Marketing OS

One skill. Fourteen modules. The full surface a marketer touches.

Built by tearing down the most-starred marketing skill repos on GitHub — from 44K-star collections to 100-star craft pieces — keeping the structures that worked, and adding the three things every one of them lacked: scores on everything, artifacts instead of advice, and an explicit "what I couldn't determine" in every report.

What's inside

Module Covers
audit Website/funnel audit → weighted 0-100 + prioritized fixes
geo AI-search citability (ChatGPT, Perplexity, AI Overviews)
copy Generate 15-20 variants → expert-panel score → de-slop
hooks 18-tactic hook engine → hook matrix → diagnostic funnel
paid-ads Concept-level fatigue diagnosis → ranked production brief
email Welcome/nurture/launch sequences, written in full
social LinkedIn/X posts that survive the feed
launch Launch playbook incl. Product Hunt
positioning Positioning, offer design, pricing strategy
competitive Competitor teardowns from public sources
app-store ASO diagnosis + metadata + screenshot sequences
analytics Honest data reads, proper test design
+ slop-patterns The AI-tell catalogue, run on all prose
+ rubrics/specs Scoring bands, engine playbooks, store field rules

Ads briefs don't stop at paper: if an ad-generation MCP is connected (e.g. Arcads), the skill generates the briefed assets directly — diagnose → brief → produce, in one session.

Install

Claude Code (one command):

/plugin marketplace add Yuzzyuk/marketing-os
/plugin install marketing-os@marketing-os

Claude (web/desktop): Settings → Skills → + → upload marketing-os.zip from the latest release. Requires code execution enabled.

Manual / Codex / Cursor / other agents: copy skills/marketing-os/ into your agent's skills directory (e.g. ~/.claude/skills/marketing-os/).

First five minutes

Copy brand-context.template.md → brand-context.md, fill it in, drop it in your project root or .claude/. Every module reads it. Skip this and the output is competent and interchangeable — the one thing marketing can't be.

Design principles

  • Progressive disclosure: the router loads first (~2k tokens); each task loads only its module. Fourteen modules cost nothing until used.
  • Subagent-native: multi-dimensional work (audits, teardowns, variant generation) fans out in parallel when the host supports it.
  • No executable code. Pure markdown. Nothing to review before trusting it.
  • Honesty spine: no invented proof, no fake precision, no winners declared on ten conversions, and every score labeled as the heuristic it is.

MIT.