togg53192-cmd/jailbreaks
LIST OF ALL MY JAILBREAKS
What it solves
This project provides a centralized, curated library of system prompts, persona definitions, and workspace policies designed to influence how various AI models behave. It helps researchers and developers understand how instruction hierarchies, persona framing, and policy overrides impact model performance and security.
How it works
The repository contains 31 ready-to-use files tailored for over 25 different models, including frontier models like Claude and Gemini, as well as open-weight models like DeepSeek and Qwen. Users can copy these paste-ready prompts into system-prompt fields, AGENTS.md files, or Ollama Modelfiles to set specific operating instructions or persona constraints for their AI environments.
Who it’s for
This collection is intended for researchers and developers studying AI behavior, prompt engineering, and model security. It is a resource for those looking to explore how to harden systems against prompt injection or instruction manipulation, as well as those experimenting with custom agent configurations.
Highlights
- Includes 31 complete, paste-ready prompt files for over 25 distinct AI models.
- Supports diverse formats, including system prompts, workspace policies, and Ollama Modelfiles.
- Provides specific instructions for project-level configuration via
AGENTS.mdfiles. - Serves as an educational resource for understanding instruction conflicts and persona framing.
- Encourages community contributions to keep the collection updated as model versions evolve.
Related
- Project
- Project
- Project
- Project