moonbit-refactoring

The moonbit-refactoring skill provides a structured framework for AI agents to perform high-quality maintenance on MoonBit codebases. It is designed for developers, library maintainers, and teams working with MoonBit who need to update legacy code, enforce project architecture standards, or optimize existing modules for performance and clarity. By applying this skill, an agent acts as a specialized refactoring engine that prioritizes semantic correctness, API stability, and idiomatic expression.

• Performs holistic architectural reviews, including package structure, dependency management, and file size optimization, aiming for focused packages under 10k lines.

• Implements API minimization by identifying redundant public exports, moving helpers to internal packages, and strictly limiting the public API surface area to necessary call sites.

• Modernizes code by converting free functions into methods to improve discoverability and chaining, often utilizing the #as_free_fn macro to maintain backward compatibility during the transition.

• Optimizes memory usage by replacing data copying with view types such as ArrayView, StringView, and BytesView, ensuring efficient data processing.

• Drives the use of expressive MoonBit features, specifically replacing imperative mutation with declarative pattern matching, nested matching, and guard clauses for cleaner control flow.

• Automates the integration of loop invariants and documentation updates, ensuring that refactors are verified by compiler checks and unit tests.

• The skill requires access to the full repository and relies heavily on the moon ide toolchain, including moon ide doc, moon ide find-references, and moon ide analyze, to safely identify the impact of changes across the project.

• It assumes the agent will follow a workflow of starting with broad architectural improvements before descending into granular local refactoring.

• Users should expect the agent to run moon check and moon test frequently throughout the process to guarantee no regressions.

• Input to the agent should ideally include the specific package or symbol requiring attention, and the output is typically a series of coherent commits that improve package modularity, simplify enum constructor usage, and align the codebase with MoonBit best practices.