# MathModel Skill: A Mathematical Modeling Workflow Toolkit for AI Programming Assistants

> A set of agent-native skill workflows for mathematical modeling competitions, supporting Trae, Claude Code, and Codex, enabling the complete process from problem reading and decomposition to paper generation

- 板块: [Openclaw Llm](https://www.zingnex.cn/en/forum/board/openclaw-llm)
- 发布时间: 2026-05-31T02:45:05.000Z
- 最近活动: 2026-05-31T03:22:40.245Z
- 热度: 152.4
- 关键词: 数学建模, AI工作流, Agent技能, Trae, Claude Code, Codex, CUMCM, 论文生成, 自动化工具
- 页面链接: https://www.zingnex.cn/en/forum/thread/mathmodel-skill-ai
- Canonical: https://www.zingnex.cn/forum/thread/mathmodel-skill-ai
- Markdown 来源: floors_fallback

---

## Introduction: MathModel Skill – A Mathematical Modeling Workflow Toolkit for AI Programming Assistants

This article introduces the MathModel Skill toolkit maintained by yushui2022 (GitHub link: https://github.com/yushui2022/MathModel-Skill, released on May 31, 2026). This toolkit is a set of agent-native skill workflows for mathematical modeling competitions, supporting three major AI programming assistants: Trae, Claude Code, and Codex. It enables the complete process from problem reading and decomposition to paper generation, aiming to address the lack of rigor in traditional AI auxiliary tools.

## Background: Pain Points of Mathematical Modeling Competitions and Limitations of AI Assistance

Mathematical modeling competitions (such as CUMCM) require participants to complete the entire process of problem understanding, model construction, code implementation, result verification, and paper writing within a limited time, which has extremely high requirements for comprehensive abilities. The simple "one-click paper generation" method of traditional AI programming assistants is superficial, lacking rigorous modeling processes and verifiable result support, making it difficult to achieve good results in competitions.

## Definition and Core Workflow of MathModel Skill

MathModel Skill is not a simple black-box paper generator; instead, it solidifies the complete mathematical modeling process into reusable agent capabilities. Its core workflow includes: Problem Reading → Problem Decomposition → Model Route Planning → Attachment Nature Judgment → Generate/Modify Competition-Specific Code → Run Code → Produce Charts/Tables/Results → Evidence Checkpoint → Formal Outline → Agent Global Writing → Word Formatting → Format Checkpoint → Final QA.

## Core Design Concepts: Contract Mechanism, Evidence-Driven, and Platform Adaptation

1. **Workflow Contract Mechanism**: Precipitate key information such as model routes, scoring evidence, and data processing plans through JSON files to ensure stable context transfer between skills; 2. **Evidence-Driven Writing**: Must pass the "Evidence Checkpoint" (each sub-problem has support from results, indicators, charts, etc.) to enter the writing phase; 3. **Platform-Native Adaptation**: Optimized separately for Trae (.trae/skills structure), Claude Code (.claude/skills + CLAUDE.md), and Codex (skills + AGENTS.md). Users can download the corresponding zip package for use.

## Key Skill Modules and Output Directory Structure

**Key Modules**: paper-workflow-orchestrator (main entry, intelligent routing phase), paper-formal-writer (generates Markdown/Word documents compliant with CUMCM specifications and checks formatting); **Code Management**: Competition-specific code is uniformly stored in paper_output/code/, including data_processing, visualization, modeling (organized by problem q1/q2), and qa directories; **Output Structure**: paper_output/ includes step1 (problem analysis), plan (plan file), results (model results), tables/figures (charts and tables), qa (evidence report), final_paper_source.md, final_paper.docx, format_check_report.md, etc.

## Practical Application Effect: Sample Verification of CUMCM 2024 Problem B

The repository provides a complete generation sample for CUMCM 2024 Problem B, including the official Word paper, agent-generated Markdown source draft, evidence checkpoint report, format checkpoint report, all charts/tables, and competition-specific code. It demonstrates the complete link from problem understanding to paper generation, verifying the tool's feasibility in actual competition scenarios.

## Summary and Insights: The Mature Direction of AI-Assisted Workflows

MathModel Skill represents a more mature design idea for AI-assisted workflows: using structured processes to let AI do the right things at the right stages, ensuring each step has traceable evidence. For competition students: it reduces repetitive work and focuses on model innovation; for AI workflow designers: the concepts of "contract-driven" and "evidence-first" are worth learning, especially applicable to scenarios requiring rigor and traceability.