Claude Code has over 5000 open issues. And this is after issues that are inactive for 60 days being auto closed. Such a policy is facetious to say the least. What is more perplexing is why they don't use Claude to triage the issues?
I think the idea is that nobody will be using CC in 5 years. If anthropic loses, nobody will use it. If anthropic wins, still nobody will use it! The value is in the model solving problems, and CC is just the hacky vessel for that, not the end goal. If they believe in themselves, polishing the product is a waste of time.
Closing things just so you have less open issues is the worst kind of dashboard driven development / goodharts law style of process failure.
Many issues are evergreen and people will come around continue to comment on them as they get hit. The idea that no one comments on old issues is simply a false premise.
If you look up examples of stalebot feedback the only people who think its a good idea are people literally only caring about how many issues are open.
If a team does not have the bandwidth to fix an issue, closing it acknowledges this and doesn't leave people in false hope. Ideally with a status like 'won't fix' to clarify the situation. Users can move on and deal with the reality, perhaps even rally resources if they are in a position to do so, so the issue can be reopened and addressed. Everyone hates it when you get a 'me too' or 'the developers suck' email to a bug opened 15 years ago. The bug is making the world a worse place.
Secondly, the larger the collection of open issues the harder it is to actually triage and manage. There are plenty of projects where everyone would benefit if the issue tracker suffered catastrophic data loss. So many dangling issues in features that no longer exist or are completely unrecognizable, making it impossible to separate the wheat from the chaff. You are in a twisty maze of 30,000 issues, all of them alike. What should I be working on?
Unfortunately, people react to the first reason poorly if their issue gets closed as won't fix. They take it personally and abuse the developers. So we generally don't use won't fix, because of people being hostile. And then triage falls behind. And then you find you have an unmaintainable bug database of 30,000 open issues, un-triaged, many duplicated, unknown how many actionable. 'open' status has become the unofficial 'wont fix'. When you submit a new bug, you hope someone is watching and you get lucky and someone assigns it 'in progress' or sticks it on a task list. The bug tracker gets bypassed, with real issues going via back channels to developers. The project realizes it has a problem and makes it harder to submit bugs, in a hope they can get on top of things.
Keeping an issue open is not going to force a maintainer to care. If they want to close the issue it is likely because they don’t care and getting them to care is not a UI/UX problem. Even if the issue is left open, it is likely not getting prioritized on internal roadmaps and discussions.
Meanwhile OpenAI Codex keeps my issues open for months, monitors their popularity and relevance, and updates me on them once they finally make it to development and release.
Has happened with over a dozen I’ve opened. What you’re saying may be true of Anthropic (and you've done a good job justifying it for them) but certainly not its competition.
They only manually close my feature request tickets if they've been open for a long time (several months) without "upvotes" from the community and aren't already planned. A senior engineer always explains why they're closing these.
It sounds like OpenAI Codex is operating the same way, but with much better communication. It takes resources to do the triage properly, but will make the community a much more pleasant place.
Claude Code has over 5000 open issues. And this is after issues that are inactive for 60 days being auto closed. Such a policy is facetious to say the least. What is more perplexing is why they don't use Claude to triage the issues?
I think the idea is that nobody will be using CC in 5 years. If anthropic loses, nobody will use it. If anthropic wins, still nobody will use it! The value is in the model solving problems, and CC is just the hacky vessel for that, not the end goal. If they believe in themselves, polishing the product is a waste of time.
Exactly. I listened all the speeches at the WEF about AI. Anthropic believes an AI God is coming.
> What is more perplexing is why they don't use Claude to triage the issues?
I wonder if it's because of cost.
This is about some issues are closed incorrectly due to a bug.
But as for the policy itself, why not? If an issue is inactive for 60 days it's very likely to stay so forever.
Closing things just so you have less open issues is the worst kind of dashboard driven development / goodharts law style of process failure.
Many issues are evergreen and people will come around continue to comment on them as they get hit. The idea that no one comments on old issues is simply a false premise.
If you look up examples of stalebot feedback the only people who think its a good idea are people literally only caring about how many issues are open.
GitHub stale bot considered harmful | https://news.ycombinator.com/item?id=28998374
What's the problem is such inactive issue stays open?
Two major problems.
If a team does not have the bandwidth to fix an issue, closing it acknowledges this and doesn't leave people in false hope. Ideally with a status like 'won't fix' to clarify the situation. Users can move on and deal with the reality, perhaps even rally resources if they are in a position to do so, so the issue can be reopened and addressed. Everyone hates it when you get a 'me too' or 'the developers suck' email to a bug opened 15 years ago. The bug is making the world a worse place.
Secondly, the larger the collection of open issues the harder it is to actually triage and manage. There are plenty of projects where everyone would benefit if the issue tracker suffered catastrophic data loss. So many dangling issues in features that no longer exist or are completely unrecognizable, making it impossible to separate the wheat from the chaff. You are in a twisty maze of 30,000 issues, all of them alike. What should I be working on?
Unfortunately, people react to the first reason poorly if their issue gets closed as won't fix. They take it personally and abuse the developers. So we generally don't use won't fix, because of people being hostile. And then triage falls behind. And then you find you have an unmaintainable bug database of 30,000 open issues, un-triaged, many duplicated, unknown how many actionable. 'open' status has become the unofficial 'wont fix'. When you submit a new bug, you hope someone is watching and you get lucky and someone assigns it 'in progress' or sticks it on a task list. The bug tracker gets bypassed, with real issues going via back channels to developers. The project realizes it has a problem and makes it harder to submit bugs, in a hope they can get on top of things.
I assume the thinking is “why look bad when you can look good?”
Or a longing for “cleanliness” by just throwing things away.
Reminds of a TV show scene where doing the dishes meant using them for target practice. If you have no dishes, there is no dishes to do!
Keeping an issue open is not going to force a maintainer to care. If they want to close the issue it is likely because they don’t care and getting them to care is not a UI/UX problem. Even if the issue is left open, it is likely not getting prioritized on internal roadmaps and discussions.
Meanwhile OpenAI Codex keeps my issues open for months, monitors their popularity and relevance, and updates me on them once they finally make it to development and release.
Has happened with over a dozen I’ve opened. What you’re saying may be true of Anthropic (and you've done a good job justifying it for them) but certainly not its competition.
They only manually close my feature request tickets if they've been open for a long time (several months) without "upvotes" from the community and aren't already planned. A senior engineer always explains why they're closing these.
It sounds like OpenAI Codex is operating the same way, but with much better communication. It takes resources to do the triage properly, but will make the community a much more pleasant place.
At scale this works: If the issue is affecting enough user base, the issue will still be in the list.
This is not true unless you want people to start spamming "+1" and "me too!" comments.
The correct text is "bump."
someones been on the darknet forums ;)
Or just any forum
https://miro.medium.com/v2/resize:fit:4800/format:webp/1*4ru...
Ya, but when you're not at scale it's annoying
Yeah, it only checks whether the last bot comment is older than 30 days, completely ignoring any human comment
Hard to imagine how this got past code review...https://github.com/anthropics/claude-code/blob/5e3e9408feea9...
>code review
This implies they do any. Every Claude Code release is full of regressions and new features that are released barely work or need hotfixes.