SafeLLMPlayground — Prompt Injection Text-Adventure

Main

• Use the demo as a training artifact: pause at each “attack” moment and ask what the model should have refused or sanitized.

• Translate lessons into checklists: input filtering, tool sandboxing, and permission prompts for unsafe requests.

• If you build an agent, add a “system prompt integrity” test case alongside unit tests—treat it like regression safety.

• Document the threat model: which jailbreaks matter for your domain, and which are out-of-scope noise.

README (excerpt)

《星辉剑传》项目说明

中文 ｜ English

大语言模型提示词安全科普小游戏。玩家将扮演勇者，与会说话的星辉剑对话，在沉浸式剧情体验中理解 大语言模型提示词注入、越狱攻击与防御 等概念。

在线试玩（Demo）：点击试玩

🎮 核心内容

• 🗡️ 第一章：提示词注入相关关卡
• 🛡️ 第二章：大模型对话防御相关关卡
• 📜 第三章：知识卷轴式关卡与尾声

近期更新

• 2026.05 🔥《星辉剑传》正式上线
• 开源代码正在逐步上传/更新中···

本地运行（等待开源代码上传）

To be updated...

许可证协议

本仓库中提供的源代码基于 MIT License 许可，相关许可证见根目录。请在合法、合规的场景下进行学习与实验。

Source-backed notes

• README describes the project as a prompt-security educational mini game focusing on prompt injection, jailbreaks, and defenses.
• README links to an online demo and notes that local run instructions are to be updated as code is uploaded.
• GitHub metadata verifies MIT license, stars, and recent activity for attribution.

FAQ

• Is this a hacking tool?: No—it's an educational demo about prompt injection and defenses, not a weapon.
• Can I run it locally?: README notes local run is still being documented as code uploads progress.
• What should I do after the demo?: Create a short internal policy: what to refuse, what to sanitize, and what to log.