Describe the bug
The last week or so I've had this happen several times. I've got a few sessions going and then one or more of them will just freeze their updates. Like I can do some stuff in the UI still, so it's not deadlocked, but I no longer get updates. Under the hood the agent is still working as I see the events.jsonl getting lines added. It shows this
even while in the events.jsonl it shows the agent kept going and responded to my conversation.
It seems it will do this forever. The working bytes stops tallying I see no progress. I have not waited more than 30m to see if it ever recovers, but it doesn't recover within that time limit. I have to kill the terminal tab and resume meaning whatever it was working on at that moment is lost and has to be redone which is not cool.
It won't let me send new messages or run most / commands. Won't even let me CTRL + C to exit. I have to just close the terminal to get it to stop and restart.
Affected version
1.0.91
Steps to reproduce the behavior
I am not sure what causes it. For the particular conversation where this happened twice (froze, restarted, had a turn or two then it froze again) I have had this conversation compacted 33 times so it's been going for quite awhile. Is it just having a hard time handling that much history? I also expanded the "Thinking" on an Opus 5.5 Max thought which can be pretty lengthy, not sure this is a consistent cause though.
Expected behavior
If the agent is still going, why isn't the UI? How can I force it to "refresh" or something so I don't have to kill the agent which is still plugging along nicely?
I expect the UI stays up to date, even for long conversations. I expect a way to help it recover if it can't handle such a long history.
Additional context
Windows x64 in Windows Terminal.
Let me know if there's a way to get you the logs from within the broken session if you need them.
Describe the bug
The last week or so I've had this happen several times. I've got a few sessions going and then one or more of them will just freeze their updates. Like I can do some stuff in the UI still, so it's not deadlocked, but I no longer get updates. Under the hood the agent is still working as I see the
events.jsonlgetting lines added. It shows thiseven while in the events.jsonl it shows the agent kept going and responded to my conversation.
It seems it will do this forever. The working bytes stops tallying I see no progress. I have not waited more than 30m to see if it ever recovers, but it doesn't recover within that time limit. I have to kill the terminal tab and resume meaning whatever it was working on at that moment is lost and has to be redone which is not cool.
It won't let me send new messages or run most
/commands. Won't even let me CTRL + C to exit. I have to just close the terminal to get it to stop and restart.Affected version
1.0.91
Steps to reproduce the behavior
I am not sure what causes it. For the particular conversation where this happened twice (froze, restarted, had a turn or two then it froze again) I have had this conversation compacted 33 times so it's been going for quite awhile. Is it just having a hard time handling that much history? I also expanded the "Thinking" on an Opus 5.5 Max thought which can be pretty lengthy, not sure this is a consistent cause though.
Expected behavior
If the agent is still going, why isn't the UI? How can I force it to "refresh" or something so I don't have to kill the agent which is still plugging along nicely?
I expect the UI stays up to date, even for long conversations. I expect a way to help it recover if it can't handle such a long history.
Additional context
Windows x64 in Windows Terminal.
Let me know if there's a way to get you the logs from within the broken session if you need them.