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
I found that if it run long build it uses llm to watch build process, and seems if there is long build with a lot of data just eat out tokens literally for nothing. I was at 10% of context and when it did docker compose up --build it become to "automatically compact context" , before build done (it also failed it, but may be because I finally stopped conversation)
Suggestion:
Limit head of build process, on "watch mode" (not sure how you call such task) just take "tail" not all data, for check step, and may be not place it to history (make agent decide ask it to record only errors and final lines to main thread ?) .
Activity
[-]A way to limit commanlne tokens[/-][+]A way to limit command executon tokens[/+]etraut-openai commentedon Mar 29, 2026
This feature request hasn't received enough upvotes, so I'm closing it.