[CI] ci: add ai workflow for fixing nightly failures - #716
Conversation
github-actions
Bot
changed the title
SimonHeybrock
left a comment
SimonHeybrock
left a comment
There was a problem hiding this comment.
This seems very complicated. Why not just tell Claude to open a PR if it finds a fix, else and issue with results of investigation?
We could do that. (I think I would prefer that too.) The advantage of this approach is that the agent only has read-only access to the github repo, except for the one PR that is opened automatically if it leaves changes. 🤷 |
|
Okay I think the new version @SimonHeybrock suggested where the surrounding logic (context creation, PR submission, etc) is removed and the agent is told to handle it is much better. |
f2c4bd3 to
2a2980b
Compare
|
It would be nice to get this experiment up and running. We will likely have to tweak things based on how it does. |
SimonHeybrock
left a comment
SimonHeybrock
left a comment
There was a problem hiding this comment.
Seems reasonable. I suspect we will get cases where Anthropic API is down (or similar issues), resulting in silent Nightly fails that receive neither issue nor PR, but we can address that later based on how often it happens.
20af4e0 to
d466bd6
Compare
2 participants
This is a suggestion for how we could set up a workflow that lets claude investigate nightly test failures and suggest a fix.