AI Policy

AI assistance is welcome and optional. It can help with learning, implementation, tests, debugging, documentation, and review. Contributors remain responsible for the work they submit.

Core Rule

Understand, review, test where practical, and maintain every contribution during review, regardless of how it was produced. Do not submit work you do not understand. Be able to explain what changed, why the approach fits the project, what tests cover it, and its risks, dependencies, and maintenance cost.

AI suggestions are input for human judgment. Confidence, compilation, passing tests, or an automated review do not establish correctness. Check suggestions in context, reject incorrect advice, and ask maintainers when requirements are unclear. AI review does not replace human review, security review, or maintainer judgment.

Repository Instructions and Validation

Follow applicable AGENTS.md, Copilot and area-specific instructions, contribution guidance, and this policy. Repository requirements and maintainer review take precedence over an AI tool's suggestions.

AI-assisted work must meet the same standards as other contributions: correct, readable, maintainable, consistent with the architecture and coding rules, and free of unnecessary abstractions, invented APIs, and security or privacy defects. Use the normal build, test, formatting, linting, CI/CD, and GitHub Code Quality checks. See development setup and testing guidance for the current workflow.

Add or update tests when appropriate. Review generated tests against requirements, not just the implementation; for bug fixes, prefer a test that fails before the fix and passes afterward. Record expected behavior changes in an issue, PR, test, architecture note, or project document. Applying AI review feedback carries the same responsibility to verify it as any other change.

Disclosure

Routine autocomplete, wording or grammar fixes, exploratory questions, commit-message help, and private code explanations need no disclosure.

Mention AI assistance when materially relevant to review, including mostly AI-generated or agent-authored PRs, architecture proposals, compatibility analysis, and generated tests that define important behavior. A short note describing the assistance and the review and validation actually performed is enough. Do not claim checks that were not run.

Autonomous Agents and Bots

Agents must remain accountable to a human contributor. Do not run unattended agents that mass-create issues, PRs, comments, reviews, or discussions.

The contributor who starts an agent must review its PR before requesting maintainer review. Remove mistakes, irrelevant or noisy changes, broken formatting, and false explanations first.

Do not give agents broad write access, secrets, production credentials, private reports, or sensitive data without explicit maintainer approval of that setup. Maintainers may limit, close, or block agent activity that creates security risk, review or moderation burden, or low-quality noise.

Decompiled Source and Original Client Material

OpenGameBuilder independently reimplements and extends MyGameBuilder. Decompiled original-client source exists outside this repository; it is not implementation input.

Do not use AI to copy, translate, port, rewrite, adapt, refactor, or mechanically convert decompiled or proprietary source into OpenGameBuilder work. Do not supply that source to AI tools to explain it for recreating its implementation or to derive equivalent classes, methods, names, control flow, or architecture. The ban also covers source-derived tests, comments, documentation, and pseudocode. Intermediate AI output does not make copied or source-derived work acceptable.

Use independently observed behavior for compatibility work: public user-facing behavior, safe screenshots or recordings, independently written notes, expected input/output examples, behavior tests, project-created safe fixtures, and community knowledge. Describe what the original client did rather than how its source implemented it.

Do not use private original-site data, unapproved original assets, or copied comments, names, structures, or implementation details from proprietary material as compatibility inputs. If an important behavior can only be justified by decompiled source, ask maintainers before implementing it; this does not permit using AI to derive an implementation from that source.

Privacy and Sensitive Information

Do not put private, sensitive, identifying, confidential, or non-public project information into AI tools without explicit maintainer approval of that use. This includes passwords, tokens, API keys, user information, private reports, sensitive historical material, unpublished security details, maintainer discussions, infrastructure details, and anything unsuitable for a public issue or PR. When in doubt, do not send it.

Licensing and Attribution

AI does not remove licensing, attribution, or source-origin responsibilities. Submit copied code, assets, documentation, or text only when its license permits the use and proper attribution is included. If generated output appears copied, verify the source is compatible with the project's license and attribution requirements before submitting it. Do not use AI to hide or launder its origin.

Documentation, Issues, and Evidence

Documentation must describe the actual project accurately and usefully. Remove filler, invented commands or architecture, fake citations, and unsupported certainty. Verify AI-generated bug reports yourself before submitting them; include reproduction steps, expected and actual behavior, relevant logs or screenshots, and enough context to act. AI speculation is not a verified finding.

Enforcement

Maintainers may require explanations, revisions, tests, rewrites, or removal. They may reject or close work that is unreviewed or not understood by its contributor, contains invented facts or behavior, creates unnecessary maintenance burden, violates security, privacy, licensing, attribution, or source-material rules, generates excessive agent noise, or repeatedly ignores this policy.