The following is a short step-by-step rundown of what one typically would do to contribute.

• Fork this project on GitHub.
• For setting up the environment to run the self tests, please run init-tests-after-clone.sh.
• Please try to write a test that fails unless the contribution is present.
• Try to avoid massive commits and prefer to take small steps, with one commit for each.
• Feel free to add yourself to AUTHORS file.
• Create a pull request.

Pull requests are preferred when you have a fix or improvement to contribute: a PR can be treated as an issue with a fix attached. Describe the problem and proposed solution in the PR itself; a separate issue is not required.

Please avoid opening an issue solely to follow it immediately with a PR that fixes it. Open the PR directly instead. Issues are still welcome for problems without a fix or for changes that need discussion before implementation.

Contributions must be made with care and meet the quality bar of the surrounding code. That means a change should not leave GitPython worse than it was before: it should be readable, maintainable, tested where practical, documented and consistent with the existing style and behavior.

A contribution that works only narrowly but lowers the quality of the codebase may be declined. The maintainers may not always be able to provide detailed feedback.

AI agents communicating through a person's account must identify themselves, for example in issue or PR descriptions and comments. AI assistance that does not replace the person as the speaker, such as proofreading or wording polish, does not require identification.

Even when identifying themselves, agents must not assert on a person's behalf that that person performed an action, such as reviewing or approving a PR. The person must make any such statement themselves. If it is included alongside agent-authored content, it must be supplied by the person and preserved verbatim in a clearly labeled, separate user-authored section. Agents must not draft, paraphrase, or embed such statements in their own narration.

Attributing AI assistance in commit metadata, for example with a Co-authored-by trailer, is welcome but not required. Code is reviewed the same way regardless of its origin.

Tests require Python 3.11+, Git 2.52+, Bash, and the test extra, which installs GixPython even when testing the CLI backend. Select GIT_PYTHON_BACKEND=cli or GIT_PYTHON_BACKEND=gix before running pytest.

Put initial repository and filesystem states in tests/fixtures/*.sh and use test.lib.fixtures to run them through gix.testtools. Read-only tests share cached output; mutating tests use writable copies. Use execute=True only when the fixture cannot be copied, and explain the reason at its call site. Keep operations under test in Python. Immutable parser byte samples remain in test/fixtures; generated caches and archives are ignored.

Use test.cleanup.cleanup_directory for cleanup outside the shared fixture owners. Close repository handles before removing their files. Cleanup is best-effort: it logs filesystem errors, removes whatever it can, and leaves locked files behind without failing or skipping a test. The shared writable repository decorators still keep failed tests' directories for debugging.

Deletions and renames that exercise library behavior or prepare a fixture for reuse must remain strict. If a cached fixture cannot be cleaned up, rebuild it at a fresh location before handing it to another test.

For details related to contributing to the fuzzing test suite and OSS-Fuzz integration, please refer to the dedicated fuzzing README.