When is an AI agent's approval prompt a security boundary?gist.github.com2 points·nrig··0 commentsOpen articleSaveView on HN