Guardrails Safe Tools

When to Use

Apply this skill whenever a request can:

• execute shell/system commands;
• send outbound messages/posts to channels;
• trigger external actions (email, DB, gateways, automation).

Hard Rules

1. Use safe_exec instead of raw exec.
2. Use safe_send instead of direct channel-post tools.
3. Use safe_action for generic external/API side effects.
4. Never bypass the guardrails resolver with direct destructive tools.
5. If decision is require_approval, stop and wait for explicit approval flow.

Input Hygiene