Thiago Alves Luiz

Web developer that likes a lot JavaScript and studies different front-end techniques to improve user experience

[Hanako @hanakoxbt]

10 agent evals for AI engineers:

(explained with usage)

1) golden set
→ a fixed set of cases you never edit, run on every single change.
→ use as the baseline that tells you whether anything moved at all.

2) llm as judge
→ a second model scores the output against a written rubric.
→ use when the answer is open-ended and there is no string to match against.

3) rubric scoring
→ one number per dimension: correctness, tone, safety, cost.
→ use when a single score hides which part actually got worse.

4) trajectory eval
→ grade the path the agent took, not only the answer it landed on.
→ use when the right answer for the wrong reason is going to bite you later.

5) tool unit tests
→ test each tool on its own, with fixtures, no model in the loop.
→ use always. most agent bugs are tool bugs wearing a costume.

6) regression suite
→ replay past runs against the new prompt or model and diff the results.
→ use before every prompt change, because prompts have no type system.

7) a/b in prod
→ split live traffic between two versions and compare outcomes, not vibes.
→ use when offline scores stopped predicting what users actually do.

8) human review
→ sample a slice of runs and have a person grade them honestly.
→ use to calibrate your judge, because a judge nobody checks quietly drifts.

9) shadow run
→ the candidate runs on real traffic in parallel and its output is shown to nobody.
→ use before a risky rollout, when one bad answer would be expensive.

10) red team
→ deliberately attack it: jailbreaks, injection, exfil, tool abuse.
→ use before anyone external can reach it, not after.

offline evals tell you it works. online evals tell you it still works.

both sides matter, but not all ten do. run the two that would have caught your last outage.

save this. then read the full breakdown on loop engineering below.

[Sunil Joshi @suniljoshi19]

🚀 Introducing https://t.co/rIWfC8Ik3I

A collection of top quality free & open-source dashboard components, blocks and layouts built with shadcn/ui & @base_ui

✅ Open-source dashboard components & blocks
✅ Beautiful admin dashboard template
✅ Copy & paste with shadcn/ui
✅ Pre-Built AI Prompts
✅ MCP Server
✅ Production-ready code

Huge thanks to @shadcn for creating such an amazing ecosystem ❤️

🔗 https://t.co/NhPytA8IN1
Would love your feedback.

[beamnxw ./ @beamnxw]

New useful skill that tracks informative X accounts and collects key AI news, articles, projects and other important materials

Works with Codex, Claude, Hermes, OpenClaw and others

3 steps:

1\ Set up X MCP - X API: https://t.co/hSUFXSPvKr

2\ Install the skill from here: https://t.co/ZoRpeqkJUP

3\ Run this prompt:

"Use the x-agent-intelligence skill to create an autonomous local feed based on my X MCP connection; ask me for source handles if needed, save the result as feed.html and check it"

It generates a clean HTML artifact you can customize however you want

After that you can schedule or automate it to run daily or at any other interval. The author runs it every 4 hours

You select the X accounts yourself (a few solid options are already in assets). You can also ask your agent to configure the feed exactly the way you need

Storing the feed in wiki format is possible but optional

[Gary Simon @designcoursecom]

I'm testing a new library (img2threejs) converts pictures of objects into usable, performant three.js scenes.

👇

[Thiago Alves Luiz @devthiago]

I am back to Twitter - sorry, back to X - and the point is: who cares???

[TANSTACK @tan_stack]

Adding voice messages to your AI chat app? 🎙️

useAudioRecorder() → record → get a ready-to-send audio part. That's it.

One API across React, Vue, Solid, Svelte & Angular.

[Tom Dörr @tom_doerr]

Identifies unique visitors reliably without cookies or local storage

https://t.co/pFlHtYOCj9

[Oliver @oliviscusAI]

Google just open-sourced a collection of agent skills that connect your coding tools directly to Stitch.

It's called Stitch Design Skills.

→ Convert React, Vue, or HTML code directly into Stitch designs
→ Generate multi-page websites from a single prompt
→ Export Stitch designs back to React or React Native components
→ Works with Claude Code, Codex, Cursor, and Gemini CLI

100% Free.

[Oliver @oliviscusAI]

Turn any google street view into a 3d gaussian splat 🤯

Just paste a link from google maps and it generates a fully immersive 3d scene.

100% Free.

[Abraham John 🦄🦓 @Abmankendrick]

Web Designers and Developers, this might be the next-level animation library

ReactBits gives you a growing collection of smooth, modern UI animations you can plug straight into your projects.

Website → https://t.co/ZL01TgWqIv

[Adam Pietrasiak @pie6k]

I have been working with CSS for ~20 years, and this guy keeps amazing me by showing what can be done with a few lines of it

[Fred Moon 🚀 @fw3d]

i built this in @hiunicornstudio & @framer

the cool thing about controlling the webgl uniforms inside a framer component → you can smoothly control your scene between pages (view transition API)

see how the URL changes?

[Chong-U @chongdashu]

Here's how to vibe code your very own 3D game characters for iOS.

> nano banana pro
> meshy for rig
> export to glb
> threejs skill to import
> capacitor js for iOS
> GPT 5.3 codex

Full prompts + workflow in 🧵

[@levelsio @levelsio]

This guy is rebuilding Microsoft Flight Simulator in the browser with Opus 4.6

Very impressive

https://t.co/ri1nsxarxZ

[Adam Argyle @argyleink]

img:hover {
scale: 1.2;
clip-path: inset(
10% 10% 10% 10%
round 20px
);
}

nice, n' simple

https://t.co/X0pDcf4GBy

[Thiago Alves Luiz @devthiago]

Decided come here and start using back. It’s been a while. Well, I don’t have a topic, to be honest.

[Adam Argyle @argyleink]

Wild how far you can push "customizable select" with just #CSS

[Thiago Alves Luiz @devthiago]

🫥

[Thiago Alves Luiz @devthiago]

looking https://t.co/RyWZGAHsX8

[Thiago Alves Luiz @devthiago]

Let's see https://t.co/VkoxLKL5SQ