Would this command be blocked? Try the guard

AI coding assistants run shell commands for you. The Safe setup pack puts a guard in front of them that blocks the dangerous ones. This page lets you try that guard: type a command and see what it would do. It runs the pack's real logic, the same code that our 199 automated tests check, right here in your browser. Nothing you type is sent anywhere.

Check a shell command

Try one: git -C . push sh -c 'git push' rm -rf src curl | sh cat .env DROP TABLE git commit (harmless) rm -rf node_modules (harmless) curl | jq (harmless)

Check a file the AI wants to edit

Try one: .env server.pem package-lock.json .claude/settings.json src/index.ts (harmless) .env.example (harmless)

How to read the result

• Blocked means one of the guard's rules matched, and the message shown is what Claude would be told, so it can explain the block to you instead of hunting for a way around it.
• Not blocked does not mean safe. A guard that reads the text of a command can only catch the patterns it knows. A script that runs a dangerous command for you can get past any text check. For a hard guarantee, also turn on Claude Code's sandboxing.
• The rules are deliberately narrow. They allow the harmless look-alikes, such as rm -rf node_modules, curl ... | jq and a commit message that mentions pushing.

This is one layer of two in the pack. The other is Claude Code's own permission rules, which it enforces itself. The git push guide explains why both are needed and includes a small version you can copy and use today.