internetmaximalism/lean-reinforced-concrete

This repository is a small collection of Claude skills, packaged instructions that extend what an AI coding assistant can do, focused on formal verification using the Lean 4 proof language. Each skill lives in its own folder with a description file and optional helper scripts. The first skill, lean-formal-audit, helps translate a real system, such as a cryptographic protocol, smart contract, distributed system, or business rule set, into formal mathematical statements in Lean, so its behavior can be proven correct or compared against alternative designs. The second skill, lean-security-audit, is described as the offensive counterpart: it takes an existing formal model built by the first skill and actively tries to break it, questioning the assumptions behind its safety proofs, generating edge case test scenarios from the specification, and searching for concrete counterexamples that disprove claimed properties. Using either skill means copying or linking its folder into a project's local Claude configuration folder, or into a user-wide folder to make it available across every project. There is no application to run and no separate installation process beyond that copy step. This fits developers and security researchers who already use Claude for coding and want structured help applying formal methods, the mathematical technique of proving software correct, to audit real systems for hidden assumptions or bugs. The two skills are meant to work together as a pair: the first builds confidence that a system's design is sound by proving properties about it, while the second actively tries to tear that confidence down by hunting for gaps in the proof, unfinished steps, or variables that were never properly constrained. Treating those gaps as bug signals rather than minor loose ends is the core idea behind the project, and it gives a team a repeatable way to stress test both the system being audited and the formal argument used to justify it.