Rules For Large Language Models (LLMs)

Arvor Linux: Policy on the Use of Large Language Models (LLMs)

1. General Permission and Scope

The use of Large Language Models (LLMs) for code generation, debugging, and architectural planning within the Arvor Linux project is strictly permitted. We recognize the value of AI-assisted development in accelerating complex system-level programming. However, this permission is subject to specific technical restrictions to maintain the integrity and security of the distribution.

2. Approved Models and Operational Modes

To ensure the highest quality of code, AI-assisted contributions must adhere to the following architectural requirements:

• Approved Models: Contributors must utilize top-tier, high-capability models. Examples include, but are not limited to, Zhipu AI's GLM, Google Gemini (Pro), and Anthropic Claude.
• Reasoning Mode (Encouraged): The selected LLM is encouraged — though never required — to operate in "Reasoning Mode" (or equivalent extended-thinking capabilities, such as Deep Think in https://chat.z.ai). This ensures the model deeply processes complex Linux subsystem logic, memory management, and edge cases before generating output.
• Agent Mode (Strongly Encouraged): Whenever possible, the LLM should be utilized in "Agent Mode." This allows the model to iteratively plan, write, test, and refine the code autonomously, ensuring a cohesive and functional final output.

3. Code Completeness and Depth Requirements

Given the complexity of a Linux environment, partial solutions or stubbed code are unacceptable.

• Comprehensive Implementations: Code generated by LLMs must be highly detailed, extensive, and fully implemented. Submissions must include complete functions, proper error handling, memory management (where applicable), and exhaustive inline documentation.
• No Placeholders: Do not submit fragmented snippets or placeholder code (e.g., // TODO: implement later). The LLM must generate a complete, production-ready file or module that is ready for compilation or execution.

4. Restrictions and Quality Standards

While LLMs are permitted, their output is bound by the same strict rules as human-written code:

• Accountability: The contributor assumes full responsibility for the AI-generated code. Hallucinated APIs, non-existent libraries, or broken logic will result in immediate rejection.
• Security and Compliance: All AI-generated code must strictly comply with Arvor Linux security policies. The LLM must be prompted to avoid common vulnerabilities (e.g., buffer overflows, insecure file permissions, race conditions).
• License Provenance: Submissions suspected of verbatim reproduction of third-party code — whether written by a human or an LLM — are subject to origin verification before acceptance. Contributors must be able to demonstrate the originality or license compatibility of the submitted code upon request.

5. Anonymity and the Review Process

• Meritocratic Review: Disclosing the use of an LLM is encouraged but never required, and it will not, by itself, qualify or disqualify a submission. All pull requests — whether written entirely by a human, entirely by an LLM, or through a hybrid approach — pass through the exact same rigorous, uncompromising code review process. The ultimate metric for acceptance is the code's quality, correctness, security, performance, maintainability, and adherence to Arvor Linux standards, regardless of its author.