gstack-guard

To prevent destructive commands during production debugging, you need a guard mode that intercepts operations like rm -rf, DROP TABLE, and force-push. This skill enforces warnings before executing dangerous commands to reduce the risk of accidental damage in live environments.

Yes, you can restrict file edits to a specific directory subtree during live maintenance. This skill enforces directory-scoped edit restrictions by storing your chosen boundary path, actively blocking any file modifications that occur outside that designated safe subtree.

The best way to set up a maximum safety workflow is activating a one-shot safety posture that combines destructive-command warnings with strict directory-scoped edit restrictions. This ensures changes only occur within a chosen boundary during production maintenance.

To activate guard mode, tell the AI you want full safety mode and then provide the exact directory path to restrict edits. This triggers pre-tool hook checks wired to careful and freeze guard scripts, storing the chosen boundary to block external edits.

Yes, this directory restriction approach works without additional dependencies. It operates standalone by wiring pre-tool hook checks to internal careful and freeze guard scripts, enforcing both destructive-command warnings and edit boundaries without requiring external packages.