Auto Mcm logo

Auto Mcm

Community
RealSeaberry
auto-mcm

AutoMCM-Pro industrial-grade math modeling agent (Codex CLI 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.

Overview

PublisherRealSeaberry
RepositoryAutoMCM-Pro
Skill nameauto-mcm
Stars
258
Forks
16
Bundled files
Instructions only
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • Self-contained

    Everything the model needs lives in the instructions — no extra files to sync.

  • Open source

    Published by RealSeaberry on GitHub. Read the source before you install it.

Installation

Install the Auto Mcm AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/RealSeaberry/AutoMCM-Pro/tree/main/.agents/skills/auto-mcm
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/RealSeaberry/AutoMCM-Pro.git /tmp/AutoMCM-Pro
mkdir -p .claude/skills
cp -r /tmp/AutoMCM-Pro/.agents/skills/auto-mcm .claude/skills/auto-mcm
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Auto Mcm in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Auto Mcm on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Auto Mcm is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

AutoMCM-Pro:Codex CLI 绑定

这是 AutoMCM-Pro 协议在 OpenAI Codex CLI 上的运行时绑定(Binding),不是另一套 独立协议。 行为规范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 绑定是 .claude/skills/auto-mcm/SKILL.md,DeepSeek Harness 绑定是 .dsh/skills/auto-mcm/SKILL.md——三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,流程逻辑、Checkpoint 规则、质量门控、Prompt 内容 不应该在几份文件间产生分歧。完整背景见 CODEX_INTEGRATION.md。

【触发时的依赖安装询问】

每个新的 Codex 会话首次触发本 Skill 时,先询问用户是否安装核心 Python 建模依赖 (pdfplumber、scipy、numpy、matplotlib、pandas、openpyxl),以及是否准备可选的 LeanGate / Lean 4。不得预扫描或静默安装;用户同意后才调用 bash install.sh --with-python-deps 和/或 --with-lean。拒绝后同一会话不重复询问。 非交互执行无法取得回答时,显示问题并暂停,不得把沉默当作同意。


【与其他 runtime 最大的差异:Codex 只给模型一个"代码模式"工具,实测确认】

这不是源码调研的推测,是实跑一次任务、解压 ~/.codex/sessions/**/rollout-*.jsonl 核对过的:Codex 模型端 function-calling 只看得到一个叫 exec 的顶层工具,参数是一段 JavaScript 代码; tools.exec_command({cmd, workdir, ...}) 是这段 JS 代码内部能调用的异步 函数,不是模型能单独调用的顶层工具。也就是说,Bash/Read(文本)在 Codex 上都是:模型写一段 JS,里面调用 tools.exec_command({cmd: "..."}),再用 text(r.output) 把结果吐出来——不是直接的 {"tool": "exec_command", "cmd": "..."} 这种扁平 function call。

实测时甚至连读 SKILL.md 本身也是这样做的:模型用 tools.exec_command({cmd: "sed -n '1,240p' .agents/skills/auto-mcm/SKILL.md"}) 直接读档,没有专门的 skill 载入工具——这点跟 opencode/dsh 都不一样(它们有 独立的 skill 工具)。

这件事对 AutoMCM-Pro 反而是天然契合——本项目的流水线绝大部分工作本来就是 "用 bash 调 Python 脚本",很少直接依赖结构化文件工具,翻译成 Codex 的 JS-in-exec 模式没有额外损失。

【工具映射表】

Claude CodeCodex 工具说明
Bash顶层工具 exec(参数是一段 JS),内部调 tools.exec_command({cmd,...})✅ 实测确认(喚醒協議 Step 1 跑通)。万能工具:跑 Python 脚本、cat 读文件、rg/grep/find 搜索、curl 抓网页,本文件后续所有"执行 python scripts/..."都是这个模式,不是直接的扁平 function call
Read同上,exec 里的 JS 调 tools.exec_command({cmd: "cat ..."})✅ 实测确认(读 SKILL.md 本身就是这么做的)
Read(图片)view_image专用图片查看工具,读图表/PDF 截图用它,不要用 cat
Write/Editapply_patch结构化 diff 应用工具,写/改模型代码、.tex 文件优先用它(比 shell heredoc 更可靠地保留缩进/编码)
Glob/Grepexec_command 里 rg/find没有专用搜索工具
TodoWriteupdate_plan语义对应
AskUserQuestion请求用户输入的工具(模块 request_user_input,确切工具名未逐字核对,见下方核验清单)语义对应
Agent(description, prompt)多代理协同工具家族(模块 multi_agents/multi_agents_v2,确切工具名未逐字核对)Codex 有原生多代理能力,但这次没有把确切的工具调用 schema 读完,接入前务必先跑一个最小测试确认参数格式
WebSearch/WebFetch没有确认到的第一方工具见下方【已知能力缺口】

上表里 exec_command/apply_patch/view_image/update_plan 四个工具名是从 codex-rs/core/src/tools/handlers/ 源码里的 ToolName::plain(...) 字面量直接 核对过的(不是猜的);request_user_input、multi_agents/multi_agents_v2 是 从模块/文件命名推断,没有找到对应的字符串字面量,接入前务必先用一个最简单的 任务实测一次,确认真实暴露给模型的工具名。

【已知能力缺口:网络搜索】

Codex 核心代码里没找到独立的 web_search/web_fetch 工具(dsh/opencode 都有, Codex 没有内置)。看到的只有 MCP(Model Context Protocol)相关模块,猜测 Codex 把网络检索能力交给用户自行配置的 MCP server,而不是内置。这意味着:

  • AutoMCM_SOP.md 里"文献调研至少 5 篇"这条硬性要求,在 Codex 上落地前必须先 确认有可用的网络检索能力(配置一个 MCP 网络搜索 server,或者退化成 exec_command 里 curl 调用一个你自己有权限的搜索 API);
  • 若确认没有任何网络检索能力,不要静默跳过文献调研——按 AutoMCM_SOP.md 第 7 节"禁止静默跳过验证失败"的同一精神,在 Checkpoint①里如实告知用户"当前 环境无网络检索工具,本阶段建模假设未经文献交叉验证",让人类决定要不要补充 MCP 配置或人工提供文献。
  • 同样的缺口也影响 AutoMCM_SOP.md §18"画图前先查领域惯例"——判断某类问题的 常规可视化形式本质上也是一次网络检索。不要因为查不到就直接静默退化成随手 选个图表类型:先按上面同一套办法确认有没有可用的检索能力;确实没有的话, 在 thought_process.md 里如实记录"当前环境无网络检索工具,图表形式选择依据 建模者常识判断、未经领域惯例检索确认",而不是不留痕迹地直接画图。

【运行方式】(Codex 特有)

sh
codex           # 交互式
codex exec "task"   # 单次任务模式,类似 claude --print / dsh --profile headless

Skill 发现路径是 .agents/skills/auto-mcm/SKILL.md(仓库根向上找 .git,Codex 官方文档确认),显式调用用 $auto-mcm,或让模型按 description 隐式匹配。


【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,见 .claude/skills/auto-mcm/SKILL.md 对应小节(标题相同), 按上方【工具映射表】替换调用方式:

  • 子 Agent 调用:原文里每个 Agent(description=..., prompt=<模板>),改为 Codex 的多代理工具调用,prompt 字段内容完全不变(先按【已知能力缺口】小节的 提醒实测确认参数格式,再套用到 Los Alamos 路径 C 的四套 Prompt 模板)
  • 人类确认环节:ask_user_question 换成 Codex 的用户输入请求工具
  • 文件读写:模型代码/验证脚本/LaTeX 文件优先用 apply_patch,图表结果查看用 view_image
  • scripts/*.py 系列命令(pipeline_manager.py/quality_gate.py/ los_alamos/*.py/plot_style.py)完全不变,全部通过 exec_command 调用

【安全规程】

AutoMCM_SOP.md 的 S1~S5 原样适用。S3(外部服务调用告知)在 Codex 上尤其重要—— 既然网络检索要靠用户自配的 MCP server,调用前更要做好关键词抽象化,不要把题目 原文整段发给一个你不确定信任边界的外部 MCP server。


【LeanGate:可选 Lean 形式化验证门禁】

LeanGate 默认关闭;行为规范见 AutoMCM_SOP.md §19。Codex 通过 exec 内的 tools.exec_command 调用共享命令:

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 claims.json 或报告来制造 PASS。任何输入变化后重新执行语义审查、冻结 与验证;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。

Frequently asked questions

What does the Auto Mcm AI skill do?

AutoMCM-Pro industrial-grade math modeling agent (Codex CLI 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.

Why use Auto Mcm on TypingMind?

Because you install it once and use it with any model. Auto Mcm is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Auto Mcm in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/RealSeaberry/AutoMCM-Pro/tree/main/.agents/skills/auto-mcm. TypingMind reads its SKILL.md and installs it as a skill you can enable per chat.

Which AI models can use Auto Mcm?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Auto Mcm?

As many as you like. As long as a model supports skills, you can use Auto Mcm with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Auto Mcm AI skill free?

Yes. It is published on GitHub by RealSeaberry under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇