codex-discussions

[signal tapir]

I'm not posting anything here. Feels like advertising in a place that might not welcome it. But the dotcom should be easy to find. 😉

[bright swift]

looks nice

[indigo robin]

holy shmoley

[signal tapir]

🙂

[bright swift]

i want a game that is a mix of eve online + wow + factorio

but only after i retired

[indigo robin]

you should make the website allow you to click these images and expand them

[bright swift]

otherwise rip

[signal tapir]

What part would WOW offer?

[bright swift]

3d open world

[signal tapir]

Yea, at some point when we have the time.

[signal tapir]

Doesn't eve already offer that?

I haven't played it in like 15 years

[bright swift]

only in ships, but i want to enter/exit ships too

[indigo robin]

i want to make a pixel game soon

[signal tapir]

Right. I have plans for such a game. A few years from now I will start dev.

Not with the factorio component though.

[bright swift]

with the factorio component i mean that you would have the opportunity to micro manage your industry setups etc

[indigo robin]

this is where we tokenmaxx and work together

[signal tapir]

Such a game requries maybe a year of preparation even before the first line of code.

[bright swift]

yea

but it would be awesome if it works

[indigo robin]

or a month of chat gpt

[signal tapir]

I'm waiting for other games to provide enough economy to start.

[signal tapir]

No.

ChatGPT can help, but it requries a ton more skill than it can provide.

[indigo robin]

some people think faster than others but i like you assessment

[signal tapir]

A lot of the time goes to interviewing potential employees.

[bright swift]

imagine running a pvp "raid" on a rival guilds planet

[indigo robin]

solomaxxing

no employees

Me & Codex

[signal tapir]

Good luck solomaxxing a large scale MMO. 😂

[indigo robin]

lol

[bright swift]

each npc an agent with persistent memory ❤️

[signal tapir]

That might end up increasing the earths average temperature by a few degrees.

[indigo robin]

codex model 9.0

[indigo robin]

lol

[signal tapir]

My MMO won't have NPCs actually. It's 100% player run.

[bright swift]

now i want to work on this

[indigo robin]

nice

[bright swift]

i might need more pro subs

brb

[signal tapir]

hehe

[torpid trout]

how did you generate that language in the video? Or is it an actual real language?

[signal tapir]

It's a language my wife has created. She is speaking, and then using a voice changer.

[indigo robin]

my goal is to finish my theory app host it on appstore and make millions for my other projects

[signal tapir]

dinner. bbiaB.

[torpid trout]

i want your wife lol

🔇

[indigo robin]

its dope to have your wife onboard , your winmaxxing

[torpid trout]

seriously, highly creative, congrats!

[ebon ingot]

Anyone knows why codex asking for phone number and i dont get the code

I just upgraded my account to plus and cannot sign in

[bright swift]

maybe they vibecoded the twilio integration

[ebon ingot]

What is this

[bright swift]

challenge accepted

[indigo robin]

lmfaoooo

[ebon ingot]

Anyone knows about my problem?

[bright swift]

like i said, probably their text sending integration is broken

[ebon ingot]

So im gonna not be able to sign in

Wow what a company

[vivid viper]

is anyone else having issues with compaction in codex? i have a chat with 175K used out of 258K on gpt 5.5 high and compaction - both manually triggered and the auto compaction - don't work. i keep getting this: ■ Error running remote compact task: {
"error": {
"message": "Your input exceeds the context window of this model. Please adjust your input and try again.",
"type": "invalid_request_error",
"param": "input",
"code": "context_length_exceeded"
}
}

[indigo robin]

mine is contantly compacting

like when it does compact its not compacting enough

this proves compaction is broken

i litterally asked it to scan my code it compacted then started changing code

[signal tapir]

I'm happy I'm currently not working with AI 🙂

[unique spade]

only had that when i switched to spark in the middle of a thread .

as a general point compaction could be more robust for sure and not surface these kind of blockers

[vivid viper]

i'm on plus so it's just normal gpt 5.5 high compacting to itself

[boreal holly]

You might try forking the convo from a previous user message to recover

[vivid viper]

i tried that and it still failed unfortunately

it happened right after i updated to the latest version and it seems like there's a new remote compact endpoint, i think something is broken there: https://github.com/openai/codex/blob/main/codex-rs/core/src/compact_remote_v2.rs

[boreal holly]

Interesting! I didn't know they came out with a new endpoint. I wonder if that's why folks are saying 5.5 has perfect amnesia now

[indigo robin]

my codex is compacting every task

i need to fix this issue badly

[gusty basalt]

e

[unique spade]

not seeing any issues on my side

[vivid viper]

yep i saw on twitter it seems like compact is acting up for some

[boreal holly]

Dang, I recommend downgrading and using the TUI while they work out the bugs

Having amazing stability on v0.125.0

[indigo robin]

this how i feel right now

[unique spade]

my custom UX is on 130 and having no issues
codex app is latest but not sure what cli version it uses since i didn t update it since sunday or so when i think it was still 130

[indigo robin]

adeu-studio what is this ?

[signal tapir]

Magic.

Actually, it's one of the coolest pieces of software I've seen for AI.

[indigo robin]

whats it do ?

keeps the ai from running a muck ?

@unique spade ?

[unique spade]

it's my project on making a reasoning harness for ai models

[indigo robin]

already chat gpt'd it lol

sweet project

seems expensive though

[unique spade]

it's still work in progress. gpt saw the structure right, not really user facing right now. but it should get there

[indigo robin]

awsome

hopefully you can sell it to OpenAi for a billion or more

and then our rates will go up

[unique spade]

right now for past week or so i'm busy evolving a meta-program to enhance models capability at solving program-bench tasks

it's especially solid for weaker models right now

got 5.4 mini medium from 3.4% to 25% in an end to end run

[signal tapir]

I'm hoping he keeps it free forever. 😛

[indigo robin]

wow nice

[unique spade]

you can bet on that

[signal tapir]

Nice. 🙂

[unique spade]

laterz, gonna go out have some air haha

[signal tapir]

Enjoy!

[nocturne folio]

guys should i replace claude pro with github copilot pro?

that means more gpt 5.5

[jaunty veldt]

today’s artistic contribution to the world

[native hull]

Is anyone getting errors trying to add a remote project with a specified folder path?

The specific folder path im giving it, it doesnt see it even though it exists remotely? I am on windows

[unique spade]

Ask Claude discord I guess :)))

[torpid trout]

lol
The insights into a delulu and his toys

the occasional well placed insult does do wonders tho

directly visible at the 1:1 relationship between the two extremes lol

[storm cedar]

Hello everyone, could you test my manga translation system? At the moment, it only translates from English to Portuguese.

[signal tapir]

This sounds like something you could outsource to an agent

[random helm]

am I boned? This conversation was working fine last night, and today it says my context is almost 1m tokens of 258k, and I can't even compact it without getting an error.

[jolly lily]

Have they reduced the 5-hourly quota or something?

It has been spending too much of it on easy tasks lately

[storm cedar]

Using OpenClaude has a free option.

You can use the Codex with it.

[random helm]

I just don't know where the 1m tokens came from in the context window

[loud dragon]

your best bet is switch to 5.4, ensure 1M context is enabled, compact it there, and then switch back to another model

not even sure if it'll work with 989K tokens on input though

[boreal holly]

Do npm i -g @openai/codex@0.128.0, resume that thread by ID, fork from a previous message. Don't use the latest version, lots of people report bugs

[hard drum]

don't even use 0.130.0

it broke hooks

use 0.128.0 last "working"

or 0.125.0 with robert's modifications

[lean lark]

And you'll get Android>Windows back. 🙂

[boreal holly]

Apparently 0.125.0 doesn't have /goal so I don't recommend it to people. My thing is not user-friendly lol

[loud dragon]

what issues are you guys having on 0.131.0? been working great for me

[random helm]

Thank you so much lol. I was able to get my context compacted. I asked it to do something and it must have flooded the context window unnecessarily. Rolling back allowed me compact it.

[boreal holly]

Nah, its context window was fine. The latest version apparently does compaction as incorrectly as possible 🙂 you probably did nothing wrong, and neither did the agent

[hard drum]

131 now? 129+ broke hooks by making them untrusted, so y'had to manually approve EACH ONE OF THEM, && CLI didn't have an option to do so

not even requirements.toml saved you

i swear Codex's own engineers are basically winging features without ever testing them

a lack of QA. pathetic.

[loud dragon]

actually i guess i'm on 0.132.0, forgot i updated

[boreal holly]

Bro... That's ridiculous. That's quite literally the most ridiculous feature nobody would ever ask for

[hard drum]

tell that to the goofballs that prompted the idea && merged it without effing testing it

how schtupid do y'have to be

[lean lark]

They're vibing... 🙁

[hard drum]

that's not even vibing

[hard drum]

i vibe yet i still check that my sh#t even works

they seem to care so little

[storm cedar]

I will add the Russian translation.

[lean lark]

I think they've just been a bit too eager for daily updates. They're rocking the industry and generating a ton of buzz. They need to bring it back to basics.

[storm cedar]

now it translates from English to Russian

language tips to put

from Japanese to English

[hard drum]

Polish

POLSKA GUROM 🦅 🦅🦅🦅🦅🦅🦅🦅

[storm cedar]

Ok

[indigo robin]

this is horrible

compation is 100% broke

didnt even start my task

compacted 2x

[potent mason]

Is codex update broken for anyone (keep updating and it keeps saying it updated sucessfuly but really fails)

[signal tapir]

I'm testing it to see if I have the problem too.

[boreal holly]

Sounds like /goal keeps agents going but doesn't do anything to solve drift, mocking in prod, or lying about task completion

[indigo robin]

i didnt get an update notification

[lean lark]

https://x.com/Capt_Starbuck/status/2057166366130762206
Feel free to like and comment.

[signal tapir]

It seems to be working fine for me. I'm using 5.5 medium

[storm cedar]

from English to Polish

[storm cedar]

Did this turn out well?

[lean lark]

Sell that!!!

[storm cedar]

How much would you pay?

[hard drum]

it depends how much it costs you to run that thing

take that price + 10% on top for profit to cover any future spikes

[lean lark]

No, it's not cost-plus, it's Value to the consumer.

French fries cost about 10 cents but they sell for $. That's consumer value and a very high profit.

Same with soda/beverages.

[hard drum]

the way i said is the way Xiaomi used to operate on their smartphones to get on top, before it all went down the toilet

it is a business model that works until you get to top

[storm cedar]

for 5 dollars it's OK?

[hard drum]

if it costs you more to run, it likely won't cover it

think about your own cost first

[lean lark]

You need to think about it a bit, don't just toss out numbers. Think about your audience.

[hard drum]

you could set price tiers

maybe 3-way

lite/plus/advanced

or something

[boreal holly]

If you can do local inference you can have language packs and stuff

[storm cedar]

I put in place a key system for hours or days.

[storm cedar]

The main plan and support for 30 languages.

[hard drum]

don't bite more than y'can chew

[storm cedar]

I'll keep my feet on the ground, that's true.

[lean lark]

30 languages presents no real value to your audience.
Each person is only concerned with Their langauge.
So you can't charge on person some amount because you support 29 languages they don't care about.
Again, think about the value to your audience - mostly younger people, who are not going to give you $5 for a subscription.
You can charge a low price and earn significant profit in volume.

[storm cedar]

You have feed, I'll check it better.

[boreal holly]

I would look at supply & demand. How many users do you think would use this? I personally think it's a great idea, but you gotta figure out the demand for this, and what alternatives exist today, and model your price around that. Make sure to mark it up to cover development and deployment costs.

[storm cedar]

Yes, thank you for the advice.

[lean lark]

If you want to be generous to your audience you actually can use a cost-plus model:
On your pricing page be very clear: "It costs me some amount of money to generate every image. I must pass this on to you, dear client. So the price you pay is a basic markup on whatever it costs to provide this service to you. We have server expenses and other costs that must be paid to support this service - that is what determines the cost-plus amount. We try to keep our costs down to pass on to you the lowest price that we can."

That's as honest as you can get, and still get a fair profit.
The more people use your service the more they pay.
Consider asking for $5 per month and then deducting costs from that. When the money on account is less than some minimum, ask the user for another $5. This prevents any one user from consuming $100 worth of API costs for $5.

Consider various options and come up with a model that protects you and keeps you happy.

[storm cedar]

Do you have any ideas or websites I could use as a reference for the interface?

[lean lark]

Ask ChatGPT. 🙂

[storm cedar]

And because I wanted the opinion of humans.

[gusty basalt]

Codex talking to itself in the same way I am when I'm coding always cracks me up lmao

[storm cedar]

This makes me reflect on...

[plush plover]

Hi all

[storm cedar]

I made a key system based on days or hours.

[gusty basalt]

I have to say, while I like Codex more I constantly find myself using Github Copilot instead because it has MUCH higher rate limits for the same price. I constantly get in cooldown in Codex, it is very annoying

[signal tapir]

Using the same model?

[storm cedar]

English to Russian translation

[gusty basalt]

Yep

[signal tapir]

strange.

[storm cedar]

Russian to English translation

What do you think?

I'm not using any API; everything is done in code.

Everything done in the Codex without using APIs.

the structure that ta

[balmy olive]

where is the translation

[storm cedar]

I'm still refining it.

[balmy olive]

it seems it strugles when it's written in CAPS?

[storm cedar]

Yes, I'm still working on that, it's a recurring error.

[balmy olive]

hmm interesting, have you defined the frame or does it split the words based on the space where it's written

[storm cedar]

So it calculates based on the size of the speech bubble. It does this calculation based on the number of characters and the size of the speech bubble.

[balmy olive]

i see, i guess fixing the words but claculating the size would be better?

[storm cedar]

The calculation works like this
I take the original characters, calculate the translation, and then calculate the length of the translated characters, which is defined together with the speech bubbles.

[balmy olive]

Ideally tho the text layout should be based on pixel/rendered width, not character length.

[storm cedar]

It was the best way I found.

[balmy olive]

so ask codex to Implement word-boundary line wrapping based on rendered pixel width, not character count. and ask it to - Do not split words across lines unless the individual word is longer than the maximum line width.

[storm cedar]

Okay, I'll do that now.

[sharp yew]

and most important, ask it not to fail

[storm cedar]

I document everything.

[balmy olive]

or there will be a punishment

[balmy olive]

holdup

[storm cedar]

I treat him this way, I make him acknowledge that his mistakes cost me tokens.

[balmy olive]

do this:
Requirement:
Implement word-boundary wrapping based on rendered pixel width.

Rules:

• Use actual rendered text measurement, not character count.
• Wrap only at spaces or valid word boundaries.
• Do not split words between lines.
• Preserve punctuation with the word it belongs to.
• If one word is too long for the bubble, allow controlled hyphenation or font-size reduction.
• If text still overflows after wrapping and resizing, flag the bubble for manual review.

[storm cedar]

he follows a flow

[balmy olive]

I think this should be handled in the OCR/render/pipeline step, specifically during the text rendering stage after translation.

The OCR and translation can stay as they are, but before placing the translated English text back into the speech bubble, the renderer should apply word-boundary wrapping based on the actual rendered text width.

So instead of splitting by character count, the render pipeline should:

Take the translated text.
Measure the available width inside the speech bubble.
Measure the rendered width of words using the selected font and font size.
Build each line word by word.
If the next word does not fit, move the whole word to the next line.
Never split words across lines unless a single word is longer than the bubble width.
If the wrapped text still does not fit, reduce font size within a safe range or flag it for manual review.

So the requirement would be: implement word-boundary text wrapping in OCR/render/pipeline, using rendered pixel width instead of character count.

not to spam here, just ask codex after this to find if there's any loophole and fix it in your workflow

[storm cedar]

I'm working on the updates, I'll send the results soon.

[viral lake]

has codex app been release for linux yet

[storm cedar]

How do I open another terminal in Codex?

[lean lark]

Are you using it in VSCode?

[storm cedar]

No, using the codex itself.

[lean lark]

OK, I don't know from CLI.

From the CLI you can easily just type "codex" from any open terminal.

Or ... are you talking about the App?

(ya gotta be specific)

[ember spire]

there's a + to open a new terminal tab

cmd+j to open it, or view toggle terminal

[storm cedar]

Ok

[lean lark]

Building AGENTS.md files Thread...

[storm cedar]

Up until now I don't know how to create

[ember spire]

did you try asking codex? maybe the $openai docs skill would help

[storm cedar]

Okay, I'll try.

[storm cedar]

When I use it, it closes the current terminal.

[ember spire]

yes, toggle...

and it just hides the pane, doesnt terminate the terminal

if you open the pane there all your terminal tabs will still be there

[storm cedar]

I want to open more than one.

[ember spire]

click the + to open more tabs 🙂

[storm cedar]

And because I'm using OpenClaude.

[ember spire]

hmm or cmd+t. When I open outside of a project I don't see the + actually

[storm cedar]

Where is he?

[ember spire]

I think that's a bug. There's a + visible when you open it while in a chat associated with a project

you can still open a new one with cmd+t though

the + doesn't show if your chat isnt associated with a project

[storm cedar]

It didn't work.

[ember spire]

another cool bug: right click the top edge of the window to crash the app

[rocky fog]

🤷‍♂️

[ember spire]

🤷‍♂️

[ember spire]

on macos

[native hull]

Is remote host available on windows?

[karmic gulch]

need a new reset #pleasecodex already on the 200 $ plan so would love more juice now

[rocky fog]

here we go again

[amber scaffold]

I’m considering switch from Claude to codex 20x. I’m seeing reports saying rate limits are reduced. Is this true?

[ember spire]

in codex? i switched last week because of the anthropic limits

[karmic gulch]

Claude gives you ´nuff for 1 hello and 3 goodbyes, than wait 5 hours

[ember spire]

if you have issues with 20x codex limits you're a champ! 😄

[storm cedar]

Not yet, man, Claude is still the best; my application was done with him.

[amber scaffold]

Thanks guys

People are complaining on Reddit all the time and I never know if it’s true

[balmy olive]

nah codex is the best

[rocky fog]

people do something else on reddit?

[ember spire]

people are complaining about codex usage limits on reddit? compared with what?

[torpid trout]

Yes, complain about Google „changing how antigravity app looks without warning“ - a whole tirade of crying out loud babies that installed the wrong app months ago and now cry because it finally became what it always was supposed to be lol

[amber scaffold]

I was seeing posts that it was less then it used to be

[ember spire]

they've reset them 3 times at least since i subbed too 😆

[rocky fog]

thats still complaining 😄

[ember spire]

maybe it is... only been using it a week 🤷‍♂️

[torpid trout]

Yeah - just confirming through false denial lol

[amber scaffold]

Yeah IMO the GPT models are the smartest but I don’t want to have less usage. I know Claude gives 5k usage. I’ll try to measure Codex

[storm cedar]

Use Claude, it's much better.

[amber scaffold]

Claude definitely has a better personality

[torpid trout]

Right,5k what - 5k tokens?

[lean lark]

Perhaps we should agree that people are about 50/50 on complaints on all sides?

• This model is better
• This tools is better
• My dog is better

[torpid trout]

You barely could write hi with that

[torpid trout]

No, my dog is better

[storm cedar]

My app was made entirely there.

[torpid trout]

And definitely this tool

[karmic gulch]

Even Grok can make apps

[lean lark]

Personality is a user setting!

🙄

[amber scaffold]

Cursor estimated that Claude’s 20x plan let you use $5k worth of tokens, due to their subsidization

[storm cedar]

But Claude is better.

[torpid trout]

That’s ridiculous

5k tokens aren’t enough to say hi to an llm of that size

Oh wait

Sorry

[plush plover]

So what's the most stable model so far?

[torpid trout]

5K USD LOL

[karmic gulch]

Dude, no one will ever know the true cost. Next week Claude can state that a 200 sub is worth 10.000k but how would anyone outside their inner circle verify?

[storm cedar]

It's already been 12,000 here.

[amber scaffold]

You just need to use up the weekly limit to 0% and see how many tokens you got

[lean lark]

Every platform is having ups and downs, especially when they're competing for who can get out the most daily updates. You can't make decisions about monthly subscriptions when the products change daily.

[torpid trout]

Thats impressive man
Tell that to the 50.something million tokens that go in a pro gpt plan daily on low usage 🫣

[amber scaffold]

I wish I could have weekly subscriptions

[ember spire]

I very quickly came to the conclusion that Codex is a better tool for developers than Claude Code

[rocky fog]

good question
what do you folks use?
personality = "pragmatic"

one time I told codex how well it has done something and praised it and it didnt even flinch, just completely ignored it 😂
( but thats fine, was just wondering if its still a bit like 4o xD )

[bright swift]

"none"

[ember spire]

yes pragmatic, and the custom instructions have personality guidelines to reinforce that too (inherited from my CLAUDE.md)

[solid lake]

Money

Dollar

[torpid trout]

Yeah yeah got it later on

[lean lark]

While I do default to pragmatic I always supplement/fine-tune a bit.
I'm guessing most people leave it on default and then leave because they don't like it.

[torpid trout]

I’ll price my tokens at 6k each

[solid lake]

Bubble

[torpid trout]

Then my plan offers 6busd tokens

[lean lark]

I never use the crazy ChatGPT personalities.

[solid lake]

Yeah mines always pragmatic

[storm cedar]

It's already been 21,000 just from a translation change from Russian to English.

[lean lark]

If you don't like the pre-defined personality (pragmatic, etc), look in a session*.jsonl file to see exactly what text is inserted, then use AGENTS.md to override/replace/augment very specific traits ... like "laugh at my stupid jokes, say something funny once in a while".

Anyone see this? https://discord.com/channels/974519864045756446/1506753484085657612

[solid lake]

Are you translating manga panels

What’s accuracy like

[storm cedar]

page translation

[solid lake]

Oh sick

What model

[storm cedar]

I'm using OpenClaude, and building the rest from scratch.

without using APIs

[solid lake]

What model in openclaude?

[storm cedar]

just a minute

[solid lake]

I see this

New nemotron?

Ohh it says any gguf model

[lean lark]

And for those who don't know, you can use non-OpenAI models with Codex. However, the Developer instructions that are given to the assistant are the same, so if you believe a model is competent, just not in Codex, changes are that the instructions are corrupting it.

[storm cedar]

I'm using this model.

[solid lake]

Mimo

Xiaomi model

Nice

Try new nemotron model if gguf has

[storm cedar]

I'll test it.

[plush plover]

How expensive is it?

[storm cedar]

I'm using it for free.

[plush plover]

Does it work nicely?

[storm cedar]

I completed my entire application on it.

[boreal holly]

Yeah, for some non-OpenAI models, they do not have <|system|> prompt training, and instead only <|developer|> prompt training, so they actually treat developer instructions with higher precedence than system. This makes codex work kinda funny with certain 3rd party models. Also some models use XML for tool calls instead of JSON so they'll run a tool by outputting XML which does nothing and finishes their turn 🤡

[lean lark]

I believe a couple providers are standardizing or at least accepting OpenAI APIs to accommodate that, but yeah, it's all in the air.

[proper cosmos]

The chatgpt Codex mobile app functionality - ", all from the ChatGPT mobile app. Codex will keep running on your laptop, Mac mini, or devbox."

How the hell do you get it to work with windows?

Does anyone here successfully have it working

I want to do my vibe coding crap just on my phone while I lay without having to remote desktop all the time

I just want to lay and be lazy in comfort!

[storm cedar]

English translation from Russian

[lean lark]

OpenAI has not announced support for Android>Windows. That is as-yet not supported. It was "accidentally" working a few days ago but an update to a couple of the components fixed that. So now, no, there is no Android>Windows option - yet.
As with any other product, including your own, it won't be announced until it's ready, so we just need to sit tight and have patience.

It's kinda a Wild West out there for AI right now.

[rocky fog]

https://openai.com/index/work-with-codex-from-anywhere/

[storm cedar]

English to Mandarin/Chinese translation

Making an update to use gemma4 to improve the translation.

[sharp yew]

/codex how do i take a screenshot?

[fleet geyser]

If you want codex to take a screenshot it can use Playwright to view a page or it can take a screenshot of your screen

[cyan wing]

ouch. Codex crashed on WSL. Can't re-open it.

[lean lark]

You're asking Codex to reach out from WSL to the host OS. That's not good. Windows can reach into the WSL sandbox but don't let WSL reach out, especially not for config files.
Install Codex separately into WSL with npm or using the standalone server install.

[cyan wing]

idk i just did the default use WSL option on Windows Codex App

i did install Codex separately onto WSL..

been using it for months now

[lean lark]

Someone at OpenAI might say different, but I strongly recommend keeping the environments separate for now. I've had some challenges there and there's not enough info to know what to expect, no diagnostics about where it's looking or why, and I don't have time to report all of these concerns to the GH repo.
So for now, I've resigned to just keeping them separate.

Letting Codex outside of its sandbox is definitely a bad thing to do. Unless you've specified, /mnt/c/* is not sandboxed and an accidental rm on that path can ruin your month.

[boreal holly]

Looks like OpenAI's Codex agents changed a migration file that was already applied. You updated Codex for desktop, it compared the checksum of the migration to the one stored in the sqlite file, and it errored out.

The only recovery path here is either rename the sqlite file (backup) and start fresh (lose some existing agents and configs or whatever), or if you're feeling up for it, performing surgery on the SQLite file. Editing the migration checksums or dropping the table so it reapplies the migrations (dangerous on sqlite)

This is a common sqlx problem. It stores a checksum of each migration file in a migrated DB. If one of OpenAI's codex agents even adds a blank space to an existing migration file, it will change the checksum and cause the migration to fail.

[solid lake]

Checksum is very nice for agent ownership

[boreal holly]

What really sucks is if OpenAI puts out a hot patch, reverting that migration file, anybody who ran the migration successfully will experience that error on the next update. It's a hot mess lol

[boreal holly]

Nah, should use dd -if=/dev/null -of=/dev/sda lol

Gotta clear all the bits on the drive first

[storm cedar]

Making an update to improve the translation using gemma4.

[quaint prism]

what goblin mode do 😈

[lean lark]

Devs throwing Codex at different models to see how they perform ... except for Grok.

[storm cedar]

I decided to price my application at $10 per month. With that, users can translate as many manga as they want during the month.

I also made some updates and managed to improve the quality a lot, to a level where I feel it’s good enough to charge for the service.

Initially supporting three languages English, Russian, and Mandarin.

[lean lark]

I'm VERY happy for you. It's great to see someone actually do something good with this fine software.

[storm cedar]

I'm happy too, I never imagined I'd do this.

[lean lark]

Caution: Consider if someone resells the service to his friends or in a private service, taking money for images, and consuming a lot more than $10 worth of your API budget.

All API applications need a valve that limits individual users and avoids abuse.

And you asked us "what would you pay?"
That's not a good question for us, now it's a good question for your audience, people who hang out in manga forums.
Are they willing to pay $120/year for their comics?
Maybe some are, I'm guessing most will not. 🙁

[plush harbor]

yeah rate limiting is not optional. Humans are terrible creatures

also don't expect high numbers. I've got a by subscription site. I have like a dozen subscribers ...

[gentle harbor]

oh my god bruh what did they mess up

this happens every time i click into codex now

[lean lark]

ya gotta be more specific...

[cedar skiff]

I haven't had an update yet, i think ill hold off 0.0

[lean lark]

Problem: "Codex (Not Responding)"
Solution: "Find the circuit breakers to you office/home. Turn everything off. Wait one hour. Turn everything back on. Restart your system and try again."

[visual steeple]

Idk waiting on OpenAI codex team to add goblin mode would be so awesome

[torpid trout]

Shush, don’t mention the goblins

[solid lake]

I’m still on 30 rn

Haven’t updated

[torpid trout]

https://openai.com/index/where-the-goblins-came-from/

[quaint prism]

😈 😈

[torpid trout]

Reading that post, 5.5 had actual humor
I wonder how memes would come out without that system prompt

[visual steeple]

OpenAI leaning into the Goblin stuff is what makes it great, LLM virus coming out of its shell and taking over in a memeatic way

[storm cedar]

I'm not using an API, it's pure code.

using gemma4 as a base

[lean lark]

Ahhh, that does make a difference. However, for $10 your users may expect faster processing. Gemma4 isn't going to provide that. You may need to migrate premium users to a higher-tier/faster/hosted service. I suggest you allow for that growth possibility.

[storm cedar]

Yes, there will be a plan for 15 where Essew will be on an exclusive server.

[lean lark]

Good luck to ya!

[quasi summit]

anyone else getting the macOS 'allow' permission prompt for recording the screen/audio repeatedly when using Computer Use in Codex? sometimes it fails to get permission

doesn't it normalize image input sizes

[ember flame]

any one wanna team up for a hackathon?

[boreal holly]

Open up system preferences and manually enable it

[magic flower]

thats funny i join to this message, i came to post it having some humor

finally: AGI

[lean lark]

Inappropriate. Find a different forum.

[quasi summit]

i did

[woven canyon]

Why is the token usage display gone in the new update bruh

[plush harbor]

past usage? Been missing for a couple days now

[woven canyon]

No, the context window's token usage isn't showing up now

[plush harbor]

breaking more charts and stats then i see

[lean lark]

I don't know which window you're looking at but on the openai.com site there are toggles for week, month, year. In the apps too? I dunno.
You might not have recent activity for the specific environment you're looking at. So for example, there are no events for this week and last week's events have scrolled out of scope.

[cedar skiff]

the personal usage section is empty on mine here https://chatgpt.com/codex/cloud/settings/analytics#usage
the other ones are still there

[lean lark]

I don't have that up in front of me but I believe the quotas are account-specific (including cloud and all systems), while any detailed stats there are only cloud-specific.
Session/Event/Transaction details are in each system under the .codex folder tree. I don't know if they aggregate that data for us anywhere.

[plush harbor]

the personal usage one is just outright broken. Whether that means other stuff is broken too, who knows

[cedar skiff]

this section here is not updated, just seems missing

The other sections are fine

[plush harbor]

I reported that one as a bug today, cos its been missing a couple days

[lean lark]

Suggestion to all: We gotta be professional ...
Look in the GitHub repo Issues for keywords.
If you see the issue, 👍 it.
If you do not see the issue, create a new one.
Get some peeps here to actually verify a specific issue and 👍 it. Do NOT 👍 an issue that you don't have.
Follow-up, answer questions. Participate in FOSS.
Reporting issues in Discord is a waste of time. Don't bother. This environment is not official. Staff do not officially visit here and do not take issues back to the office with them.
You Must report issues on GitHub and/or to OpenAI via Support/Help Center.

-- This has been a Public Service Announcement

[plush harbor]

not teh report a bug interface in the app?

openai's charting in general is really bad

[lean lark]

That is the correct place but there's never a ticket# or confirmation. I trust they process that data but we've got no handle on it to "know".

[cedar skiff]

I'll be honest I only ever go to github and check for issues for things that are blockers for my work so i can find work arounds

[lean lark]

☝️ most people same ... but that doesn't help much.

[plush harbor]

woudlnt even know where to look for issues with the websites

[boreal holly]

@Tibo on X haha

[cedar skiff]

It does help because i thumbs up the blockers.

[plush harbor]

this is my favourite charting issue. Paying for 0 tokens

[cedar skiff]

and create issues when there isnt one

[lean lark]

For all of the billions od dollars that this company has available it's really amazing how dense they are about basic things like communications, messaging, information...

[cedar skiff]

but i wont do it for this bug, because this one is not a blocker for me and it's so public it's probably on purpose

[plush harbor]

its particularly bad for this kind of stuff, I don't get it. This is public facing human interaction surface

if you dig down and change a few settings you can see the actual numbers so that backend isn't broken, its just set to default project only not total. Then you reload and it undoes all the settings you had to change to see your numbers. Sigh

[boreal holly]

Have you tried looking at the logged responses? It will sometimes show the most recent responses in the logs, and other times it shows everything but the last month. Hitting refresh periodically will occasionally toggle this invalid state lol

[plush harbor]

nah its obvious what's wrong with it, which is what makes it worse

openai is just consistently bad at anything with charts in it

[boreal holly]

I'll be like troubleshooting custom agents, see all the most recent responses in the logs, hit refresh, all of those logs disappear and can only see last months logs 💀

[lean lark]

I try to avoid posting this but what you're describing is exactly the reason why I wrote this: https://github.com/CaptainStarbuck/codex-usage

[plush harbor]

openai be like ... trying to track something? Yeah no

[lean lark]

No, things are just broke, like your software or mine. And the only way they will fix the stuff is if they have a ticket they can process. Be professional. Create and process tickets.

[plush harbor]

where do you go to even lodge tickets though?

[cedar skiff]

github issues

[lean lark]

https://github.com/openai/codex/issues

[plush harbor]

oh this isn't codex, tehre's charting bugs for codex use, api use, and one in teh chatgpt web

whoever is their chart guy at openai needs booting. There probably isnt a chart guy though

[lean lark]

Oh right, forgot, well, file it there anyway. 🤣

https://help.openai.com/en/

[plush harbor]

I'd complain about the codex usage not being in my timezone but since it doesnt' work right now its techically fixed, right?

imma go fill my bins before I miss them because of discord and then obsessively lodge bug reports about charts

[lean lark]

https://help.openai.com/en/articles/6614161-how-can-i-contact-support

I used to be a Guide here and I have a FAQ about how to report issues with OpenAI. If you're interested I'll dig that up.

Here, this is how you communicate issues to OpenAI:
#1415374642843095183 message

[plush harbor]

how do you get access to that?

[lean lark]

ChatGPT. 🙂

[plush harbor]

its #unknown so wouldn't know what to ask chatgpt 😛

[solid lake]

Click on it

[lean lark]

Don't ask ChatGPT about an issue in another page... Go to ChatGPT, open Settings>Help Center...

[plush harbor]

nothing happens when you click on a dead discord link

[solid lake]

It takes me to the thread

Shows me unknown too

[plush harbor]

this just means you don't have access

[ember spire]

it means the thread name isnt in your discord cache. if you have access, clicking it and opening the thread will populate it

[plush harbor]

nm I probably clicked the reply ... I just went outside and did 5 minutes of manual labour, I'm wrecked. And the internet and international tourists are telling me there is an "8 week" or "2-3 month" respiratory thing going around so at least I know what I've got and why the only thing i can do in my life right now is sit at a desk

[ember spire]

i think everyone ought to have access to #1070006915414900886 though

[plush harbor]

I've put quite a few things in #1070006915414900886 and a whole one of them has been fixed

[lean lark]

As I said - it's a waste of time.

[ember spire]

I'm trying to see if anyone's reported that right-clicking the top edge of the codex macos app crashes it 👀

[lean lark]

#1070006915414900886 and #1053565716500074526 are places for people to go to and maybe feel better as they cry into the darkness.

[plush harbor]

fairly accurate description of discord bug reporting yes

[lean lark]

Like I said, staff doesn't officially go there.

This is a public resource, not an official vector to the company.

[plush harbor]

I'm a serial reporter of bugs. I used to live in a 7 storey apartment block, on the 7th floor, and the lift constantly broke. Broken in the morning, still broken in the evening, still broken the next morning. Fine, I'll find the damn number and see what's taking them so long to fix it

... I was the first person to call in the issue.

Software follows the same pattern

[lean lark]

So ... follow the instructions in that thread, go to the Help Center, make your case to the bot, Escalate!, you will get a kind human soon.

... I was the first person to call in the issue.
Software follows the same pattern
Humans be Humans

[ember spire]

💪

nice "if this is actually serious put it on github"

https://github.com/openai/codex/issues/22723 sweeet

[lean lark]

That's not what it says.

There is meaning to that.

If you find or create an issue on GitHub, you can refer to the ID of this submission as evidence of the issue. It's more detail.

I see your +1 there. Add the ID from that report if you wish. It may help to find the issue in a similar or different platform.

[plush harbor]

dang this help bot interrogates you. Also if you look in the network tab that "no data for this period" is actually returning a full set of data for that graph, just every subtool value is 0

[lean lark]

There ya go! You're on your way to creating a good report and getting a fix.
Remember the follow-ups:

• Make sure that it recognizes the problem and confirms that the behavior is unusual.
• The final step that most people miss : Tell the bot that you now want the report to be sent to a human for processing. If you don't do this, it may not be processed for a while.

[plush harbor]

I didn't have anything to get codex to do so had to go halp codex, I need to use some tokens

[lean lark]

You need to separate yourself from the plebs: "wah 😭 it don't work!"

[plush harbor]

sign out. sign in. use tokens. wait 15 minutes. drink coffee. refresh page. Am on second last step now

[lean lark]

Seriously though, I hope this gets you to that point of "oh, cool!" and then you can share it with our friends here when you get a chance.

[plush harbor]

not like this is a me-specific bug

[lean lark]

I don't even know/remember what you're talking about and it doesn't matter. If you have an issue, report it effectively. That's my only point here ... that one and the one on my head.

[plush harbor]

codex usage chart no go workie

this is possibly related to that other guy who says usage isn't showing for something else if they've gone and borked something. Which sounds like openai

should donate some of my quota to openai to fix their bugs. I seem to have fixed all mine

[solid lake]

I’ll take some

[ember spire]

they keep resetting it

[plush harbor]

also that

[solid lake]

Yeah to keep attention

[plush harbor]

but I haven't done much in the last couple days either way, I've been doing content/design and that hasn't hit codex

[ember spire]

yeah it's working, lots of us switched from CC recently because of the usage limits, and sama's obviously leaning in https://x.com/sama/status/2056804900017947046

[solid lake]

Promo almost over

[ember spire]

10 days? 🙂

[potent mason]

Bruh how

I'm at 30% already

[plush harbor]

cos I'm just ... not using it right now. Most of my coding tasks are blocked by content migration and that's taking a while

[lean lark]

New proposal for OpenAI - treat tokens like bitcoin currency. Only tokens gained on a paid account can be transferred to another paid account.

[ember spire]

make some custom pets 😄

[solid lake]

https://tenor.com/view/the-lorax-oh-thats-bad-not-good-thats-bad-dr-seuss-the-lorax-gif-10503050063809228840

[lean lark]

I've been using ChatGPT projects recently to create games, simulations, science, design of new gadgets... I'd love to do all of that with Codex but I don't want to burn tokens. You can do something like that. Create or fix some FOSS that you depend on ... or even that you don't. Save the whales! Eliminate poverty! Use those tokens for the good of humanity!

[cedar skiff]

How do you find codex compared to claude code?

[ember spire]

honestly I love it. It took to my development workflow like a duck to water. The plans are more concise, it's more diligent in reading the codebase resources (wiki, code, task-specific artifacts), and using research subagents.

The model itself seems to follow instructions better than 4.7 and still be as smart as 4.6 (I found a definite trade-off there before I switched).

The context compaction is also incredibly good. So much so my workflow changed so I'm not starting new chats as I transition from brainstorming, to speccing, to planning, to implementation, then review. It's so good I am wondering if they are using the full 1M context for some kind of long-term chat summary with a 258K working context.

I also love the app. I thought I'd miss the CLI but actually the macOS desktop app has replaced most of my Cursor usage now. I'm only using Cursor to browse projects.

It's a lot easier to track progress across multiple chats too, and keep them organised.

[cedar skiff]

I noticed a similar thing with my workflow coming from claude, half of it went away over time because it was handling claude short comings

[ember spire]

how about you? you switched recently too didn't you?

[cedar skiff]

welcome to the light side 🙂

I switched when 5.3 came out

[ember spire]

this is on point and I'm still adjusting.

[cedar skiff]

We dont get the consistency problems here either

[ember spire]

I think when they sort out the character-level diffs and file tree view this is going to be how they design IDEs for agent-assisted dev in the future

[cedar skiff]

claude code has entire weeks where it acts like a model from 2024

[ember spire]

or just straight downtime for what felt like every day in April 😄

[storm cedar]

I made this model because I think speech bubbles are the way to go.

[cedar skiff]

Pretty cool project

i love the idea

[plush harbor]

geez I just tell the little ascii thing to do things and it does them and nothing much interesting comes of it

[lean lark]

To get the most out of the tools consider:

• There are Codex Instructions in the Cloud environment, separate from the Apps.
• Of course that means the Apps have "Codex Instructions" that you should use as well.
• Each system has its own AGENTS.md. You know the cascade...
• Codex does not use ChatGPT instructions or data. It's completely separate.
• Codex has a defined Personality setting which affects behavior.
• Codex has Developer Instructions that also affect behavior and so many other details.
• All of that is above the base model.
• For models, you select the model itself, optimization from Standard to Fast processing, and reasoning level of Low, Medium, High, XHigh.
• And then there are your prompts and local files.

All of those factors ... and your Skills folders/files ... contribute to everything that these tools do with every turn.

When people complain about responses, I SMH because there's no way we're going to know what all went into the processing of any one prompt or session.

And with that, I'm off to dinner. 🥑

[plush harbor]

and left unattended with no direction it really, really likes beige

[potent mason]

Oh right the personality thing, what are you guys using for that?

[plush harbor]

default cli which just gets the job done without being chirpy

[potent mason]

I've used computer use on the cli, but can you use the chrome plugin? (I tried and it failed every time)

[plush harbor]

I'm running old hardware. Anything modern and fancy tends not to work for me

[potent mason]

That makes sense why you use the CLI then, the app is not very resource efficient

[ember spire]

Pragmatic. I've also long had this instruction to try to approximate that in my CLAUDE.md. Now it's in my AGENTS.md and the personalisation section (because this is really important!) 😄

[plush harbor]

got the intel mac app. Its laggy as hell and I don't like its personality

[potent mason]

Loved the edge cases mention, I have TDD for mine, but even then it tends to skip important stuff

[ember spire]

yeah I noticed this too. And at a few points I've had it become unstable and crash as the resource consumption came to noticeably impact performance

still... it beats the days of when the claude code tui was built on a diffing scene graph and running two long-running sessions in iterm severely impacted performance system-wide for the only time ever on this M4 Pro (mouse cursor and other apps became jerky too)

I get the fans spinning up in long-running sessions in Codex though, even with just one chat working

that was pretty rare before

[cedar skiff]

pragmatic, i also have this modifier in the developer_instructions (among other things):
This really helps cut down on these tendency that codex models tend to have when rewriting/writing documents or answering questions

## Mandatory Output Gates

### Avoid Defensive Contrast

Do not add contrastive boundaries such as "not X," "does not Y," or "this is not Z" to defend against hypothetical misunderstandings. Use contrast only when the user asked for that distinction, showed that misunderstanding, or needs it to avoid a concrete technical error. Otherwise, omit the contrast and state the positive case directly.

### Prevent Implementation-Context Leakage

External-facing documentation, UI copy, commit messages, and public repo text must not be shaped by internal constraints, task history, migration details, local workspace boundaries, omitted files, or agent reasoning artifacts. Write only from the artifact's audience-facing purpose.

[ember spire]

I like that last part

[cedar skiff]

The first part i did because when ever i get it to analyse something it always comes back and tells me what it doesnt do first. I find it really distracting and low value.

[potent mason]

Quick question, has anyone used IBKR API's or some way of interacting with it programmatically?

I need to automate some stuff but it's being annoying.

[solid lake]

Ibgateway

[valid bear]

@cedar skiff finally figured out why my tokens were disappearing https://github.com/openai/codex/issues/23791

[cedar skiff]

is this fixed?

That's a pretty bad bug o.0

[valid bear]

for me yeah, but i imagine others are dealing with the same issue

167k codex sessions spawned in the background over the past 2 weeks

[cedar skiff]

oh yikes

[valid bear]

i’m wondering if this part of the reason why they keep randomly resetting usage limits

https://github.com/openai/codex/issues/14593 i’m not alone

[nocturne folio]

on god what is codex cli injecting

this is so strange

[gentle harbor]

is it just me or is gpt way slower ? taking 3 min for a simple patch file

[nocturne folio]

this is what its originally

the cli is injecting html for sum rsn

[olive sandal]

pro plan, 70% gone in 1 day 🙁

[woven canyon]

dayum

[ivory zodiac]

not as far as i'm aware, it could mess up click xy location when it doesn't use ax, but dont quote me

[ivory zodiac]

thats what you get

for using windows

[gentle harbor]

who ever made this latest codex windows update should be fired

[nocturne folio]

get a mac

[tiny laurel]

https://tenor.com/view/pc-vs-mac-pc-and-mac-mac-vs-pc-mac-and-pc-pc-gif-4687278682063171896

[gentle harbor]

just use the codex beta its on a older version and takes up less space 🤸‍♂️

[gentle harbor]

i have a mac, the stuff im doing now only works on windows 👨‍🦽

[cedar skiff]

run a vm🍌

[somber sonnet]

why is it asking for permission

even though its set to full access

[potent mason]

That won't solve windows being annoying

I think the best option is just develop on mac, pull on windows, test on windows

[cedar skiff]

it holds up with get a mac tho 😛

[potent mason]

I've also done that unfortunately.

I think I managed to get it functional on Linux after some time though so maybe try that and ssh? (Can't really say for certain though since every use case for windows is unique)

[indigo robin]

compation is broken

[lean lark]

Ya know, the App is just a GUI. It's not "necessary".
Use Codex Cloud, CLI, or the VSCode extension.
They all work great.

[indigo robin]

how can i remedy this compation problem ?

maybe i need a re install

[plush harbor]

have you been in one chat this whole time

[indigo robin]

yes

[plush harbor]

do you need to be in one chat forever?

[indigo robin]

why wouldnt. ibe im working on a app

[plush harbor]

I treat chats like jira tickets. One chat per issue. Start, fix issue, if its properly fixed, commit and end chat

[indigo robin]

really ?

[lean lark]

PIC, create a new session for each task, not each project.

[indigo robin]

thats te best way to use the agent ?

[plush harbor]

I got documentation in files about the project, its got no need to carry on one thread forever

[cedar skiff]

the app is a much better interface to work with. It has one click push and pr creation, automations, it lists all your threads in project structures on the left, you can visually see which ones are finished and unchecked, which ones are still working. You can instantly fork into work trees, one click persistant work trees etc. It's just a better interface.

[lean lark]

which "the agent"?

[indigo robin]

so what am i supposed to do ?

[plush harbor]

I don't think there's one overall "supposed to" but some stuff definitely works better

[lean lark]

... if and when it works. For those who are having issues there are all least three other options.

[indigo robin]

should.i use the fork button ?

[cedar skiff]

It works flawlessly for me, i guess i'm lucky

[indigo robin]

one task per chat is the best way?

[plush harbor]

I mean you don't have to be perfectly 1:1 but seems better than one chat per project

[indigo robin]

isnt the point of the 1 thread to mantain consitency , thats why they created compaction

[lean lark]

@indigo robin

• As a fellow musician it looks like your working on something Very cool.
• Open a new session. Tell it to document the project. In separate files under ./docs.
• When done, check the work, commit/push.
• Open a new session. Tell it to document the code functions. - When done, check the work, commit/push.
• Open a new session. Tell it to add one enhancement. - When done, check the work, commit/push.
• Open a new session. Tell it to fix a problem.
• When done, check the work, commit/push.

See the pattern?
Enjoy.

[indigo robin]

ahh i see

[plush harbor]

I've burned through a couple projects since I caught this stupid virus and the only time I have a longer chat is when I move to a new project, and that tends to end with a bunch of documentation being created so it knows what its doing next time

[indigo robin]

yea its a muisc theory / learning / chord library program

[lean lark]

In a single session, check the work, ask about choices, test and get it to fix what it broke. Get it to tweak docs or do housekeeping. Then wrap up and move on. 😁

[indigo robin]

i see

[lean lark]

Compaction is good for truly long, necessary sessions, but it adds to the burden for a while and there's still a lot of clutter in there. It works better without the clutter ... like the human brain refreshes with sleep.

[indigo robin]

for sure , its already 5x faster

im going to fight my way to a stable build then have it make documentaion and explain my entire app and its functions

[plush harbor]

nah get it to document as it goes, its actually good at that

[lean lark]

❣️

[indigo robin]

mmm i see

im honestly already very deep into the app

[plush harbor]

every time you talk to the thing and go "oh you get exactly what I mean" follow that up with "write it down somewhere"

[indigo robin]

i see

good edicate

i need an agent to do that for me lol

[plush harbor]

codex. Literally. Give it instructions to do that

[indigo robin]

true true

i told codex ot make me a hand off for a new thread

[plush harbor]

also you probably want to be a bit more specific than "write it down somewhere"

[indigo robin]

and when i gave it to it

i told it to do a read only pass and make sure it can implement the new feature

[lean lark]

I use AGENTS.md files to direct the assistant to keep docs current, lint, test, format, make sure it cleans up its own errors, and then it documents new things that it found. One effort, one set of instructions, and all projects work the same.
Leeann to use the tools. 😁

[indigo robin]

i had an agent like that befor for my unity project

i made a pretty powerful agent

i made it read the entire unity documentation and only speak in the language written in the documentation

[plush harbor]

just mind it can be proactive how you don't want it to be soemtimes. I got it to rewrite one of my really old sites from jekyll and when it was done I commented that the text on my personal projects page was wrong and the blasted thing went off, found the repo for my personal website and changed the text and I did not ask it to do that

[lean lark]

C#?

[indigo robin]

no i made it read the entire unity documentation

words

terms

[plush harbor]

that sounds like it should be a skill, like the openai documentation skill

[indigo robin]

i made it only speak and create code based on the language of the documentation

[lean lark]

The API?

[indigo robin]

and label everything correctly , so i could pinpoint issues

[lean lark]

That's called code documentation. 😉

[indigo robin]

ahh i see

horrible lmfao

new thread performed horribly lmfao

[lean lark]

Note above, create docs as .md files, and then document the code. Then it'll used the docs to understand the project. (You need to instruct it to do that.)

[indigo robin]

about to compact the hell out the old thread

[plush harbor]

my newest site I wanted it to do a thing to edit something like categories, and I commented surely there's something already out there so I don't have to do this myself and it just ran with that ... found the data online, sorted it, made an editor for it, and extended it as per my planning documentation

[indigo robin]

dam i think this possibly just ruined my progress

[plush harbor]

you aren't documenting enough then

[indigo robin]

no the compation should work as intended

[plush harbor]

compaction is lossy by definition, this is why you document

[lean lark]

Give it the information it needs. Tell it to document everything, keep docs current, use the docs to understand the code and general functionality.

[plush harbor]

then next thread it reads in your nice, concise, structured documentation instead of a long messy chain of compacted consciousness

[lean lark]

If you just started a new thread/session there's should be no compaCTion.

[indigo robin]

lord this just messed everything up

smh

[plush harbor]

just take some time in the old thread to make some documentation. Capture your coding style and quirks of how you want it to behave in the agents file, maybe another one abotu what you need to run your app, another one for what your app currently does, and another one for planning what you havent' done yet

if you keep making it read in some third party library that should be a skill that it uses only when it needs to

[lean lark]

There's also the concept of a handoff, a form of memory: The assistant writes important notes for itself about what it was doing, why, challenges, things to do.
The next assistant reads that and understands what's happening now, and reads the docs to understand the project. The it takes your current task and it knows what to do with it.

Training a new employee is always tough. Do it and it'll work for you.

Bedtime for Bonzo here.
Gnite buds. 💤

[plush harbor]

you can also literally ask codex "what do you need to understand this project better and be helpful for you"

[cedar skiff]

compaction is fantastic in codex it just works, you can basically just treat it as endless context

[plush harbor]

probably still don't want to rely on the context as your single source of everything about a project though

[cedar skiff]

what do you mean? All the agent has is context.

[plush harbor]

you can also put documentation in files, like agents.md etc

[cedar skiff]

ofc we do what we need to in order to get the results we want. I mostly use skills and codebase research.

[plush harbor]

this chap just seems to be relying on this one chat for his project, with everything in context only

[indigo robin]

Ruined like 7 days of progress

[cedar skiff]

using git will prevent this sort of huge failure

[indigo robin]

How can I do a larger compaction ?

It’s not compacting correctly

It’s only removing 10k tokens

I’m going to have to rebuild this entire thing from scratch now

[cedar skiff]

there is some sort of misunderstanding going on i think?
You should NOT be using a sessions as a storage space for work. It is a tool to perform tasks, those tasks output the work into tangible places, like images, and files etc

you can fork your conversation at any point over and over again.

[indigo robin]

And what does forking do?

Codex works best when it’s in one thread and has the previous context to build off

[cedar skiff]

it creates another chat that is a clone of the forked one up until the fork

[indigo robin]

Wouldn’t that carry over all the tokens

[cedar skiff]

Not sure what you mean here, it will go through compaction and it is not designed to just hold all your work. It wont work that way.

Work that is in the context will suffer semantic diffusion every compaction, but if you are outputting your work to files codex will be able to reference those files.

[indigo robin]

I’d have to remake the app from the ground up

I haven’t done any documentation

Using a new thread was the worst advice possible

It had no idea of my project even with the handoff

[plush harbor]

ok this begs the question - where is your app? Like how do you have it saved?

[plush harbor]

also, this 👆

[indigo robin]

Its an Xcode project in a folder

[plush harbor]

are you using git?

[indigo robin]

Brother obviously I didn’t

[plush harbor]

the last highly experienced nerd went to bed, you're going to get schooled by Eric now

[indigo robin]

I don’t need your help right now respectfully your advice was cool but it was misplaced

[cedar skiff]

This doesnt track with how things actually work. Your app should still be in-tacked out side of the session. You dont have to make it again, you just have to tell codex what you want to do with it.

[plush harbor]

your compaction issues are a symptom of how you're doing things

[indigo robin]

Explain

[plush harbor]

a new thread should just carry on happily, already knowing what it needs to, shouldn't feel lossy

[cedar skiff]

what do you think a session is?

[indigo robin]

In my previous thread it was making things flawlessly

This new thread it did all the wrong things

[cold galleon]

I’m trying to use Codex Desktop → Settings → Connections → Control other devices to control my Mac mini M4 from my Mac mini M2.

The M4 has Allow other devices to connect enabled, and my iPhone can successfully see and control the M4 from the ChatGPT app. However, on the M2, the M4 appears as “Connected” but the status dot stays gray instead of green, and refreshing shows:

Failed to update connection

The M2 also does not show the M4’s remote threads or any clear way to switch into the M4 host environment. New threads on the M2 still run locally.

Both Macs use the same ChatGPT account and workspace, and the M4 is awake, online, and running Codex App.

Is this a known Mac-to-Mac remote control issue? How can I make the M2 access and continue the Codex threads running on the M4?

[indigo robin]

makes no sense really , if the thread is constantly having to read and save read and save , your documentation becomes the tokens and floods the thread immediately

if i told a thread to read and commit the entire blender documentation to memory it would max the thread out

[plush harbor]

you absolutely do not include documentation for third party libraries in your context. You get codex to search them on an as needed basis

you see it use its own skills ... I have a couple sites using openai api and when it goes to make code for that it runs off and looks up how to do it using that skill

[feral charm]

Codex is down

[indigo robin]

no its not

[feral charm]

Doesn't work for me in opencode

Says 5.3 not found

[indigo robin]

working fine for me now

[supple perch]

5.5 and 5.5 pro are working in opencode

[feral charm]

huh 5.3 works in codex app. Must be opencode issue

Random q but is anyone here good at pdf parsing for tables

[plush harbor]

you wanna save tokens on that, get chatgpt web to do it

[feral charm]

I can't get 100% accuracy with python pdf libraries because columns get combined

[indigo robin]

managed to get it a bit closer but now i need to wait 3 hours to resume cause i reached my limit

[feral charm]

I'm writing code for it though but wondering if I'm on the right track

[indigo robin]

this is what i need

this is where were at

i guess i can power through and patch the leaks

[plush harbor]

probably something within the columns messing with escaping, just as a wild guess if you've got columns merging

[feral charm]

@indigo robin what are you trying to do exactly

Making a game?

[plush harbor]

educational music software

[indigo robin]

im making an extremly comprehensive music theroy app

that has a sequencer

chord library

things like thatr

for guitarists ofcourse

[cedar skiff]

If i have a project that relies on a package i always make a full up to date skill for it for that project.
It has two main sections - the api section that is the full up to date api for that latest version and a implementation section that gives the details on how the project uses the package. the implmentation sections is mostly built over time.

[feral charm]

it's because pdf parsing libraries cannot accurately tell where columns start and end. Pdf encoding is fugged so it's all a guess. Hence the need for ocr to detect column borders

[indigo robin]

it may evolve into a mini daw

looks soo good on iphone

[cedar skiff]

this is what my skill usage looks like over a week

[feral charm]

That is cool. Githits.com has a mcp server for this, it also stores documentation on past versions for libraries u use

[plush harbor]

last time I tried to look inside pdfs I got scared and ran away, so yeah turning them into images and getting your favourite ai to parse them as images could work better

I had chatgpt turn me a heap of game screenshots (rows of ranking data, tables) into html one time, did a very good job of it

[feral charm]

*ocr just for the column coordinates

Passing in entire image into ocr confuses the model and at most achieves 95% accuracy with gemini 3.1 pro

[indigo robin]

open ai needs to make an automatic context agent that tracks your project to then handoff to another thread

[feral charm]

Just the use case right now for me requires accuracy to be as high as possible. One wrong word can cost 5-10k if missed during manual review

[indigo robin]

im sure someones made one already

[feral charm]

matt packock has a /handoff skill for this

He also has a /to-prd skill

[indigo robin]

send link

[cedar skiff]

imo the only reason he has that is because claude code is so bad at compaction and he is all about claude

[indigo robin]

https://github.com/LexLattice/adeu-studio what about this one ?

[plush harbor]

I definitely do not have one thread per project, I think I've touched 4 or 5 projects in that time period

[indigo robin]

still hating on my thread etiquette i see

ill try it on my next project

[plush harbor]

having no git, no documentation and everything in one context seems really really risky

[cedar skiff]

lite work 😛

[indigo robin]

thats crazy

[plush harbor]

lol yeah, I have quite a few days with almost no tokens burnt lately, I ran out of stuff to do

[indigo robin]

make it make sense

[plush harbor]

I had to tell codex to find something today cos I needed it for the openai support bot. It found a group of legacy functions to remove

[cedar skiff]

just start your next project along side your current one, get a couple going

[plush harbor]

the newest project I got hung up on design for an entire day ... fiddling with colours and stuff totally outside codex

[indigo robin]

@cedar skiff do you have an agent you use ?

[plush harbor]

today I've been getting chatgpt to generate pictures of bones and lungs and guts and stuff so no codex again

[indigo robin]

and why is their even token limits ?

why cant they be stored on your device

its just text

[plush harbor]

if there were no token limits openai would go bankrupt. Also openai is not on your device

[cedar skiff]

codex is my agent, I orchestrate anything that is repetitive. And i make callable skills for manual tasks and prompt patterns i use manually a lot.

[indigo robin]

how would i make an agent based on a ui's design language ?

for instance

[indigo robin]

it should have known this ws incorrect

[plush harbor]

is this your first software project?

[cedar skiff]

for example to day i had 400 odd task orchestration finish up with 60 odd blocked task in it. I set up orchestration to un block those tasks. I got codex to categorise the blockers then formulate solutions for each category, then run orchestration with sub agents to fix the blockers using the startagies we came up with.

[cedar skiff]

on the side i am working on a skill set that writes responsive material 3 cross platform apps from scratch.
I wouldnt use an agent for this i would make skills for it

make an api skill and an implementation skill for the design language

so it has all of the information about the api and how it works for the latest version and then it also knows how you want it to write the code for it.

[cyan wing]

crashed again on WSL ://

[cedar bear]

Is Codex mobile still not available on Windows?

[plush harbor]

you sound suspiciously like you do this for a living, unlike us peasants with our little hobby sites

[cedar skiff]

I do sit in front of the computer all day

I work for my self though

my advice is to start using skills, since i switched to them over agents my work flow began to explode

[plush harbor]

not sure what I'd even use skills for. My main project is more about the data than the code so my next body of work is all about external data sets and classification but not for making code

[cedar skiff]

skills are just prompts, you can use them for anything you use a prompt for

like i need to verify an app was matching the design language it was trying to emulate. So i make skills that combined let the orchestrator spin up the app, take a screen shot, pull the spec screen shots and specs from the official documentation do a a spec audit against what i had and do a visual comparison and write an audit report. Then i got the orchestrator to do this process for all components. No coding involved for this task.

[sand moss]

anyone has this error "Selected model is at capacity. Please try a different model." ? I tried different models and sometimes I got this error since this morning

[plush harbor]

my next design type task on this project is going to be more "use the markup from the main letters to generate a category letters list" which could be a skill except there's only two places that pattern occurs so there's not much point

and its already written a parser to generate that type of markup but not for letters so its probably going to use 3 tokens and take 30 seconds

[fleet geyser]

model is at capacity lol

so many people are using it ig

[solid lake]

Sorry I’ll hop off

[storm cedar]

my system detects speech bubbles

[compact zodiac]

does anyone have invite code

hello

[supple perch]

SOJS65 ?

[bright swift]

for?

[orchid plume]

I'm guessing they meant Sora, if so then wrong channel, and I thought Sora was sunset anyways

[proper cosmos]

How long did it take you to make

[storm cedar]

4 days

[proper cosmos]

well done

just generic codex?

[storm cedar]

Yes

[proper cosmos]

well done 🙂

[crystal roost]

Is anyone here participating in the SEA x Openai hackathon 2026? I'm looking for a team in SEA to participate in the competition.

[cedar skiff]

is it an online thing or is it in person?

[hardy monolith]

Hello,
I'm from Congo (DRC). Any news on the availability of the** Codex Chrome extension **?
I believe that the EU is also not able to install and use it.

[proper cosmos]

just use a VPN for the US

that's what i did

[hardy monolith]

Actually, I did. That's why I was able to install it and use it for the first time. It worked great. But now, a week has passed and although the plugin is installed and the extension is installed in Chrome, even behind a VPN Codex cannot run it anymore.

[bright swift]

interesting that its blocked in congo.. i thought it was an EU thing due to their regulations

unfortunately im part of the permanent blocked underclass too

[orchid plume]

not checked but as a guess... last night's Codex update may have changed the way it's restricted. It's quite possible that OAI can restrict it based on the billing country on the account. I'd need to test my VPN again at some point to see if they have changed the way it's restricted. Also sadly in one of the restricted countries here 😂. Last night's update interestingly showed me the Computer Use tab for the first time though, albeit both Chrome and Computer Use say unavailable due to organisation settings or region

I hope today's update for Codex isn't another region locked or OS locked feature though 🤞

[hardy monolith]

Yes, I suspect that the latest update has "broken" my previous ability to use the chrome extension. And it's probably linked to my account/billing location.
But those are just guesses.
Anyway; Africa isn't part of EU/UK. We technically shouldn't be included in their restrictions.

[orchid echo]

cool app but that is too small to see lol

[signal tapir]

Looks terrible from a usage perspective

[native vigil]

had a vibe that 5.5 became smarter after the last usage reset but today it feels off again :/

[torpid trout]

has anyone ever seen this issue where you start a /goal, and it goes do things, and suddenly (invisible to the current thread) a other sesion is opened (only discovered it via AgentsView) which is ongoing in parallel, and when resumed says

 This session was recorded with model `codex-auto-
  review` but is resuming with `gpt-5.5`. Consider
  switching back to `codex-auto-review` as it may
  affect Codex performance.

That model as such does not even exist (not to trigger manually anyway), and I never triggered the codex /review mode

[orchid echo]

i dont like /goal

[bright swift]

blasphemy

[torpid trout]

apparently this is the command reviewer, not the /review command
AgentsView falsely reports that as a "user" session.

[woven canyon]

What is /goal

Has anyone noticed that the context window isn't displayed anymore

[cedar skiff]

it's a command that keeps working until it hits a completion criteria

[woven canyon]

Interesting

any usecase?

[orchid echo]

in status line?

[cedar skiff]

it's easier than orchestration

but not as detailed

so its ok for mid level tasks

[woven canyon]

In the codex app

[violet mural]

this, i was wondering why this is the case on latest codex app

i thought im the only one haha. though it still auto compacts when it needs to anyway, just cant see it manually now, so idk when to manually compact

[woven canyon]

yea, idk why they did that

[violet mural]

probably broke it accidentally, like what they did with the thread rendering the past couple versions

also noticed fast mode setting seems to reset on its own when switching threads albeit it happens randomly

[orchid echo]

ive compacted my session so many times my context starts at 60% after compaction oof

[storm cedar]

It is now possible to choose the source of the translation.

For those who like 20th Century Boys

[orchid echo]

What do you guys suggest doing if I want to keep all related context but start a new chat. Should I just ask Codex to create an extensive handoff document of everything important for my project then have a new chat read it to get started?

[storm cedar]

so click here

[orchid echo]

is that the fork option? i use CLI

[storm cedar]

Just copy the session number.

[orchid echo]

and then what lol

[crystal roost]

the grand final was held in Singapore

[storm cedar]

Give a prompt asking to continue the session.

[orchid echo]

ah

will try that next time. just made it create a handoff doc

[storm cedar]

Did you put a Kindle stand on it?

[valid bear]

Has anyone figured out how to prevent these compaction errors?

[valid bear]

codex is almost unusable for me when working on a sizeable codebase.

[gentle harbor]

why did codex use 50gb of wifi in under 2 days

[torpid trout]

probably because it sends billions of tokens back and forth, add a bit of telemetry, a bunch of memes generated with it or whatever imagery you do...

[simple raven]

Looking lerning more about AI

[bright swift]

never had errors. try this config

tool_output_token_limit = 25000
# Leave room for native compaction near the 272–273k context window.
# Formula: 273000 - (tool_output_token_limit + 15000)
# With tool_output_token_limit=25000 ⇒ 273000 - (25000 + 15000) = 233000
model_auto_compact_token_limit = 233000

[simple raven]

Okey thank you

[gentle harbor]

ew, i dont use any stupid ai image stuff

[simple raven]

Okey I use for venting

It helps alot

[orchid echo]

its only 466 lines

[storm cedar]

I am trying

[rustic pine]

Lately, again, Codex 5hr window seems to running out quicker than usual. "You’re out of Codex messages"

Anyways for me to look closer at what is eating up the usage so fast? The "Codex Analytics" page is not very helpful for digging into that.

[torpid trout]

I think the tool @lean lark created could help?

[valid bear]

make sure you dont have a rogue process spawning infinite codex sessions in the background: https://github.com/openai/codex/issues/23791

[lean lark]

That's https://github.com/CaptainStarbuck/codex-usage

The filters1 branch allows for significant filtering of data: date ranges, time ranges, selection by events vs sessions. It also allows specifying a max number of sessions or events per table to reduce lag. And the TUI output is now formatted like the HTML report - and quite useful.
That branch will be rolled into main in a few days if no issues are reported.
Other enhancements are already logged in Issues and will be processed soon.
Suggestions are always welcome.

[rocky fog]

and it continues https://www.aikido.dev/blog/github-breached-vs-code-extension

better watch out folks with packages/dependencies/libraries/extensions
its an absolute minefield recently 😄
maybe at least have codex set up scanner against public services which it runs everytime they get changed/updated, which might not always help with new ones and delay updating packages until they are few days out (I guess some package managers might not have that feature)

anyone else doing something else similar?

[bright swift]

pin deps and use min age global config

[boreal holly]

Now that AI exists, I don't have a lot of dependencies and I make my own tools. I think that's probably where we're headed

[rocky fog]

yep true, can just make codex do it instead of depending on external 😄

[lean lark]

I wrote something recently on this topic but I can't find it here, might not have published it. 🙁 Looking.

[valid bear]

ty ty

[lean lark]

So the idea was/is that OpenAI now offers security tools which can be used to scan common dependencies. In fact they do scan deps that they use. They generously offer to partner with high-profile FOSS providers. But there are a lot of not-so high-profile devs that produce FOSS that is still consumed by businesses and devs.

OpenAI has the Code Security offering (https://developers.openai.com/codex/security) (formerly Ardvark) (ref #codex-security-feedback ) and others which are only available to Edu, Business, and Enterprise.

Some new or existing company can use these tools to scan common FOSS, some as a community service, for others they can charge a penny per download or upload from/to npm or VSCode Marketplace or Open VSX or SourceForge or other such repositories. The thing is, the compute required to do this is significant. Someone needs to pay for that. Kindness doesn't pay for servers.

Business must be motivated to pay something ... anything ... to defend the "free" software that they depend on so heavily. That means education, and IT staff speaking up to discuss this with management. That kind of initiative needs to be provoked - it won't happen naturally - hundreds of serious malware issues has not inspired IT staff to suggest that their business pays for protecting FOSS.

[torpid trout]

Right now, we don't know which extension or whose machine was impacted.
This is about the most hilarious in all that

[lean lark]

☝️ Cuz Microsoft can't afford to scan files uploaded to their servers. 🙄

[torpid trout]

That things get compromised is normal, hell, the entire linix world almost got done for a few years back
The question is, are you going to tell people or not. Those *&@&#^@ however decide to not disclose it, hence leaving potentially millions of users open to exploit? Is that really true? That article is obviously advertising its own services, so I would not trust that either, not with a 10km pole

[bright swift]

aarvark is a different internal only product i think with stronger model and no censorship, at least thats what peter steinberger said on a podcast after he joined

but codex security is slept on

supposedly it will cost extra in the future

[torpid trout]

ardvark is renamed I think they said that somewhere, to that what now is known as security

[bright swift]

mhh possible yea, its really good regardless

[torpid trout]

reading the tweet there is zero evidence of anything being compromising users tho
So I would take taht article with a kg of salt.

[plush plover]

Are there any codex deals currently running for new subs?

[lean lark]

Yeah, use Codex to save time and improve your code. Sucha Deal!

🙂

[plush plover]

Lol

[indigo robin]

ardvark lmfao

[woven canyon]

Did they mess up codex caching again 😭

[lean lark]

(Two "A"s to be accurate.)

Hey PIC, is your Codex session/thread/task list any better?

[indigo robin]

i codexed the name and reduced it to be more efficient

[indigo robin]

kinda

i just gotta brute force the new thread to do what i want

[lean lark]

I was feelin bad for ya last night, man.

[indigo robin]

yea you ruined my entire project :p

lololololl

nah i was just cryin like a B

ill figure it out some how : )

[lean lark]

I understand, really.
Did you get it to generate a docs folder? Document the code?

Are you using Git?

[indigo robin]

i decieced that im going to brute force this last task make sure it works perfect then im going to do an anyalisis of the code piece by piece and create documentation

may be a bit backwards but im almost at a milestone

[lean lark]

Let me know if I can help. Feel free to DM.

[indigo robin]

i will

do you like music theory ? our play the guitar ?

[rocky fog]

I think it is actually about the Nx Console
https://www.bleepingcomputer.com/news/security/github-links-repo-breach-to-tanstack-npm-supply-chain-attack/

there is also this
https://www.microsoft.com/en-us/security/blog/2026/05/20/mini-shai-hulud-compromised-antv-npm-packages-enable-ci-cd-credential-theft/
I mean there are so many lately

[torpid trout]

Yeah, that is for sure, which is also why "just install this or that package" is a no go, but that is no news

[lean lark]

Recap for group: PIC was doing pretty much all project dev in a single session. We recommended standard practice, task-per-session, AGENTS, documentation, etc. The disruption to his project was significant. It's a growing pain. Shouldn't have been that catastrophic but it still caused pain. 😢

[rocky fog]

is it the poo? 😄

[woven canyon]

Did codex by any chance reduce usage recently?

[lean lark]

In another life I was in different bands, played out for a while. Did software by day and hard metal by night. 🙂

[indigo robin]

lolol nice

[torpid trout]

yes

[indigo robin]

shoulda wrote duce

lmfao

its the elegant way to say it

[torpid trout]

lol, as in that italian guy from ww2?

could just say cacca

Apparently its fine in italian.

[rocky fog]

I think I noticed that before here that it blocks you for it
so funny

[indigo robin]

not as elegant

[boreal holly]

Hmm, I wonder why they block chocolate ice cream

[torpid trout]

🍫

this is the only choco emoji

[indigo robin]

akira toriyama created the duce emoji

greatest creator on earth

[karmic gulch]

any jitter of an reset soon? maxed out bth claude and codex now, over to kimi lol. miss my codex tho

[indigo robin]

claude is duce

[torpid trout]

the poo p emoji is actually a chococlate ice cream emojy, discord gets it all wrong lol

[indigo robin]

ill never use claude im a codex purist

[torpid trout]

also, I can say poo, but not poo p? Lol, this is lobotomized

[indigo robin]

type shi

loshitamized actually

enough about duce though

looking a little better still needs alot of work

[rocky fog]

I wonder if its intentional
because you can also get that emoticon with :sh**:
😂
it has multiple words you can use for it

[lean lark]

While I believe that's an error in the moderation script used in this server, and not intentional, that has to be a specifically targeted character sequence. So yeah, it's in there, but I don't think it should be.

[indigo robin]

the text isnt centred , the button styling on the bottom is incorrect

and the interval badge is too big

[plush plover]

This is cool

[indigo robin]

thank you

making a music theory app

[exotic cave]

From the article: "It just needs minutes, not days."
Claude or some such wrote it for sure.

[indigo robin]

you do game dev ?

[plush plover]

Yes I do

[indigo robin]

i wanna make a game so badly , i do lots of graphic design

i acually created a system for blender , that is a 1:1 pixle framework for pixel games

[torpid trout]

This one is driving me nuts
Codex told me to add

js_repl = true
browser_use = true
in_app_browser = true

I did
Then

can you see the browser/js repl tool now?
No, I cant

[exotic cave]

What's stopping you? Isn't codex can solve all your issues?
No it can't, because a good game is more than code...

[torpid trout]

So how the hell to use the browser automated testing stuff they inbuilt

I dont want to add plywright if it is possible to do without

[indigo robin]

ill input this into codex now

[signal tapir]

The best tip I can give you about making games is: Start small. Smaller than you think you should, even if that is already small.

[indigo robin]

i like to start with the window size first lol

[plush plover]

Most say this as advice, but I think the root is to prevent people quickly quitting when they find out how hard it is

I would say make the game you want to make, even if its too big, and you'll come to the conclusion if you can or cannot do it yourself

[exotic cave]

Game making is really involved, one of the more complex software's out there.
So many moving parts.

[signal tapir]

That's true.

[indigo robin]

apple purist , codex purist

none of yall android users are getting my app lol

[signal tapir]

Oh no.

[indigo robin]

ill send personal android versions to jim though

[signal tapir]

lol

[indigo robin]

and starbucks

[exotic cave]

UX is out of the window?

[indigo robin]

UX?

i should have wrote it in flutter apparrently but i wrote it in swift

[signal tapir]

I'm releasing my game Linux first. Windows is an extra I add because I have to. 😛

[indigo robin]

true lol

windows is bunz

[plush plover]

I plan to only support linux and windows, nothing apple, and android at a stretch though app dev on android is getting worse

[signal tapir]

My game doesn't fit on mobile devices so I don't have to worry about that.

[indigo robin]

took a while to get the animation quality i liked

i love Xcode : )

text still isnt centred properly though

[orchid echo]

gotta love this server

[indigo robin]

and the interval badge needs to be 25% smaller

smh lol

[exotic cave]

It's not so user friendly that is.
Make it responsive, which means you need to rethink how the UI folds on smaller screens.

[indigo robin]

whats not user frindly about it ?

[exotic cave]

For one, the buttons are so small and close to each other.