[BUG] Improve plan mode by allowing non-side-effect modifications

Bug Description

It would be very helpful to allow non-side-effect changes, such as debug statements, comments, and break indicators, while in plan mode.

I have found that my workflow and results are much better when restricting Claude to only altering code by adding these non-side-effect statements.

It has been a challenge to implement this restriction using context, agents, or prompts.

Perhaps we can discuss how this may be implemented in this thread.

I am calling this a BUG because I spend 8-10x more time trying to fix Claude's bad assumptions than developing my concepts. Claude goes down these "rabbit holes". Every time I require debug statements only, and then feed the debug back in, my issues are solved without the hackfest. So my idea is a plan mode that allows these non-side-effect modifications would enhance planning.

Or maybe just create a new "mode" called Trace Mode or such. IDK. 🤷‍♀️