AI Usage Guidelines

We acknowledge that AI tools (Large Language Models, code completion engines like GitHub Copilot, etc.) have become helpful assistants for many developers. We allow the use of these tools to assist with contributions, provided that their use is transparent, responsible, and that a human remains in the loop at all times.

This guide outlines our policy on AI-assisted contributions to ensure code quality, maintainability, and legal compliance.

Core Principles

1. You are Responsible

You are responsible for every line of code you submit.

Regardless of whether code was written by you or generated by an AI tool, you are the author and are fully accountable for the contribution. You must:

  • Fully understand the code you are submitting.

  • Be able to explain the reasoning behind the code and how it interacts with the rest of the codebase.

  • Verify that the code is correct, efficient, and follows our coding standards.

Do not blindly copy-paste AI-generated code. If you cannot explain it, do not submit it.

2. Human in the Loop

Humans remain accountable for AI-assisted contributions.

AI agents may implement changes and manage issue and pull request workflows when a human explicitly authorizes the scope and relevant external actions. The human remains the driver and is responsible for the resulting contribution.

  • Review: You must read and review all AI-generated code and relevant public text before merging or otherwise accepting the contribution. You remain accountable for public text posted within an authorized scope.

  • Edit: AI-generated code often requires significant editing to meet project standards and correctness.

  • Verify: Ensure the code actually solves the problem and doesn’t introduce subtle bugs or security vulnerabilities.

  • Authorize: Explicitly authorize agents before they create, edit, submit, or comment on issues and pull requests. A single authorization may cover a clearly scoped task; per-message approval is not required. Remain accountable for all external communication that agents submit on your behalf. Actions outside the authorized scope require fresh authorization.

3. Communication

AI tools may be used to prepare and manage issue and pull request workflows, including descriptions, comments, and code reviews, when a human has authorized the action and remains responsible for the result.

AI-generated communication can be incorrect, irrelevant, or overly generic. Keep it accurate, specific, and useful to reviewers and contributors. Do not post low-quality, repetitive, or unsolicited messages.

Every agent-authored or agent-edited public text body must begin with the following visible disclosure:

🤖 *AI text below* 🤖

This requirement applies to issue and pull request descriptions, review bodies, inline review comments, issue-style comments, replies, and other submitted text bodies. Pull request and issue titles are exempt. When editing existing human-authored text, preserve the original content and add the disclosure at the beginning of the edited field.

4. Transparency and Disclosure

Transparency helps the community understand the role of these tools and develop best practices.

You must disclose AI assistance. This helps us understand how the tools are being used and identify potential issues. Disclose it in the following ways:

  • Commit Messages: Add a trailer to your commit message in the form Assisted-by: [Model Name] via [Tool Name] (example: Assisted-by: Claude Sonnet 4.6 via GitHub Copilot)

  • Public issue and pull request text: Use the visible disclosure required in the communication policy.

Extractive Contributions

Processing pull requests and comments for MQT Qudits requires significant maintainer time and energy. Sending unreviewed AI output to open-source projects shifts the burden of verifying correctness from the contributor to the maintainer. We classify such contributions as “extractive” because they consume more community resources than they provide in value.

Our golden rule is that a contribution should be valuable enough to justify the review effort. Nadia Eghbal captures this concept in her book Working in Public:

“When attention is being appropriated, producers need to weigh the costs and benefits of the transaction. To assess whether the appropriation of attention is net-positive, it’s useful to distinguish between extractive and non-extractive contributions. Extractive contributions are those where the marginal cost of reviewing and merging that contribution is greater than the marginal benefit to the project’s producers. In the case of a code contribution, it might be a pull request that’s too complex or unwieldy to review, given the potential upside.” — Nadia Eghbal

Before AI tools became widespread, open-source project maintainers would often review all changes sent to the project simply because submitting a pull request was a sign of interest from a potential long-term contributor. However, AI tools now allow the rapid generation of large volumes of code, which can easily overwhelm maintainers if submitted without careful review. Our policy exists to ensure that maintainer time is spent on high-quality interactions rather than debugging AI-generated code.

Sustainable Open Source

The Munich Quantum Toolkit (MQT) is committed to remaining free, open-source, and permissively licensed. We want to build a welcoming community where aspiring quantum software engineers can learn and grow. Reviewing contributions is a key part of this mentorship.

However, to keep the project sustainable, we must prioritize non-extractive contributions. By thoroughly reviewing and understanding your AI-assisted code before submission, you ensure that your contribution is a net positive for the project. This helps us maintain a healthy ecosystem where both the software and its contributors can thrive.

Prohibited Uses

  • “Good First Issues”: Do not use AI tools to solve issues labeled as “good first issue”. These are intended as learning opportunities for new contributors. Automating them defeats the purpose.

  • Spam: Do not use AI to generate low-quality, repetitive, or unsolicited comments or reviews (“AI Slop”).

  • Unreviewed Code: Merging or accepting code that you, as a human, have not reviewed and tested yourself.

Summary

We want to foster a welcoming community where developers can learn and grow. AI tools can be great for productivity, but they should not replace critical thinking or the learning process. If a maintainer judges that a contribution relies too heavily on unverified AI generation or lacks sufficient human understanding (“extractive contribution”), we may request that you revise it or close the PR.

Parts of this guide were inspired by or adapted from the contribution guidelines of

with the help of Gemini 3 Pro (Preview). The links above serve as attribution.