/review #748
/review #748
Conversation
added openAI
Simple prompt engineering, one response per user per question.
better initial prompt
update prompt
original context indexable across repos, issues and pull requests
comments etc
waiting for more input
pulls from sources using the hashtag in issue body
total token usage now displayed
feed gpt the spec and diff for it to analyze
will only spec check once now fixed tsc issues
sort import
allow it to be set in org and repo config
reset default
unused dep
no longer undefined
set to 8k in config 4k by default
improved pr-review and context for both ask and pr-review commands
added gpt pr review
✅ Deploy Preview for ubiquibot-staging ready!
To edit notification comments on pull requests, go to your Netlify site configuration. |
|
It could be interesting to plug this into CI on push so that it won't pass until the bot thinks it passed. Until we implement automatically closing pull requests due to poor performance though, we should not automatically invoke the |
clear error message
conflict resolved
remove suffix if its an error
using hashtag text
good point. we can create an ai-review ci/cd running on each pull request. |
|
I'm really looking forward for this one to get merged in because I think it will be able to help us merge the other pull requests faster. @Keyrxng any updates? By the way, https://github.com/Keyrxng/UbiquityBotTesting/pull/41#issuecomment-1719534993 is my favorite review result. |
I felt that I was waiting for you guys but I will stop work on Blame and just ensure everything is operating inline with the favourite you mentioned which I'm sure it is. So long as it is my work with this for now is done unless there are further requests. Will let you know in a couple hours. |
So the response is short and sweet if it's meeting the spec while when it's failing it's going into depth with things and suggesting how to meet the spec or what is missing from the current implementation. Done my best to remove any type of overview or listings, I think that ratio you mentioned is better in the more recent stuff than the one you said was your favourite so far. |
|
What would be so ideal is if it could be tested on closed prs or something as that amount of real data would be so much better to work with than my shitty gpt-assisted test issues 🤣 Hope it's looking better tho, lmk and I'll wrap this up tomorrow. |
better signal to noise more concise
Remove blockquote from footer
change from token limit to max tokens
|
Does it make sense to eventually turn all the AI functions into https://docs.netlify.com/functions/background-functions/ @0xcodercrane rfc |
| @@ -8,6 +8,8 @@ | |||
| "disable-analytics": false, | |||
| "comment-incentives": false, | |||
| "register-wallet-with-verification": false, | |||
| "openai-api-key": null, | |||
| "openai-max-tokens": 8000, | |||
There was a problem hiding this comment.
1/ Why put the null value in default.json here?
2/ @pavlovcik What's your recommendation value? If you put it in the comment, I can directly push to speed up the merge.
The background function is a new feature of netlify and I think it's worth to take the 15 mins execution timeout benefit of that feature. but my concern from their docs is the pricing plan. they just say it may not be available on all the pricing plans. |
|
Okay I'm back in the game All good to pull things from this pr and start with a clean commit history from the most recent dev branch, seems cleaner and I can avoid dealing with all of the conflicts? |
Resolves #746
invoked via /review
The implemented prompts differ from the original spec but after experimenting with variations, I believe the current setup provides uniform formatting and covers all requirements of the spec.