REA
Reverse engineer anything from your terminal or agent with one CLI and MCP server.
An entire marketing department as one Claude skill. 14 modules: audits scored 0-100, an 18-tactic hook engine, copy graded before you see it, ad diagnosis, GEO, email, launches, pricing. Works in Claude, Codex and Cursor.
如果你是作者,可把这段代码贴到 README。徽章会链接回本页。
[](https://mcp.popnx.com/en/skills/yuzzyuk-marketing-os)在同一工作中把 AI 连到服务的 MCP
Reverse engineer anything from your terminal or agent with one CLI and MCP server.
Codebase intelligence for TypeScript and JavaScript, exposed as structured MCP analysis tools.
MCP server for terminal commands, file operations, and process management
同类工作,排序与 Skill 列表相同
| 本项 marketing-os | rea | andrej-karpathy-skills | graphify | MoneyPrinterTurbo | |
|---|---|---|---|---|---|
| 作者 | Yuzzyuk | morluto | multica-ai | Graphify-Labs | harry0703 |
| Skill 数 | 1 | 1 | 1 | 1 | 1 |
| GitHub 星标 | 579 | 468,733 | 234,332 | 150,287 | 144,566 |
| 最近代码更新 | 2026年8月17日 | 2026年11月19日 | 2026年4月20日 | 2026年11月18日 | 2026年11月18日 |
| 许可证 | MIT | MIT | – | Apache-2.0 | MIT |
Yuzzyuk 提供的 1 个 Skill(指令集),AI 读取后照做 · 不是 MCP server
MCP = 手
把 AI 连到外部服务或数据,例如读取 Notion 或开发票。需要有程序或服务器在运行。
Skill = 手册
教 AI 怎么做某项工作的文本文件,不连接任何服务。
有些会重叠:少数 Skill 带有调用工具的脚本,少数 MCP 只提供知识。请从你想完成的工作出发。
Claude CodeCodexCursor
依据开发者 README 和 repo 中的文件夹
GitHub account Yuzzyuk (person), 1 years old
★ 579 stars · code last changed 2026年8月17日
No, plain text files the AI reads and follows
No scripts, so it does not go online by itself
No hooks or writes to machine settings found
Checked automatically from the repo files at commit bb67dff (read only, nothing was run). Not a safety guarantee: read it and decide for yourself
第一步:你用哪个 AI?
选择你用的应用,本页只显示该应用的安装方法
Uploading to claude.ai would fail: the description is over 200 characters (1022). Use "Let the AI install it" instead
复制这句话,粘贴到 Claude Code 的对话中
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
重新打开 Claude Code,输入 /marketing-os,或直接提出相关任务,Claude 会自动选用
Checked against official docs 2026年11月12日 · Source: code.claude.com
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
正在检查命令 · 通过后才会显示复制按钮
摘自开发者 README,会下载并运行开发者的代码
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
正在检查命令 · 通过后才会显示复制按钮
ChatGPT 桌面应用内置 Codex,Codex 从 ~/.agents/skills 文件夹读取 Skill
在 ChatGPT 桌面应用中打开 Codex,把这句话粘贴到对话中
The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
把 marketing-os 文件夹放到 ~/.agents/skills/marketing-os
~ 是你的主文件夹($HOME)
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
正在检查命令 · 通过后才会显示复制按钮
Checked against official docs 2026年11月15日 · Source: learn.chatgpt.com
不能使用你自己安装的 Skill,只能使用目录中插件自带的 Skill
想在 ChatGPT 中使用这个 Skill,请按上面的方法使用 ChatGPT 桌面应用
Checked against official docs 2026年11月15日 · Source: learn.chatgpt.com
复制这句话,粘贴到 Codex 的对话中 · 适用于 ChatGPT 桌面应用中的 Codex、Codex CLI 和 IDE 扩展
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
Codex 会自动发现新安装的 Skill;如果没有出现,请重启 Codex · 使用方法:在 ChatGPT 中输入 @ 并选择 Skill · 在 Codex CLI 或 IDE 扩展中运行 /skills,或输入 $ 加 Skill 名称
Checked against official docs 2026年11月15日 · Source: learn.chatgpt.com
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
正在检查命令 · 通过后才会显示复制按钮
复制这句话,粘贴到 Cursor 的对话中
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
在 Cursor 的 Agent 聊天中输入 / 并选择 marketing-os
Checked against official docs 2026年11月12日 · Source: cursor.com
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
正在检查命令 · 通过后才会显示复制按钮
适用于其他能读取 Skill 的应用。这句话列出了 Claude Code、Codex 和 Cursor 的路径,AI 会按你的应用选择 · Copy this sentence into the app’s chat
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
Checked against official docs 2026年11月12日 · Source: code.claude.com, learn.chatgpt.com, cursor.com
其他支持 Agent Skills 的应用(如 Gemini CLI、GitHub Copilot)使用不同文件夹,请查看该应用的指南,再把含 SKILL.md 的文件夹复制过去。
已于 2026年11月11日 对照 Claude Code、Codex 和 Cursor 官方指南核实 skill 文件夹
以下文字来自开发者的 repo,不是 PopNx MCP 的指令
显示前 8,000 个字符 · 在 GitHub 打开完整文件
---
name: marketing-os
description: A complete marketing department in one skill. Website and landing-page audits with weighted 0-100 scores, copywriting with panel scoring and AI-slop removal, an 18-tactic ad hook engine, GEO/AEO for getting cited by ChatGPT/Perplexity/AI Overviews, paid-ads creative diagnosis and production briefs, email sequences, LinkedIn/X writing, launch playbooks, positioning and offer design, competitor teardowns, app store optimization, honest analytics and test design. Use for ANY marketing task — audit, write, rewrite, diagnose, score, plan, launch, position, price, analyze — whenever the user mentions marketing, growth, conversion, copy, ads, hooks, CPM, ROAS, SEO, GEO, email, social, landing pages, funnels, launches, competitors, brand, positioning, pricing, or app stores, or says "my landing page sucks", "nobody's converting", "why are my CPMs up", "AI doesn't recommend us", "write me 20 hooks". Route via the table inside; fan out subagents for multi-dimensional work. Not for pure engineering, legal, or finance.
license: MIT
metadata:
author: marketing-os
version: "1.1"
---
# Marketing OS
One skill, fourteen modules, the full surface a working marketer touches. Built by tearing down the most-starred marketing skill repos on GitHub (44K-star collections down to 100-star craft pieces), keeping what worked, and fixing what every one of them got wrong.
Three rules hold across every module, because they are what the existing ecosystem uniformly lacks:
1. **Score everything.** Findings without a number are hard to act on and easy to argue with. Every audit ends in a weighted 0-100.
2. **Ship artifacts, not advice.** Write the replacement headline, the JSON-LD block, the email, the screenshot caption. "Your headline is vague" is worthless; the rewritten headline is the deliverable.
3. **State what you couldn't determine.** Every report ends with an explicit gaps section. A stated gap is credible; a silently filled one destroys the document.
## Setup — always do this first
**Read `brand-context.md`** if it exists (working directory, `.claude/`, or `.agents/`). It holds the product, ICP, positioning, proof, voice and constraints, and it changes nearly every judgement below. If absent: proceed, say the output is un-contextualised, and offer to generate the file from what you learn — `brand-context.template.md` in this skill is the blank.
**Identify the task type**, then open ONLY the module file(s) needed. Do not load all references — the routing below exists so you load ~1 file, not 13.
## Routing
| The user wants to... | Module | Also often needed |
|---|---|---|
| Audit/review/score/roast a website, landing page, funnel; "why isn't this converting" | `references/audit.md` | `audit-rubric.md` |
| Get cited by ChatGPT/Perplexity/AI Overviews; GEO, AEO, llms.txt, "AI doesn't recommend us" | `references/geo.md` | `geo-engines.md` |
| Write/rewrite anything: headlines, ads, pages, "make this punchier", "sounds AI-written" | `references/copy.md` | `copy-frameworks.md`, `slop-patterns.md` |
| Hooks for ads/video: "write me 20 hooks", thumbstop problems, hook batches per segment | `references/hooks.md` | `paid-ads.md`, `slop-patterns.md` |
| Diagnose paid ads: CPM up, ROAS down, fatigue, "what to test next", competitor's ads | `references/paid-ads.md` | `ads-diagnostics.md`, `hooks.md` |
| Email: welcome/nurture/launch sequences, subject lines, deliverability | `references/email.md` | `slop-patterns.md` |
| LinkedIn/X posts, personal brand, content that doesn't read as AI | `references/social.md` | `slop-patterns.md` |
| Launch a product, feature, or Product Hunt run | `references/launch.md` | `copy.md` |
| Positioning, category, offer design, pricing page strategy | `references/positioning.md` | — |
| Tear down a competitor: site, ads, positioning | `references/competitive.md` | `paid-ads.md` |
| App Store / Google Play: listing, screenshots, keywords, install rate | `references/app-store.md` | `store-specs.md` |
| Read performance data honestly, design a test, "did this work?" | `references/analytics.md` | — |
Multi-part requests load multiple modules. "Audit my site and rewrite the homepage" = `audit.md` then `copy.md`, carrying the audit findings forward rather than re-researching.
Every de-slop pass — copy, email, social — runs `slop-patterns.md` before delivery. No exceptions. A reader who clocks output as AI-written discounts the claim, not just the prose.
## Subagent fan-out
When subagents are available and the task is multi-dimensional, parallelize. This is the difference between a 15-minute audit and a 2-hour one.
**Full marketing audit** — spawn six, one per scoring dimension (messaging, conversion, search, competitive, trust, growth), each with the URL set and its slice of `audit-rubric.md`. Synthesize their sub-scores into the weighted total yourself; never delegate the synthesis, because the pattern across dimensions is the product.
**GEO audit** — spawn one per target question to query engines and record who gets cited, plus one for on-page extractability.
**Competitor teardown** — one per competitor.
**Copy generation** — one per angle family (problem/outcome/contrarian/identity/mechanism/offer) generating 3-4 variants each; you run the scoring panel on the merged set.
**Paid ads** — one per concept cluster for classification; you do the fatigue diagnosis on the merged concept table.
Rules for fan-out: give each subagent its exact reference slice and output schema; launch all in one turn; never let a subagent write the final report. If subagents are unavailable, work the dimensions sequentially in the order listed — the sequence is deliberate.
## Shared output standards
**Reports** follow this skeleton, adapted per module:
```
# [Deliverable] — [subject]
[date] · Score: XX/100 (where applicable) · Basis: [what you had access to]
## The one thing
[The pattern behind the findings, one paragraph. If they read nothing else.]
## Scorecard / Findings
## Do these first
[3-5 items, each with the actual fix written out, effort S/M/L, confidence H/M/L]
## What's already working
[Never skip. All-negative reports read as generated.]
## What I couldn't determine
```
Write reports to files (`[module]-[subject]-[date].md`), not into the chat — these are documents people forward.
**Copy deliverables** lead with the copy, reasoning after. Recommended option first, scored runners-up, then the single sharpest test contrast.
## Honesty spine — applies to every module
- **All scores are heuristics** from marketing judgement, not measured performance or anyone's internal ranking data. Say so in the report, every time.
- **Never invent proof.** No fabricated statistics, testimonials, customer names, or case studies — not as placeholders. Write `[NEED: figure]` and flag it. A plausible fake number in marketing copy is how a client ships a false-advertising claim.
- **Never declare winners on small samples.** If the data can't support the claim, say the result is directional and state what volume would settle it. `analytics.md` has the discipline.
- **Do not anchor** on scores, numbers, or conclusions the user supplies. Form an independent read first, then compare and say where you differ.
- **Say when the problem isn't the deliverable.** If the offer is weak or the positioning is undifferentiated, better copy won't fix it. One uncomfortable sentence saves a wasted quarter.
- **Verify anything time-sensitive** (platform rules, character limits, engine behavior, crawler user-agents) with search before shipping it, when search is available. These change on a scale of weeks.
- Never handle credentials or touch live campaigns/accounts. Diagnose and prescribe; the human executes in-platform.
## Module directory
```
references/
├── audit.md Website & funnel audit workflow
├── audit-rubric.md Scoring bands for the six dimensions
├── geo.md AI-search citability workflow
├── geo-en提问、分享技巧或报告无法使用。按“有帮助”数量排序。