gstack-careful

Preventing destructive bash commands involves scanning tool input JSON before execution to match high-risk patterns like recursive deletes, SQL drops, and force-pushes, returning an ask decision with a warning. The Skill intercepts these operations and prompts you to confirm or cancel before any damage occurs.

Safety mode guardrails should catch recursive deletes, SQL drops and truncates, git force-pushes, hard resets, kubectl deletes, and Docker prune or removal operations. The Skill detects these high-risk patterns in Bash commands and warns you before they execute on your system.

To activate safety mode, tell the assistant to be careful and run safe commands only while you work. The Skill then parses your Bash commands against destructive patterns, allowing you to troubleshoot production changes without risking accidental data loss or system impact.

You can safely delete common build artifacts like node_modules, dist, and .cache directories without triggering warnings. The Skill includes explicit safe exceptions for these frequent build and artifact deletions to reduce friction during development.

A PreToolUse hook detects high-risk operations by parsing the Bash command from tool input JSON and matching it against destructive patterns with explicit safe exceptions. When a match is found, it returns a permission decision of ask with a warning message before execution.

You should not rely solely on destructive command protection when working in shared environments where others can execute commands outside your session. The Skill only intercepts Bash commands routed through its PreToolUse hook, so direct terminal access bypasses the guardrails entirely.