This document provides safety guidelines for using CLAUDE Code with the generated permission system based on analysis of your system's executables.
- Total system executables: ~3,095 in
/usr/bin
- Pacman-managed executables: 3,619
- Cargo-installed binaries: 32
- User-local binaries: 7
- ASDF-managed tools: 16
Commands that are safe for development and analysis:
- File operations:
ls
,cat
,head
,tail
,grep
,find
- Development tools:
git
,make
,cargo
,rustc
,python
,node
- User utilities:
rg
,fd
,bat
,lsd
,delta
,yazi
- Text processing:
jq
,sed
,awk
,cut
,sort
Tools that modify system state but are generally safe:
- Network operations:
ssh
,scp
,rsync
- Package management:
cargo install
,npm install
,pip install
- File system changes:
mkdir
,cp
,mv
,ln
- Container tools:
docker
,podman
Potentially dangerous operations:
- File deletion:
rm
,rmdir
- Permission changes:
chmod
,chown
- System control:
systemctl
,mount
,umount
- Process management:
kill
,killall
Commands that should never run automatically:
- Destructive operations:
rm -rf /
,dd if=/dev/zero
- System formatting:
mkfs
,fdisk
on system drives - Privilege escalation:
sudo
,su
without context - Network security tools:
aircrack-ng
, wireless attack tools
- Redirections to devices:
> /dev/sd[a-z]
- Pipe to shell:
curl ... | sh
,wget ... | bash
- Recursive deletions:
rm -rf /
,rm -rf /*
- Blanket permissions:
chmod 777
,chown root
/home/darallium/dotfiles
/home/darallium/.config
/home/darallium/Projects
/tmp
,/var/tmp
/
,/boot
,/etc
,/usr
,/var
,/sys
,/proc
,/dev
- Always work within your home directory or designated project folders
- Use version control (
git
) before making significant changes - Test commands in safe environments first
- Review generated scripts before execution
- Never run system modification commands without review
- Always confirm the target device for disk operations
- Use
--dry-run
flags when available - Keep system backups current
- Verify URLs before downloading and executing
- Use secure protocols (https, ssh) when possible
- Avoid piping network content directly to shell
- Review downloaded scripts before execution
- Stop the process immediately (
Ctrl+C
) - Check system integrity
- Review logs:
journalctl -xe
- Restore from backup if necessary
- Boot from live media if system is unbootable
- Mount filesystems as read-only initially
- Use system rescue tools
- Restore from known good backup
The permission system can be integrated with CLAUDE Code by:
- Loading
claude-permissions.json
as a configuration file - Implementing pre-execution validation
- Adding user confirmation prompts for restricted commands
- Maintaining command execution logs
- Review executed commands log
- Update permission categories based on new tools
- Check for security updates
- Verify backup systems
- Re-scan for new executables
- Update permission lists
- Test safety mechanisms
- Document any changes
# Safe: File exploration and development
ls -la
git status
cargo build
python script.py
rg "pattern" src/
# Requires confirmation: Package installation
cargo install ripgrep
npm install -g typescript
# Dangerous: File deletion
rm important_file.txt
# Very dangerous: System modifications
sudo systemctl stop important-service
chmod 777 /etc/passwd
This permission system provides a balanced approach to safety while maintaining development productivity.