Persistent Chain-of-Thought (CoT) and Action Logs for Debugging

Currently, during a run, the Chain-of-Thought (CoT) is only visible while the model is thinking but disappears once the run is complete, making it impossible to review or debug the model’s reasoning. I am requesting that CoT be persistently logged and saved per run, along with a full log of all actions taken (tool calls, API requests, etc). This would allow users to inspect exactly how and why a task succeeded or failed, improve prompt tuning, and diagnose failures more effectively. Having a permanent, viewable record of both CoT and actions would significantly improve transparency, debugging, and agent development workflows.