Skip to content
gohawk

AI policy

gohawk was developed with assistance from AI tools. AI-assisted contributions are permitted, but contributors must clearly disclose how AI was used in their pull request.

Using AI does not change the project’s quality requirements. Contributors remain responsible for understanding their changes and ensuring they follow the repository’s contribution guidelines. In particular, analyzer changes must meet gohawk’s standards for precision, testing, documentation, and supported analyzer scope.

AI-generated output should be reviewed and validated before submission. Pull requests must include appropriate tests and demonstrate that reported diagnostics are actionable and have a low risk of false positives.

Human readability is a priority. AI-assisted code must remain clear, idiomatic, and maintainable by contributors without relying on AI to explain it. Avoid unnecessary abstractions, excessive indirection, and generated complexity that makes the implementation harder to review or modify.

The project owner may close low-effort or insufficiently reviewed pull requests that do not follow the contribution guidelines, regardless of whether AI was used.

How AI was used to develop gohawk

gohawk has been developed with substantial assistance from LLM-based coding tools, primarily OpenAI Codex. AI assistance has been used to:

  • Explore analyzer designs and implementation approaches.
  • Write and refactor portions of the codebase.
  • Draft tests, documentation, and diagnostic wording.
  • Review suspected bugs and potential false positives.
  • Construct reproductions and investigate behavior in real-world projects.
  • Run and interpret validation, dogfooding, and integration tests.

AI-generated output is not accepted without review. The project owner determines the project’s direction, reviews changes, evaluates whether diagnostics are sufficiently precise, and remains responsible for the resulting code.

Analyzer behavior must be supported by focused fixtures, appropriate regression tests, representative dogfooding, and CI validation regardless of whether AI contributed to the implementation.