Call for commenting on GitHub about a Claude Code inaccessibility issue

Hi,

Thanks for bringing this to the community, I commented on the issue and I encourage everyone that either use or want to use claude-code, or who just cares about making software accessible to everyone, to comment the issue as well.

Anyway, what's for dinner?

If you don't care that's fine but come on, don't just clutter up the page with silly comments like that.

I replied to the page and even though i'd not tried the coding part; I had tried claud today and reporting my findings. I honestly don't think it wil be fixed but you never know.

I actually care very deeply about this subject, in a very negative way, as in this nonsense can't go away soon enough. It's hyped by people who over-invested in a technology that while scientifically interesting isn't really adding any real value to the world even in its current heavily subsidized form, and those people are engaged in mass deception hoping to get rid of the liability before it blows up on their faces. The boot lickers below them are busy deceiving each other about alleged trade secret productivity boosting accomplishments that so far don't really have anything to show, and contrary to their ridiculous claims that AI would come for software engineering jobs, some people allege that the sector is rebounding.

Does anyone remember a couple of years back, when, "I", was the one getting all kinds of flack from users here on AppleVis, with regards to my negative viewpoints on AI and AI trends?
God it's nice to see others realizing what I realized back then. 😁👋

Man, please, do not change. You made my day with this single random comment.
@Everyone. I am a junior junior cs student. -10 worries about job security. Period.

It's still pretty pathetic that these tools do not have good accessibility, which I think is the point of this post. Whatever your feelings on AI coding I don't think it's reasonable to just be rude to posters for wanting equal access.

Feel free to raise another ai/vibe coding post if you want to debate the wider views on it though.

I've only used codex so far and actually from the terminal it works pretty well, box drawings aside. I will try Claude code when I get some time and will comment on this if it doesn't work for me too.

And most ncursus interfaces are nicer to user after in terminal.
But yeah, accessibility is the last thing to be done poorly after public pressure as always.

Oh wait - I never thought to actually ask it to stop with the box drawing symbols - I just presumed it would humour me. Well that's brightened my day even if it doesn't; help the OP much.

The #1 legendary Apple employee, the only person in the world with an Apple employee ID lower than Steve Jobs', probably the most famous engineer who shares the same kind of passion for elegant technology that I do, the one and only Steve Wozniak, Apple's co-founder, just made some comments on this subject, which makes it sound like my lack of enthusiasm for this technology aligns with his own. Not that I need validation, but to me it's an honor to know that this specific person's views on the hype surrounding this subject align with mine.

There are also those who slam people for having negative opinions of AI. Mustafa Suleyman, CEO of Microsoft's AI group, called public criticism of the tech "mind-blowing," Nvidia's Jensen Huang says the negativity is hurting society, and Nadella has pleaded to move the conversation beyond "AI slop".

This is my favorite part of the article. 😇

I think part of the problem is, all of these bigshot CEO's have dumped something like over $700 billion into AI research, and they don't want to see that investment go to waste. Mike Buckley (from Be My Eyes) talks about this in a recent Double Tap podcast.

Edited because I forgot to switch to markdown. Oops!

Thanks for sharing.
Cool the issue was fixed. I'll try it when I'm obsolete in 6 month. Since 2022.

@mr grieves I’m just curious—how do you interact with Codex’s interactive menus in the terminal? When Codex pops up a menu with options, how do you use VoiceOver to select the one you want?

Most of the time it just does what I ask without any questions. Really on the whole the only time it asks me is when I want to commit my changes to git, otherwise there's not much prompting. But when there is I usually want the default option so press enter. There's usually esc if you want to ask for something else. The other options do have a key shortcut to use but I've not needed that in a while.

When I first installed codex - and it's been updated about a billion times since - it was a bit confusing trying to authenticate as I didn't know what option I was choosing. I happened to get the right one and I've not had problems since.

Codex assured me that it was going to remember not to say this forever.

It lied.

I take it back. AI won't come for the big coding jobs, or at least I odn't think so. I'm not a person who undersand code that bwel but, if i were; I'd use ai along side my knolledge. Let's say I have a basic framework,, I have my libraries loaded and if statements and so on, , AI would be nice to be used to check over it for mistakes, or better ways to make the code flow better.

AI can help with coding and things but more as a tool, not a replacement. Will it truely replace humans when it comes to coding? I think so but that day isn't today or this year or even 2 years from now, but even when it becomes so smart it can write code with no errors, that will be a good day because it will allow humans to do other things. People are dooming and glooming about AI taking jobs and honestly I find it kind of funnny. AI is a tool, it's up to you how you use it.

about to comment on the issue and this is exactly why I gave up on the termainal version and use the code tab in the Claude Desktop app, and it works great! also they need to fix all those emoji things! it's so distracting and I can't tell what Claude is doing.

This is probably not the thread to go here, but AI is much more capable at coding than you might think. It can go off the rails a little bit so it needs you to keep an eye on it, but it is pretty efficient. At my work we use it for code reviews and it is actually scarily insightful and often catches things that have been missed. Sometimes I am shocked by what it can find.

In Jira - our ticketing system - there is literally an option to get AI to make the change for you using Atlassian's AI. A colleague of mine has a ChatGPT skill he can use and just points it at a ticket and it does the work for him, so he just needs to review it.

I have used it on some things and it does a good job, but you need to recognise when it doesn't and call it out.

I don't really want to become someone that just reviews code and barks orders - the bit I actually like is writing code not reading it, but it is making me feeling quite obsolete. I still write the majority of my code but it is a hell of a lot quicker than I am.

I do think there are risks with unreviewed vibe coding in anything but a small private, personal project. But actually it is very good at doing very very specific things just for you. How often do you get an app that kinda does what you want plus a billion more things you don't care about, and you need to adapt to what it wants. I think there will be a trend of just saying the exact thing you do want and just getting it and nothing else. And this is doable right now.

But also, as I've said before, coding isn't just one thing. So mileage varies depending on what you are doing. For example writing a little game or utility for yourself is quite different to writing something that sits in the cloud handling millions of requests which is different again to writing an operating system or something that runs off a tiny IOT device.

I think it's quite similar to driverless cars - I still find it hard to think of them as anything but science fiction but there are parts of the world where they are real and probably doing a better job at driving than a lot of humans.

I know this is a can of worms and really I should know better than to write this as I can feel what might come next. But I am very anxious about where all this is going even as I benefit from it. But then you can say that about most AI things.

The only people who should fear getting replaced by AI are the engineers who never really provided much value because they aren't that skilled anyway.

For the last few decades, software engineering has been paying relatively well, however the overwhelming majority of the people employed in the field only pulled their own weight at best, because the economic potential attracted a lot of people who don't really love or even care about the field, meaning that the quality of software has been degrading exponentially, and the only reason why end-users don't notice this problem that much is because hardware evolution has been compensating for all the inefficiencies and even security problems. To these people who never really cared about deepening their knowledge, I can understand how they might fear getting automated out of their jobs, as well as how they might feel inclined to suggest alternative workflows in which they think that they can still provide value, however the reality is that AI isn't really bringing anything new to the table, because all the problems people are using AI to solve now were already possible to solve using abstractions over battle tested libraries, so the only reason many people are afraid is because all of a sudden they see their low-effort and low-skill engineering jobs being rendered obsolete, which they already were, but in the past they could scam their employers into thinking otherwise.

One thing AI is pretty bad at when it comes to software development is coming up with and understanding code architectures, so commits that clearly miss the point by failing to follow implicit conventions and even reinvent the wheel by implementing functionality that is already available elsewhere are a very distinguishable tell tale sign of AI code slop. The problem with this code slop is that complexity undermines reasonability, so the messier the code, the harder it is to catch potential problems in review for both humans and AI alike. This specific problem is one of the reasons why I do not approve of people using AI to generate code, but am perfectly fine and actually recommend using AI to review that code. Creating is a task in which humans still excel by a huge margin, and catching pattern outliers is a task that computers are much better equipped to perform, so by making computers do the former while humans do the latter, we are reversing the roles in a way that is uncomfortable to both sides for no good reason.

I took a look at the accessibility issue you described with Claude Code and VoiceOver. To me, the problem does not seem to be the terminal itself, but the self-rendered terminal UI layer: custom cursor behavior, box-drawing characters, visual panels, and dynamic screen updates that are difficult for VoiceOver to track reliably.

Because of that, I thought this might be useful as a possible workaround, at least for testing:
https://github.com/Seregawpn/terminal-voice-bridge

This is a fully open public GitHub repository. You can inspect the code, download it, run it locally, and see whether the approach is useful in your own workflow.

The idea is not to “fix” the underlying accessibility problem inside Claude Code itself, but to provide a more accessible layer on top of noisy terminal UI. The project can:

read the active Terminal.app session;
extract the last user input, last assistant output, or last exchange;
reduce noisy terminal UI elements such as box drawing, status chrome, and other decorative output;
speak back the meaningful text instead of forcing the user to parse the raw terminal screen.
So I would describe it as a workaround rather than a true upstream fix. But if the current blocker is that the terminal UI is too noisy or too hard for VoiceOver to follow, this approach may still be worth trying.

I added it as a public GitHub repo so anyone interested can review it openly. If you decide to test it and have feedback, I would be glad to keep working on it and improve it based on real usage.

If you want a slightly shorter version for a forum reply, use this:

I saw the Claude Code / VoiceOver accessibility discussion and wanted to share something that may help as a workaround.

I put together a fully public GitHub repo here:
https://github.com/Seregawpn/terminal-voice-bridge

You can freely inspect the code, download it, and try it yourself.

The idea is not to fix the core accessibility issue inside Claude Code, but to add a more accessible layer around noisy terminal UI. It can read the active terminal session, extract the most useful content, reduce UI noise like box drawing and decorative terminal output, and speak back the meaningful result.

If anyone here wants to test it, I would be very interested in your feedback and happy to keep improving it based on real accessibility use cases.

This is both amusing and sad at the same time. These people can't even bother to proofread their AI-generated comments, it's just prompting and pasting the generated answer to the forum verbatim, as a perfect illustration of what I was talking about earlier, but applied to more mundane tasks. These people are digging their own career graves by collectively making themselves dumber.

The following quote was taken from the bottom of the previous comment at the time of this writing. Can you spot the tell tale sign of AI slop?

If you want a slightly shorter version for a forum reply, use this:

I saw the Claude Code / VoiceOver accessibility discussion and wanted to share something that may help as a workaround.

I put together a fully public GitHub repo here: https://github.com/Seregawpn/terminal-voice-bridge

You can freely inspect the code, download it, and try it yourself.

The idea is not to fix the core accessibility issue inside Claude Code, but to add a more accessible layer around noisy terminal UI. It can read the active terminal session, extract the most useful content, reduce UI noise like box drawing and decorative terminal output, and speak back the meaningful result.

If anyone here wants to test it, I would be very interested in your feedback and happy to keep improving it based on real accessibility use cases.

That part was my mistake. I was trying to offer a quick workaround and posted too fast without cleaning up the text properly. English is not my first language, so I sometimes use AI to help me phrase this kind of post more clearly. In this case, that was a stupid mistake on my part, and I should have reviewed it properly before posting. I appreciate you pointing it out. I will be more careful going forward and will write these replies myself instead.