Speed up your Issue Creation with Copilot: Feedback

[Akash1134]

To help you create higher quality issues with greater ease, you can now create issues by chatting with Copilot. This experience is currently in public preview.

Key Features

• Prompt to create an Issue: Create an issue by prompting Copilot. Copilot will transform your prompt into a fully created issue. Simply review and refine it before clicking create.
• Start with an image: Create an issue starting with an image. Copilot uploads the issue to the image.
• Review template guidance: View Copilot’s suggested templates based on your prompt. Make changes and Copilot will reformat the issue.

Still to come...

As this experience is in public preview, you may run into some bugs and missing features (let us know when you do). Some of areas we are working on include:

• Multi Issue Creation: Creating multiple issues with greater ease, rather than needing to click the Create button for each individual issue that Copilot drafts.
• Sub Issues: Nesting sub issues into parent issues.
• Updating Issues: Leveraging Copilot to update existing issues.

Try it out

To get started, go to https://github.com/copilot and try a prompt like …

• Create me an issue for..
• Create me a bug for ..

Feedback

We want to hear from you! Post your comments, questions, likes, and dislikes below.

[dserodio]

Is this also available in Copilot Free?

[KB01111]

Im not seeing it

[elliebennett]

@KB01111 Thanks for letting us know. Can you share your steps for trying out this feature? Which URL were you on when you tried and what prompt you initially tried?

[Stevenrusski395]

//:run download token

[laniigar]

Im not seeing it

Nothing is showing damn it

[T1-Ly]

Im not seeing it

[gsuberland]

I would like to know the following:

• Are the created issues marked as being generated by an LLM?
• Can maintainers opt out of this feature being available on their repos?

Thanks.

[SouthFresh]

Hi @gsuberland thanks for the questions

Can maintainers opt out of this feature being available on their repos?

Issues can only be created in repos where the user has permission to do so. At this time, there is no opt out feature.

So the only alternative would be to opt out of Github?

[ecnepsnai]

@elliebennett

At this time, there is no opt out feature.

What is the justification for the decision that this feature is good to proceed with public preview without controls for opting out?

[gsuberland]

@elliebennett My question was not hyperbolic, could you provide an answer please?

[xmatthias]

I guess the best way is to give some of the load back to github by reporting such issues / users as "spam / inauthentic behavior".
an AI generated issue IS inauthentic behavior (especially if it contains bullshit code) - so i think that's the expected approach.

(yes, this is very sarcastic - but it'll at least drive some of the load back at github).

[Daruma1]

Ran do KY 🐸

[jfalcou]

Nobody asked for this useless feature, especially with no opt out. I don't have time as a maintainer to sift through idiotic PRs written by a stochastic parrot trained on stolen data.

[Jackenmen]

Please consider #159749

[bcoconni]

A nightmare is coming true.

In case someone did not notice, issues are already written in natural language, or close to. So basically your idea is to prompt Copilot in natural language (because it's intuitive, you know) to produce something that is otherwise written in natural language ? And what is making you think that users that are not capable of stating clearly their issue without LLMs will be able to provide a useful prompt to Copilot ?

Which brings my second point: in the example you have provided, Copilot is taking the user statement for granted, and is not challenging anything. So if the user has actually misspelled the MarkDown syntax, well that's too bad because Copilot is doing what LLMs are doing best: making up facts to fill the gaps. In your example, Copilot is assuming that the user has used the correct table syntax without having any single evidence supporting that assumption.

Which brings my third point: the example that Copilot is providing to "reproduce" the bug is not tested, neither by the user nor by Copilot. So the example may or may not be relevant. And when the example is made up by Copilot, well then the project maintainer will be delighted to find that the so-called "example to reproduce the bug" is actually working.

But, oh wait, it's up to the user to check that the report that Copilot has made up prepared is correct. We're covered ! Well, GitHub may be covered but we, project maintainers, are definitely not. This wishful thinking will violently be hit by reality.

I guess you've read real life bug reports and therefore know what the average issue/ticket looks like ? I'd be very surprised that the kind of users who will use that "feature" are the people who are writing well detailed, well explained tickets with tested examples to reproduce the bug. IMO, the users who will most likely use that "feature" are the users who are not providing any example and are just telling "It doesn't work". Are you deluding yourself in assuming that they will copy the example written by Copilot and check that it is actually triggering the bug that they are reporting ?

For all these reasons, it would be appreciated that -as some other people have suggested- these issues are properly tagged as having been generated by AI, so that project maintainers know that these should be read with a pinch of salt. That would really be the bare minimum. However I am not deluding myself in thinking that this suggestion will be taken into account, I'm a small fish maintaining an obscure, low profile GitHub project (and spare me the public relationship message "GitHub is taking very seriously the feedback from all of its users.").

Well, I guess your performance at GitHub is evaluated by how much AI you are pushing into the website no matter what. So project maintainers will have to live with good and not so good AI features.

[lux-capacitor]

Copilot is not able to make pull requests. It has complete access to all our repos and when I ask it something super simple, it says directly that it cannot make pull requests. The documentation says it can though so what is the trick to enable this? I am chatting with the current repo I am asking for the change in as the context as you'll see in the 2nd screenshot. What am I missing?

In the convo below, if I say "proceed" it just makes the code in a snippet and hands it to me. I want it to make a pull request, what do I have to do to make it make a pull request?

[Jackenmen]

You need copilot pro+ tier which gives you access to copilot agent. Then, as a maintainer, you can assign issues to copilot.
Alternatively, there's "agent mode" that you can open from issue view (and maybe directly from vs code) to interactively prompt to generate changes that you may later decide to PR.

[andrei-ivanovici]

Can we please stop throwing AI at every step ? Please ? Pretty please ? This feature feels extremely rushed and not well thought through. No project with even a moderate degree of popularity would see this useful. I can already imagine the army of trolls swamping projects with useless PRs and Issues. I don't understand how GitHub or Microsoft thinks this is a good idea? Is the use of AI this desperate that determines you to alienate your user-base over ? Keep this on and GitHub will truly become an AI platform. With AI as both the content producer and consumer and Zero users to care about it

[LunaBorowska]

I hate how this feature can be used to make up issues that aren't even real, while making them look plausible.

And frankly, I don't think it's useful for anything else, the thing is, issues are already written in natural language, so anything that doesn't come from input is pretty much bullshit, outright lies that waste maintainers' time (as copilot cannot compile the project, and actually check what happens).

[Matiam0]

So there is no opt-out. We are for the most part unpaid and unwilling beta-tester's for a L.L.M. that is not market ready in the slightest. To top it off there is no way short of increasing work-load for maintainers to moderate its use. That is my understanding of this as it stands, if I see none of the above issues resolved I'll be moving to alternative or self hosted services. Encouraging others to do the same.

[taladar]

I don't see how this would even be a good idea if Copilot was a hyper-intelligent AGI. It still wouldn't be able to make up a sensible issue from the limited and potentially inaccurate information a user provides.

[dimitrilongo]

Create me an issue for ... prompt can work in VS CODE ? or only through https://github.com/copilot ?
Thanks

[just-joshing]

This particular announcement is just for https://github.com/copilot.
I expect the MCP server will eventually let you do this in VSCode as well though.

[CommanderStorm]

Crosspost on the github maintainer discussion board:

• https://github.com/community/maintainers/discussions/551

For maintainers who don't know about this => https://maintainers.github.com/

[bcoconni]

So I've used the feature with a real-life case: I've copied the whole text (verbatim) from a not so helpful report from a user in our project and asked Copilot to create an issue to see if we'd have had a better report:

On the positive side, Copilot indeed followed our issue template while the user did not 👍

Note that Copilot is not adding any useful information. Instead it is repeating ad nauseum the wording from the user and making up facts:

• there are no clear error messages - the user is telling that he/she has encountered problems but we do not know what he/she means by that. Copilot made up the fact that JSBSim was encountering an error but did not report it to the user.
• The documentation is not explicit regarding Simulink integration steps - The user does not complain about the documentation.
• ideally without having to edit the XML files - At no point the user is telling he/she does not want to edit XML files. He/she is asking if XML files have to be modified.

There is a word for that: noise. Copilot is increasing the noise and does not improve the signal. It is degrading the signal to noise ratio. All in all, I am vastly preferring the original user report over Copilot's.

[bcoconni]

Also trying to be constructive (and given the lack of feedback from GitHub it's not easy, believe me 😞). The report prepared by Copilot is missing some data which are identified by (please fill in). I would have expected that:

• At the bare minimum, Copilot would have mentioned that some information are lacking and that the user is expected to fill them in.
• Or, even better, that Copilot would have asked the user the information "I will need some further information to complete the draft. Could you please specify the operating system that you are using ? And what version of JSBSim you are using ?".

Generally speaking, I was expecting some follow up questions from Copilot to help it prepare the report but instead of that, it jumped straight to the report that I have displayed above, making up information in the process to fill the gaps.

[sshaw]

Is the only way to see the code diff for the to be PR via the GFM rendering of code blocks in the PR body‽

[sensei-woo]

Yo, to be honest, CoPilot overall is just a really poorly put together product.The online experience is completely ass. I'm basically using ChatGPT(which is connected to my GitHub so it could actually query correct data about repositories from GitHub, similar to how the MCP server would) and RooCline, Claude, and whatever other MCP clients to learn about repositories along with the RepoMix MCP and the GitHub MCP and even Devon. So in terms of learning, it's bad. In terms of coding, cursor tab is still better. In terms of providing a quick environment to make a project or component or interface(think v0, chatgpt or claude canvas, bolt.new), It's bad. So my original reasons for going with Copilot over Cursor were essentially My understanding of how integrated GitHub already was into everybody's development workflow and how much data they had that only they had access to. So I assume that their co-pilot models would be better. They own VS Code which has the most data out of anybody competing. They have unlimited funding. At some point they basically owned like half of OpenAI. And I'm talking about Microsoft in general, but not just GitHub Copilot. But that should let you know that this isn't some small player trying to make some really complex software. This is Microsoft, the company that practically invented computers and like 80% of the world's operating systems are Microsoft right now. And basically 99% of the code editors are Microsoft. You know, it just doesn't make sense how this product isn't the best at anything at all. And yet I'm paying $40 a month for it. It's like I could buy a cloud subscription and cursor for the same price. And I will literally be saving money at the end of the day because I won't be paying for a cloud code anymore because then it would be covered by the subscription. You know, it's like there's no incentive to stay here and the agent isn't good either.

[sensei-woo]

I also just remembered why I actually came to this forum or ended up at this forum and it was because I can't even do issue creation right. This is something that Claude, cline, goose, or any of them does perfectly in my IDE, or with the command line or with... Even Postman's AI assistant would probably be more helpful than what copilot just did:
It made the draft of the issue and then I told it how to change it and what revisions to make and it decided to make a markdown file.

[obryckim]

When creating an issue, copilot can see that we have templates and I can choose one of them. NOTE: the templates are not in the repo we're working in, but in our .github repo for the organization.

When I select the template, copilot says it cannot get the contents of the template.
I then have to tell it that the template is in the .github repo.
Once I tell it that, it can then see/use the template.

I would expect copilot to get the template contents from the org's .github repo if not found in the repo we're working in.

[huyenltnguyen]

I just tried this out, and I noticed that the repo selection is always pre-filled by Copilot. This makes it difficult for me to use the feature as I'd need to be extra cautious to make sure the issue goes to the right place.

I’d prefer if the repo field could be left empty by default so I can manually set it myself. The preferred / default repo can be set via the personal instructions instead.

[obryckim]

Copilot is now refusing to create issues stating that a template is required, even though I have a template selected:

[zeier]

I'm seeing the same behavior: https://github.com/orgs/community/discussions/163917#discussioncomment-13571943

[cagesellchen]

We shipped a fix for this today -- please let us know if you're still seeing this bug. Thank you!

[obryckim]

I tested this out and was able to create an issue now w/ my template.
Thank you!

[tsal]

Please give us a way to completely block this from our repos, or I'm going to be forced to stop using github enterprise.

[Daruma1]

I’m uncertain what the comment was originally ? Refresh my memory? 🔱 Sent from Gmail Mobile

…

On Sun, Jun 22, 2025 at 6:51 AM Michael K ***@***.***> wrote: Please give us a way to completely block this from our repos, or I'm going to be forced to stop using github enterprise. — Reply to this email directly, view it on GitHub <#159075 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/A27HHZDY43EEY2HSJIE4E2T3E2YGXAVCNFSM6AAAAAB5AJIRB2VHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTGNJUGI3TQOA> . You are receiving this because you commented.Message ID: ***@***.***>

[Daruma1]

How about now?

…

On Sun, Jun 22, 2025 at 3:22 PM laniigar ***@***.***> wrote: Im not seeing it Nothing is showing damn it — Reply to this email directly, view it on GitHub <#159075 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/A27HHZBVUGNIB3ZDZKNLBN33E4UDLAVCNFSM6AAAAAB5AJIRB2VHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTGNJUGUYDGNY> . You are receiving this because you commented.Message ID: ***@***.***>

[thibaudcolas]

Got to try it out today and I noted two accessibility-related issues:

• It generates Markdown with "bold text as headings" rather than proper headings. This feels like a missed opportunity to steer people towards more accessible content.
• It includes URLs without making them links. Same as above, while it works for some users, it’d be more accessible with real link text

Both of those I managed to fix with further prompting. And not really a surprise but it hallucinates. The issue I wanted was to demo performance-related template features (caching). Of two such features, it referenced one and made up another that doesn’t exist. It didn’t seem capable to browse the URL I provided. Adjusted a bit according to further prompting but clearly not able to reference the docs while creating an issue.

Generated output for ref

This issue proposes adding an example to demonstrate template fragment caching in Wagtail, as outlined in the Wagtail documentation: https://docs.wagtail.org/en/stable/topics/writing_templates.html#template-fragment-caching

**Proposal:**
- Add a new template or modify an existing one in bakerydemo that clearly implements Django's `{% cache %}` template tag for fragment caching.
- Optionally, provide a comparison or explanation in the README or as code comments about the benefits of template-level caching for performance.
- Ensure the implementation follows best practices as recommended in the Wagtail documentation.

**Acceptance criteria:**
- A visible example of template fragment caching is present in the demo site.
- Documentation or comments clarify how to use and extend this feature for real projects.

[WesleyTheGeolien]

When splitting screen and reviewing the drafted issue it is unclear on how to modify / talk with copilot as the chat pane is hidden. Until making the window full screen I didn't realise there was a chat pane available

[just-joshing]

When this happens do you see this button?

That button should hide the workbench so you can continue conversing with chat.

[MillerBoat]

The interface seems to be stuck in a loading state. Although it responds to other messages or actions, the main content doesn't load properly. This feature is really useful, but unfortunately, it's not working as expected right now. I hope this can be fixed soon.

[dipievil]

After create a full IA Driven markdown document including all details for every issue, it creates the issues correctly using the Claude Sonet 4. It also included more information and details for development all related to the planning. Amazing!

The gpt's (anyone) creates poor issues. And even after request change it, it didn't changed anything.

Using gpt:

Using claude:

Guys... Just use Sonnet 4.

Justing adding that the issue was created with copilot and it already started working on it! 😮

[kellyarwine]

When I created an issue, it used template "Report a bug". It wasn't a bug so I checked the templates and didn't know which to pick, there was no "blank option". From there, two things happened that weren't ideal:

• I clicked report a bug again to make the menu disappear ("I guess that's the best option?"). But my clicking unselected the option. That felt surprising.
• I believe it unapplied the template but left the type and label as "Bug". That also felt confusing.