Biggest f-ups by your Agent

A lot of us are working with or working on AI Agents so it is helpful to know more error scenarios that aren't the obvious -- rm -rf * or git reset.

What sort of irreversible changes/actions Agents have made which cost you a lot of time.

I'm looking for first party reports but you're also welcome to share stories you saw on another network.

Bonus: be productive, what guardrails could've/should've existed to protect you from the situation.