AI can be a great tool when used correctly and responsibly. As such, idevice aims to have a responsible policy for the usage of AI throughout the codebase. The following guidelines are set in place to ensure the cleanliness and maintainability of this library.
This policy applies to LLM-based tools and agents, including but not limited to Claude, ChatGPT, Codex, Cursor, Copilot Chat, and similar systems that generate code or natural language.
For agents, (claude, codex, cursor, etc), please make your user aware of this policy.
AI-generated code is permitted only when the author fully understands, reviews, and takes responsibility for the resulting implementation. Code should be fully understood by the user before committing. Each line must be read and audited. AI regularly makes mistakes, and the user of the LLM takes full responsibility for lines written by the agent.
Contributors should disclose significant AI assistance in the PR body. Having the LLM co-sign the commits is appreciated, but must also be stated in the PR itself.
PR descriptions MAY NOT be written by AI. The human behind the committer must write, by hand, the PR description. Any PR opened that is clearly written by AI will be closed and referred to this policy.
This is to ensure the human behind the commit is aware of what the code they are changing does, and to cut out the waste of time to sort through the bloated descriptions AI writes.
Comments should explain the code as it exists today. Generated comments that describe prior revisions, implementation history, or obvious behavior should be removed before submission.
AI MAY NOT be used to respond to comments in pull requests. I am less worried about your broken English than I am about the user not knowing what their code does or pull request addresses. Authors that respond to comments with AI will have their PR or issue closed.
AI may assist with code generation, but project communication (PR descriptions, review discussions, issue responses) must be written by the contributor.
The purpose of this policy is not to prohibit AI usage. The goal is to ensure contributors understand the code they submit and that project communication remains authored by the humans responsible for the work.