careful

You can get a warning before running destructive bash commands by enabling a PreToolUse hook that pattern-matches risky commands like rm -rf, git force push, and SQL DROP, then gates execution until you confirm. This prevents accidental data loss during production deployments.

To prevent accidental SQL DROP or TRUNCATE during database maintenance, you need a pre-command safety gate that intercepts destructive database statements and requires explicit confirmation before execution. This reduces the risk of irreversible data loss in production environments.

Yes, you can block kubectl delete and docker system prune by activating a careful mode that uses a persistent state file to intercept cluster and container cleanup commands. It prompts for confirmation before allowing these high-risk operations to proceed.

Careful mode stays active across sessions by using a persistent state file as a gate. This ensures continuous protection against destructive commands with effectively zero overhead when disabled, so you do not need to re-enable it every time.

A bash safety guardrail intercepts destructive command patterns including rm -rf, DROP/TRUNCATE/DELETE SQL statements, git force-push and reset, kubectl delete, and docker system prune, while explicitly allowing safe exceptions like deleting node_modules or dist directories.

To start using careful mode before making production changes, tell your agent to run the enable command before you begin debugging or deploying. This activates the PreToolUse hook to intercept and warn about destructive shell or database commands.