rust-expert

Rust Expert

Use this as the default Rust engineering router. Inspect the repo first, load only the references needed for the task, then use compiler/test feedback as design evidence.

Operating Model

1. Read the local guidance and manifests first: AGENTS.md, Cargo.toml, Cargo.lock, rust-toolchain.toml, .cargo/config.toml, CI, justfile, mise.toml, and nearby tests.
2. Classify the task before editing: compile failure, implementation, review, refactor, dependency, public API, release, performance, or security.
3. Fix the root cause. Avoid silencing Rust with reflexive .clone(), Arc<Mutex<_>>, broad trait bounds, unwrap, or unsafe.
4. Use repo-native gates first. When command surface is unclear, run scripts/discover-rust-gates.mjs <repo-root>.
5. Research current docs/source when facts can drift: dependency additions or upgrades, version-sensitive APIs, Tauri/plugin support, public API/release decisions, security, unsafe/FFI, or novel crates.

Domain Routing

Signal	Prefer
Command-line app, subcommands, flags, config precedence, JSON/stdout contracts	rust-cli-clap
Terminal UI, interactive dashboards, Ratatui widgets/layout/events	rust-tui-ratatui
Tauri v2, src-tauri, commands, capabilities, IPC, updater, mobile layout	rust-tauri-apps
HTTP APIs/services, Axum, Tower, Tokio server runtime, DB pools	rust-web-services
Broad architecture review or multi-domain Rust planning explicitly requested	rust-mega-eng

If no specialist owns the task, stay in rust-expert. Keep rust-mega-eng explicit-only; broad wording alone does not trigger it.

Reference Map

Load the smallest needed set:

• references/toolchain-cargo.md: editions, MSRV, resolver, lockfiles, features, workspace policy, verification commands.
• references/ownership-async-errors.md: borrow checker, ownership, lifetimes, async Send, task ownership, typed errors.
• references/crate-selection.md: preferred crates, dependency decision rules, feature hygiene, source-refresh triggers.
• references/testing-quality.md: unit/integration/doctest/property/snapshot/ benchmark checks and reviewable test design.
• references/performance-security.md: measurement, allocations, unsafe/FFI, supply chain, cargo-deny/audit, secrets and subprocesses.

Toolchain Policy

Existing repos: preserve detected contracts first. Do not change edition, resolver, rust-version, lockfile policy, or exact toolchain pins unless the task is explicitly a migration or the repo is missing required metadata.

Greenfield defaults:

• edition = "2024".
• Set [workspace] resolver = "3" explicitly for virtual workspaces.
• Add an explicit rust-version and document the policy.
• Commit Cargo.lock by default; for public libraries, also test latest dependency resolution because consumers use Cargo.toml.
• Avoid exact rust-toolchain.toml version pins unless reproducibility, nightly, embedded/custom targets, or release policy requires them.

Verification Ladder

Start narrow, then broaden:

cargo check -p <crate>
cargo test -p <crate> <test_name>
cargo fmt --all -- --check
cargo clippy --workspace --all-targets --all-features --locked -- -D warnings
cargo test --workspace --all-targets --all-features --locked
cargo test --doc --workspace --locked

Use cargo nextest run when configured. Add cargo hack, cargo deny, cargo audit, cargo semver-checks, or release tooling only when the repo or change surface warrants it.

Helper Scripts

• scripts/discover-rust-gates.mjs [repo-root]: print likely Rust gates.
• scripts/check-reference-links.mjs <skill-dir...>: validate local skill links.
• scripts/check-trigger-evals.mjs <skill-dir...>: validate trigger eval JSON.