Agent skill

AutoMCM Math Modeling Agent

by RealSeaberry in RealSeaberry/AutoMCM-Pro

Runs a staged workflow for math modeling contests such as CUMCM and MCM/ICM, with checkpoints, verified solver code and a LaTeX paper, on DeepSeek Harness.

MITAuto-check passedResearch & Science

SKILL.md written in Chinese; this summary is our English description.

Install AutoMCM Math Modeling Agent

skills CLI
$ npx skills add RealSeaberry/AutoMCM-Pro --skill auto-mcm -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install RealSeaberry/AutoMCM-Pro auto-mcm --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/RealSeaberry/AutoMCM-Pro.git skills-src && mkdir -p .claude/skills && cp -r skills-src/.dsh/skills/auto-mcm .claude/skills/auto-mcm && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
auto-mcm
GitHub stars
257
Token cost
~2.3k tokens
SKILL.md length
566 words
Files
1
Skills in repo
7
Repo updated
First seen
Licence
MIT

At a glance

Runs a staged workflow for math modeling contests such as CUMCM and MCM/ICM, with checkpoints, verified solver code and a LaTeX paper, on DeepSeek Harness.

  • Works in 2 steps: Step 2a 改成"一次性把信息塞进任务指令",而不是指望中途能问 → web_fetch 缺失时,正式认可的替代路径:bash + curl,附处理清单
  • Running an AI-assisted entry for a math modeling contest such as CUMCM or MCM/ICM
  • SKILL.md covers 【工具映射表】—— 全文遇到左列写法,按右列替换, 【唤醒协议】每次被调用时必须首先执行, 【Checkpoint 执行模板】 and 【流水线执行】, plus 5 more sections
  • Calls python, curl and npx; needs DEEPSEEK_API_KEY

What it does

This is the DeepSeek Harness binding of the AutoMCM-Pro protocol, covering both the Chinese CUMCM and the English MCM/ICM contests. It is not a separate protocol: the authoritative rules live in AutoMCM_SOP.md, and this file mainly maps Claude Code tool names to their DeepSeek Harness equivalents, such as subagent, ask_user_question, web_search, bash and the file tools.

Two modes are supported, one led by the AI and one led by the human's specification, with mandatory GitOps checkpoints, forced self-verification of all solver code before it enters the LaTeX paper, and human cross-validation at each pipeline stage. On first activation in a session the agent asks, never silently, whether to install the core Python modeling dependencies and the optional Lean 4 component, then checks pipeline state with scripts/pipeline_manager.py status before continuing or starting the first-run protocol. The skill text is mostly Chinese.

When your agent uses it

  • Running an AI-assisted entry for a math modeling contest such as CUMCM or MCM/ICM
  • Keeping solver code verified and checkpointed in Git throughout a modeling project
  • Resuming a modeling pipeline that already has saved state

Example prompts

  • “Start an AutoMCM run for this year's MCM problem in AI-led mode.”
  • “Resume the modeling pipeline and tell me which stage we are at.”
  • “Set up AutoMCM in manual mode so my own spec drives the modeling steps.”

Requirements

  • DeepSeek Harness with its tool bundles
  • Python modeling dependencies, installed through install.sh only with your consent

Workflow steps

2 steps, taken from the step headings in SKILL.md.

  1. Step 2a 改成"一次性把信息塞进任务指令",而不是指望中途能问
  2. web_fetch 缺失时,正式认可的替代路径:bash + curl,附处理清单

What it can do on your machine

Read from SKILL.md and the folder at commit 90c4727. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    Shell commands in SKILL.md call:

    • python
    • curl
    • npx
    • pnpm
    • bash

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md. Its commands use curl, npx and pnpm, which can reach the network depending on how they are called.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names these keys or tokens, usually read from environment variables:

    • DEEPSEEK_API_KEY

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

AutoMCM Math Modeling Agent loads about 2.3k tokens when it runs. Until then it costs about 94 tokens; SKILL.md has 566 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~94
When it runs · the whole SKILL.md, loaded when a task matches
~2.3k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from RealSeaberry/AutoMCM-Pro at commit 90c4727, republished under its MIT licence (© RealSeaberry). 566 words, ~2,291 tokens.

Download SKILL.mdSave it as .claude/skills/auto-mcm/SKILL.md (or your agent's skills folder).
name
auto-mcm
description
AutoMCM-Pro industrial-grade math modeling agent (DeepSeek Harness binding). Supports AP (AI-led) and Manual (human-spec-led) dual modes with mandatory GitOps checkpoints, forced self-verification of all solver code before LaTeX inclusion, and structured human cross-validation at each pipeline stage. Use for both CUMCM (Chinese) and MCM/ICM (English) competitions.
whenToUse
用户提供数学建模竞赛题目(CUMCM/MCM/ICM)并希望端到端自动/半自动完成建模、 编码验证、论文撰写时使用。

AutoMCM-Pro:DeepSeek Harness (dsh) 绑定

这是 AutoMCM-Pro 协议在 DeepSeek Harness 上的运行时绑定(Binding),不是另一套 独立协议。 行为规范的权威来源仍是仓库根目录的 AutoMCM_SOP.md(工具无关,原样 复用);可选探索层见 LOS_ALAMOS_DESIGN.md。Claude Code 上的对应绑定是 .claude/skills/auto-mcm/SKILL.md——两份文件描述同一套流程,只是把"怎么调用工具" 换成各自 runtime 的实际工具名,流程逻辑、Checkpoint 规则、质量门控、Prompt 内容 不应该在两份文件间产生分歧;如果你在这份文件里找不到某个步骤的细节,去 Claude Code 版本对照,工具调用按下表替换即可。


【工具映射表】—— 全文遇到左列写法,按右列替换

Claude Code 工具dsh 工具来源包
Agent(description, prompt)(一次性子 Agent,等待返回)subagent(description/prompt,不传 run_in_background 或传 false)dsh-tool-subagent
需要背景常驻、事后可追加指令的子 Agentsubagent 传 run_in_background: true(返回 started subagent <id>),之后用 send_message({subagent_id, message})追加、list_agents 查看状态、interrupt_agent({agent_id})中断——这个 boolean 参数是同步/常驻的唯一开关,不传或传 false 会同步执行、不产生可追踪 id,send_message 对不存在的 id 会报 Error: subagent "xxx" is unavailable(实测踩过这个坑)dsh-tool-subagent-control
AskUserQuestionask_user_questiondsh-tool-ask-user
WebSearchweb_searchdsh-tool-web
<!-- ⚠ 见下方【headless profile 的重要限制】:ask_user_question / web_fetch 在
     headless profile 下实测未挂载,上面两行是 web/交互式 profile 的理论映射 -->

| WebFetch | web_fetch | dsh-tool-web | | Bash | bash | dsh-tool-bash | | Read | read(图片用 read_image) | dsh-tool-fs | | Write | write | dsh-tool-fs | | Edit | edit | dsh-tool-fs | | Glob | glob | dsh-tool-fs-search | | Grep | grep | dsh-tool-fs-search | | TodoWrite | todo_write | dsh-tool-todo | | (载入其他 skill) | skill | dsh-tool-skill |

以上工具名取自 dsh 仓库 docs/tool-catalog.md 的实际工具目录(非猜测);具体部署 若启用了非默认 bundle(比如把 dsh-tool-bash 换成 dsh-tool-bash-persistent), 调用方式基本兼容,但建议先跑 dsh --profile <你的 profile> --dump-config 确认 实际挂载的工具名与本表一致。

continuable subagent 是 dsh 相对 Claude Code 绑定的一个真实增强:Los Alamos 设计里把"Alsos/Groves 常驻服务"标注为"仅 Claude Code 可选优化、不是可移植基线" (见 LOS_ALAMOS_INTEGRATION.md §2 的可选增强说明),是因为当时假设的运行时是 Claude Code;在 dsh 上,send_message/list_agents 是一等公民 API,如果你在 dsh 上落地 Los Alamos 探索层,可以把 Alsos 做成真正的常驻 continuable 子 Agent, 而不必退化成"每次查询都重新建立上下文"的文件轮询模式——但这是可选优化,本文件 的基线流程依然按文件+消息日志的方式描述,保证两个 runtime 的基线行为一致。


【唤醒协议】每次被调用时必须首先执行

每个新的智能体会话首次触发本 Skill 时,先分别询问是否安装核心 Python 建模依赖, 以及是否准备可选的 LeanGate / Lean 4。只询问,不预扫描、不静默安装;用户同意后 才调用 bash install.sh --with-python-deps 和/或 --with-lean,拒绝后本会话不再 重复询问。headless profile 无法等待回答时,显示问题并暂停,不得把沉默当作同意。

与 Claude Code 版本完全一致的三步判断,只是命令执行工具换成 bash:

bash: python scripts/pipeline_manager.py status
  • 退出码 0(已初始化)→ 读取当前阶段和状态,跳到【流水线执行】
  • 退出码非 0(未初始化)→ 执行下面的首次启动协议

首次启动协议(全程自然语言,用户零命令):

  1. 用 ask_user_question 询问:题目文件路径、附件数据位置、AP/MANUAL 模式(默认 AP)
  2. 用 read 读取题目(PDF 走 bash 调 pdfplumber/pypdf),自动推断竞赛类型/子问题数/数据情况
  3. 静默执行初始化:
    bash: python scripts/setup_workspace.py
    bash: python scripts/pipeline_manager.py init --mode {AP|MANUAL} --contest {CUMCM|MCM|ICM} --problems {N} --git
    bash: cp PROBLEM_PATH CUMCM_Workspace/data/
  4. 自然语言告知用户就绪状态,随后按【按需依赖处理】和【流水线执行】继续

**按需依赖处理:**不要在唤醒时扫描或安装。实际命令因缺少 Python 包、LaTeX 或字体 而失败时,报告缺失项;只有用户明确同意后才安装。


【Checkpoint 执行模板】

与 Claude Code 版本语义完全一致:

bash: python scripts/pipeline_manager.py request-review \
  --stage "<stage>" --summary "..." --results "..." --concerns "..." --next "<next-stage>"
  • AP 模式:request-review 会自动写入绑定阶段与轮次的自评批准;随后用 bash 执行 pipeline_manager.py advance <stage>,自然语言汇报成果并直接开始下一阶段
  • MANUAL 模式:展示汇报摘要,用 ask_user_question(而不是终端等待输入)明确 询问"继续,还是需要修改?",拿到回复后再决定 approve 还是 rework

Checkpoint LA 例外(仅 Los Alamos 模式,见 AutoMCM_SOP.md §9.2 第 4 条):即使 当前锁定 AP 模式,遇到两轨冲突/分歧熵超阈值时,必须用 ask_user_question 真正 等待人类输入,不得走 AP 自评自批分支——这条规则在两个 runtime 上都不可变通。


【流水线执行】

阶段定义、状态机、Checkpoint 编号(①~⑤)、质量门控(quality_gate.py)、 Los Alamos 探索层(路径 C)、图表风格规范(plot_style.py)、Andon 紧急停止 (pipeline_manager.py andon-pull/andon-clear/andon-status)、Go/No-Go 发射前检查 (quality_gate.py launch-check,final_compile 前强制)、Skunk Works 轻量模式 (pipeline_manager.py init --skunk-works)、Track2 的 RAND Delphi 多轮收敛 (adjudicate.py delphi-summary)、Kaizen 质量打磨循环 (pipeline_manager.py kaizen-assess/kaizen-round-start/kaizen-status)、工作日志 (worklog.py append/tail,单文件简体中文完整记录,唤醒协议 Step 0)、文献引用 真实性核验+共享池(cite_check.py register/verify/list/export-bibitems)、写作 风格打磨(style_check.py scan,latex_draft 固有规范非可选 addon)、官方格式 合规(quality_gate.py anon-check、ai_usage_doc.py generate/cite-format/ mcm-entry、compile_pdf.py 编译后页数提醒,AutoMCM_SOP.md §17)、画图前先查 领域惯例(AutoMCM_SOP.md §18) 全部内容与 Claude Code 版本一致,只替换工具调用:

  • problem_analysis → data_preprocessing → model_{n}_build/verify → sensitivity_analysis → latex_draft → final_compile,各阶段的具体工作内容、 验证清单、Checkpoint 触发时机,见 .claude/skills/auto-mcm/SKILL.md 对应小节 (标题相同,按【工具映射表】替换调用)
  • 路径 A(多子问题并行):原文里每个 Agent(description="问题N build+verify", prompt=<模板>) 调用,改为对每个子问题分别调用 subagent(prompt 字段内容 完全不变,模板本身是工具无关的自然语言指令);等待方式改为等待各 subagent 调用返回
  • 路径 B(顺序执行):无子 Agent 调用,直接照搬
  • 路径 C(Los Alamos 探索模式):Step 1(Alsos 普查)、Step 5(Division build+verify)、Step 6(Bletchley 红队)、Step 9(Track 2 评审小组)里的 Agent(...) 调用同样按上表替换为 subagent;scripts/los_alamos/*.py 系列 命令、quality_gate.py 新增门控、消息报文协议完全不变(这些是 bash 调用 的 Python 脚本,与 runtime 无关)。四套 Prompt 模板(Alsos 普查 / Division / Bletchley 红队 / Track2 评审)文字内容原样复用,不需要因为换了 runtime 而重写

Show full SKILL.md (198 more words)Show less

【Manual 模式附加规程】与【Rework 执行规程】

与 Claude Code 版本一致,唯一差异:人类确认环节一律用 ask_user_question 收集 自然语言回复,而不是等待终端输入。


【安全规程】

AutoMCM_SOP.md 里的 S1~S5 规则原样适用。S3(外部服务调用告知)在 dsh 上对应 web_search/web_fetch 调用前的关键词抽象化处理,规则不变。


【headless profile 的重要限制,实测确认,务必读】

这不只是 headless 的限制——dsh --profile web(交互式 Web UI)用 --dump-config 核对过完整插件树,同样没有 tool-ask-user(ask_user_question 的来源包),也没有任何 web-fetch provider 包。 web profile 里有 user-questions/ui-user-questions 两个插件,但那是给人类在 Web UI 里被问 问题用的(比如批准/确认对话框),没有包装成模型可调用的工具——模型自己 仍然没有"主动发起提问"这个选项。tui profile 这个版本没有预装。也就是说, 这个 dsh 版本(0.1.1-rc.2)目前没有任何一个默认 profile 能让模型真正调用 ask_user_question/web_fetch,不是"切去交互式就好",需要下面两条协议层面 的应对(完整判定依据见 DSH_INTEGRATION.md §3.1):

1. Step 2a 改成"一次性把信息塞进任务指令",而不是指望中途能问

在 headless 下,【首次启动协议】Step 2a 不得调用 ask_user_question——用户 发起 dsh --profile headless "<task>" 时,<task> 字符串本身就必须包含题目 路径、数据位置、AP/MANUAL 模式这三项信息(例如:"题目在 CUMCM_Workspace/data/problem.pdf,数据在同目录,用 AP 模式")。若用户第一句任务 指令没给全这三项,Agent 应该:

  • 在纯文字回复里列出缺的字段,不调用任何工具,让本次 headless 调用直接 结束(不要瞎猜、也不要卡住等一个不会来的回答);
  • 用户补全信息后再发起一次新的 dsh --profile headless "<补全后的task>"。

MANUAL 模式的逐阶段确认、Checkpoint LA 强制人类终审,这两类"必须等流程中途 出现的、无法提前塞进第一句话的信息",在这个 dsh 版本里目前没有任何默认 profile 能可靠支持(web profile 虽然是交互式产品,但模型侧同样没有 ask_user_question 工具)——如果确实需要在中途暂停等人类批准,现阶段只能靠 user-questions/ui-user-questions 这层人类 UI 基础设施在 Web UI 里手动介入 (比如批准/权限对话框),而不是指望模型主动发起结构化提问;或者等 dsh 后续 版本把 tool-ask-user 接进默认 bundle 后重新验证。

2. web_fetch 缺失时,正式认可的替代路径:bash + curl,附处理清单

不再是"模型自己想办法"的临时应急,而是协议认可的正式 fallback——遇到需要读取 某个具体网页全文时:

bash
curl -sL --compressed --max-time 30 -o /tmp/fetch_target.html "<URL>"
# --compressed 让 curl 自动处理 gzip/deflate/br 压缩,跳过手动 gzip -dc 这一步
file /tmp/fetch_target.html   # 确认拿到的是文本而不是仍被压缩/是二进制

拿到纯文本后再用 read/grep 处理。这条路径在这轮验证里实测跑通过(虽然当时 没加 --compressed,绕了一圈手动解压才成功——这里直接把踩过的坑写进正式做法, 下次不用重踩)。


【运行方式】(dsh 特有,Claude Code 绑定没有这部分)

sh
# 交互式(Web UI)
npx @deepseek-ai/dsh web

# 零命令/单次任务模式(对应 Claude Code 的 `claude --print`)
pnpm dsh --profile headless "读取 problem.pdf 并开始建模"

需要 Node.js ^22.19 或 >=24,以及 DEEPSEEK_API_KEY(或部署配置的其他模型 provider)。Python 依赖(pdfplumber/scipy/numpy/……)与 Claude Code 绑定 完全一样,只在用户对触发时的询问明确同意后调用安装器准备。

详细的架构对照与已知差异见 DSH_INTEGRATION.md。


【LeanGate:可选 Lean 形式化验证】

LeanGate 默认关闭;共享规范见 AutoMCM_SOP.md §19。DSH 用 bash 调用同一实现:

bash
python scripts/leangate.py doctor
python scripts/leangate.py --workspace CUMCM_Workspace init --policy selected
python scripts/leangate.py --workspace CUMCM_Workspace status --problem-n 1
python scripts/leangate.py --workspace CUMCM_Workspace verify --problem-n 1
python scripts/quality_gate.py formal --problem-n 1

LeanGate 启用后,离开 problem_analysis 前必须逐问题登记适用性并执行 leangate.py seal-assessment;安全边界、整数可行性、守恒、不变量、递推、收敛、 求解器证书或决策关键结论必须用 --risk-category 标注,自动成为 mandatory。

不得直接编辑权威状态或用 runtime 审查意见升级 FORMALLY_PROVED。输入变化后必须重新 审查、冻结和验证;Lean FAIL、UNKNOWN、STALE 或 Python FAIL 都是 NO-GO。 LeanGate 启用后的 model_N_verify 推进必须传入工作区内真实验证脚本: pipeline_manager.py advance model_N_verify --verify-script CUMCM_Workspace/src/verifications/verify_*.py。 required claim 还必须使用人类文件中的 [APPROVED LeanGate ...] 与 [REQUIRED LeanGate ...] 标记。

© RealSeaberry, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

Just SKILL.md in .dsh/skills/auto-mcm of RealSeaberry/AutoMCM-Pro.

Open the folder on GitHubat commit 90c4727

Compare with similar skills

AutoMCM Math Modeling Agent next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

AutoMCM Math Modeling Agent compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
AutoMCM Math Modeling Agent this skillRealSeaberry/AutoMCM-Pro257—~2.3kAutomated safety check: PassMIT
Backward Traceabilitylingzhi227/agent-research-skills390—~802Automated safety check: PassNone
LaTeX Research PostersK-Dense-AI/claude-scientific-writer2.4k12 repos~4.1kAutomated safety check: NotesMIT
Academic Paper Writing PipelineImbad0202/academic-research-skills51k—~16kAutomated safety check: PassCustom licence
Econ Writehanlulong/econ-writing-skill6512 repos~14kAutomated safety check: PassMIT
NSFC Grant Rationale Writerhuangwb8/ChineseResearchLaTeX2.9k—~945Automated safety check: PassMIT

Similar skills

  • Backward Traceability

    lingzhi227/agent-research-skills

    Makes each number in a LaTeX paper link back to the code line that produced it, using hypertarget and hyperlink tags and compile-time `\num` formulas.

    390 GitHub stars~802 tokensUpdated 7 mo ago
    Research & ScienceAuto-check passed
  • LaTeX Research Posters

    K-Dense-AI/claude-scientific-writer

    Builds conference-size scientific posters in LaTeX with beamerposter, tikzposter or baposter, including figure preparation, compilation and print preflight checks.

    2.4k GitHub starsUsed in 12 repos~4.1k tokens
    Documents & OfficeAuto-check: notes
  • Academic Paper Writing Pipeline

    Imbad0202/academic-research-skills

    Runs a 12-agent pipeline that plans, drafts, cites, reviews and formats academic papers, with modes for revision, rebuttals, abstracts and citation checks.

    51k GitHub stars~16k tokensUpdated yesterday
    Research & ScienceAuto-check passed
  • Econ Write

    hanlulong/econ-writing-skill

    Expert economics paper writing assistant synthesizing advice from 50+ top guides by Cochrane, McCloskey, Shapiro, Head, Bellemare, Goldin, Glaeser, Kremer, and other leading economists.

    651 GitHub starsUsed in 2 repos~14k tokens
    Research & ScienceAuto-check passed
  • NSFC Grant Rationale Writer

    huangwb8/ChineseResearchLaTeX

    Writes, restructures, reviews and polishes the rationale section of NSFC research grant applications in LaTeX, with backups and a diff before every write.

    2.9k GitHub stars~945 tokensUpdated 6 days ago
    Research & ScienceAuto-check passed
  • NSFC Budget Justification Writer

    huangwb8/ChineseResearchLaTeX

    Writes a submission-ready NSFC budget justification as a LaTeX project and renders budget.pdf from your grant proposal text and supporting materials.

    2.9k GitHub starsUsed in 1 repo~1.4k tokens
    Research & ScienceAuto-check passed

More from RealSeaberry/AutoMCM-Pro

  • AutoMCM-Pro for Codex CLI

    RealSeaberry/AutoMCM-Pro

    Runs a math modeling contest pipeline for CUMCM and MCM/ICM entries in Codex CLI, with git checkpoints, verified solver code and human review at each stage.

    257 GitHub stars~1.6k tokensUpdated 29 days ago
    Auto-check passed
  • AutoMCM-Pro for opencode

    RealSeaberry/AutoMCM-Pro

    The opencode binding of the AutoMCM-Pro math modeling pipeline for CUMCM and MCM/ICM contests, with tool mappings, install prompts and checkpointed runs.

    257 GitHub stars~1.2k tokensUpdated 29 days ago
    Auto-check passed
  • CUMCM Math Modeling Agent

    RealSeaberry/AutoMCM-Pro

    Drives an end-to-end workflow for the CUMCM math modeling contest: reads the problem and data, researches, codes and verifies models, then writes a LaTeX paper and PDF.

    257 GitHub stars~1.6k tokensUpdated 29 days ago
    Auto-check passed
  • Draw Image Diagrams

    RealSeaberry/AutoMCM-Pro

    Generates diagrams, flowcharts and conceptual illustrations with OpenAI's gpt-image models, while leaving data plots and result figures to real plotting code.

    257 GitHub stars~1.9k tokensUpdated 29 days ago
    Auto-check: notes
  • MCM/ICM Autonomous Modeling Agent

    RealSeaberry/AutoMCM-Pro

    Runs an MCM/ICM math modeling competition end to end: collects contest metadata, builds and verifies models and code, then generates an English LaTeX paper and any required memo.

    257 GitHub stars~2.8k tokensUpdated 29 days ago
    Auto-check passed
  • AutoMCM-Pro Math Modeling Agent

    RealSeaberry/AutoMCM-Pro

    Runs a math modeling competition entry end to end, in AI-led or human-led mode, with Git checkpoints and self-verified solver code before it enters the LaTeX paper.

    257 GitHub stars~8.5k tokensUpdated 29 days ago
    Auto-check passed

Questions about AutoMCM Math Modeling Agent

What does AutoMCM Math Modeling Agent do?

Runs a staged workflow for math modeling contests such as CUMCM and MCM/ICM, with checkpoints, verified solver code and a LaTeX paper, on DeepSeek Harness. This is the DeepSeek Harness binding of the AutoMCM-Pro protocol, covering both the Chinese CUMCM and the English MCM/ICM contests.md, and this file mainly maps Claude Code tool names to their DeepSeek Harness equivalents, such as subagent, ask_user_question, web_search, bash and the file tools.

When should I use AutoMCM Math Modeling Agent?

AutoMCM Math Modeling Agent fits situations like: running an AI-assisted entry for a math modeling contest such as CUMCM or MCM/ICM; keeping solver code verified and checkpointed in Git throughout a modeling project; resuming a modeling pipeline that already has saved state.

How do I install AutoMCM Math Modeling Agent in Claude Code?

Run `npx skills add RealSeaberry/AutoMCM-Pro --skill auto-mcm -a claude-code`. Or copy the skill folder (.dsh/skills/auto-mcm in RealSeaberry/AutoMCM-Pro) into .claude/skills/auto-mcm in your project. Claude Code loads it when a task matches its description.

How do I install AutoMCM Math Modeling Agent in Codex?

Run `npx skills add RealSeaberry/AutoMCM-Pro --skill auto-mcm -a codex`. Or copy the skill folder (.dsh/skills/auto-mcm in RealSeaberry/AutoMCM-Pro) into .agents/skills/auto-mcm in your project. Codex loads it when a task matches its description.

Can I use AutoMCM Math Modeling Agent in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add RealSeaberry/AutoMCM-Pro --skill auto-mcm -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/auto-mcm, .gemini/skills/auto-mcm, .github/skills/auto-mcm and .opencode/skills/auto-mcm in your project.

What does AutoMCM Math Modeling Agent need to run?

Going by SKILL.md and its folder, AutoMCM Math Modeling Agent needs the command-line tools its instructions call (python, curl, npx, pnpm and bash) and credentials named DEEPSEEK_API_KEY. Our summary lists: DeepSeek Harness with its tool bundles; Python modeling dependencies, installed through install.sh only with your consent.

Does AutoMCM Math Modeling Agent access the network?

SKILL.md contains no URLs. Its commands use curl and npx, which can reach the network depending on how they are called. This is read from the text; nothing was executed.

Is AutoMCM Math Modeling Agent safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does AutoMCM Math Modeling Agent use?

AutoMCM Math Modeling Agent is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does AutoMCM Math Modeling Agent use?

About 2.3k tokens (SKILL.md is roughly 9.2k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to AutoMCM Math Modeling Agent?

Skills that share tags, products or a category with AutoMCM Math Modeling Agent: Backward Traceability (lingzhi227/agent-research-skills, 390 stars), LaTeX Research Posters (K-Dense-AI/claude-scientific-writer, 2.4k stars), Academic Paper Writing Pipeline (Imbad0202/academic-research-skills, 51k stars) and Econ Write (hanlulong/econ-writing-skill, 651 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains AutoMCM Math Modeling Agent?

RealSeaberry (a GitHub user) maintains it in RealSeaberry/AutoMCM-Pro, which has 257 GitHub stars. The repository holds 7 skills in this directory. The repository was last updated on September 10, 2026.

Source: RealSeaberry/AutoMCM-Pro on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.