post mortems / bug hunting -- pinpointing what part of the logic was to blame for a certain problem.
Then look at the code, the session will only confuse. To read an LLM's explanation is to anthropomorphize what will just be a probabilistic incident.
this is what granular commits are for, the kilobytes long log of claude running in circles over bullshit isn't going to help anyone