This post is both a project diary and a design sketch for "rusty-cpp": a static analyzer that tries to bring some of Rust’s memory-safety ergonomics into C++ without requiring a custom compiler. The motivation is familiar to anyone who has fought C++ pointer bugs: segmentation faults, use-after-free, and the endless temptation to wrap everything in shared_ptr just to sleep at night.
Instead of asking teams to rewrite large codebases in Rust (or to adopt a nonstandard compiler), the project takes a more incremental approach. C++ code is annotated using comments like @safe and @unsafe, and the analyzer enforces a simple boundary: @safe code can only call other @safe code directly, and anything unannotated (including the STL and legacy code) is treated as unsafe unless explicitly wrapped. The author argues this creates an audit-friendly seam: you can start by making a small “safe island” in an existing codebase and expand it over time.
The other interesting angle is process: the author describes using Claude Code to iterate on the prototype (code + tests), gradually hardening it as failures were found. Even if you’re skeptical about whether Rust-like guarantees can be bolted onto C++, it’s a concrete example of LLM-assisted engineering being used to bootstrap a non-trivial tooling project.