Marketing OS bundles fourteen marketing modules into one AI agent skill

Marketing OS is an MIT-licensed Claude skill that packs fourteen marketing modules into a single install. The repository, published on 17 August 2026 by GitHub user Yuzzyuk, has gained 453 stars in the two weeks since. It works inside Claude, Codex and Cursor.

The skill is pure Markdown with no executable code. A router module loads first, around 2,000 tokens, and each of the fourteen sub-modules is loaded only when a task needs it. The design lets a team keep all fourteen modules installed without paying for the tokens until they ask for a specific job.

What the fourteen modules cover

The list, from the project's own table: a website and funnel audit that returns a weighted score from 0 to 100 with a prioritised fix list, a generative engine optimisation module for citability in ChatGPT, Perplexity and Google AI Overviews, a copy module that generates 15 to 20 variants and grades them before showing the writer, an 18-tactic hook engine, paid-ads concept-fatigue diagnosis, email sequences for welcome, nurture and launch, social copy for LinkedIn and X, a launch plan covering Product Hunt, positioning and offer design, detailed competitor analyses from public sources, App Store Optimisation, honest analytics reviews, plus a catalogue of low-quality writing patterns that runs across every prose output.

The GEO module is the part most directly relevant to teams building software: it checks whether the site's pages carry the passage-level structure and named-entity signals that answer engines quote from.

The three principles it commits to

The project states three design commitments up front. The first is scoring on every output, so a marketer looking at an audit or an ad brief sees a heuristic number rather than a verdict. The second is artefacts rather than advice, so the modules generate the actual copy, the actual brief and the actual scoresheet rather than telling a marketer what to do. The third is an explicit "what I could not determine" section in every report, so a missing data source is visible instead of being quietly ignored.

If an ad-generation MCP server is connected, the paid-ads brief is passed to it and the skill produces the assets in the same session.

A working developer running Claude Code or Cursor is a plausible marketer for their own product, and the difficulty is usually the form of the ideas more than the ideas themselves: what does the hook look like, what does the landing page fail on, which competitor already uses the phrase you were going to use. A skill that turns each of those into a scored artefact runs in the same session as the code changes and does not need a separate account or a second subscription.

The GEO module in particular is the reason a technical audience should look. Answer engines now decide who gets cited, and a page that reads well to a human can still be invisible to Claude or Perplexity because it has no passage the model can quote. Auditing that from inside the coding agent is quicker than paying for a separate GEO tool.

The trade-off is that the whole thing is markdown prompts. The scores are heuristics stated as heuristics, and the useful output is whatever a specific project already had the raw material to produce. A brand with no clear positioning to start with does not get one from a skill.

Source

• Yuzzyuk/marketing-os on GitHub