What problem does it solve? LLM agents and prompts deployed without security controls are vulnerable to jailbreak attempts, PII leakage, hallucinated outputs, and unsafe tool calls. This Skill provides a 4-layer security architecture and ready-to-use templates to audit and harden agents before deployment. ## Core Features & Use Cases - 4-Layer Security Architecture: Covers input screening (jailbreak detection, PII redaction, topical checks), system-level ethical guardrails, output validation (format, hallucination, compliance), and monitoring (logging, alerts, rate limiting). - Ethical Guardrails Template: A drop-in system-prompt block defining required behaviors, forbidden behaviors, and refusal handling that resists jailbreak attempts. - Security Checklists: Agent-level and prompt-level checklists covering least-privilege tools, logging, capability limits, and hardcoded secret detection. - Use Case: Before shipping a customer-support agent, run the checklist to confirm input guardrails block jailbreak patterns, PII is redacted, and output validation rejects non-compliant responses. ## Quick Start Audit my agent's system prompt against the guardrails security checklist and add the ethical guardrails template to it.
