safety-guard

To block destructive commands in production, safety-guard intercepts harmful operations before execution. It enforces protective rails by warning users or autonomous agents and prompting for confirmation to prevent accidental data loss during migrations or deployments.

You can lock file edits to a specific directory tree by activating Freeze Mode. This safety mechanism prevents autonomous agents from modifying files outside the intended scope, ensuring operations remain strictly contained within authorized boundaries.

The best way to protect production systems during autonomous workflows is Guard Mode, which combines Careful and Freeze protections. This provides maximum safety by intercepting destructive commands and locking file edits to a specified directory tree simultaneously.

Yes, safety-guard logs blocked destructive actions to ~/.gemini/safety-guard.log. This audit trail records intercepted commands and actions, allowing you to review what operations were blocked during sensitive maintenance tasks or autonomous agent runs.

To resume operations after a destructive command is blocked, you must use the explicit unlock command provided by safety-guard. This ensures intentional confirmation before proceeding with sensitive operations that were previously intercepted.

You need safety rails for production maintenance tasks like migrations or deployments where accidental data loss is a risk. Enabling protective modes ensures destructive commands are intercepted and file edits are scoped correctly to prevent unintended changes.