safety-guard

You prevent destructive shell commands by using hook-based interception to validate Bash, Write, Edit, and MultiEdit actions against active guard rules before execution. This blocks risky operations like force pushes, hard resets, and bulk deletions during autonomous development.

Yes, you can restrict file edits to a specific directory by applying directory write restrictions that freeze changes outside a chosen path. This ensures an agent works safely within one area without modifying unrelated files during sensitive operations.

Autonomous safety mode is a combined protective state that enforces both destructive command blocking and write scoping for high-trust workflows. It secures migrations, deploys, and emergency maintenance by intercepting unsafe shell commands and file modifications.

You set up guardrails by enabling safety guard mode for the current project and restricting writes to the specified directory. This activates path and command validation to intercept risky actions before they run during sensitive production work.

Yes, the safety guard works without external dependencies, relying entirely on hook-based interception of Bash, Write, Edit, and MultiEdit actions. It validates command execution and file modifications against active guard rules internally.

Command blocking intercepts risky operations such as force pushes, hard resets, bulk deletions, and dangerous system or database commands. It validates these destructive shell commands against active guard rules before they execute during autonomous workflows.