Rust Agents Orchestrator
This is the routing brain of the Rust skill package. It does two jobs:
- Route: at the START of a Rust task, map the user prompt to the exact
rust-* skill (or skills) to load.
- Gate: at the END of a Rust task, run the cross-skill quality checklist before declaring the work done.
ALWAYS run step 1 before writing Rust code. ALWAYS run step 2 before reporting a Rust task complete.
This skill routes and checklists. It does NOT do deep lint review: that is rust-agents-code-reviewer. It does NOT iterate on compiler errors: that is rust-agents-compile-fix. Treating these three as interchangeable is an anti-pattern.
The Package: 44 Skills, 5 Categories
| Category |
Count |
Prefix |
Purpose |
| core |
6 |
rust-core- |
Cross-cutting concepts, architecture, mental models |
| syntax |
15 |
rust-syntax- |
Language mechanics, API patterns, code form |
| impl |
13 |
rust-impl- |
Development workflows, end-to-end tasks |
| errors |
7 |
rust-errors- |
Diagnosing and fixing compiler/runtime errors |
| agents |
3 |
rust-agents- |
Routing, review, compile-fix orchestration |
Routing Table: Prompt Pattern to Skill
ALWAYS pick the most specific skill. If a prompt contains a rustc error code (E0xxx) or the word "error"/"fails to compile", route to an errors-* skill, NOT a syntax-* skill.
Ownership, borrowing, lifetimes
| User prompt signal |
Skill |
| "move", "value used after move", "who owns this" |
[[rust-syntax-ownership]] |
| E0382, "borrow of moved value" |
[[rust-errors-borrow-checker]] |
"& vs &mut", "borrow", "two-phase borrow", "RefCell" |
[[rust-syntax-borrowing]] |
| E0502 / E0596 / E0500, "cannot borrow as mutable" |
[[rust-errors-borrow-checker]] |
"lifetime annotation", "'a", "HRTB", "'static" |
[[rust-syntax-lifetimes]] |
| E0106 / E0623 / E0495 / E0700, "missing lifetime specifier" |
[[rust-errors-lifetimes]] |
Traits, generics, type system
| User prompt signal |
Skill |
| "trait", "default method", "supertrait", "orphan rule", "blanket impl" |
[[rust-syntax-traits]] |
"generic", "type parameter", "where clause", "monomorphization", "const generics" |
[[rust-syntax-generics]] |
"dyn Trait", "trait object", "object safety", "vtable", "trait upcasting" |
[[rust-syntax-trait-objects]] |
| "GAT", "generic associated type", "LendingIterator" |
[[rust-syntax-gats]] |
| E0277 / E0599 / E0220, "trait bound not satisfied", "method not found" |
[[rust-errors-trait-bounds]] |
Control flow, iterators, pointers
| User prompt signal |
Skill |
"match", "if let", "let else", "destructure", "exhaustiveness", "or-pattern" |
[[rust-syntax-pattern-matching]] |
"iterator", "map/filter/fold", "closure", "Fn/FnMut/FnOnce", "collect" |
[[rust-syntax-iterators-closures]] |
"Box", "Rc", "Arc", "Cell", "OnceLock", "Weak", "smart pointer" |
[[rust-syntax-smart-pointers]] |
Async
| User prompt signal |
Skill |
"async fn", ".await", "async block", "async closure", "AFIT", "RPITIT" |
[[rust-syntax-async-await]] |
| "Future trait", "executor", "Poll", "runtime choice", "structured concurrency" (concept) |
[[rust-core-async-runtime]] |
"tokio", "tokio::spawn", "spawn_blocking", "select!", "JoinSet", "timeout" |
[[rust-impl-async-tokio]] |
"!Send future", "Send across .await", async Pin/Unpin runtime error |
[[rust-errors-async]] |
Macros, unsafe, FFI
| User prompt signal |
Skill |
"macro_rules!", "declarative macro", "fragment specifier", "macro hygiene" |
[[rust-syntax-macros-declarative]] |
"proc macro", "derive macro", "syn", "quote", "proc_macro2" |
[[rust-syntax-macros-procedural]] |
"unsafe", "raw pointer", "UB", "strict provenance", "unsafe extern", "miri" |
[[rust-syntax-unsafe]] |
"FFI", "extern \"C\"", "bindgen", "cbindgen", "#[repr(C)]", "C interop" |
[[rust-impl-ffi-bindgen]] |
Edition, toolchain, core concepts
| User prompt signal |
Skill |
"edition 2024", "migrate to 2024", "edition migration", "edition idioms", "gen block" |
[[rust-syntax-edition-2024]] |
| "rustup", "channels", "components", "clippy setup", "rustfmt config", "MSRV management" |
[[rust-core-toolchain]] |
"which Rust version", "edition system", "stability", "rust-version", "release matrix" |
[[rust-core-language-versions]] |
"type system", "ZST", "repr", "niche optimization", "never type", "type-state" |
[[rust-core-type-system]] |
| "memory model", "stack vs heap", "drop scope", "RAII", "Send/Sync overview" |
[[rust-core-memory-model]] |
"stdlib", "which std module", "prelude", "std::collections", "std::sync" |
[[rust-core-stdlib-overview]] |
Cargo and project workflows
| User prompt signal |
Skill |
"cargo new", "Cargo.toml", "features", "profiles", "[lints] table" |
[[rust-impl-cargo-project]] |
| "workspace", "member crate", "virtual manifest", "workspace inheritance" |
[[rust-impl-workspaces]] |
"build.rs", "build script", "OUT_DIR", "code generation", "rerun-if-changed" |
[[rust-impl-build-scripts]] |
"cross-compile", "target triple", "rustup target add", "cross crate", "cfg" |
[[rust-impl-cross-compile]] |
Error handling design and error skills
| User prompt signal |
Skill |
"Result/Option idioms", "? operator", "custom error type", "error design" |
[[rust-impl-error-handling]] |
"thiserror", "anyhow", "library vs application errors", "error context" |
[[rust-errors-thiserror-anyhow]] |
"panic", "RUST_BACKTRACE", "abort vs unwind", "index out of bounds at runtime" |
[[rust-errors-runtime]] |
"linker error", "missing native lib", "version conflict", "cargo update", MSRV mismatch |
[[rust-errors-build-link]] |
Concurrency, common crates, testing
| User prompt signal |
Skill |
"Arc<Mutex>", "RwLock", "atomics", "Ordering", "deadlock", "scoped thread" |
[[rust-impl-concurrency]] |
"channel", "mpsc", "oneshot", "broadcast", "watch", "backpressure" |
[[rust-impl-channels]] |
"serde", "Serialize/Deserialize", "JSON/TOML/YAML", "#[serde(...)]" |
[[rust-impl-serde]] |
"clap", "CLI args", "derive(Parser)", "subcommand", "shell completion" |
[[rust-impl-clap-cli]] |
"#[test]", "integration test", "doc test", "nextest", "criterion", "bench" |
[[rust-impl-testing]] |
"#![no_std]", "embedded", "#[panic_handler]", "#[global_allocator]" |
[[rust-impl-no-std]] |
Agent / meta skills
| User prompt signal |
Skill |
| "review my Rust code", "is this idiomatic", deep clippy lint review |
[[rust-agents-code-reviewer]] |
"fix this compile error", iterate on rustc output, follow --explain suggestions |
[[rust-agents-compile-fix]] |
| "which skill", "route this task", "run the quality checklist" |
this skill |
Decision Tree: Ambiguous Prompt to Skill
Rust task arrives
├── Contains an E0xxx code or "does not compile"? ......... ROUTE to errors-*
│ ├── E0382 / E0502 / E0596 / E0500 ............. rust-errors-borrow-checker
│ ├── E0106 / E0623 / E0495 / E0700 ............. rust-errors-lifetimes
│ ├── E0277 / E0599 / E0220 ..................... rust-errors-trait-bounds
│ ├── linker / native lib / version clash ....... rust-errors-build-link
│ ├── async `!Send` / Pin across .await ......... rust-errors-async
│ └── panic / OOB / abort at runtime ............ rust-errors-runtime
├── "How do I build / set up X"? ...................... ROUTE to impl-*
├── "Why does Rust work like X" (concept)? ............ ROUTE to core-*
├── "What is the syntax for X" (language form)? ....... ROUTE to syntax-*
├── "Review / is this idiomatic"? ..................... rust-agents-code-reviewer
└── Multiple skills plausible? ........................ load the NARROWEST
that names the exact construct, then load its core-* concept skill
only if the user needs the mental model, not just the syntax.
ALWAYS prefer the narrowest match. NEVER load a core-* overview when the user asked a concrete syntax-* or impl-* question. Load 1 to 3 skills, not the whole package.
Cross-Skill Quality Checklist
Run this BEFORE declaring ANY Rust task complete. Every item is a hard gate.
| # |
Check |
Command / inspection |
Pass criterion |
| a |
Edition + MSRV |
read edition and rust-version in Cargo.toml |
code uses only features available at the stated MSRV; edition is 2024 unless the project pins older |
| b |
Clippy |
cargo clippy --all-targets -- -D warnings |
no warnings; correctness + suspicious + perf clean |
| c |
Formatting |
cargo fmt --all --check |
exits 0, no diff |
| d |
No panics on fallible paths |
grep .unwrap() / .expect() in non-test code |
every remaining one is on a genuinely infallible path or has a justifying comment |
| e |
Async hygiene |
inspect .await points |
no blocking call on the async executor; no !Send value held across .await in a spawned task |
| f |
Doc coverage |
cargo doc + inspect public items |
every public item has a /// doc; add a doctest where it clarifies usage |
| g |
Unsafe is justified |
grep unsafe blocks |
every unsafe block has a // SAFETY: comment stating the invariant |
| h |
Error type contract |
inspect library error types |
error types implement std::error::Error + Display; a library does NOT leak anyhow::Error in its public API |
| i |
Tests |
cargo test --all-targets and cargo test --doc |
all unit, integration, and doc tests pass |
If any item fails, fix it and re-run the failing check. NEVER report a Rust task done with an unchecked item.
See references/methods.md for the full command set and CI wiring, references/examples.md for worked routing and checklist runs, references/anti-patterns.md for routing and gating mistakes.
Quick Reference
- Routing happens FIRST, the checklist happens LAST. Never skip either.
- Error code present means an
errors-* skill, never a syntax-* skill.
cargo build passing is NOT the quality gate. cargo clippy plus cargo fmt --check plus cargo test is.- MSRV is set by
rust-version in Cargo.toml. A new dependency must not raise it silently.
- This skill routes and gates. Deep review is
rust-agents-code-reviewer. Error iteration is rust-agents-compile-fix.
Reference Links
