Hi all,
I know there's been a lot of discussion/concern and policy decisions around usage of LLMs for authoring content in Wikimedia projects.
I wonder if there's been similar discussion around AI coding for technical editors? I'm not sure I've seen it "go by". Can someone link to where these discussions are happening?
I also wanted to share a policy I wrote for WP1: https://github.com/openzim/wp1/blob/main/CONTRIBUTING.md#usage-of-llmsai-cod...
Cheers, -Travis
Hi,
I'm not aware of any discussions, but would add that I've been finding that patches for MediaWiki entirely or mostly written by AI (as opposed to AI used to assist) require a lot more code review and often really miss the mark to solve the problem on the first pass. At least one didn't even solve the problem described in the ticket (and wouldn't have even if they had addressed other comments)
I think if we adopt a similar policy or guideline, then it would need to be careful to avoid discouraging new developers learning by doing (i.e. using AI to help them learn)
Would be interested to see what input others have on this
Dreamy Jazz
On Sun, 25 Jan 2026, 21:06 Travis Briggs via Wikitech-l, < wikitech-l@lists.wikimedia.org> wrote:
Hi all,
I know there's been a lot of discussion/concern and policy decisions around usage of LLMs for authoring content in Wikimedia projects.
I wonder if there's been similar discussion around AI coding for technical editors? I'm not sure I've seen it "go by". Can someone link to where these discussions are happening?
I also wanted to share a policy I wrote for WP1: https://github.com/openzim/wp1/blob/main/CONTRIBUTING.md#usage-of-llmsai-cod...
Cheers, -Travis _______________________________________________ Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
I agree with Dreamy, I've recently come across a lot of new users uploading AI-gen patches (including one that hallucinated an API param and was effective DoA).
I feel like there is a need for a project-wide AI/LLM coding assistant policy of some kind that I can direct people to when I see obviously AI-gen pull requests. (My personal compass is that it's fine to use AI tooling as long as you adhere to "Own the code you commit")
Regards, Sohom Datta --- Open-source contributor @Wikimedia, @Chromium
On Sun, Jan 25, 2026 at 4:16 PM Dreamy Jazz via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
Hi,
I'm not aware of any discussions, but would add that I've been finding that patches for MediaWiki entirely or mostly written by AI (as opposed to AI used to assist) require a lot more code review and often really miss the mark to solve the problem on the first pass. At least one didn't even solve the problem described in the ticket (and wouldn't have even if they had addressed other comments)
I think if we adopt a similar policy or guideline, then it would need to be careful to avoid discouraging new developers learning by doing (i.e. using AI to help them learn)
Would be interested to see what input others have on this
Dreamy Jazz
On Sun, 25 Jan 2026, 21:06 Travis Briggs via Wikitech-l, < wikitech-l@lists.wikimedia.org> wrote:
Hi all,
I know there's been a lot of discussion/concern and policy decisions around usage of LLMs for authoring content in Wikimedia projects.
I wonder if there's been similar discussion around AI coding for technical editors? I'm not sure I've seen it "go by". Can someone link to where these discussions are happening?
I also wanted to share a policy I wrote for WP1: https://github.com/openzim/wp1/blob/main/CONTRIBUTING.md#usage-of-llmsai-cod...
Cheers, -Travis _______________________________________________ Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org
https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
I'm of the opinion if people can tell you are using an LLM you are using it wrong. Its still expected that you fully understand any patch you submit. I think if you use an LLM to help you nobody would complain or really notice, but if you blindly submit an LLM authored patch without understanding how it works people will get frustrated with you very quickly.
-- Brian
On Sunday, 25 January 2026, Travis Briggs via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
Hi all,
I know there's been a lot of discussion/concern and policy decisions around usage of LLMs for authoring content in Wikimedia projects.
I wonder if there's been similar discussion around AI coding for technical editors? I'm not sure I've seen it "go by". Can someone link to where these discussions are happening?
I also wanted to share a policy I wrote for WP1: https://github.com/ openzim/wp1/blob/main/CONTRIBUTING.md#usage-of-llmsai-coding-assistants
Cheers, -Travis
Agreed, my policy includes:
Basically, you should be able to understand, explain, and justify any
code that an AI tool contributes to your PR.
Cheers, -Travis
On Sun, Jan 25, 2026 at 6:20 PM Brian Wolff via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
I'm of the opinion if people can tell you are using an LLM you are using it wrong. Its still expected that you fully understand any patch you submit. I think if you use an LLM to help you nobody would complain or really notice, but if you blindly submit an LLM authored patch without understanding how it works people will get frustrated with you very quickly.
-- Brian
On Sunday, 25 January 2026, Travis Briggs via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
Hi all,
I know there's been a lot of discussion/concern and policy decisions around usage of LLMs for authoring content in Wikimedia projects.
I wonder if there's been similar discussion around AI coding for technical editors? I'm not sure I've seen it "go by". Can someone link to where these discussions are happening?
I also wanted to share a policy I wrote for WP1: https://github.com/openzim/wp1/blob/main/CONTRIBUTING.md#usage-of-llmsai-cod...
Cheers, -Travis
Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
Hello,
Perhaps this essay is worth reading & taking inspiration from, regarding activities in general whether or not they involve AI: https://en.wikipedia.org/wiki/Wikipedia:Competence_is_required. Especially when experienced volunteers are stretched thin, patience and time for coaching are limited resources, and a decision may be made to end someone's involvement.
My personal opinion regarding products in general is aligned with Dreamy's "I think if we adopt a similar policy or guideline, then it would need to be careful to avoid discouraging new developers learning by doing (i.e. using AI to help them learn)" along with Sohom's "(My personal compass is that it's fine to use AI tooling as long as you adhere to "Own the code you commit"", with the addition I'm opposed to expecting volunteers or a nonprofit's staff time to be used reviewing low quality contributions where those on the reviewing end haven't given informed consent in advance; also see https://www.runtime.news/ai-slop-is-overwhelming-open-source/, quoting from LLVM maintaners: "While new tools enable more development, it shifts effort from the implementor to the reviewer, and our policy exists to ensure that we value and do not squander maintainer time".
Balance is required between coaching, quality, velocity, and resource allocation issues.
Thanks for raising this question, Travis.
Pine🌲
On Sun, Jan 25, 2026 at 6:34 PM Travis Briggs via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
Agreed, my policy includes:
Basically, you should be able to understand, explain, and justify any
code that an AI tool contributes to your PR.
Cheers, -Travis
On Sun, Jan 25, 2026 at 6:20 PM Brian Wolff via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
I'm of the opinion if people can tell you are using an LLM you are using it wrong. Its still expected that you fully understand any patch you submit. I think if you use an LLM to help you nobody would complain or really notice, but if you blindly submit an LLM authored patch without understanding how it works people will get frustrated with you very quickly.
-- Brian
On Sunday, 25 January 2026, Travis Briggs via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
Hi all,
I know there's been a lot of discussion/concern and policy decisions around usage of LLMs for authoring content in Wikimedia projects.
I wonder if there's been similar discussion around AI coding for technical editors? I'm not sure I've seen it "go by". Can someone link to where these discussions are happening?
I also wanted to share a policy I wrote for WP1: https://github.com/openzim/wp1/blob/main/CONTRIBUTING.md#usage-of-llmsai-cod...
Cheers, -Travis
Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org
https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
Hi,
I agree with the points that state that there is a need for some documentation that we can point newer contributors to in case we spot completely AI-generated code/PRs, as has been the trend for some of the more recent code I've reviewed. I think the same applies for commit messages or phab task comments which sometimes look very obviously AI generated.
Something that fits along the lines of what Travis said - "able to understand, explain, and justify any code that an AI tool contributes to your PR" - should be a good resource!
Thanks, Parthiv Menon
On Sun, 25 Jan, 2026, 18:51 Pine W via Wikitech-l, < wikitech-l@lists.wikimedia.org> wrote:
Hello,
Perhaps this essay is worth reading & taking inspiration from, regarding activities in general whether or not they involve AI: https://en.wikipedia.org/wiki/Wikipedia:Competence_is_required. Especially when experienced volunteers are stretched thin, patience and time for coaching are limited resources, and a decision may be made to end someone's involvement.
My personal opinion regarding products in general is aligned with Dreamy's "I think if we adopt a similar policy or guideline, then it would need to be careful to avoid discouraging new developers learning by doing (i.e. using AI to help them learn)" along with Sohom's "(My personal compass is that it's fine to use AI tooling as long as you adhere to "Own the code you commit"", with the addition I'm opposed to expecting volunteers or a nonprofit's staff time to be used reviewing low quality contributions where those on the reviewing end haven't given informed consent in advance; also see https://www.runtime.news/ai-slop-is-overwhelming-open-source/, quoting from LLVM maintaners: "While new tools enable more development, it shifts effort from the implementor to the reviewer, and our policy exists to ensure that we value and do not squander maintainer time".
Balance is required between coaching, quality, velocity, and resource allocation issues.
Thanks for raising this question, Travis.
Pine🌲
On Sun, Jan 25, 2026 at 6:34 PM Travis Briggs via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
Agreed, my policy includes:
Basically, you should be able to understand, explain, and justify any
code that an AI tool contributes to your PR.
Cheers, -Travis
On Sun, Jan 25, 2026 at 6:20 PM Brian Wolff via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
I'm of the opinion if people can tell you are using an LLM you are using it wrong. Its still expected that you fully understand any patch you submit. I think if you use an LLM to help you nobody would complain or really notice, but if you blindly submit an LLM authored patch without understanding how it works people will get frustrated with you very quickly.
-- Brian
On Sunday, 25 January 2026, Travis Briggs via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
Hi all,
I know there's been a lot of discussion/concern and policy decisions around usage of LLMs for authoring content in Wikimedia projects.
I wonder if there's been similar discussion around AI coding for technical editors? I'm not sure I've seen it "go by". Can someone link to where these discussions are happening?
I also wanted to share a policy I wrote for WP1: https://github.com/openzim/wp1/blob/main/CONTRIBUTING.md#usage-of-llmsai-cod...
Cheers, -Travis
Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org
https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org
https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
Yes I linked to the full policy I wrote for WP1 in my OP, but I'll quote here for people who missed it
Cheers, -Travis
*Usage of LLMs/AI coding assistants* In general, we are okay with contributors using AI tools to help them write code. We do it ourselves. However, the maintainers of this project are not interested in wasting time reviewing low effort or low quality PRs in general. Unfortunately, AI tools can sometimes make it easy to create such PRs. When using these tools, make sure they are in service of your own contribution efforts. That is, the tool should assist you not replace you.
To be completely explicit, please do not point an LLM at one of our issues, have it write a PR that you didn't look at, ask us to review it, and then simply feed our feedback back into the tool. We could do that ourselves, thank you anyways.
Basically, you should be able to understand, explain, and justify any code that an AI tool contributes to your PR.
And if you use an AI tool, please be honest about it when asked. Attempting to mask the use of AI and "pass it off as one's own" is distasteful to say the least, and will make us not want to work with you.
On Sun, Jan 25, 2026 at 8:02 PM Parthiv Menon via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
Hi,
I agree with the points that state that there is a need for some documentation that we can point newer contributors to in case we spot completely AI-generated code/PRs, as has been the trend for some of the more recent code I've reviewed. I think the same applies for commit messages or phab task comments which sometimes look very obviously AI generated.
Something that fits along the lines of what Travis said - "able to understand, explain, and justify any code that an AI tool contributes to your PR" - should be a good resource!
Thanks, Parthiv Menon
On Sun, 25 Jan, 2026, 18:51 Pine W via Wikitech-l, < wikitech-l@lists.wikimedia.org> wrote:
Hello,
Perhaps this essay is worth reading & taking inspiration from, regarding activities in general whether or not they involve AI: https://en.wikipedia.org/wiki/Wikipedia:Competence_is_required. Especially when experienced volunteers are stretched thin, patience and time for coaching are limited resources, and a decision may be made to end someone's involvement.
My personal opinion regarding products in general is aligned with Dreamy's "I think if we adopt a similar policy or guideline, then it would need to be careful to avoid discouraging new developers learning by doing (i.e. using AI to help them learn)" along with Sohom's "(My personal compass is that it's fine to use AI tooling as long as you adhere to "Own the code you commit"", with the addition I'm opposed to expecting volunteers or a nonprofit's staff time to be used reviewing low quality contributions where those on the reviewing end haven't given informed consent in advance; also see https://www.runtime.news/ai-slop-is-overwhelming-open-source/, quoting from LLVM maintaners: "While new tools enable more development, it shifts effort from the implementor to the reviewer, and our policy exists to ensure that we value and do not squander maintainer time".
Balance is required between coaching, quality, velocity, and resource allocation issues.
Thanks for raising this question, Travis.
Pine🌲
On Sun, Jan 25, 2026 at 6:34 PM Travis Briggs via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
Agreed, my policy includes:
Basically, you should be able to understand, explain, and justify any
code that an AI tool contributes to your PR.
Cheers, -Travis
On Sun, Jan 25, 2026 at 6:20 PM Brian Wolff via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
I'm of the opinion if people can tell you are using an LLM you are using it wrong. Its still expected that you fully understand any patch you submit. I think if you use an LLM to help you nobody would complain or really notice, but if you blindly submit an LLM authored patch without understanding how it works people will get frustrated with you very quickly.
-- Brian
On Sunday, 25 January 2026, Travis Briggs via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
Hi all,
I know there's been a lot of discussion/concern and policy decisions around usage of LLMs for authoring content in Wikimedia projects.
I wonder if there's been similar discussion around AI coding for technical editors? I'm not sure I've seen it "go by". Can someone link to where these discussions are happening?
I also wanted to share a policy I wrote for WP1: https://github.com/openzim/wp1/blob/main/CONTRIBUTING.md#usage-of-llmsai-cod...
Cheers, -Travis
Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org
https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org
https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org
https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
Manual testing is an important aspect of working with AI too. The other day I saw a patch submitted to the PageTriage repo that was AI-generated and that a reviewer determined didn't work, which was not a good use of the reviewer's time. Inexperienced patch writers may be tempted to just feed tickets into AI/LLM to generate a patch, then submit the patch without manual testing.
Anyway, I've taken a stab at documenting some of this by creating the page https://www.mediawiki.org/wiki/Using_AI_in_MediaWiki_patches just now. Feel free to edit.
Hope this helps.
Novem Linguae
Hi Novem Linguae
Anyway, I've taken a stab at documenting some of this by creating the page https://www.mediawiki.org/wiki/Using_AI_in_MediaWiki_patches just now. Feel free to edit.
I would suggest REQUIRING that users MUST tick a "I used AI to write this code" box if this was the case, at time of upload of their patch. This may help the reviewer to prioritise the search of hallucinations in the code provided.
Sveta
I don't think that's currently possible due to the wide diversity of interfaces we use (Wikimedia Gerrit can be controlled by us, but what about Github). Also the code upload is done through a third-party CLI which is going to be hard to integrate with.
Regards, Sohom Datta --- Open-source contributor @Wikimedia, @Chromium
On Tue, Jan 27, 2026 at 11:29 AM Svetlana Tkachenko via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
Hi Novem Linguae
Anyway, I've taken a stab at documenting some of this by creating the page https://www.mediawiki.org/wiki/Using_AI_in_MediaWiki_patches just now. Feel free to edit.
I would suggest REQUIRING that users MUST tick a "I used AI to write this code" box if this was the case, at time of upload of their patch. This may help the reviewer to prioritise the search of hallucinations in the code provided.
Sveta _______________________________________________ Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
Good point about diverse interfaces, @Sohom. It made me think that adding something to the commit message guidelines https://www.mediawiki.org/wiki/Gerrit/Commit_message_guidelines might work. We direct people there even if they use a different tool to commit. And we could link to all the useful thoughts here on this thread.
Something like "using LLMs to help contribute is allowed, we ask that you please include [...this note...] in your commit message and read the following advice on coding with LLMs..."
On Tue, Jan 27, 2026 at 11:32 AM Sohom Datta via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
I don't think that's currently possible due to the wide diversity of interfaces we use (Wikimedia Gerrit can be controlled by us, but what about Github). Also the code upload is done through a third-party CLI which is going to be hard to integrate with.
Regards, Sohom Datta
Open-source contributor @Wikimedia, @Chromium
On Tue, Jan 27, 2026 at 11:29 AM Svetlana Tkachenko via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
Hi Novem Linguae
Anyway, I've taken a stab at documenting some of this by creating the page https://www.mediawiki.org/wiki/Using_AI_in_MediaWiki_patches just now. Feel free to edit.
I would suggest REQUIRING that users MUST tick a "I used AI to write this code" box if this was the case, at time of upload of their patch. This may help the reviewer to prioritise the search of hallucinations in the code provided.
Sveta _______________________________________________ Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org
https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
There is also the possibility of adding an AGENTS.md file to the root of a repo, where we can give specific instructions to the agent itself, which can include things like "When submitting a pull request or Gerrit patch, ALWAYS disclose that it was made with the help of an AI agent," or even "Prior to doing any work, ALWAYS ask a Yes or No question whether the user has read our Commit Message Guidelines at this URL: ..." (<< I have confirmed that this works, at least with Claude Code)
Now, the extent to which the agents will *honor* the instructions is not at all clear, and perhaps the presence of an AGENTS.md file could be perceived as an invitation or endorsement for using an agent, but it still could be an option to consider.
On Tue, Jan 27, 2026 at 12:39 PM Dan Andreescu via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
Good point about diverse interfaces, @Sohom. It made me think that adding something to the commit message guidelines https://www.mediawiki.org/wiki/Gerrit/Commit_message_guidelines might work. We direct people there even if they use a different tool to commit. And we could link to all the useful thoughts here on this thread.
Something like "using LLMs to help contribute is allowed, we ask that you please include [...this note...] in your commit message and read the following advice on coding with LLMs..."
On Tue, Jan 27, 2026 at 11:32 AM Sohom Datta via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
I don't think that's currently possible due to the wide diversity of interfaces we use (Wikimedia Gerrit can be controlled by us, but what about Github). Also the code upload is done through a third-party CLI which is going to be hard to integrate with.
Regards, Sohom Datta
Open-source contributor @Wikimedia, @Chromium
On Tue, Jan 27, 2026 at 11:29 AM Svetlana Tkachenko via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
Hi Novem Linguae
Anyway, I've taken a stab at documenting some of this by creating the page https://www.mediawiki.org/wiki/Using_AI_in_MediaWiki_patches
just
now. Feel free to edit.
I would suggest REQUIRING that users MUST tick a "I used AI to write this code" box if this was the case, at time of upload of their patch. This may help the reviewer to prioritise the search of hallucinations in the code provided.
Sveta _______________________________________________ Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org
https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org
https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
FYI, related news that might interest participants in this thread:
1. https://arstechnica.com/ai/2026/03/ai-can-rewrite-open-source-code-but-can-i...
2. https://arstechnica.com/ai/2026/03/after-outages-amazon-to-make-senior-engin...
Pine🌲
On Tue, Jan 27, 2026 at 11:41 AM Dmitry Brant via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
There is also the possibility of adding an AGENTS.md file to the root of a repo, where we can give specific instructions to the agent itself, which can include things like "When submitting a pull request or Gerrit patch, ALWAYS disclose that it was made with the help of an AI agent," or even "Prior to doing any work, ALWAYS ask a Yes or No question whether the user has read our Commit Message Guidelines at this URL: ..." (<< I have confirmed that this works, at least with Claude Code)
Now, the extent to which the agents will *honor* the instructions is not at all clear, and perhaps the presence of an AGENTS.md file could be perceived as an invitation or endorsement for using an agent, but it still could be an option to consider.
On Tue, Jan 27, 2026 at 12:39 PM Dan Andreescu via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
Good point about diverse interfaces, @Sohom. It made me think that adding something to the commit message guidelines https://www.mediawiki.org/wiki/Gerrit/Commit_message_guidelines might work. We direct people there even if they use a different tool to commit. And we could link to all the useful thoughts here on this thread.
Something like "using LLMs to help contribute is allowed, we ask that you please include [...this note...] in your commit message and read the following advice on coding with LLMs..."
On Tue, Jan 27, 2026 at 11:32 AM Sohom Datta via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
I don't think that's currently possible due to the wide diversity of interfaces we use (Wikimedia Gerrit can be controlled by us, but what about Github). Also the code upload is done through a third-party CLI which is going to be hard to integrate with.
Regards, Sohom Datta
Open-source contributor @Wikimedia, @Chromium
On Tue, Jan 27, 2026 at 11:29 AM Svetlana Tkachenko via Wikitech-l < wikitech-l@lists.wikimedia.org> wrote:
Hi Novem Linguae
Anyway, I've taken a stab at documenting some of this by creating the page https://www.mediawiki.org/wiki/Using_AI_in_MediaWiki_patches
just
now. Feel free to edit.
I would suggest REQUIRING that users MUST tick a "I used AI to write this code" box if this was the case, at time of upload of their patch. This may help the reviewer to prioritise the search of hallucinations in the code provided.
Sveta _______________________________________________ Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org
https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org
https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org
https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
-- Dmitry Brant https://dmitrybrant.com _______________________________________________ Wikitech-l mailing list -- wikitech-l@lists.wikimedia.org To unsubscribe send an email to wikitech-l-leave@lists.wikimedia.org https://lists.wikimedia.org/postorius/lists/wikitech-l.lists.wikimedia.org/
wikitech-l@lists.wikimedia.org