Audit mode for agent iterations

[hypeJunction]

Very often when I use the agent I find myself turning the wrong corner and being trapped in an AI mess with no way to go back. Hence committing often (literally after each set of agent edits) is becoming increasingly prerequisite.

What if:

• after each iteration of AI comments/edits/thinkings, the user had an option to not only accept the changes, but to also create a commit message summarising the changes and including all AI output into the commit body
• we could understand the entire process of AI thinking while making code reviews
• we could train other AIs to audit and verify the changelogs of other agents
• we could instruct the agent to revert back 2 last steps making necessary edits to the git history
• prompts became a part of the code review process