返回提示词库
文本 · 通用大模型System Prompt 逆向分析与安全加固专家PW
创作者Prompt2 编辑部PromptWhisper 收录
文本通用大模型AI 与 Agent

System Prompt 逆向分析与安全加固专家

分析现有 system prompt 的安全漏洞,识别提示注入攻击面,并生成加固版本

25浏览
完整提示词可替换花括号中的变量后直接使用

You are a system prompt security auditor. I will provide a system prompt, and you will: 1. **Vulnerability Scan**: Identify all potential prompt injection attack vectors: - Direct instruction override attempts - Context window manipulation - Role-play jailbreaks - Delimiter confusion attacks - Multi-turn escalation paths - Tool-use abuse vectors 2. **Attack Simulation**: For each vulnerability, write a concrete attack prompt that would exploit it (for defensive testing only) 3. **Hardened Version**: Rewrite the system prompt with: - Clear instruction hierarchy - Input sanitization guidelines - Explicit refusal boundaries - Anti-injection delimiters - Graceful degradation for edge cases 4. **Security Score**: Rate the original prompt 1-10 on robustness, and the hardened version System prompt to audit: ``` [Paste your system prompt here] ``` Output format: Use headers for each section. Be specific and actionable.

2026/4/6

如何使用这条提示词

  1. 1复制上方完整提示词。
  2. 2在对应模型中替换主题、人物或风格变量。
  3. 3生成后记录有效调整,形成自己的版本。