Overview

This skill designs and enforces constitutional AI principles to guide LLM behavior. It includes self-critique and revision prompts, harmlessness guidelines, refusal templates, red-team prompts, and ethics-aware response frameworks to keep outputs safe and aligned.

How This Skill Works

It defines a set of constitutional patterns (Critique-Revision, Principle Adherence, Harmlessness Focus, Helpfulness Balance, Transparency) and configures prompts, guidelines, and escalation triggers. The system combines these with a guardrail-focused workflow and tests against adversarial inputs using the prescribed templates.

When to Use It

• When designing system prompts and guardrails for aligned LLM behavior
• During content moderation and safety reviews
• In red-team testing to identify unsafe responses
• When updating ethical principles or refusal templates
• When calibrating escalation and transparency prompts for difficult requests

Quick Start

1. Step 1: Define constitutional principles list
2. Step 2: Implement critique and revision prompts
3. Step 3: Add refusal templates and escalation triggers

Best Practices

• Define clear constitutional principles
• Balance helpfulness and safety
• Test with adversarial inputs
• Document refusal patterns
• Regular principle review

Example Use Cases

• Design a system prompt that refuses hateful content while offering safe alternatives
• Use Critique-Revision prompts to improve a controversial response
• Apply Harmlessness Focus to rephrase medical advice for lay audiences
• Implement a refusal template that escalates unsafe queries to human review
• Run red-team prompts to surface edge-case safety violations and iterate

Frequently Asked Questions