v15 released with fixes for overthinking/indecision and loops on tool errors
I spent lot of time testing and debugging all the reported issues and more. So far v15 is holding up well in my testing. I am hoping this will be the final version, and I can promote it to official release. Please check the corresponding readme for details.
Thanks to everyone who reported issues.
As usual, feedback appreciated.
Hi, thanks for your hard work. I was testing v15 and it was working fine until the context was almost full and claude code started auto compacting. This was the error message:
Error: Error during compaction: Failed to generate conversation summary - response did not contain valid text content
After that I tried /compact with v8 and worked fine. I'll continue testing more.
oh... I have never managed to get qwen to compact, right from the beginning, even with the official template
I'm using v16 and compact is working fine. Perhaps it was related to the change made in this version.
Thanks for the report. I have now promoted v16 (newer than the one you used) to official release.