Natural language is often ambiguous. Writing code directly with LLMs can be brittle and hard to verify. This skill explores a middle ground: math-style specs (sets, relations, invariants) that agents can turn into architectures, APIs, and tests.