Permissions

Introduction

When you're working with an agent, you need to think carefully about risk versus reward. Give Claude Code too much power, and it might accidentally delete your entire file system. That's why Claude Code has a detailed permissions model that's very strict by default about what it allows the agent to do.

You have complete control over what Claude Code can access. You can approve actions one at a time, create rules for entire categories of commands, and even share these permissions with your team so everyone gets the same safe setup.

Steps To Complete

Understanding the Permissions Flow

[ ] Run a safe command with Claude Code

Ask Claude Code to run a safe command and observe that it executes immediately without asking for permission.

run echo hello

[ ] Run a command that requires approval

Ask Claude Code to run a type check command. Notice that it now shows you:

The exact command it wants to run
The reason it wants to run it
Three options: approve this time, approve all similar commands, or deny

run a type check on the project

Working with Permission Files

[ ] Open .claude/settings.local.json in your project

This file stores all the permissions you've granted. Look for a permissions property with allow and deny arrays.

[ ] Add a permission manually to the allow array