Matt Galligan

@RhysSullivan @backnotprop It's been something I've been building towards for a while now actually! More opinionated than just a lib, but intent is the same. Still working through a few things before it's "ready" but close! Here's the repo: github.com/outfitter-dev/…

@backnotprop nah executor doesn't do this different things, might be worth to make a lib for it but i feel like theres one out there already

it's worth calling out for any companies wanting to do this, the best way is an OpenAPI spec that automatically gets converted into an SDK, MCP, and CLI easiest way to acheive this

@thomaspaulmann @raycast 1. A leader key / prefix mode 2. Dedicated "Search Apps" command 3. Custom per-app key remapping. Set a hotkey, then remap that hotkey to others on a per-app basis. Example: hotkeys for "Next/Prev Tab" which can always take on that meaning for every app you run.

What are your wishes for @raycast?

@joshpuckett @Nico__Chuan As someone who grew up in IL but live in PA now, I concur.

@Nico__Chuan Sometimes I miss the quiet open roads like these of the Midwest (tho can also believe this is PA!)

@peduarte @raycast Just submitted the feedback! Included screenshots and a (hopefully) pretty detailed description of what's happening.

For sure. Thanks, should have looked there first! While I'm at it, I realized it would be really cool if Raycast had a built in agent that could directly manage Raycast itself. So if I wanted to make a broader set of changes, or wanted to try out some aliases, shortcuts, snippets, etc it could just hammer through and get it all done. Could also make suggestions for customization based on my usage.

@raycast @peduarte I've been having some trouble w/ getting Raycast Beta to recognize keyboard shortcuts when trying to trigger them from my Glove80 keyboard (uses ZMK firmware), or my Logitech MX Master S4. I've tried to diagnose as much as possible but it seems like it's in how Raycast is interpreting modifier keys, and only triggering on key up. In most of the non-standard keyboard setup stuff that will mean Raycast rarely sees the full mod+keys. For instance, I have a key on my keyboard to send Hyper + Space in one shot, but Raycast never sees the space. On my Mouse, I have a similar thing to send an F key, but Raycast only ever sees the Globe key (Fn) but never the F number. Strange! Happy to go deeper and diagnose/debug with y'all because I'd love to make more use of the beta for all this stuff.

@simonbs Add something like that in your global AGENTS file. Suggest subagents when exploring codebases, doing reviews, or executing on larger tasks where delegating well-defined portions would be feasible. Codex will act more proactively w/ even just that.

It’s embarrassing how often I forget to ask Codex to launch subagents. Codex CLI needs to be more like Clippy: “It looks like you gave me a list of tasks. Want me to spin up some subagents?”

Rest in peace, friend. @om you were one of a kind. om.co/2026/06/24/196…

@thsottiaux I'd love to have a way to provide custom thread naming instructions. It would make it so much easier to skim the list of threads for the one you want. Something like: "[$PROJECT] Brief description of current work"

What should we improve in the Codex app. What's not delightful?

@peduarte Should be a first-class feature!

what

@romainhuet @thsottiaux @jxnlco Wanted to share a strange thing that's been happening for me w/ Codex lately within the iOS and iPadOS apps: Codex is running on my Mac Studio, and I'm connected to it on the mobile apps (w/ built-in, not ssh). Any time there's a response, it comes in super slowly. One word per second basically. It will keep doing this until the full response is generated, except that when the full response comes in, the UI does this even weirder thing, where it disappears, only to then start re-drawing the response again. Backing out of the thread, and going back in doesn't seem to solve it either. But a force quit and reopening the thread will show it there in full, no problem. Needless to say it's been a *lot* of force-quitting as of late. It's been like this for at least a week, can't remember how much longer before that. And the behavior is consistent regardless of the internet connection. At home the Mac Studio is connected over 2.5Gbps Ethernet directly to the router, then my devices are connected w/ WiFi 7 so everything is super fast w/ low latency. However, same behavior when I'm on 5G cellular. Happy to provide any system reports or whatever you might need to get this problem diagnosed!

@nielr1 From what I've been seeing in agentic builder land, there has been a huge swing towards using the Codex desktop app. True for me too. Before: 90/10 Claude Code CLI, Codex CLI. Today: 70/30 Codex desktop, Claude Code

Are we all converging on OpenClaw and Claude as the main two tool harnesses everyone uses? Is Codex going to give us a third option anytime soon? Feels like they are signaling this way but not really embracing the idea yet.

I really liked using Canvas before, but this is a nice upgrade. The iOS copy-to-clipboard feature still has an annoying pre-existing bug: copying does not preserve Markdown syntax. Top-level list item asterisks are the only exception. The only workaround is to open ChatGPT in a web view and copy from there.

Long-form writing works better with a little more space. Now you can edit longer pieces in full-screen and save them to your Library to come back to later.

@Dimillian @awwstn One small papercut is that when you're in a Codex thread, swiping back from the left edge of the screen takes you back to ChatGPT, leaving codex entirely. If I'm in the context of working w/ Codex, I'd prefer for it to stay there and just go back to the threads list.

@awwstn tell me more, if you mostly use Codex/exit the app while in Codex the app should resume in Codex

codex needs a separate ios app from chatgpt so bad. i’m using claude chat more than ever because it’s such a pain to switch out of codex to chat

There seems to be a weird interaction between the installed Codex CLI and macOS apps when it comes to available tools. Even w/ both at latest stable. At first it was the "automation_update" tool, now the new thread tools. I've tried every combo of latest CLI using npm and standalone installers, as well as the one that ships w/ the macOS app. I just can't seem to get to a state where Codex agents can use the new thread tools, as was the case w/ automation tools the other day (though that's now working). Which got me wondering: is there a "blessed" or "best" setup when installing and actively using the Codex CLI and macOS apps, so that all the latest features work in both as described? cc @OpenAIDevs @romainhuet

Is 42B+ tokens a lot w/ Codex? …also pretty easy to spot when the macOS desktop app shipped. The CLI's been good, but Codex really shines on the desktop.

@DennisonBertram Never got it up and running fully, but I thought this looked like it could be cool: paperclip.ing

I need help finding the perfect agent orchestration platform. I have agents running my codebase, company, design, operations, but none of the tools like Codex make it really convenient to manage in one spot. Suggest me some tools I should check out.

Anyone else seeing with the Codex macOS app that the thread automation / "automation_update" tool seems to have disappeared / regressed? Previously my agents could set timer-based automations within threads as reminders, but it has seemingly vanished now.

The Codex macOS app is truly a pleasure to use, so very little in the way of complaints there. For my use cases, I tend to fire up multiple windows, which is nice…but they're all just named "Codex." That can make some workspace automation and scripting hard. @romainhuet @embirico it would be awesome if they might be able to carry a name, so that there's more visibility into what's going on in each window!

@thsottiaux Good plan. Good plan indeed. The one thing that could make it an even better plan might be easing some limits on us poor souls who plow through 100% of their max'ed plan's weekly limit in 3 days. The cause of which is good releases, and one may want to enjoy the subsequent ones.

Our master plan is to release better and more efficient models. And also to release better products, week after week. Oh and get more compute too. Together with spending too much time on x. How good is this plan?