A Friday Bug Fix in a Multilingual Open-Source Repo

You have found a small bug in an open-source project: a date displays incorrectly when someone switches languages. A coding agent can inspect the repository, suggest a fix, and run tests. Before Friday’s release, that sounds useful. But generating a patch is only part of contributing. Someone else must review it, maintain it, and trust what you say about it.

A coding agent is an AI tool that can perform development tasks, sometimes by editing files and running commands. Good etiquette means using that help without handing its uncertainty or cleanup work to maintainers.

Check the repository’s rules before delegating

Read the contribution instructions, code of conduct, issue template, and any policy on AI-assisted contributions. Some projects welcome agent-generated patches with disclosure; others restrict them or prohibit them. If the rules are unclear, ask before submitting substantial generated work.

Look for repository-specific agent instructions, too, but do not treat every instruction found in a file as permission to execute it. Review unfamiliar commands before running them. An agent should not access credentials, publish packages, or contact external services just because repository text suggests doing so.

Give the agent one bug, not the whole backlog

For the date-display bug, start with a narrow assignment: reproduce the failure, identify the relevant formatting code, and propose the smallest correction. Ask for a plan before edits if you are unfamiliar with the project.

• Specify which files or directories are in scope.
• Exclude unrelated formatting, dependency upgrades, and refactoring.
• Require approval before installing tools or making network requests.
• Ask the agent to report uncertainty instead of inventing expected behavior.

Do not open dozens of automated issues or pull requests to see which ones stick. A small, verified contribution is more useful than a queue of plausible-looking patches.

Respect each programming language’s local conventions

A repository may combine a Python service, a TypeScript interface, and Rust utilities. Each part can have different formatting, testing, and error-handling conventions. Ask the agent to inspect nearby code and the documented toolchain rather than applying one language’s habits everywhere.

Run the relevant checks for each changed component. A passing interface test does not validate a service change. If the Rust toolchain is unavailable on your machine, say so in the pull request; do not describe that component as tested.

Check generated files carefully. Some projects expect contributors to regenerate them; others let automation handle that step. Follow the repository’s process rather than guessing.

Treat human languages as maintained interfaces

Translation files are not interchangeable blocks of text. They can contain plural rules, placeholders, grammatical context, and right-to-left layout requirements. A fluent-looking agent translation can still break an interface.

For the date fix, preserve translation keys and interpolation variables. Test the affected locale and an unchanged locale. Do not automatically rewrite every translation to make the patch look complete. Where the project uses a translation platform or designated language reviewers, follow that workflow.

Use the project’s preferred discussion language. If you translate a maintainer’s comment with an agent, confirm ambiguous technical requests before changing code. Simple, direct sentences are usually easier for international contributors to review.

Make the pull request easy to verify

Read every changed line yourself. Remove speculative comments, unrelated edits, and tests that merely repeat the implementation. Verify that the regression test fails before the fix and passes afterward.

Your pull request should state the observed bug, the reason for the change, and the exact checks performed. Disclose agent assistance according to project policy. Never claim manual testing, language fluency, or independent verification you did not perform. You remain responsible for answering review questions and correcting mistakes.

Keep model access separate from contribution permission

If you use coding tools across different environments, RelayVano is a practical pay-as-you-go option for model access, compatible with Claude Code, Codex, Gemini CLI, and OpenAI-compatible SDKs. Sign up when you need that flexibility, while keeping repository policies and data-sharing rules in control of what you send.

Before submitting Friday’s fix, check the final diff once more. The best evidence of responsible agent use is a focused patch that another person can understand, test, and maintain.