mathmodel-skill (rootkiller6788)
A skill for math modeling contests CUMCM, MCM/ICM and the Electrical Cup: 10 stages, three roles, five gates and a LaTeX paper
- 73 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The skill runs scripts and LaTeX compilation in the working directory and writes files to state/, results/ and paper_output/
- Generated formulas, code and references need manual checking before submission
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/mathmodel-skill-rootkiller6788Detects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The skill runs scripts and LaTeX compilation in the working directory and writes files to state/, results/ and paper_output/
- Generated formulas, code and references need manual checking before submission
This is third-party code. Review the repository files before installing.
What it does
The skill guides a team through a 72–96 hour math modeling contest, from picking a problem to the final check before submission. It supports CUMCM (72 hours, in Chinese), MCM/ICM (96 hours, in English) and the Electrical Cup (72 hours, in Chinese). Work splits into ten stages: start, topic choice, breakdown, model selection, foundation, a recursive loop over subproblems, robustness and sensitivity, evaluation and generalization, paper assembly and a final review. All state lives in one file, `state/decision_log.json`, and Codex and Claude Code can hand work to each other in the same workspace. Three roles (modeler, programmer, paper writer) have their own guides. At deliverable boundaries five independent gates, M1, P1, P2, W1 and W2, run in separate read-only sub-agents, and a failure means rework. The fast, standard and championship modes set the depth of feedback. The bundle includes material on seven algorithm classes, a tool that builds figures from data, publication search through OpenAlex and AnySearch, PDF build with `scripts/render_paper.py` and four LaTeX pre-checks: environment, references, keywords and PDF anonymity. The `render_ai_usage.py` script prepares the AI-use disclosure that CUMCM and MCM require. Start it with the phrase "开始建模" or by naming the skill.
Who it is for. Student teams in math modeling contests who work with Claude Code or Codex and want to go through every stage from the problem to a PDF within the rules.
Good fit when
- A team prepares for CUMCM and needs a 72-hour plan with roles and checks at every stage
- MCM/ICM needs a LaTeX paper within the page limit and a report on AI use
- The answer to the second question changed and the conclusions must be rebuilt so the third does not cite old results
- Before submission you need to check references, keywords and PDF anonymity
Not a fit when
- You need a finished paper to submit: the author calls generated papers reference material and not ready to submit as is
- You need a guarantee the model is correct or a prize forecast: the skill orders the process and does not rate chances
- The task is not math modeling and there is no contest: ordinary code and paper skills are enough
Example request
开始建模: use mathmodel-skill for CUMCM problem B with a 72-hour deadline, build the plan and prepare the paper template.Limitations
A generated paper serves as reference material. Structure and format must follow the current year rules, and the team checks all AI-created formulas, code, facts and references. Pass statistics are only a guide: for CUMCM they rely on 59 sample papers, and for MCM and the Electrical Cup there is no sample (`n=0`).
Contest rules change: the `competitions/<contest>/current_rules.md` files keep the last check date and must be verified against the official announcement before submission. The LaTeX engine depends on the contest: xelatex for Chinese, pdflatex for MCM. Word is available only on explicit request. The last commit is from September 8, 2026.
How to disable. Delete the `~/.claude/skills/mathmodel-skill` or `~/.agents/skills/mathmodel-skill` folder. Project working directories with `state/decision_log.json`, `paper_workspace/` and results stay and are removed by hand.
FAQ
How do I install mathmodel-skill?
Copy the whole repository folder into the skills directory: `~/.claude/skills/mathmodel-skill/` for Claude Code or `~/.agents/skills/mathmodel-skill/` for Codex. The project has no separate install command.
How do I start?
In Claude Code type "开始建模" or name the skill, for example "使用 mathmodel-skill 开始 CUMCM 建模". The skill collects five start fields: contest, problem number, team strengths, deadline and the path to the problem PDF.
What are the five gates M1, P1, P2, W1 and W2?
Independent read-only sub-agents accept deliverables at stage boundaries. If a gate fails, the deliverable goes back for rework.
Which contests are supported?
CUMCM (72 hours, Chinese), MCM/ICM (96 hours, English) and the Electrical Cup (72 hours, Chinese). Each has its own pack of rules, anti-patterns and writing hints.
Can I submit the generated paper?
The author treats such work as reference material that is not ready to submit. The format must be checked against the year rules, and the team verifies formulas, code, facts and references.
Discussion
No questions or reviews yet.
Related
SkillsA workflow for mathematical modeling contests: ten stages, ten gates and frozen results before the paper is written.
Two skills for the CUMCM and MCM/ICM math modeling contests: problem breakdown, model choice, code and the paper
A three-stage workflow for math modeling tasks: analysis and model choice, Python or MATLAB computation, and paper assembly in Word or LaTeX
A skill that carries problem B of the Chinese CUMCM mathematical modelling contest from reading the task to a LaTeX paper.
A skill set for math modeling contests: problem parsing, method selection, Python and MATLAB code, and paper writing
A skill that takes problem C of China's CUMCM mathematical modelling contest from data to a checked LaTeX paper.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| mathmodel-skill (rootkiller6788) (this card) | no | 73 | |
| Math Modeling Skill | no | 1,924 | a neighboring math modeling skill, while this one adds three roles, five independent gates and a shared state file |
| latex-paper-skills | no | 275 | helps format a paper in LaTeX, while this one runs the whole contest chain from problem to PDF |
README in short
The description explains why contests fail from lost context and out-of-sync results and introduces a shared state file and independent gates. It then gives the capability matrix, quick start, workspace layout and the LaTeX workflow.
README badge
Original SKILL.md
---
name: mathmodel-skill
description: CUMCM 国赛、MCM/ICM 美赛与电工杯数学建模竞赛的端到端协作工作流，以及一般数学建模项目。Use when a user works on a modeling competition (cumcm/mcm/icm/diangong/国赛/美赛/电工杯) or asks to run/review a modeling-competition paper from problem selection through modeling, solving, robustness, writing, compliance, and final submission review. 采用 10 阶段项目流程为脊柱、建模手/编程手/论文手三角色为阶段内执行引擎、5 个独立门禁质检，并集成双引擎论文搜索、出版级可视化、LaTeX/PDF 编译与预检、Excel/PDF 等工具。Do not trigger for generic model selection, ordinary data analysis, or non-competition paper review.
---
# mathmodel-skill — 数学建模三竞赛工作流（10 阶段 × 三角色 × 5 门禁）
本 Skill 把 72–96 小时的竞赛协作组织为可恢复、可检查、可交接的流程：**10 阶段项目时间线**（选题→拆题→建模→求解→稳健→写作→终审）为脊柱，**建模手 / 编程手 / 论文手**三角色负责阶段内的实际执行，**5 个独立门禁质检**（`M1/P1/P2/W1/W2`）在每个交付物边界验收，L1–L4 反馈层驱动打分与定向精修。用户回答关键问题，agent 维护状态与脚本。
- **10 阶段**：`references/stage_00..09_*.md`；状态统一存 `<cwd>/state/decision_log.json`。
- **三角色**：`references/roles/{建模手,编程手,论文手}/SKILL.md`，映射见 `references/角色-阶段映射.md`。
- **5 门禁**：`references/Subagent调度.md` 定义固定质检协议。
- **全工具**：`tools/{figure,paper_search,docx,xlsx,pdf,latex}` + `scripts/latex_check/` 预检。
- **生成论文仅供用户参考**，非可直接提交作品；结构格式以当届官方规则为准。
**使用前**：把 `使用指南.md` 复制到工作区 `<cwd>/`。
## 强制执行协议
用户点名本 Skill 或任务命中时，严格执行，不降级为建议：
1. 首次进度更新中回显：已激活本 Skill、`<cwd>`、当前阶段、目标竞赛与届次、计划读取的角色和工具入口。未确认的官方规则标为待核验。
2. 进入每个阶段前**实际读取**该阶段角色 `SKILL.md`；使用 PDF、Excel、论文搜索、DOCX 或 LaTeX 时再读取对应工具 `SKILL.md`。知道路径 ≠ 已执行。
3. 严格调用本 Skill 提供的脚本和模板；已有初始化、转换、编译或校验工具时，禁止手写替代实现。
4. 任何校验预警视为未完成；只有当届官方规则或用户明确要求允许偏离时，记录“规则来源、偏离项、理由”后继续。
5. 环境缺少引擎、搜索源或依赖时，报告阻塞并继续完成仍可验证的部分；禁止静默换工具或用较差产物冒充完成。
6. 按 `references/Subagent调度.md` 在 `M1/P1/P2/W1/W2` 节点立即派发独立质检，禁止等全流程结束后才首次派发；作者自检不能替代独立验收。除固定质检外不主动派发其他 Subagent。
7. 交付前运行当前阶段的全部完成门禁；任一命令未运行、退出码非零、门禁未通过时，不得声称“已完成”。
8. 最终回复列出实际读取的入口、实际运行的关键命令与退出码、门禁状态、核心质量指标和仍存在的阻塞；不得只说“已检查”。
## 三竞赛 × 三模式矩阵
| Competition | 时长 | 语言 | LaTeX 引擎 | 规则基线 | 经验数据 |
|---|---|---|---|---|---|
| cumcm | 72h | 中文 | xelatex / 原创 ctexart | CUMCM 2026 | 91 来源 / 59 可提取样本 |
| mcm | 96h | English | pdflatex / article | COMAP 2027 | `n=0`，无论文分位 |
| diangong | 72h | 中文 | xelatex / ctex | 官网 2026-03-21 | `n=0`，无论文分位 |
| Mode | 反馈层 | 用途 |
|---|---|---|
| fast | L1 单次 | 选题试跑 / sanity check |
| standard | L1 + L2 | 默认主流程 |
| championship | L1+L2+L3+L4 + red-team | 终稿前深度评审 |
模式按距 deadline 自动推荐（>60h standard；<6h 直进 Stage 9 championship）。
## 10 阶段索引
| # | 阶段 | 承担角色 | 门禁 | 时长 | 关键产物 |
|---|------|---------|------|------|---------|
| 0 | 团队启动 + 资料预扫 | — | — | 1h | 启动字段、环境、规则基线 |
| 1 | 选题（多题对比 → 1） | — | — | 2-4h | 选择理由、放弃项、task_type |
| 2 | 问题拆解 | 建模手 | — | 2-3h | 子问、变量、约束、依赖图 |
| 3 | 模型选型 | 建模手 | — | 2-4h | 候选模型、证据、淘汰理由 |
| 4 | Foundation | 建模手 | **M1** | 1h | `题目分析报告.md` + `术语表格.md` |
| 5 | 递归子问题循环 Q1..Qn | 编程手 | **P1 → P2** | 按题分配 | 代码、`results/`、`figures/`、`复现清单.json` |
| 6 | 稳健性 / 灵敏度 | 编程手 | — | 2-3h | 稳健区间、失败边界 |
| 7 | 模型评价 + 推广 | 建模手 | — | 1-2h | 优缺点、改进、迁移条件 |
| 8 | 论文装配与写作 | 论文手 | **W1 → W2** | 12-30h | `paper_workspace/*.md` → LaTeX/PDF |
| 9 | 提交前终审 + Panel | 论文手 | — | 2-6h | 最终 PDF、支持材料、Panel 记录 |