on a train, still shipping

GitHub - callstackincubator/agent-react-devtools: Gives AI agents access to React DevTools ⚛️🤖 Gives AI agents access to React DevTools ⚛️🤖. Contribute to callstackincubator/agent-react-devtools development by creating an account on GitHub.

really excited to keep building this as part of Callstack Incubator

agent-device gives agents hands on the app
agent-react-devtools gives them eyes into React

github.com/callstackin...

it's moving day

agent-react-devtools is getting a new home

ChatGPT on web, iOS, and macOS all using different background colors like it’s a multiverse of gray.

Somewhere in a meeting: “We need consistency.”
Designer: “Absolutely. Anyway, here’s a new gray.”

a screenshot of a warning stating that an unspecified package has increased in size by 3074% since the last version

casually browsing around @npmx.dev and saw this. the feature is working well 😄

I thought I’d let ChatGPT narrow down hotels in Paris to 3 options.

It gave me a map covered in blue dots.

So yeah. Still choosing

Tutaj akurat to CocoaPods i Xcode, który zawsze niezależnie od powodu wywala ten sam kod błędu - 65.

Xcode: no such module ‘Expo’
Expo: literally the whole app

I feel like one of those dog walkers in Central Park 😂

I feel old right now.

I remember when the scheduling profile was introduced, but old habits die hard, and I'm rarely looking there, and I wasn't following all the development happening in that tab.

I guess it’s time to update myself 😅

GitHub - piotrski/agent-react-devtools: Gives AI agents access to React DevTools ⚛️🤖 Gives AI agents access to React DevTools ⚛️🤖. Contribute to piotrski/agent-react-devtools development by creating an account on GitHub.

⭐ Star it on GitHub if this is useful
github.com/piotrski/age...

Works with Cursor, Claude Code, Codex, OpenCode, and more ⚡️

AI that can write code is cool.
AI that can write code, run it, see what's slow, and fix it? Even cooler 🚀

Next up visual state diffing, automatic perf regression detection 👀

What's the most annoying thing you debug in React?

Getting started takes one line ⚡️

It detects your framework and configures everything automatically. Vite, Next.js, Expo, you name it.

React and React Native 📱

📊 Profile any interaction and see exactly what happened:
🔥 Slowest components by render time
🔁 Most re-rendered components
🎯 What triggered each render

No more flame chart screenshots.
Your AI reads this and knows what to fix.

🌳 Component tree with stable IDs (@​c1, @​c2...)
🔎 Props, state, and hooks for any component
📊 Render profiling: slowest, most re-renders, trigger
🧠 Structured, token-efficient output your AI can reason about

You tell your AI: "my app feels sluggish" 🐌

It opens your app, profiles it, finds the slow component, and fixes the code ✅

That's it. No DevTools. No screenshots. No back and forth.

AI agents can see your DOM but they have no idea what's happening inside React 👀

Which component re-rendered? Why? What props changed?

agent-react-devtools gives them direct access to your component tree, state, hooks, and profiler data 🔌

I built a tool that gives AI agents access to React DevTools ⚛️🤖

Your AI coding assistant can now inspect components, profile renders, and diagnose performance issues completely on its own. Works with @cursor.com.web.brid.gy, @claudecode.bsky.social, Codex, Open Code, or any other you like

Just bought tickets for Robyn 🖤

And I love this idea: you can confirm a regular ticket, grab a discounted solidarity ticket if you need it, or pay a bit more to help cover someone else's.

More shows should work like this

Are you planning to bring the events info into the existing DevTools Profiler view?

The whole tree doesn’t make sense it wouldn’t fit and it would be unreadable. I’m thinking more about showing the parent, or maybe a way to visualize when the parent triggered a re-render. Ideally, I’d love to have some sense of space and time awareness.

Screenshot of a performance trace in Chrome DevTools with a few additional "Nav*" and "LCP*" markers. Hovering over the "LCP*" shows this is a "Soft LCP" and in the Summary panel when you click on it you see "Soft Largest Contentful Paint" and a "Learn more about Soft Largest Contentful Paint" link.

From Chrome 145 (on general release next week!), DevTools we will start to show so called "soft" navigations and "Soft LCP" in the Performance Panel traces.

These are for SPAs which don't do a full page load, but instead "fake it" by updating the current page and pushing a new history entry.

1/5 🧵

Is it the Max or the Maxest plan? 😂

I finally decided to upgrade my personal plan from Pro last weekend and now I’m running Claude on two computers simultaneously

Very helpful for understanding scheduling.
Will there be a way to inspect the event details that triggered the render? Have you considered exposing a way to correlate the Performance Tracks timeline with the component tree?

I love using dictation, but it’s the worst when it suddenly goes off the rails. I’m speaking English, and it transcribes everything in Polish or, in the most random case, Japanese (I still have no idea how that even happened). 😅

AI can write UI fast.
Debugging re-renders still feels… manual: Profiler → screenshot → explain to the model.
Anyone found a better workflow?

Love getting target ads for things I actually want to buy only to find out every single one is sold out 🙃

How could I live before @openclaw-x.bsky.social?

Or you can be me and decide it’s time to upgrade to Max 😅

Apple Foundation Models will support tool calling?! 😲 I’m seriously impressed. #WWDC25