You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Now that the big restructuring is in from last October would you like me to add Repo Assist here? I think we can use it to make progress on the feature backlog and technical debt, and generally take this library forward.
The assist really does three things
Labels issues
Puts analysis comments on issues (usually at most one comment)
Proposes PRs by solving issues or generally "taking the repository forward". It never merges PRs of course.
It can be customized to the specifics of this repo if needed.
Let me know what you think. I've found it makes repo maintenance really quite enjoyable and would be happy to help out here.
Tbh i am a bit torn on this. This easily leads to many PRs and just shifts the backlog from implementing the Features to reviewing ai-generate PRs. But if you want to try go ahead @bvenn
Stats can be a bit different to structured programming however, as the coding can be more subtle, the types less help, the conditions around epsilons and divergence and so on trickier.
Equally the models have clearly read a vast amount of stats code and will almost certainly "know" the implementation technique for every stats method imaginable. And if they don't you can just point at a solid implementation in another language and they will duitifully transfer-learn it over
@bvenn I will add Repo Assist on a daily cadence initially.
Are you OK with me co-maintaining or would you both like to maintain instead of me (I'm OK either way, but would only merge the more basic things as my stats knowledge is not super-deep).
Also are there any other maintainers we should clear this with? It's important to get everyone onboard.
@bvenn I've added the workflow, on daily cadence, with 3 initial runs.
Initially it will label existing issues, then start to respond with comments to old issues. Comments should always be checked.
Occasionally we can get negative reactions to AI comments on old issues. That's only happened once, but you can see why people might find it a bit surprising to see an AI analysis of an issue from years ago. If it happens just let me know and I'll record it at fsprojects/.github#58 and ask for further feedback. Realistically the AI comment will only have improved the chances of the issue being actioned and analyzed. But some people react negatively to any AI assists.
@dsyme I had a look into some of the topics and they look quite promising. I curious having a deeper look, but am unable until thursday. If you want you could start, otherwise I start on thursday 👍
And regarding to your offer to co-maintain FSharp.Stats @dsyme: I'm happy having you onboard 🚀 Unfortunately, I havent had the time I hoped for maintenance and developing new features, so I'd love seeing effort put into it.
@dsyme I had a look into some of the topics and they look quite promising. I curious having a deeper look, but am unable until thursday. If you want you could start, otherwise I start on thursday 👍
You make a start at your own pace, I'll act as junior here, since my stats knowledge is weak.
I'll nudge and help with procedural things.
Note you can write "/repo-assist DO SOMETHING" in comments or issues to get extra things done beyond the automation.
@dsyme I've started reviewing some pull requests and they look really promising! Some could be merged right away; others, of course, require more research. However, you could definitely reduce the frequency, as I cannot facilitate a daily review cycle.
Feel free to pick whichever sort you like. Perhaps @LibraChris or @muehlhaus have students who would benefit from these kinds of reviews to gain insights and improve FSharp.Stats simultaneously. No pressure, just an idea that came to my mind while scrolling through the PRs 🙂
@dsyme I've started reviewing some pull requests and they look really promising! Some could be merged right away; others, of course, require more research. However, you could definitely reduce the frequency, as I cannot facilitate a daily review cycle.
Would you like me to merge them or would you like to do it? Thanks :)
@bvenn@kMutagene Checking in again. I'd like to start getting some velocity on the fixes and improvements in the repo. Are you ok with me working to action some of the fixes? Thanks
Hey @dsyme sorry for the long silence, just some busy stretches in academia. I have been heavily into local agentic coding in the last month and now finally have a better understanding on how to use and manage a tool like this.
I have a few general questions about repo assist:
does it respect and use a AGENTS.md file? (i have read your post, it does)
does it use skills, and if so which format (i guess this bridges also to the next question)
does it use skills, and if so which format (i guess this bridges also to the next question)
We need to write a full guide on this for GH-AW, I'll get back to you.
The easiest path for now is not to use skills in the repo and just put what you need in AGENTS.md or the local copy of the repo-assist workflow itself (which in a way is 11 different skills = tasks - there is not that much difference). If you want to factor things out you can use imports in GH-AW too (https://github.github.com/gh-aw/reference/imports/)
@kMutagene@bvenn Could we make progress on this again? We have lots of agentic fixes coming in, and I think they're pretty solid. Should we just start reviewing/merging them?
Activity
bvenn commentedon Mar 9, 2026
Hi @dsyme
I'd love to give it a try. We have many backlog-topics for which the assist may be a great support.
kMutagene commentedon Mar 20, 2026
Tbh i am a bit torn on this. This easily leads to many PRs and just shifts the backlog from implementing the Features to reviewing ai-generate PRs. But if you want to try go ahead @bvenn
dsyme commentedon Mar 22, 2026
After using Repo Assist to clear the Deedle backlog I'm pretty convinced it will work well here. https://bsky.app/profile/fsharponline.bsky.social/post/3mhom3pg7c22u
Stats can be a bit different to structured programming however, as the coding can be more subtle, the types less help, the conditions around epsilons and divergence and so on trickier.
Equally the models have clearly read a vast amount of stats code and will almost certainly "know" the implementation technique for every stats method imaginable. And if they don't you can just point at a solid implementation in another language and they will duitifully transfer-learn it over
dsyme commentedon Mar 22, 2026
@bvenn I will add Repo Assist on a daily cadence initially.
Are you OK with me co-maintaining or would you both like to maintain instead of me (I'm OK either way, but would only merge the more basic things as my stats knowledge is not super-deep).
Also are there any other maintainers we should clear this with? It's important to get everyone onboard.
dsyme commentedon Mar 22, 2026
@bvenn I've added the workflow, on daily cadence, with 3 initial runs.
Initially it will label existing issues, then start to respond with comments to old issues. Comments should always be checked.
Occasionally we can get negative reactions to AI comments on old issues. That's only happened once, but you can see why people might find it a bit surprising to see an AI analysis of an issue from years ago. If it happens just let me know and I'll record it at fsprojects/.github#58 and ask for further feedback. Realistically the AI comment will only have improved the chances of the issue being actioned and analyzed. But some people react negatively to any AI assists.
dsyme commentedon Mar 23, 2026
Here's a typical response - seems pretty strong and would allow us to close out the issue with confidence
#284
[-]Repo Assist?[/-][+]Repo Assist (discussion thread)[/+]dsyme commentedon Mar 24, 2026
@bvenn Woukd you like to start actioning the items in #357 or would you like me to do it? Thanks!
bvenn commentedon Mar 24, 2026
@dsyme I had a look into some of the topics and they look quite promising. I curious having a deeper look, but am unable until thursday. If you want you could start, otherwise I start on thursday 👍
bvenn commentedon Mar 24, 2026
And regarding to your offer to co-maintain FSharp.Stats @dsyme: I'm happy having you onboard 🚀 Unfortunately, I havent had the time I hoped for maintenance and developing new features, so I'd love seeing effort put into it.
fyi @muehlhaus
dsyme commentedon Mar 24, 2026
You make a start at your own pace, I'll act as junior here, since my stats knowledge is weak.
I'll nudge and help with procedural things.
Note you can write "/repo-assist DO SOMETHING" in comments or issues to get extra things done beyond the automation.
dsyme commentedon Apr 4, 2026
@bvenn Checking in
Thanks! :)
bvenn commentedon Apr 7, 2026
@dsyme I've started reviewing some pull requests and they look really promising! Some could be merged right away; others, of course, require more research. However, you could definitely reduce the frequency, as I cannot facilitate a daily review cycle.
Feel free to pick whichever sort you like. Perhaps @LibraChris or @muehlhaus have students who would benefit from these kinds of reviews to gain insights and improve FSharp.Stats simultaneously. No pressure, just an idea that came to my mind while scrolling through the PRs 🙂
dsyme commentedon Apr 20, 2026
@bvenn I've reduced to weekly. You can still invoke manually via "/repo-assist COMMENT" on any issue or PR comment
dsyme commentedon Apr 20, 2026
Would you like me to merge them or would you like to do it? Thanks :)
dsyme commentedon May 8, 2026
@bvenn @kMutagene Checking in again. I'd like to start getting some velocity on the fixes and improvements in the repo. Are you ok with me working to action some of the fixes? Thanks
kMutagene commentedon May 11, 2026
Hey @dsyme sorry for the long silence, just some busy stretches in academia. I have been heavily into local agentic coding in the last month and now finally have a better understanding on how to use and manage a tool like this.
I have a few general questions about repo assist:
does it respect and use a(i have read your post, it does)AGENTS.mdfile?what model is currently used for these workflows?(it currently is sonnet 4.6)Thanks a lot for providing this to fslab. I appreciate it a lot more now and will try to get a handle on the review work.
dsyme commentedon May 18, 2026
Great to hear @kMutagene
We need to write a full guide on this for GH-AW, I'll get back to you.
The easiest path for now is not to use skills in the repo and just put what you need in AGENTS.md or the local copy of the repo-assist workflow itself (which in a way is 11 different skills = tasks - there is not that much difference). If you want to factor things out you can use imports in GH-AW too (https://github.github.com/gh-aw/reference/imports/)
dsyme commentedon Aug 24, 2026
@kMutagene @bvenn Could we make progress on this again? We have lots of agentic fixes coming in, and I think they're pretty solid. Should we just start reviewing/merging them?