I’m getting this problem all the time when using Deepseeker in Chat. Its a real productivity killer and I swear other people have been reporting similar, though I just can’t find the thread about it now.
Hey, this can happen sometimes with Markdown outputs in thinking models, but we are working on improving the support on this in the editor soon!
For now, starting a new Composer will stop the output from being mangled, but there is still the risk it gets funky again.