Most Claude Code security failures stem from misunderstanding where permission modes end and sandbox boundaries begin. This post breaks down the four modes, `--allowedTools` whitelisting, and how bubblewrap restrictions work together to prevent unauthorized access.
