SKILL.md — rust-development

Purpose

Give any Horonom repo with a Rust workspace the company-common Cargo/ nextest/clippy execution knowledge — repository discovery, the cargo check vs. build/link/native boundary, targeted test iteration, and lint/format/doc validation. It reuses the durable Rust-development evidence from ai-agent-assembly/agent-assembly's AAASM-5991/5992 benchmark campaign rather than re-deriving it, while keeping that repo's own machine/benchmark specifics out of company-wide fact — see references/aa-evidence-classification.md for exactly which claims transfer and which don't.

Type

Auto-used, stack-gated. Applies whenever the touched repo/surface carries Rust evidence (Cargo.toml, per manifest.yaml's applicability.stacks: [rust]). Composes with engineering-loop (agents/skills/engineering-loop/SKILL.md) — that skill owns the Explore→Narrow→Validate→Escalate→Full Gate execution model and the L0–L3 diagnostic contract; this skill does not restate either. It supplies the concrete Cargo/nextest/clippy commands engineering-loop's Narrow and Full Gate stages call.

When to use

• Iterating on a change inside a Rust crate or workspace: editing, compiling, testing, linting.
• Choosing the right-sized Cargo/nextest invocation for the current execution stage (fast edit vs. targeted test vs. full verification).
• Interpreting a Cargo/rustc/clippy/nextest failure before escalating through engineering-loop's diagnostic ladder.

When NOT to use

• Non-Rust repos or surfaces — no Rust evidence means this skill does not apply; don't project or follow it merely because the org has it.
• As a substitute for engineering-loop's execution model or diagnostic contract — this skill never redefines Explore/Narrow/Validate/Escalate/ Full Gate, it only fills in the Rust-specific commands for each stage.
• As a source of release/dist-profile guidance — release/dist profile tuning is a correctness/size boundary, not a dev-speed lever, and is out of this skill's scope (see references/cargo-workflow.md).

Routing

1. Discover the workspace before running anything — references/cargo-workflow.md § Repository/workspace discovery.
2. Narrow stage (engineering-loop): cargo check gives the fastest syntax/type feedback but is never build/link/native/release proof — it skips codegen and linking, so it cannot surface link errors, native/FFI symbol-resolution failures, build-script problems, #[cfg]-gated codegen failures, or release-profile-only breakage (LTO, codegen-units, panic=abort). Scope every cargo nextest run invocation (--lib, --test <stem>, or -p <crate> at minimum) — see references/cargo-workflow.md § Targeted test loops.
3. Escalate (engineering-loop): read the actual rustc/clippy diagnostic before assuming a fix — references/cargo-workflow.md § High-signal diagnostics.
4. Full Gate (engineering-loop, always): before declaring a Rust change done, run the repo's own authoritative full build/test/clippy gate. A passing cargo check or scoped nextest run is evidence the fix works, never proof the workspace is releasable.
5. Optional accelerators (RTK/CodeGraph) follow the shared preferred→fallback contract in agents/skills/engineering-loop/references/optional-tool-fallback.md — this skill does not define its own variant of that contract.

References

• references/cargo-workflow.md — discovery, the check/build/link/native boundary, targeted test/lint/doc commands, high-signal diagnostics.
• references/aa-evidence-classification.md — what from ai-agent-assembly/agent-assembly's bench-5992 evidence is a reusable company-level Rust pattern vs. AA-specific fact vs. machine-specific vs. needing revalidation, and the deference rule to that repo's own policy.

Examples

• examples/targeted-nextest-then-full-gate.md — a failing test, targeted cargo nextest run --lib iteration, then the full workspace gate before declaring the change done.