Vitor Buzinaro

Staff Frontend Engineer at @close.

[Vitor Buzinaro @vbuzinas]

I just built React Trace: a development-time inspector for React.

Hover any element, see which component rendered it, and jump to its source.

Preview the code, edit it in the browser, open it in your editor, or add comments and send them to your AI agent.

Here's a quick demo:

[Vitor Buzinaro @vbuzinas]

@SamGCoder @grok Any thoughts on Bend?

[Vitor Buzinaro @vbuzinas]

@marclou "The game's hourly email limit has been reached." 🥲

[Taelin @VictorTaelin]

Bend 2 is here!

It is a new programming language that blocks AI mistakes via *proof checking* - the same technique big AI labs used to solve open math problems, like Navier-Stokes.

It is also very fast, and runs on GPUs.

Watch the video. Link in the comments.

[Vitor Buzinaro @vbuzinas]

@kentcdodds Although I agree he indeed does that, in this particular case, he was specifically answering a question: "Apart from Grok 4.7 what to expect from SpaceX AI in September?"

[Vitor Buzinaro @vbuzinas]

@winglee Any plans on supporting Cursor? And specifically Cursor Cloud Agents + self-hosted ones? I was using T3 Code, but now that I'm running self-hosted Cursor Cloud Agents I'm stuck on Cursor itself, because T3 Code doesn't support them, but your app looks very neat!

[Vitor Buzinaro @vbuzinas]

@jackfriks @b_nnett There is a tool called CLIProxyAPI that automatically loads balances between subscriptions. You just need to add `ANTHROPIC_API_URL` to your `~/.claude/settings.json`. DM me if you need more info. Very simple one-time config.

[Vitor Buzinaro @vbuzinas]

@mattpocockuk I have a (bit more detailed) version of this that I've been using for a few months! On my skill, I ask the orchestrator to initiate its prompt to the subagent with `/implement` and it works like a charm (so every subagent has its own code reviewed before sending it back).

[Vitor Buzinaro @vbuzinas]

@poteto @bot Would you be open to record a S&T going through how you use the skills? I read pretty much all of them and I think I have some general idea on how to use them, but I can't be sure if my workflow can be as optimal as the person who created them. I'd love to watch (or read) it!

[Vitor Buzinaro @vbuzinas]

@b_asselstine @rough__sea I don't think that's how it works. You can use the same Durable Object for every channel, you just create an instance for each one. You can have a single worker routing every instance.

[Vitor Buzinaro @vbuzinas]

@solid_js @RyanCarniato Nice write-up! Looking forward to the next one!

[Vitor Buzinaro @vbuzinas]

@ThadDoormat @n_ullO7 @theo @fullfrankchan T3Code doesn't have a system prompt, doesn't write its own tools etc. It just uses the harness you already have under the hood. So when you send a message to Fable using the Claude Code harness, it doesn't matter if you're using T3 Code, Claude Desktop or Claude Code TUI.

[Vitor Buzinaro @vbuzinas]

@ThadDoormat @n_ullO7 @theo @fullfrankchan A harness is what provides a system prompt, tools, skills etc to the model. Opencode is a harness: for example, its "edit" tool is a code written by Opencode itself: https://t.co/jfN0KD59Dd

You can think of T3Code as a frontend for existing harnesses: Claude Code, Codex etc.

[Vitor Buzinaro @vbuzinas]

@RyanCarniato That's a great article! Very different than the ones I've seen in the past few days from Solid core team. Keep up the great work!

Btw, when the content is good, it doesn't matter if the article is long. This one is extremely long, but it's a joy to read: https://t.co/EzEtIx9Akx

[Vitor Buzinaro @vbuzinas]

@mattpocockuk This works well because Fable starts the subagent session with `/implement` which makes the subagent to run `tdd` and `code-review`. Then after the subagent is fully done, Fable does another pass to catch intent, anything that only the grilling conversation could've caught.

[Vitor Buzinaro @vbuzinas]

@mattpocockuk I have a `/to-implementation` skill that does exactly that (Fable to orchestrate the tickets with Grok high or Luna max subagents), and it's been working like a charm for months. I do `grill-with-docs` → `to-spec` → `to-tickets` → `to-implementation`, all on the same session.

[Vitor Buzinaro @vbuzinas]

@RyanCarniato If the author spent some time to edit the article and make it human-consumable I'd love to give it another try, but after reading 15-20 paragraphs, the AI sloppiness is just unbearable. I tried it once yesterday and once again today just because you recommended, but I just can't.

[Vitor Buzinaro @vbuzinas]

@penberg To be honest I like their posture (it challenges me to think a bit deeper about stuff), and usually if I'm 100% sure I just say so and they are happy to the be the wingman once again (but a lot smarter nowadays).

[Vitor Buzinaro @vbuzinas]

@rauchg @shadcn Now why was Shadcn built on React and not on Solid, for example? Obviously because of React's popularity and the fact that a lib like Radix UI already existed. It's one person building on top of another, and assuming that React is successful due to a lib is kinda embarrassing.

[Vitor Buzinaro @vbuzinas]

@rauchg @shadcn Years before shadcn, there was Material UI, but then people wanted to have more flexibility. Radix UI came in and was actually the headless lib that everyone always wanted. But of course, people like some sensible defaults, and here's where Shadcn came in.