What problem does it solve? Informal natural-language requirements leave correctness of critical algorithms, protocols, and security properties unverified. This Skill converts vague specifications into Lean 4 formal specifications, proves them with kernel-checked verification, and connects proven properties to test-first implementation so only proven claims are called guaranteed. ## Core Features & Use Cases - Property catalog extraction: Enumerates state machine invariants, MUST/MUST NOT clauses, temporal properties, and workflow-level guarantees from specs before formalization, following the principle that over-extraction is safe and omissions are dangerous. - Autoformalization and proof repair loop: Translates requirements into Lean def and theorem declarations, then iterates with lake build, tactics like simp/omega/decide/exact?, and lemma decomposition until proofs pass. - Proof authenticity checks: Verifies proofs with #print axioms to detect leftover sorry holes or native_decide trust-base expansion before claiming guarantees. - Bridge to test-first implementation: Maps each proven theorem to a property-based test that must fail first, exposing divergence between the proven model and the actual implementation. - Use Case: When implementing a protocol state machine, use this Skill to prove that error states always transition to closed and that chunked message reassembly is segmentation-independent, then generate failing property tests that catch implementation drift from the proven model. ## Quick Start Ask the AI to formalize and prove the correctness properties of your specification in Lean 4, for example: formalize this protocol spec and prove its state machine invariants with Lean.
