Permission Modes

🎥 Video lesson — video not yet migrated (source Skool video id: 1db122f7e544499faa73baa7710ea46c).

Claude Code has four permission modes — think of them as levels of trust you give your AI employee.

Level one: Default. Claude Code asks before every single action — reading a file, creating a file, running a command. You approve each one. Safest mode. Start here. Level two: Auto-edit. Claude Code can create and edit files without asking, but still asks before running commands like installing packages or running scripts. Level three: Plan Mode (covered in the previous chapter) — Claude Code plans everything first, shows you the full plan, and only executes after your approval.

Level four: "Dangerously Skip Permissions" — and the name is intentional. Claude Code executes everything without asking. No permission prompts. Full speed. Powerful when you're in a flow and trust the task, but if Claude Code decides to delete something or run a destructive command, it won't ask first. Best used on projects where you have git commits to fall back on. Experienced users often run in bypass mode for speed. But if you're new, don't start here. You need to understand what Claude Code is doing before you let it run free. Start with default, build trust, then gradually increase autonomy.