Marketing OS: A Claude Code Skill That Runs an Entire Marketing Function

Yuzzyuk/marketing-os ships 14 marketing modules inside a single Claude Code skill: audits, GEO, copywriting, hooks, paid ads, email, social, launches, positioning, competitive research, app store optimization, and analytics. It has grown fast since release, sitting at 209 stars and 48 forks as of this week, up sharply from where it stood a few days earlier. The pitch is direct: install one skill instead of the four or five separate AI tools a solo operator or small marketing team would otherwise stitch together to cover the same ground.

What's actually in the 14 modules

The module list reads like a marketing department's job list rather than a features page. The audit module runs a weighted 0 to 100 score against a website or funnel with prioritized fixes attached. GEO checks how citable a page is to ChatGPT, Perplexity, and AI Overviews, the newer discipline of showing up inside an AI answer rather than a search results page. Copy generates 15 to 20 variants of a piece of copy, runs them through an expert-panel scoring pass, and strips out what the skill's own documentation calls "slop," its term for generic AI-sounding phrasing. Hooks runs an 18-tactic engine against a hook matrix and a diagnostic funnel, aimed at the same problem catalog-reel writers solve by hand: which opening line actually earns attention. The remaining modules cover paid-ad fatigue diagnosis, full email sequences, social posts, launch playbooks including Product Hunt, positioning and pricing strategy, competitor teardowns built from public sources, app store metadata and screenshot sequencing, and an analytics module the author describes as producing "honest data reads" with proper test design rather than declared wins on small sample sizes.

Two smaller pieces sit underneath the twelve visible modules: a catalogued list of AI writing tells the skill checks all output against, and a shared set of scoring rubrics and engine playbooks the individual modules draw from so the scoring logic stays consistent across audits, copy, and competitive work.

How the installation and loading actually work

Installing it in Claude Code is a two-line command: add the marketplace, then install the skill. It also installs as a zip upload in Claude's web or desktop app with code execution enabled, or as a manual copy into a skills directory for Codex, Cursor, or other agents that support the same skill format. The design principle worth knowing before you install it is progressive disclosure: the router loads first at roughly 2,000 tokens, and each of the 14 modules only loads when a task actually calls for it, so having the full department installed does not mean paying its full context cost on every request.

The other design choice that matters in practice is what the skill calls a brand-context file. Before using any module, you copy a template, fill in details about your brand, and drop it in your project root. Skip that step and the documentation is upfront about the tradeoff: output stays competent but interchangeable, which the author frames as the one failure mode marketing work can't afford. That's a real dependency, not boilerplate. A generic audit or generic copy from a tool that promises 14 specialized functions defeats the point of installing it.

What using it actually looks like

The skill is pure markdown, no executable code, which the author positions as a trust decision, there's nothing to review before you run it beyond reading the instructions themselves. It's also built to fan work out across subagents when the host supports it, so a multi-dimensional task like an audit or a competitive teardown can run its component checks in parallel rather than sequentially. One feature extends past the skill itself: if an ad-generation MCP server like Arcads is connected, the paid-ads module can go from diagnosis to a written brief to produced ad assets inside the same session, instead of handing off a brief for someone else to execute.

Where it's thin

The repository's commit history is a single commit as of this week, meaning the entire tool arrived at once rather than being built and refined in public over time, which is worth knowing if you weight a tool's maturity by how long it's survived real use rather than by its star count alone. The 200 stars and 46 forks reflect real interest, but interest is not the same as a track record across hundreds of live marketing workflows. It's MIT licensed and free, so the downside of trying it is mostly the setup time to write a real brand-context file rather than any lock-in risk. For a solo operator or a two- or three-person marketing team who wants one system instead of five subscriptions, and who's willing to fill in the brand context properly, that's a low-risk way to find out if it earns a permanent spot in the stack. Anyone who only wants better ad copy without the rest of the surface area is better served by a narrower tool like ai-copywriter, built for exactly that one job.

github.com/Yuzzyuk/marketing-os