math-olympiad Claude Code Plugin
Claude Code PluginClaude CodeMathWeb Search, Research & LearningClaude Code Customization & WorkflowThe explanation below is AI-generated. Please verify it against the sources.
math-olympiad is a Claude plugin that solves competition math problems (IMO, Putnam, USAMO) using adversarial verification rather than self-verification. According to the README, self-verification is unreliable — it cites arXiv:2503.21934 ("Proof or Bluff"), which found that 85.7% self-verified IMO success drops to under 5% under human grading. The plugin instead uses a context-isolated verifier that sees only the clean proof (not the reasoning trace) and applies pattern-armed adversarial checks such as looking for accidental overreach or counterexamples. It is calibrated to abstain ("no confident solution") rather than produce a bluffed answer, and produces a clean LaTeX/PDF presentation after verification passes. The README reports validation results of 17/18 IMO+Putnam 2025 problems solved, 0 false positives, and 2 novel proofs found.
About the service
math-olympiad is a plugin/skill for Claude (installed via /plugin install math-olympiad@claude-plugins-official) that solves and verifies competition math proofs, auto-triggering on terms like "IMO", "Putnam", "olympiad", or "verify this proof".
What you can do with math-olympiad
- Ask Claude to solve a competition math problem, e.g. Solve this IMO problem: [statement]
- Have proofs checked by a context-isolated verifier that never sees the original reasoning trace
- Get proofs tested against pattern-armed adversarial checks (e.g. checking for accidental overreach or 2×2 counterexamples)
- Receive calibrated abstention ("no confident solution") instead of a bluffed proof when confidence is low
- Receive a clean LaTeX/PDF presentation of the proof after it passes verification
- Trigger the skill automatically using keywords like "IMO", "Putnam", "olympiad", or "verify this proof"