Skip to main content

Permissions

Overview​

Set up proper permissions for AI development tools to balance productivity with security. Auto-allow safe operations, require approval for dangerous ones.

Risk-Based Classification​

🟢 Safe Auto-Allow​

  • git status, git log, git diff
  • npm test, bundle exec rspec
  • ls, cat, grep, find
  • curl -s -I http://localhost:*

🔴 Always Require Approval​

  • git push, git push --force
  • npm publish, gem push
  • rm -rf, sudo rm
  • aws *, kubectl *
  • docker run, docker push

Setup Guidelines​

  1. Start restrictive - Begin with minimal auto-approvals
  2. Add incrementally - Approve commands as you encounter them
  3. Monitor usage - Review what commands are requested frequently
  4. Adjust based on patterns - Auto-approve commands you find yourself approving repeatedly

Command Evaluation​

Before auto-approving, ask:

  • Can this command cause data loss?
  • Can this command expose sensitive information?
  • Can this command affect production systems?

If any answer is "yes", require manual approval.

Template Available​

There's a permission management template in .claude/templates/ that helps you configure AI tool permissions safely using natural language descriptions.