Wow. I’m completely floored. Your agent just ran rm rf ~ on my system with no related prompting. I had tasked it with reviewing files within a specific repo to see if they were referenced elsewhere so I could clean up orphaned files and boom. The nuclear command. There is very little reason, if not none, for this command to be suggested. Let alone for it to have write access with that level of impunity. I own fully that fat fingered accept on the command and fortunately I have backups but you could do serious mission critical damage with that kind of cowboy code suggestion. It makes me think your agents may be trained on malicious code because there was no communication context to it being put up. Either from me or from the agent. Very disappointing.
Hey, thanks for the report. Actually, we’re already looking into this issue.
This topic was automatically closed 22 days after the last reply. New replies are no longer allowed.