What context do you preserve before switching AI models mid-debug?

A coding-tool question about preserving failing command, exact error, changed files, and failed attempts before moving a bug to another model.

Switching AI models during a long debugging session can reset the conversation if the context packet is weak. I now save the goal, failing command, exact error, environment, changed files, and fixes already tried before moving the case. Otherwise the next model repeats the same basic suggestions. What do you preserve before switching models mid-debug?