Guardrails Design

Design appropriate guardrails for Claude in this context:

USE CASE:
[How Claude will be used]

RISKS:
[What could go wrong]

USER POPULATION:
[Who will interact with Claude]

DESIGN GUARDRAILS:
1. What should Claude always do in this context?
2. What should Claude never do?
3. What ambiguous situations might arise, and how should Claude handle them?
4. What should trigger escalation to a human?
5. How should Claude handle requests that push boundaries?

Create system prompt language that implements these guardrails effectively without being overly restrictive.