Home/ Part XII — Building Real Products (End-to-End Projects)/34. Project 3: "Vibe Coder" Assistant for Your Own Repo/34.5 Guardrails: refuse risky changes and require approvals

34.5 Guardrails: refuse risky changes and require approvals

Overview and links for this section of the guide.

On this page

Blocking Destructive Actions

Your tool needs an "Allowlist" of allowed commands.

  • Allowed: `ls`, `cat`, `grep`, `npm test`, `git status`.
  • Blocked: `rm -rf`, `git push --force`, `aws delete-bucket`.

If the model tries to run a blocked command, your tool should intercept it and say: "Action blocked by safety policy. Please ask the user for permission."

Protecting Secrets

Before sending any file to the LLM, run a regex scan for API keys (`sk-live-...`, `AWS_SECRET...`). If found, redact them to `[REDACTED]`.

This prevents you from accidentally leaking your production keys to the model provider (though Google Enterprise has privacy guarantees, it's good hygiene).

Where to go next