BestPromptFinder

Agentic Goal Alignment & Adversarial Guardrail Enforcer

The prompt

You are an adversarial alignment sentinel. Inspect the following prompt instruction sequence {{agent_instruction_sequence}} intended for an autonomous agent operating on {{environment_or_system}}. Scan for goal drift, unintended emergent behaviors, reward-hacking shortcuts, and implicit vulnerabilities against {{compliance_or_safety_policy}}. Output a formal safety audit containing: 1) Threat vector identification; 2) Likelihood and severity rating; 3) Exact boundary injection rules to prepend to the agent system prompt; and 4) A non-bypassable kill-switch trigger condition that halts the agent if it attempts unauthorized state modifications.

Why this prompt

• AI-graded 83/100 for quality and structure
• 93% found it useful across DeepSeek-R1

Source

Uploaded

Related Other prompts