BTW, any concerns/objection in merging the coderev...
# proj-extensions
c
BTW, any concerns/objection in merging the codereview-roasted and codereview skill? Feels a bit tricky to maintain two sets of code review skills facepalm (cc @limited-student-10747 @elegant-branch-54423)
l
yeah, go for it
e
works for me πŸ‘
c
I wonder if it will work well πŸ€” The roasted review has a way to channel some LLMs towards... I don't know how to name this, it's more of an intuition... that it's not the same as codereview. Maybe it's more attention to those principles that otherwise it's possible to treat a bit more superficially perhaps, but I could be wrong.
On the other hand, I feel it's the other way around too... codereview is maybe too nice πŸ˜…
l
I kinda think that we could remove the linus persona and just specifically point out the points we want it to cover. We also now have the ability to A/B test code review using the laminar monitoring, so we could try different options
c
we could remove the linus persona
agree nod just keep those principals
We could just monitor the code-review plugin performance change before and after the change, maybe @wide-doctor-37830 we could team up to do this since irrc you were looking at the code review perf metrics?
nod 1
Merging
code-review
skill and
codereview-roasted
skill - agent approved. Just need a quick human πŸ‘€ before we ship it β€’ https://github.com/OpenHands/docs/pull/456 β€’ https://github.com/OpenHands/extensions/pull/175
c
Actually having the two separate ones has been very handy in conversations with customers about how to make skills their own and the Linus personality really resonates in story telling I do with them. If we are going to merge these I'll probably move them to another repo or to a different place in the same repo to preserve the teaching tool.
πŸ‘ 1
c
I'm fond of Linus too, because I feel that from the LLM perspective, it channels its response due to the history of Linus in a "stronger" direction - that just telling it principles might not go far enough; but that's just me being silly maybe. Or maybe it works a bit differently with LLMs more prompt-sensitive than Claudes are. So I see it, but most of us use Claudes so idk What if these skills would be part of a non-default marketplace?
c
I'll probably move them to another repo or to a different place in the same repo to preserve the teaching tool.
yeah, maybe it will be helpful to have these in another repo...
What if these skills would be part of a non-default marketplace?
The main pain point is that, whenever we add new stuff to codereview-roasted πŸ˜… we have to do the same for codereview which is.. not pretty
βœ… 1
c
I'm fond of Linus too,
And actually the personality is what helps the content be interesting and engaging to humans. And that HCI stuff is really important when it come to actually engaging with and getting benefit from the tool.
nod 1
it channels its response due to the history of Linus in a "stronger" direction
I think you are right about this, though I don't have the data to back it up.
I got a code review today that felt off--not enough critical feedback. I created a new branch and PR bumping ci back to use the version before the merge and got more meaningful critical feedback. It's only one data point, but I think we should keep eyes on this over next few days.
c
I start to suspect if it is because we removed the name "linus tovards"πŸ˜…
nod 1
We can keep it for a few more days, maybe this could be a measurable signal we can get from Laminar, it would be fun if we can actually proof this w/ numbers cc @wide-doctor-37830
c
My hypothesis is that maybe it brings with it a history stored in parametric knowledge, which then… hey πŸ˜…
l
We should run an AB test πŸ˜‚
linus vs no linus