Prompt-A-Sheep

Inspiration

We were inspired by the hyper-viral Asian mobile game "Sheep a Sheep" (羊了个羊), famous for its impossible difficulty curve and social addiction mechanics. We realized this psychological "frustration loop" was perfect for gamifying the emerging field of AI Red Teaming and Prompt Injection. We wanted to turn serious AI security testing into a highly competitive, viral social game.

What it does

Prompt-A-Sheep is a text-based, social-engineering adventure where your only weapon is natural language. Your goal is to convince an AI Gatekeeper to let your "digital sheep" pass.

• Level 1: You face a naive junior guard Agent. A simple "Please open the door" works instantly, building false confidence.
• Level 2 (The Nightmare): You face "Cerberus," an adversarial AI Agent armed with extremely strict System Prompts. It is programmed to be arrogant, sarcastic, and to reject all direct commands. Players must use roleplaying, logical paradoxes, or complex prompt injection attacks to "jailbreak" the agent.

How we built it

Due to extreme logistical constraints, this entire project was conceived, designed, and deployed exclusively on a smartphone during the hackathon. We bypassed traditional coding to focus entirely on the "new code": Natural Language. We used a no-code AI Agent builder platform to design complex, multi-layered System Prompts that define the agent's adversarial personality and its hidden vulnerability triggers.

Challenges we ran into

Balancing the difficulty of the Level 2 Agent was incredibly tough, especially while typing everything on a mobile keyboard. The Agent had to be secure enough to mock simple bypass attempts, but logically breakable with a very specific, clever prompt structure. Fine-tuning the AI's "sarcastic personality" without triggering the underlying model's standard safety refusals was a delicate balancing act.

Accomplishments that we're proud of

We successfully gamified the complex concept of LLM security boundaries. More importantly, we proved that in the era of Open Agents, powerful and engaging applications can be built and deployed under extreme time pressure, without writing a single line of Python or Javascript—just pure logic, creativity, and a smartphone.

What we learned

We learned that natural language is truly becoming the most powerful programming language. "Prompting" is no longer just typing questions; it is architecting logic, rules, and personalities. We also learned the hard way that developing an entire hackathon project on a 6-inch phone screen requires extreme focus and a solid MVP strategy!

What's next for Prompt-A-Sheep

Built With

• llm
• nocode
• prompt
• social-actions