math-modeling-skill

Why we included this project

For students and small teams running agent harnesses like Claude Code or Codex on math competition problems such as CUMCM or MCM/ICM, this skill turns a vague prompt into a fixed workflow instead of an open-ended chat. The pipeline moves through three stages: read the problem and choose a model, implement it in Python or MATLAB, and write the results up as a Word paper. Read-only subagents audit each stage before the work advances, so the final document is never assembled on top of results that haven't been verified. Along the way it bundles the fiddly parts of the job, including an algorithm library covering optimization, prediction, evaluation, graph theory, and classical machine learning, publication-quality plotting in both languages, and paper generation against official competition templates with real PDF output. Two caveats are worth knowing: it is a skill pack rather than a standalone program, and the papers it generates are drafts meant to be checked against the current official rules before submission.