Every so often, Noah stops and asks you to approve something before it does it. Fair question: will you actually understand what you're saying yes to?
The reason is required
You will. I made the plain-English explanation a requirement. When the AI wants to run a command, it has to hand over two things: the command, and a plain sentence saying what it does and why. The sentence isn't optional — Noah won't even offer you the command without it.
A lot of software does this backwards: it acts, and you hope it tells you why. A polite "always explain yourself" is easy for an AI to forget when it's focused on the task. So I made the explanation part of the request itself. There's no version of the command without a reason attached.
You approve a plain sentence
Behind the scenes, the command might be a line of technical code. What you see on screen, and say yes to, is something like "Delete old log files to free up space" — in plain words, at the moment you decide.
The explanation sticks around, too. Afterward you'll see "Approved: Delete old log files to free up space" in the history, or "Skipped" if you said no. Come back a week later and you'll still see, in plain words, what you agreed to.
One limit
This makes sure there's always an explanation. It can't make sure the explanation is correct. The AI writes the sentence, and it can't double-check that the sentence really matches the command. Making sure the command is actually safe is a different job — the safety checks from the first post.