Rust orchestrator

Rust Orchestrator

The single entry skill for Rust work. It locates the task on the write ↔ verify axis and delegates to one of two specialist spokes. The cross-cutting model both spokes share — the library-vs-application error strategy (thiserror vs anyhow), the type-driven design stance, and the standard cargo toolchain — lives in rust-core; read it before choosing an error type or shaping a public API.

Cluster map (intent → spoke)

• rust-patterns — idiomatic production code. Ownership & borrowing, Result/? error propagation, enums + exhaustive matching, traits/generics, Arc<Mutex<T>> and async concurrency, builder/newtype patterns, and domain-organized crate/module layout.
• rust-testing — verification. TDD RED-GREEN-REFACTOR, #[cfg(test)] unit tests, tests/ integration binaries, #[tokio::test] async tests, rstest parameterization, proptest property tests, mockall trait mocking, doc tests, Criterion benches, and cargo-llvm-cov coverage gates.
• rust-core (shared reference) — the error-strategy decision both spokes turn on, plus shared cargo conventions, the version/tooling matrix, and guardrails.

Folded spokes

• rust-coding-skill — deep-dive authoring companion to rust-patterns. Goes finer-grained on data modeling (struct/enum/newtype, &str vs String, Cow/Arc ownership choices, modeling invariants with types like NonZeroU32), impl-block organization (placement, constructor/getter/mutation grouping, builders), macros (derive, focused declarative macros, proc-macro boundaries), and build-speed tuning (mold linker, sccache, cargo check iteration, workspace splitting, dev/release profiles). Route here when the ask is specifically about how to shape the types/impls/macros or speed up compiles; route to rust-patterns for broader idiomatic review (borrow-checker fights, concurrency, error propagation) and to rust-core for the error-strategy and toolchain model both share.

Routing rules by intent

Writing or shaping code → rust-patterns

• "Is this idiomatic?" / code review → rust-patterns
• Ownership, lifetimes, Cow, "stop cloning to please the borrow checker" → rust-patterns
• Error design — thiserror for a library, anyhow for a binary → rust-patterns (model in rust-core)
• Modeling states as enums, newtypes, builders, trait objects vs generics → rust-patterns
• Concurrency — Arc<Mutex<T>>, channels, async/Tokio → rust-patterns
• Crate/module structure, pub surface, visibility → rust-patterns
• Detailed data modeling (struct/enum/newtype, ownership of each field), impl-block layout, writing macros, or build-speed tuning (mold, sccache, cargo check, workspaces) → rust-coding-skill

Proving or measuring code → rust-testing

• "Write tests for this" / add coverage → rust-testing
• TDD workflow / "test first" → rust-testing
• Async test, integration test, mock a dependency → rust-testing
• Property-based testing, fuzz-style invariants → rust-testing
• Benchmarks, coverage thresholds, CI test matrix → rust-testing

Touches the boundary between both

• Error types are designed in rust-patterns and asserted in rust-testing (matches!(err, …), Result-returning tests) — keep the variant shape consistent via rust-core.

Standard flow

1. Locate the task on the write ↔ verify axis: shaping/refactoring code → rust-patterns; proving/measuring it → rust-testing.
2. If it touches error types, public API shape, or the cargo toolchain, pull the model from rust-core first — the error strategy chosen in code dictates how tests assert on it.
3. Delegate to the spoke(s). A "build it the right way with tests" ask fans out in TDD order: rust-testing (write the failing test) → rust-patterns (implement idiomatically) → rust-testing (coverage gate).
4. Return: chosen spoke(s), the error strategy implied (lib thiserror vs app anyhow), and the next action.

Guardrails

See rust-core. In short: let the type system do the work — ? over unwrap() in production, model illegal states as unrepresentable, keep unsafe minimal and documented with a # Safety comment, and expose the narrowest pub surface. Pick thiserror for libraries and anyhow for applications, and never silently widen that boundary. Tests follow TDD, stay independent, never sleep(), and assert on typed error variants — not panic strings — wherever the code returns a Result.

Loading spokes on demand

To keep CLI startup context lean, this cluster's spokes are not separately registered as skills — only this orchestrator and its *-core are enumerated. When you route to a spoke named above, load it on demand by reading its file:

~/.agents/skill-clusters/skills/<spoke-name>/SKILL.md (or skills/<spoke-name>/SKILL.md inside the skill-clusters repo).