Решение олимпиадных задач по математике

★ 8.6 · learning

math-olympiad is a Claude Code skill that solves competition math problems — IMO, Putnam, USAMO, AIME — using adversarial verification designed to catch the errors that self-verification consistently misses. It launches 8–12 parallel solver agents, each internally iterating through a Yang-Huang-style solve → self-improve → self-verify → correct loop (up to 5 rounds), then passes only the cleaned proof to a fresh-context adversarial verifier running 5 targeted attack passes. The skill operates on pure reasoning with no computation tools; if no confident solution is reached, it says so explicitly rather than bluffing. When LaTeX is available and verification passes, a formatted PDF is compiled automatically. It handles individual proofs, full competition problem sets run in parallel, and external proof-checking requests alike.