Math Modeling Skill

Install

Works in

• Claude Code
• Codex
• Cursor
• Koda
• Coddy

npx skillfoxx add skills/math-modeling-skill

What it does

The skill splits the task into three roles: analysis and model design, code implementation and execution, and paper writing. In the first stage the agent reads the problem and attachments, breaks down subproblems, picks a model and a validation plan. In the second it writes and runs Python or MATLAB code, produces result tables and figures, and keeps a reproducibility record with seeds, input hashes and dependency versions. In the third it builds the paper from real results, in Word by default and additionally in LaTeX with real PDF compilation on explicit request. Read-only subagents run quality gates at key points, and the algorithm references cover optimization, forecasting, evaluation, graph theory, statistics and machine learning.

Who it is for. For participants of math modeling contests (CUMCM, MCM/ICM and similar) and for general modeling projects.

Example request

Run this problem through the math modeling skill and assemble the paper in Word

Limitations

References and instructions are mostly in Chinese. The finished paper is only a reference; structure and format must be aligned with the official rules and template of the specific contest year. Computation and figures need Python or MATLAB with the required libraries, and LaTeX needs an installed TeX toolchain. Literature search calls OpenAlex and AnySearch, AnySearch may need the ANYSEARCH_API_KEY key, and that step needs network access.

How to disable. Remove the math-modeling folder from your agent's skills directory. If installed as a dsh preset, delete the preset folder.

README in short

The README describes a three-stage workflow: analysis and model choice, code implementation, and paper writing, with the option to run the whole task or a single stage. It lists bilingual Python and MATLAB implementation, publication-quality figures with self-checks, reproducible runs, dual-engine literature search, and Word or LaTeX paper generation with full quality gates. Separate subagents act as reviewers at key points, with only the quality check enabled by default. The repository also ships a preset for DeepSeek Harness. Installation is via git clone or npx skills add.