Small fixes and optimisations

[CodeAndCanvas728]

While trying to get a Qwen3.6-35B based model running in SwiftLM and Pi coding harness, I ran into a few bugs and optimisation opportunities. Would it be ok for me to open a couple of small PR's addressing these things?

Genuinely like this project and happy to help out as much as I can!

[solderzzc]

Hi, @CodeAndCanvas728 , thanks for your support and contribution, please feel free to make changes to SwiftLM, Pi coding harness is also my local model agent, I chose it because of a comparision with other coding harness I've done locally before. What I have local is Qwen3.8-27B.