Permission Mode
ELI5 — The Vibe Check
Permission mode controls whether your AI agent confirms before acting, auto-approves low-risk moves, or skips all gates entirely. It's the trust dial on your AI tool. Turn it up and the agent ships fast, asks questions never. Turn it down and it's basically just a very smart autocomplete. Most developers calibrate this dial through a painful trial-and-error process involving deleted files.
Real Talk
In Claude Code, permission mode determines the agent's behavior when it needs to take potentially impactful actions — writing files, running shell commands, making network requests. The three modes are: interactive (confirm everything), auto (classify and approve safe actions), and dangerously skip permissions (no gates at all). Permission mode is set per-session or via config and directly affects how fast and how risky your vibe-coding sessions run.
When You'll Hear This
"Set permission mode to auto for this session — we're just refactoring." / "What permission mode are you running? It just deleted the migration folder."
Related Terms
Auto Mode
Auto mode is Claude Code's permission classifier — the middle ground between confirm-everything and YOLO mode.
Claude Code
Claude Code is Anthropic's agentic coding tool that lives in your terminal.
Dangerously Skip Permissions
Dangerously skip permissions is the nuclear option — bypassing all AI confirmation gates. Named 'dangerously' because the devs meant it.