Beautiful Rust skills for Claude

Asked for a small semver library without the skill, an agent in the author's test made every struct field public and put unwrap() in its doc examples, according to the README. With it, the fields stayed private and the docs listed each way parsing can fail. Beautiful Rust is three skills by Morgan Linton, MIT-licensed and not published by Anthropic. They run in Claude Code, OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI and any agent that reads the Agent Skills format. New to skills? Start with how Claude Skills work.

What's in the repository

• rust-best-practices loads whenever the agent writes or changes Rust. Its SKILL.md of about 500 lines covers ownership, API design, errors and panics, async, unsafe code and FFI, tests and performance. Then comes a checklist: cargo fmt, clippy with -D warnings, tests, cargo doc.
• rust-review reviews a file, crate or diff and sorts findings into must fix, should fix and consider. Each one gives a file:line, the rule it breaks with that rule's source, and the smallest fix. The report closes with what the code already does well.
• rust-port moves C, C++, Python, Go or JavaScript to Rust. It ports the original tests first, turns null into Option and error codes into an error enum, and writes a PORTING.md listing every deliberate difference.
• lints/ is a clippy config for CI. It flags unwrap() outside tests, unsafe blocks with no // SAFETY: comment, bool parameters, and leftover dbg! and todo!.

The rules are drawn from five published guides: Google's Rust style guide, the Rust API Guidelines, Effective Rust, Rust Design Patterns and Microsoft's Pragmatic Rust Guidelines. Where they disagree, the skill picks one and says why. A separate Bevy reference stops general advice from breaking game code, where system parameters are taken by value on purpose.

Does it help?

The author ran eight tasks on Claude Code and Codex CLI, with and without the skills, and graded the results blind. With the skills, Claude Code on Opus 5.5 went from 72 to 80 of 85 checks, and Codex went from 53 to 76. Reviews gained most. One task got worse: Claude scored 7 of 10 on a Bevy feature with the skills and 10 without. Each setup ran once and a Claude model did the grading, so treat it as one data point.

When to use it

• A library crate other people will call.
• A review pass on a Rust pull request before a person reads it.
• Porting a C or Python module that has tests, so you can prove the port behaves the same.

For review in any language, the bundled /code-review skill is already in Claude Code. Superpowers adds test-first habits that aren't tied to Rust.

How to install it

Any agent, with Node 22.20 or later:

npx skills add morganlinton/BeautifulRust

As a Claude Code plugin, so updates arrive with claude plugin update beautiful-rust:

/plugin marketplace add morganlinton/BeautifulRust
/plugin install beautiful-rust@beautiful-rust

Install all three, since the review and port skills lean on the best-practices one for style. Codex, Copilot and Gemini CLI each get their own command in the README. On Claude.ai, zip each skill folder and upload it under Customize > Skills, with code execution turned on.

Example request

Review [path, e.g. src/parser.rs] with the rust-review skill. Run cargo fmt, clippy and the tests first if a Rust toolchain is installed. Sort findings into must fix, should fix and consider, cite the rule each one breaks, and show the smallest code change for each. Don't edit the file.

Limits

• The project is young. Version 1.5.0 is dated October 30, 2026 in the changelog.
• The review runs cargo, clippy and tests only when a Rust toolchain is installed where the agent works.
• Running the bundled evals calls a real model, so you pay for the tokens.
• The privacy policy says the skills collect nothing and make no network requests. Cargo still downloads your project's dependencies as usual.