Using AI tools

AI coding tools are welcome in pySDC, and its maintainers use them too: aidecl.yaml declares which tools helped with the code and how, and the CI validates that declaration on every run. A contribution made with AI help is reviewed like any other, and a few rules keep it that way.

You are the author

• Understand every line you submit, and be able to explain it in the review. “The tool wrote it” is not an answer to a reviewer’s question.

• Check numerical code against something you trust: a known convergence order, a reference solution, a result from a paper. Passing tests are not enough when the tool also wrote the tests.

• Never let a tool make a failing test pass by loosening a tolerance, skipping a case or changing the expected value, unless you can say why the old value was wrong.

• Keep pull requests focused. A generated refactoring across many files is hard to review, so open an issue first.

Say what the tool did

• Keep the Co-authored-by: trailer that agents add to their commits, also when squashing. aidecl.yaml counts these commits.

• In the pull request’s description, mention which parts were written with AI help, e.g. “tests generated, implementation by hand”, so the reviewer knows where to look closer.

• Git AI records which lines an agent wrote, without changing your workflow. It is optional, but recommended if you contribute regularly.

Issues, reviews and discussions

Check what a tool found before you post it. An issue or a review comment with an unverified claim costs the maintainers the time to disprove it, so post only what you have reproduced, and say how.

Licenses

pySDC is under the BSD 2-clause license. Do not submit code that a tool reproduced from a source with an incompatible license, and do not paste code or data into a tool that you may not share.

⬅️ Back to publishing a new release — ⬆️ Contributing Summary