The Agent Skills Directory

[Safety Methodology Framework]: The skill establishes an 'assume-malicious' posture for auditing Rust, treating every # Safety section as a theorem and every safety comment as a formal proof. This is a defensive security practice designed to minimize undefined behavior (UB).
[Authoritative Axioms]: It correctly identifies the Rust Reference and Standard Library documentation as the sole authoritative sources for memory safety, explicitly rejecting 'folklore' or implementation-specific behavior as valid proof premises.
[Security Test Cases]: The included testcases/*.rs files (such as test1.rs through test5.rs) contain intentional memory-safety vulnerabilities (e.g., aliasing violations, uninitialized memory reads, unsound Sync implementations). These are explicitly designed as evaluation benchmarks (EVAL.yml) to test the agent's auditing capabilities and do not represent malicious intent by the skill author.
[Adversarial Modeling of Safe Code]: The skill adopts a high-security approach by modeling caller-provided safe code as adversarial but type-correct, ensuring that unsafe abstractions remain sound even if safe trait implementations (like Ord or Iterator) are semantically incorrect.
[Trust Scope Compliance]: The skill correctly attributes its foundations to trusted sources, including the official Rust documentation and Google's open-source crate audits, following the established security guidelines for identifying trusted vendor resources.

unsafe-rust-review