Directory / math-olympiad
math-olympiad
StrongSolves competition math (IMO, Putnam, USAMO, AIME) with adversarial verification: a second fresh-context agent attacks the proof before the answer ships.
Verdict
Math-olympiad solves a real failure mode in AI-assisted proofs: self-verification is biased toward agreement because the verifier sees its own reasoning chain. This skill runs the verifier in a fresh context with only the cleaned proof — no access to the reasoning that produced it — and specifically attacks known gap patterns (Riemann Hypothesis specialization, 2×2 counterexamples on general lemmas, same-gap-twice signals). The "say no confident solution rather than bluff" policy is the right calibration for competition math. No network calls; LaTeX compilation is opt-in when the local binary is present.
Install
Audit receipts
Every listing is verified against a specific commit. These are the receipts.
One skill with two optional shell scripts (check_latex.sh, compile_pdf.sh) for LaTeX-to-PDF rendering when LaTeX is installed. Core workflow is pure reasoning via the Claude agent system.