yunwei37/prompt-hacker-collections

Organizes hundreds of attack and defense techniques as structured YAML prompts across four categories: jailbreak methods (like DAN variants), reverse engineering tactics for extracting model instructions, direct prompt attacks, and defensive countermeasures. Targets LLMs broadly including ChatGPT, Copilot, and Midjourney with cross-model applicability notes, enabling researchers and security professionals to systematically test and understand prompt injection vulnerabilities.

295 stars. No commits in the last 6 months.