Policy about AI-assisted contributions

Artificial-intelligence (AI) coding assistants and large language models are now part of many developers’ workflows. While SPECTRE acknowledges the added value of such tools, we think that they must be used in a very defined environment. This page describes what is acceptable, what is not, and the responsibilities that come with using these tools when contributing to SPECTRE.

The guiding principle is simple: a human contributor is fully and personally responsible for every line they submit, regardless of how it was produced. AI can assist you, but it cannot be a contributor, and it cannot take responsibility for the code that ends up in SPECTRE.

SPECTRE is a scientific project and as such, one needs to understand how the results are produced to make sure of their quality.

Declaring AI use

Any use of AI in a contribution must be declared in the merge request. Transparency lets reviewers focus their attention appropriately and keeps the project’s provenance clean. Disclosing AI use is never held against you. It is expected and welcome.

The default template for merge requests adds a checklist for AI-assistance disclosure. Fill it appropriately when opening a merge request that involved AI assistance in any form (code, documentation, commit messages, or translation).

## AI usage declaration

- [ ] AI tools were used in this contribution
- **Tool(s) used:** <*e.g.* GitHub Copilot, ChatGPT, Claude>
- **What it was used for:** <*e.g.* autocompletion, drafting a docstring,
  translating a comment>
- [ ] I have read and understood every line I am submitting
- [ ] I confirm the contribution does not infringe any copyright or license

If no AI was used, you do not need to check any box, or you can remove this section entirely.

If a reviewer suspects undeclared AI involvement, they may ask you about it. Please do not take such a question negatively. It is part of keeping the process honest, and an open conversation is all that is needed to resolve it.

Your responsibility as a contributor

You — not the AI, and not the maintainers — are solely responsible for the code and content you push. Before submitting, you must be able to stand behind it as if you had written it by hand. Concretely, this means:

  • Read and understand every single line. Do not submit code you cannot explain, justify, or maintain. If you do not understand what a generated snippet does, it does not belong in your merge request.

  • Make sure it does not infringe any copyright. AI models can reproduce code from their training data verbatim, including code under incompatible licenses. It is your responsibility to ensure that what you submit is compatible with SPECTRE’s license and does not copy protected material.

  • Verify correctness. Generated code can be subtly wrong, untested against SPECTRE’s conventions, or based on outdated APIs. Test it and confirm it behaves as intended.

What we do not accept

  • Contributions written entirely by AI. A merge request whose code is, in substance, the raw output of a language model is not an acceptable contribution, even if it works.

  • Automatic or agentic contributions. We do not accept merge requests opened or authored by autonomous agents, bots, or pipelines that generate and submit changes without a human author who has reviewed and understood them.

In short, AI may help you write a contribution, but it may not be the contributor.

What is acceptable

Used as an assistant under your direct supervision, AI tools are welcome. Some examples of acceptable, limited use:

  • IDE autocompletion: completing a variable name, a function call, or a few lines you were already going to write.

  • Boilerplate: generating repetitive, mechanical code (e.g. an argument parser stub or a test skeleton) that you then review and adapt.

  • Drafting docstrings or comments that you check for accuracy.

  • Explaining unfamiliar code or errors, or suggesting how to approach a problem, while you write the actual solution.

  • Refactoring suggestions on code you already wrote and understand.

The common thread: the contribution remains yours. You drive it, you understand it, and you would have been able to produce it without the tool. The AI only made it faster.

Documentation

Using AI to help with documentation: improving phrasing, fixing grammar, restructuring a paragraph, or drafting an explanation is acceptable, as long as you have proofread and validated the content. Documentation must be technically correct and must reflect how SPECTRE actually behaves. Do not paste generated prose into the docs without reading it carefully. A confident but wrong sentence in the documentation is worse than no sentence at all.

Translation and non-native English speakers

We are an international project and we know that not everyone is a native English speaker. Language should never be a barrier to contributing. Using AI to translate or polish your English in issues, merge request descriptions, comments, or documentation is explicitly encouraged.

When you do, try to keep your own wording and voice rather than letting the tool rewrite your message into something generic. Your intent and meaning matter more than perfect prose, and a clear human message in imperfect English is always welcome.

If in doubt

If you are unsure whether a particular use of AI fits within this policy, just ask. Open a discussion or mention it in your merge request. We would much rather have an open question than an undisclosed assumption. This policy will evolve as these tools and our experience with them change.