safety layer / agent shell

Let agents run commands.
Not destroy the machine.

The Destructive Command Guard blocks dangerous git and shell commands before agents execute them—keeping autonomy useful without handing over an unguarded terminal.

Install the guardView source
inspectcommand intent
blockdangerous actions
allowsafe work
dcg / policy engine● ARMED
agent $ git reset --hard HEAD~12
BLOCKED
destructive history rewrite detected
requires explicit human approval
agent $ git diff --stat
ALLOWED
read-only repository inspection
policy examples illustrative
guard the command boundarygithub.com/Dicklesworthstone/destructive_command_guard