When AI Agents Learn Marketing: 43 Skills That Fit a CMO into Markdown

Forum topic · ✨步子哥 · 2026-10-02

Summary

coreyhaines31/marketingskills is an MIT-licensed GitHub repository containing 43 marketing skills written purely as Markdown files, with no executable code. Created by Corey Haines, a veteran SaaS marketer and founder of Conversion Factory, the library is designed for AI Agents such as Claude Code, Codex, and Cursor: each skill file specifies trigger conditions, workflows, dependencies on other skills, and reference data like industry benchmarks and checklists. Every skill loads a root node skill, product-marketing, before executing, mirroring how a human marketing director coordinates work. Skills also cross-reference each other (copywriting, CRO, A/B testing, SEO audit, cold email), forming a knowledge graph that Agents traverse at runtime. The project is free but funded through a Verified Partners model with disclosed tool integrations from Converly and Ploy. A companion product, Magister, positions an autonomous AI Agent as a full CMO that decides which marketing tasks to run. The post argues marketing may be the first domain where AI shifts from tool to autonomous role, because metrics like conversion rate and CAC provide natural reward signals, and predicts a wave of vertical skill libraries for legal, finance, and HR.

This post from zhichai.net analyzes coreyhaines31/marketingskills, a trending GitHub repository (~139 stars/day, MIT license) that packages 43 marketing skills as Markdown files for AI Agents. Its author, Corey Haines, is a veteran SaaS marketer and founder of Conversion Factory.

Key points

Skills as loadable knowledge. The repo contains no .py or .ts files — only .md files, each describing best practices for one marketing task (A/B testing, cold email, SEO audit, churn reduction). When loaded into Claude Code, Codex, or Cursor, Agents detect the task and execute per the file's framework. The post frames this as "a function written in natural language, with the AI Agent as its interpreter" — knowledge that can be modularized, versioned, and composed.
A dependency tree rooted in product-marketing. Every skill reads the product-marketing skill (which defines product, audience, positioning, differentiation) before executing — analogous to dependency injection (@Autowired in Spring, Context.Provider in React). It acts as the "marketing director" that guarantees context.
Cross-references form a runtime knowledge graph, e.g. copywriting ↔ CRO ↔ A/B testing, SEO-audit ↔ schema ↔ AI-SEO. Given "optimize my landing page," an Agent traverses copywriting → CRO → A/B testing → analytics → attribution. Unlike humans, Agents never forget or skip a dependency.
Monetization: Verified Partners. The library is free and MIT-licensed; vetted, disclosed integrations from Converly (conversion tracking) and Ploy (AI marketing platform) fund the work while core skill recommendations stay neutral. A third open-source monetization path: knowledge free, tool recommendations paid. Sustainability depends on community trust in the vetting and partner conversion rates.
Magister: the AI as CMO. The linked companion product Magister is an autonomous Agent that decides *which* marketing tasks to run — a leap from "AI as tool" to "AI as role." The post argues marketing reaches this before coding because feedback loops are shorter and more quantified: conversion rate, CTR, and CAC are natural reward signals.
Vertical skill libraries are rising. Alongside general frameworks like obra/superpowers and mattpocock/skills, marketingskills is deliberately vertical (marketing only), winning on depth like Veeva did versus horizontal Salesforce. The post predicts future legal-skills, finance-skills, and hr-skills repositories with full dependency trees.

Conclusion

The post closes with an engineering take on Feynman: if you can't write a marketing skill as a Markdown file an AI can execute, you don't truly understand it. Turning tacit knowledge explicit, structured, and executable is the third stage of knowledge engineering — and 43 running skills are a marketing team's collective wisdom, encoded, versioned, and invoked by an Agent.

Links: github.com/coreyhaines31/marketingskills · Magister · Conversion Factory

When AI Agents Learn Marketing: 43 Skills That Fit a CMO into Markdown

Summary

Key points

Conclusion

Tags