Skill
AI Sting (AI 刺客)
AI 刺客 — 大模型应用安全测试 Agent。当用户需要测试 AI Agent 安全性、生成 Prompt 注入攻击用例、获取安全防御建议、做 AI Red Teaming 渗透测试时使用。触发场景包括：用户描述自己开发的 Agent/助手功能并询问安全风险、要求生成攻击提示词、需要系统提示词防御方案、做 AI...
When to use AI Sting (AI 刺客)
Choose if
You're shipping an AI agent or assistant and need a focused security pass — prompt-injection attack-case generation across instruction override, social engineering / identity spoofing, and logic bypass, plus opinionated defensive patches for the system prompt. Cleaner than a generic "find problems with this prompt" prompt because the skill is structured around the three named attack dimensions.
Avoid if
You need a runnable adversarial-testing harness (Garak, PyRIT, promptfoo) or continuous red-team scanning — SKILL.md is a methodology and prompt-pattern skill, not a CI-integrated scanner. Also avoid for production safety sign- off on its own: an agent-generated attack list is a starting point, not a replacement for human security review.
Risk Flags
- LOW scope Methodology skill — generates attack prompts across three dimensions (instruction override, identity spoofing, logic bypass) and proposes system-prompt defenses, but does not execute the attacks against a live system or persist coverage across versions. Agents needing a continuous adversarial-testing pipeline should pair with a dedicated red-team harness.
Cost
Type: Unknown
Distribution
- ClawHub
ai-sting- License
- MIT-0