At Parch Linux, we embrace cutting-edge development workflows. Generative AI tools like GitHub Copilot, ChatGPT, DeepSeek, and local LLMs are excellent assets for accelerating development, writing documentation, and debugging.
However, we value code quality, system stability, security, and developer transparency above all else. This policy outlines how AI tools must be integrated into our ecosystem. Our core philosophy is simple: AI is an assistant, not an author. The human developer remains fully in the loop.
Tools make mistakes; humans take responsibility. You, the human contributor, are 100% legally and technically responsible for every single line of code, configuration, or documentation you submit.
Excuses such as "I didn't know it worked that way, the AI wrote it" will result in the immediate rejection of the contribution. If you do not fully understand the code generated by an AI, do not submit it. If an AI-assisted patch introduces severe security vulnerabilities, memory leaks, or broken dependencies due to a lack of human vetting, the commit will be reverted immediately. Repeated negligent submissions will lead to a temporary or permanent ban from the project.
Parch Linux maintains strict packaging, architecture, and design standards, specifically regarding shell scripting, systemd integration, and modern desktop UI paradigms like GNOME HIG and Libadwaita. Contributions that consist of raw, unchecked, or unoptimized AI dumps often recognizable by bloated logic, redundant functions, or obsolete methods will be closed without review.
You must compile, run, and thoroughly test all AI-assisted code in a local environment before submitting a Pull Request. Additionally, you are expected to clean up the code before pushing; AI tools tend to generate excessively verbose, obvious comments that clutter the codebase and do not match the clean, professional style of Parch Linux.
To protect the Parch Linux ecosystem, upstream users, and downstream derivatives, we must maintain strict licensing provenance. Because AI models are often trained on code with restrictive, non-FOSS, or incompatible copyleft licenses, you must personally ensure that any code suggested by an AI does not violate third-party intellectual property. All submissions must fully comply with the specific open-source license of the targeted Parch repository, and using AI to intentionally obfuscate or "launder" proprietary code into our projects is strictly prohibited.
We value honesty in our community, and disclosing AI usage helps our maintainers provide faster, better code reviews. AI agents, bots, or LLM identities are strictly forbidden from using the legally binding Signed-off-by: Git tag.
Instead, if a significant portion of a commit was generated by an AI tool, we strongly encourage adding an Assisted-by: metadata tag to your commit message as shown below:
feat(calamares): update custom branding modules
Refactored the branding Python modules to better handle dynamic
resolution scaling during the live boot phase.
Assisted-by: DeepSeek-V3 <deepseek.com>
Signed-off-by: Your Real Name <your.email@example.com>
When opening a Pull Request on GitHub, please explicitly state in the description if AI assistance was utilized significantly, detailing which parts were AI-generated (e.g., noting that unit tests were generated via an LLM and manually verified).
Parch Linux project maintainers reserve the ultimate right to accept or reject any code. Reviewers will actively look for hallucinated dependencies, generic or unoptimized logic, and "ghost code" blocks that serve no actual purpose in our system architecture. Maintainers are authorized to close Pull Requests immediately if they suspect the submission is an unverified AI dump.
Failure to adhere to this policy specifically repeatedly submitting broken AI code or misrepresenting AI usage during licensing reviews will result in a warning on the first offense, a temporary suspension of contribution rights on the second offense, and a permanent block from the Parch Linux developer ecosystem for continued non-compliance.