Math Olympiad

anthropics/claude-plugins-officialOfficial

This is a proof-oriented solver for IMO, Putnam, USAMO, and AIME problems that runs parallel solving attempts with internal refinement, then strips the thinking traces and runs fresh-context adversarial verification. The workflow matters: it launches 8-12 solver agents that each iterate through solve, self-improve, self-verify cycles, cleans the solutions to remove reasoning traces that bias verifiers toward agreement, then attacks each proof with pattern-specific checks like "does this prove the Riemann hypothesis as a special case" and small counterexample testing. No computation tools allowed, pure reasoning only. It will say "no confident solution" rather than confidently hallucinate a proof. Best for competition proofs where wrong-and-confident is worse than honest abstention. For full problem sets, run one workflow per problem in parallel with labels to track which result belongs to which problem.

install copies · 90 days. Times the install command was copied on this site in the last 90 days, bots filtered.: 5
skills.sh installs: 4.7K
GitHub stars: 46.7K
last updated: Aug 21, 2026

View on GitHub

Install

$ npx -y skills add anthropics/claude-plugins-official --skill math-olympiad --agent claude-code

Installs into .claude/skills of the current project.

Files

Math Olympiad Solver

The five things that change outcomes

Strip thinking before verifying — a verifier that sees the reasoning is biased toward agreement. Fresh context, cleaned proof only.
"Does this prove RH?" — if your theorem's specialization to ζ is a famous open problem, you have a gap. Most reliable red flag.
Short proof → extract the general lemma — try 2×2 counterexamples. If general form is false, find what's special about THIS instance.
Same gap twice → step back — the case split may be obscuring a unified argument. Three lines sometimes does what twelve pages couldn't.
Say "no confident solution" — wrong-and-confident is worse than honest abstain.

Tool policy: Solvers and verifiers use THINKING ONLY in the tight-budget workflow. Competition math is reasoning. Computation is for deep mode (§6c), and even then bounded — a recurrence that's doubly-exponential can't be computed past n~30, work mod 2^m instead.