api-audit

Start by enumerating all exported types, functions, and traits into a surface map, then rate each component against dimensions like minimal surface, type safety, and naming. This protocol structures that process into phases with human checkpoints between components.

A well-designed API uses newtypes instead of raw primitives, exhaustive enums over magic strings, and Result/Option instead of null returns. It hides implementation details, offers one canonical path per concern, and makes invalid states unrepresentable at compile time.

Yes, the protocol includes language-specific checklists for all four. Rust checks cover non_exhaustive enums and unwrap usage, Go covers error wrapping and zero values, TypeScript covers strict mode and branded types, and Python covers type hints and dataclasses.

Run it before publishing a library, during major refactors, or when no deterministic test can verify API design correctness. It suits cases where human judgment about elegance, naming, and composability is required rather than automated linting.

The audit depends on iterative human checkpoints and does not modify code itself; it only produces findings and a remediation plan. Large codebases require multiple sessions since components are reviewed one at a time to maintain reviewer coherence.