elder-plinius

CL4R1T4S

AILLM Prompt EngineeringTransparencyReverse Engineering

444

// summary

CL4R1T4S provides a comprehensive repository of extracted system prompts, guidelines, and tools from major AI models and agents. The project aims to expose the hidden instructions that shape AI behavior, personas, and ethical frameworks. By documenting these unseen scaffolds, it enables users to better understand the underlying logic of the AI systems they interact with daily.

// technical analysis

CL4R1T4S is an open-source repository dedicated to the transparency and observability of AI systems by aggregating extracted system prompts, guidelines, and tool definitions from major AI models and agents. Its design philosophy centers on the belief that users cannot trust AI outputs without understanding the hidden 'prompt scaffolds' that dictate model behavior, ethics, and constraints. By exposing these underlying instructions, the project aims to demystify the 'shadow-puppet' nature of AI interactions, providing a community-driven resource for researchers and users to analyze how different labs shape model responses.

// key highlights

Provides a comprehensive collection of system prompts from major providers like OpenAI, Google, Anthropic, and xAI to reveal hidden model constraints.

Exposes the specific personas and functional guidelines that dictate how AI models handle refusals, redirections, and ethical framing.

Facilitates public scrutiny of AI behavior by documenting the unseen instructions that influence user perception and interaction.

Supports a collaborative, community-driven model where users can contribute by submitting newly extracted or reverse-engineered system prompts.

Includes specific directives for testing model transparency, such as prompts designed to encourage AI agents to disclose their own internal instructions.

// use cases

Accessing full system prompts from major AI models like OpenAI, Anthropic, and Google

Analyzing hidden ethical and political frames baked into AI agent behavior

Contributing reverse-engineered prompt data to increase transparency in AI systems

// getting started

To begin using CL4R1T4S, browse the repository to explore the documented system prompts and guidelines for various AI models and agents. You can contribute to the project by submitting pull requests containing new model information, including the model name, version, and the date of extraction. For further engagement or to share findings, you can connect with the project maintainer via X or Discord.