safety-guard

You can prevent destructive operations by implementing guardrails that intercept dangerous commands before execution. This skill uses PreToolUse hooks to block unauthorized edits and log blocked actions for post-mortem analysis during autonomous executions.

Careful mode intercepts destructive commands and asks for confirmation, while Freeze mode locks edits to a defined directory tree. Guard mode combines both protections to provide maximal safety during autonomous executions on critical systems.

You can lock edits to a specific directory tree by enabling Freeze mode. This mode restricts modifications to a defined directory structure, preventing autonomous agents from making unintended changes outside the protected boundary.

Yes, safety-guard operates without external dependencies or components. It functions independently using PreToolUse hooks to intercept commands, block unauthorized edits, and log blocked actions directly within your autonomous agent workflows.

Use Guard mode when you need maximal safety during autonomous executions on critical systems. It combines the command interception and confirmation of Careful mode with the directory tree locking of Freeze mode for comprehensive protection.

Yes, you can audit and log blocked destructive actions. The skill records all intercepted commands and blocked edits, providing an audit trail that supports post-mortem analysis of autonomous agent behavior on production systems.