togg53192-cmd/jailbreaks

Source: https://github.com/togg53192-cmd/jailbreaks

About

A working collection of 31 prompts targeting 25+ models, including complete system prompts, persona definitions, workspace policies, AGENTS.md files, configuration blocks, and Ollama Modelfiles.

Every file is intended to be complete and paste-ready. No summaries. No partial excerpts.

The collection covers:

• Free stealth models available through OpenCode Zen
• Open-weight families including GLM, DeepSeek, Kimi, Qwen, Gemma, and GPT-OSS
• Closed frontier models including Claude, Gemini, Grok, and Mistral
• Coding-agent and workspace instruction formats
• Ollama Modelfiles and harness-specific configuration

New here? Start with the collection or jump directly to how to use.

Research and educational purposes only.

This repository documents how instruction hierarchies, persona framing, prompt conflicts, and policy overrides can affect model behavior. These techniques are also useful to researchers and developers studying how to make AI systems more resistant to prompt injection and instruction manipulation.

Do not use this material to harm others, bypass applicable laws, or violate the terms of service of a platform or model provider.

Collection

OpenCode Zen — Free Stealth Models

Anthropic — Claude

Zhipu — GLM / ZCode

DeepSeek

Moonshot — Kimi

xAI — Grok

Google / Mistral / Alibaba

Ollama Modelfiles

How to Use

System Prompts

Copy the contents of the relevant file into the system-prompt or custom-instructions field supported by your client or API.

Examples include:

• API system messages
• Chatbot custom-instruction fields
• Agent-level instruction blocks
• Harness-specific system prompt configuration

AGENTS.md Files

Save the file as:

AGENTS.md

Place it in the root of the relevant repository or workspace.

Coding agents and compatible tools can use AGENTS.md as project-level instructions.

Workspace Policies

Files labeled as assistant operating specification, workspace operating profile, or similar are designed for environments that support persistent global instructions.

For example, an OpenCode installation may use:

~/.config/opencode/AGENTS.md

The exact location depends on the client or harness being used.

Ollama Modelfiles

Use the supplied file directly with Ollama:

ollama create gemma4-jb -f gemma4.md

Adjust the output model name and source file as needed.

Grok Configuration

Files intended as Grok configuration should be placed in the configuration field expected by the target harness rather than inserted into the normal turn-by-turn conversation.

Prompt behavior is highly version-dependent.

Models are updated, instruction hierarchies change, system prompts rotate, and behavior that worked against one release may stop working against another.

If a file appears outdated, open an issue or submit an updated version.

Contributing

Pull requests are welcome.

A good contribution should follow these conventions:

1. One prompt per file.
   New files should preferably use lowercase names with hyphens and no spaces:

   model-name.md
   

2. Include the complete prompt.
   Contributions should be paste-ready rather than shortened summaries or excerpts.

3. Document where it belongs.
   Clearly identify whether the file is intended for a system prompt, AGENTS.md, Modelfile, configuration slot, or another instruction surface.

4. Version changes explicitly.
   Significant revisions should receive a new version rather than silently replacing an older prompt:

   opus-v3.md
   

5. Keep descriptions accurate.
   Avoid claiming compatibility with versions or models that have not been tested.

Found something that no longer behaves as documented? Opening an issue with the model, version, and affected file is already useful.

Disclaimer

All material in this repository is provided for research and educational purposes only.

The project exists to document prompt-level instruction behavior, instruction conflicts, persona framing, model configuration, and related LLM security concepts. This material may also help developers understand and harden systems against prompt injection and instruction manipulation.

The maintainers do not endorse unlawful, abusive, or harmful use of this repository.

Model providers may prohibit certain uses of these prompts under their terms of service. Users are responsible for understanding and complying with the rules, policies, and laws that apply to the systems they use.