Changelog

All notable changes to this project are listed here. Versions follow Semantic Versioning, and the version lives in .claude-plugin/plugin.json.

Unreleased

Added

• A Codex and ChatGPT plugin. A portable plugin.json at the repo root, with the extensions.com.openai listing details, packages the three skills for Codex and the ChatGPT / Codex Plugins Directory. .agents/plugins/marketplace.json lets Codex install it with codex plugin add beautiful-rust@beautiful-rust. scripts/package-openai-plugin.py builds the directory upload without the eval cases, and scripts/check-skill.py now checks the manifest against OpenAI's field limits and keeps its version in step with the Claude Code manifest. PRIVACY.md states that the skills collect nothing. Requested by @Castor6 in #7.

1.5.0 - 2026-10-30

Bevy support, a Claude Code vs Codex benchmark, and every Rust example compiled in CI. Thanks to @tmchow and @DylanCosto for their pull requests, and to @millionart for proposing Bevy support.

Added

• Bevy support. skills/rust-best-practices/references/bevy.md covers Bevy 0.20. It says which general rules don't apply to systems: by-value system parameters, the Bevy prelude, public fields on plain-data components, and complex query types. It lists the ECS bugs that compile cleanly: conflicting queries (B0001), deferred Commands, systems with no order between them, and unwrapped single(). It also gives the four lint allows a Bevy crate needs on top of lints/lints.toml. Every behavior it describes was confirmed with a test on bevy 0.20.0. Both the writing and review skills read it when a crate depends on bevy. Proposed by @millionart in #5.
• Two Bevy eval cases. rust-review/evals/review-bevy-arena plants a query conflict and an ordering bug in a combat plugin whose tests pass, alongside valid Bevy patterns the review should leave alone. rust-best-practices/evals/bevy-arena-regen adds a feature to the fixed plugin and checks that Bevy's conventions survive. Both pin bevy = "=0.20.0" and ship a Cargo.lock.
• A cross-agent benchmark. benchmarks/cross-agent-2026-10/ ran all eight eval cases on Claude Code and Codex CLI, with and without the skills, and graded them blind. The skills raised Codex from 53/85 to 76/85 and Claude from 72/85 to 80/85. The page includes every assertion result, the harness, and the caveats, including one case where the skills made Claude worse.
• Every Rust example is compiled. scripts/check-examples.py compiles each Rust code block in the skills, the reference digests, the README, and examples/ with cargo check, and reports errors at the markdown line. scripts/examples.json holds the surrounding code that fragments need to compile, or the reason a block is skipped. CI runs it on every push and pull request. Contributed by @tmchow in #4.
• Two more lints in lints/lints.toml. clippy::missing_assert_message flags an assert! with no message, the checklist rule from the Google Rust Style Guide's "Use context strings to clarify assertions"; it stays quiet inside #[test] functions. clippy::exhaustive_enums ([lib]) flags an exported enum without #[non_exhaustive], the lint the same guide names for that rule; binary crates export nothing, so it never fires there. lints/check/ proves both fire. Contributed by @tmchow in #8.

Changed

• The glob-import rule in both skills allows a framework's documented prelude, such as use bevy::prelude::*. Clippy's wildcard_imports already exempts these.
• Both skills now cover impl Into<T> parameters and collected return values, from the Google Rust Style Guide's "Accepting Into" and "Iterators vs. owned collections in return types". Application code takes String or &str rather than impl Into<String>, and library code returns an iterator instead of collecting into a Vec. rust-review flags both, with Google's exceptions for owned results. Contributed by @tmchow in #9.

Fixed

• The Errors section paired get with get_unchecked as a fallible API and its panicking twin, but an out-of-bounds get_unchecked is undefined behavior, not a panic. It now pairs indexing (v[i], which panics) with v.get(i). Reported and fixed by @DylanCosto in #3 and #6.

1.4.0 - 2026-10-26

The first release with outside contributions. Thanks to @mvanhorn for both.

Added

• A skill checker in CI. scripts/check-skill.py validates every skill's frontmatter, links, eval cases, and the plugin manifests, using only Python's standard library. It runs on every push and pull request. Contributed by @mvanhorn.
• Proof that every lint fires. lints/check/ is a crate with one deliberate violation for each of the 28 lints set to warn, and scripts/check-lints.sh runs clippy on it. It fails if a lint goes quiet, if something unexpected fires, or if the check crate's copy of the table drifts from lints/lints.toml. CI now runs it too. Contributed by @mvanhorn.
• Four more lints in lints/lints.toml: redundant_clone, rc_buffer, string_add, and wildcard_imports. module_inception and enum_glob_use are now allowed, matching rules the skill already states.

Fixed

• allow_attributes and allow_attributes_without_reason are silently disabled by clippy for crates whose rust-version is below 1.81. The lints config and README now say so.

Changed

• rust-best-practices is back under the 500-line guidance, with no rules removed.
• CI uses actions/checkout@v7 and runs with read-only permissions.

1.3.0 - 2026-10-25

Added

• rust-port skill. Ports C, C++, Python, Go, JavaScript, or TypeScript to idiomatic Rust with the same behavior. It ports the original tests first, maps each language's habits to Rust (NULL and None to Option, error codes and exceptions to error enums, flag ints and magic strings to typed values), keeps the port proportionate instead of reimplementing the source language's runtime, and writes a PORTING.md with API and test mappings and every deliberate difference.
• Two eval cases for it: a C config parser and a Python task scheduler, each with its original test suite. With the skill, the ports passed 22 of 22 checks; without it, 19 of 22. Both configurations ported correctly. The skill's ports were the ones that stayed proportionate: the baseline Python port was 5.8 times the original's size and reimplemented Python's string and integer parsing; the skill's was 3.1 times, with none.
• CONTRIBUTING.md and issue templates for proposing a rule and reporting a review false positive.

1.2.0 - 2026-10-25

Added

• rust-review skill. Point your agent at a file, crate, or diff and get a prioritized review: must fix, should fix, and consider. Every finding cites a file:line location, the rule it breaks with its source, and the smallest code change that fixes it. It runs fmt, clippy, and tests when a toolchain is available. On the inventory eval it passed 11 of 12 checks against 7 of 12 without the skill; the difference was rule citations, code fixes, and line references.
• Drop-in lints in lints/: a [lints] block for Cargo.toml and a clippy.toml. They make clippy enforce the machine-checkable rules, including no unwrap() outside tests, // SAFETY: on every unsafe block, reasons on every suppression, and no bool parameters. Every lint name was verified against clippy 0.1.99.
• An example review report in examples/review-inventory.md.
• A new painted README banner.
• An eval case for the review skill under skills/rust-review/evals/.
• README install instructions for agents other than Claude Code, since both skills are standard Agent Skills, and a before/after example of what the writing skill changes.
• This changelog.

Verified

• Every Rust code example in the skills, the reference digests, and the README now compiles on stable Rust 1.99.

Changed

• rust-best-practices points to the lints config when a project has no [lints] table.

1.1.0 - 2026-10-24

Tightened the skill based on a with/without comparison on three realistic tasks. With the skill, 29 of 30 assertions passed; without it, 24 of 30.

• bool parameters are out of public and shared signatures. Use a two-variant enum or two methods instead. An existing bool is a smell to remove when refactoring.
• New section on refactoring someone else's code: fix every smell on the list, keep public behavior identical, show before and after for any signature you change, and flag quirks instead of silently fixing them.
• A hand-written Ord must come with matching PartialEq and Hash.
• Added three eval cases under skills/rust-best-practices/evals/ and a Testing section in the README.

1.0.0 - 2026-10-23

First release.

• The rust-best-practices skill, distilled from the Google Rust Style Guide, the Rust API Guidelines, Effective Rust, Rust Design Patterns, and Microsoft's Pragmatic Rust Guidelines, with one reference digest per source.
• Packaged as a Claude Code plugin that is its own marketplace.
• MIT license, source attribution, and a plugin icon.