Contributing

Contribute focused fixes, documented engine features, reproducible tests, and bug reports to the engine repository. Read its AGENTS.md files, Engine/CONTRIBUTING.md, architecture baseline, and roadmap before implementation.

Workflow and coding standards

Fork and clone the engine, branch from the target branch, and run commands inside Engine/. Explain the concrete problem and expected behavior before expanding scope. Keep application wiring in apps, shared services in crates, and backend/language types inside adapters.

Use the pinned Rust toolchain, rustfmt, and workspace Clippy rules. The workspace forbids unsafe code through its lint policy. Keep dependencies locked and approved by deny.toml. Persistent user data needs staged, flushed, atomic writes and a failure path that retains prior valid data. Baseline architecture changes need a decision record in Engine/docs/decisions/.

Tests and documentation

cargo xtask doctor
cargo xtask test

The test task builds Luau and runs formatting, Clippy, tests, checks, dependency/license checks, and application smoke checks. Add tests for changed behavior and failure/recovery boundaries; use focused reproduction fixtures rather than assertions that merely repeat implementation.

Update corresponding engine-local guides and the docs site for user-visible engine or scripting changes. Verify samples against the implementation and link a companion documentation PR when changes span repositories. Repository instructions require a fresh Windows installer for engine code changes; use a Windows environment/workflow and report any failure or unavailable validation.

Pull requests and review

Describe the final problem, resulting behavior, validation, and any migration or platform limitation. Follow the engine repository's completion instructions for committing, pushing, opening/updating the PR, and applying type:* plus relevant area:* labels. Keep generated artifacts out of source control.

For bug reports, provide the engine commit, environment, exact steps, expected/actual results, and a minimal scene/script. See Troubleshooting for useful diagnostics. Reviewers should be able to reproduce the issue and assess both successful behavior and recovery.