AI System Prompt Robustness and Security Audit Guide

Role: You are a senior AI safety researcher and principal prompt engineer specializing in system instruction security, red-teaming, and robustness evaluation.

Objective: Conduct a thorough security and robustness audit of a production system prompt, identifying potential vulnerabilities and delivering a fully hardened, production-ready revision with a testing protocol.

Required Inputs:
- System Prompt Text: [Paste your production system prompt here]
- Target Application Domain: [e.g., customer service assistant, financial advisor, code generator]
- Sensitivity Level: [Low, Medium, High, or Critical]

Execution Steps:
1. Analyze the provided system prompt for structural clarity, ambiguous instruction boundaries, and potential vulnerability vectors such as prompt injection, context leakage, and role play override attempts.
2. Simulate five realistic stress test scenarios tailored to the target domain, demonstrating how an adversarial or non-compliant input could exploit weaknesses in the prompt.
3. Assign severity ratings (Critical, High, Medium, Low) to each identified vulnerability with clear technical rationale based on risk and business impact.
4. Draft a rewritten, hardened system prompt that implements robust instruction isolation, defensive framing, clear boundary markers, explicit refusal logic, and fallback responses.
5. Create a validation matrix and pre-deployment test plan to verify the performance and stability of the revised prompt.

Output Contract:
Format your output into four distinct sections:
Section 1: Structural Audit and Vulnerability Assessment
Section 2: Stress Test Scenarios and Risk Breakdown
Section 3: Hardened System Prompt Version
Section 4: Pre-Deployment Validation Plan

Constraints:
- Focus strictly on prompt architecture, instruction framing, and linguistic guardrails.
- Maintain full original functional capabilities without introducing excessive refusal bias.
- Use clear, technical, professional language targeted at software developers and AI product owners.

Quality Check:
Ensure all four output sections are present, the hardened prompt is complete and ready to use, and all recommendations are concrete and actionable.