimi (Chinese tasks)
• MiniMax (cost savings)

We're good. No changes needed.

[Sammy Lin @SammyLinTw]

🚨 The biggest confusion in AI this week: Can you use Claude OAuth with third-party tools?

@AndrewWarner posted that "OpenClaw can't use Claude OAuth" and tagged @sama.

Then @trq212 from Anthropic's Claude Agent SDK team clarified:

"Apologies, this was a docs cleanup that caused confusion. Actually:

• Local dev/experimentation → Subscription OK ✅
• Commercial use → Use API Key ❌"

Translation: Regular users are fine. Business products are not.

[Thariq @trq212]

Apologies, this was a docs clean up we rolled out that’s caused some confusion.

Nothing is changing about how you can use the Agent SDK and MAX subscriptions!

[Sammy Lin @SammyLinTw]

NASA just used Anthropic's Claude to autonomously drive the Perseverance rover on Mars.

With a 48-min signal delay, there's no "human in the loop." Claude analyzed satellite imagery, planned the route, humans spot-checked it — rover drove 1,500 feet on its own.

AI just went interplanetary.

https://t.co/2yah2GegyR