math-olympiad

Solves competition math problems with parallel solvers and adversarial proof verification.

16|1|Updated Mar 20, 2026
One-click install
npx skills add https://github.com/SocialGouv/iterion --skill math-olympiad-socialgouv
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: math-olympiad
Source: https://github.com/SocialGouv/iterion/tree/main/vendor/github.com/SocialGouv/claw-code-go/internal/tools/bundled_skills/math-olympiad/skills/math-olympiad
Command: npx skills add https://github.com/SocialGouv/iterion --skill math-olympiad-socialgouv

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? Competition math proofs (IMO, Putnam, USAMO, AIME) often contain subtle errors that self-verification misses, and confident but wrong answers are worse than honest abstention. This Skill orchestrates parallel solving agents, fresh-context adversarial verifiers, and calibrated abstention to produce proofs that survive hostile review. ## Core Features & Use Cases - Parallel candidate generation: Launches 8-12 solver agents with varied starting angles, each internally iterating through solve, self-improve, and self-verify rounds. - Adversarial verification: Fresh-context verifiers attack cleaned proofs using 12 specific failure patterns (tautological reductions, hypothesis mismatches, one-line lemmas) rather than generic logic checks, with asymmetric 4-confirm/2-refute voting. - Calibrated abstention and deep mode: Says "no confident solution" with partial results instead of bluffing, and escalates abstained problems to a focused deep-mode agent with bounded local computation. - Use Case: Given an IMO problem statement, the Skill fans out solvers, strips reasoning traces, runs pattern-armed verifiers, revises holes up to 3 cycles, and compiles the verified proof into a LaTeX PDF. ## Quick Start Ask the assistant to solve this IMO problem and verify the proof adversarially before presenting it as a PDF.

Frequently Asked Questions about math-olympiad

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I verify an olympiad math proof for hidden errors?▼

Submit the problem statement and proof for adversarial verification. A fresh-context verifier attacks the argument using 12 specific failure patterns such as tautological reductions, unsatisfied theorem hypotheses, and one-line lemmas, returning a HOLDS, HOLE FOUND, or UNCLEAR verdict.

How does this solve IMO or Putnam problems step by step?▼

It launches 8-12 parallel solver agents with different starting angles, each iterating through solve, self-improve, and self-verify rounds. The top candidate then passes up to 5 independent adversarial verifiers with a 4-confirm/2-refute asymmetric vote before revision or presentation.

Can it produce a PDF of the verified proof?▼

Yes, if pdflatex or xelatex is installed. After verification passes, a presentation agent rewrites the proof for clarity using a LaTeX template, and the compile_pdf.sh script wraps it in a minimal preamble and compiles it to PDF.

What happens when the solver cannot find a complete proof?▼

After 3 failed revise cycles it abstains with a calibrated verdict listing what was tried, what partial results survived verification, and where the proof breaks. A deep-mode agent with bounded local computation then gets one focused attempt before the abstention is finalized.

Does it use web search or computation to solve problems?▼

No. Solvers and verifiers use pure reasoning only, and web access is explicitly forbidden including in deep mode. Deep mode permits only bounded local computation such as modular arithmetic and small-case enumeration with a 60-second limit.

When should I use majority voting instead of full proof verification?▼

Use majority voting for numeric-answer problems like AIME where no proof is required. Run 5-7 solvers with varied approaches and take the majority answer; if there is no majority, verify the top two candidates by substitution.