Marketing Skills and the thousand-dollar teardown

Getting a senior product marketer to tear your pricing page apart, tell you why the demo form is leaking, and hand back a prioritized list of what to fix first costs $1,000 and about a week at Conversion Factory, which publishes its whole rate card rather than hiding it behind a discovery call. The version where somebody keeps doing that every month is $3,000 for advisory, $6,000 and up for an embedded team, $25,000 and up if the answer turns out to be a new website. The person who sells all of that has also published fifty of the checklists behind it on GitHub under an MIT license, and as of this morning about 49,000 people have starred them.

The repo is coreyhaines31/marketingskills, and it picked up roughly 2,600 stars in the last week. Corey Haines runs the agency and wrote the skills.

What is actually in it

Fifty markdown files, each one a set of instructions your coding agent reads before it starts working on a marketing task. Not software. There is nothing to run, nothing to host, no server, no database. The whole repository is text, and the LICENSE file is a plain MIT grant, copyright Corey Haines, which means you can use it commercially, fork it, edit it, and never tell anyone.

The coverage is wider than the marketing label suggests. There is cro for pages and forms, signup and onboarding for the parts after someone says yes, pricing, churn-prevention, ads, ab-testing, analytics, cold-email, emails, seo-audit, ai-seo for getting cited by the assistants people now ask instead of Google. And then there is a sales and revenue operations pair, revops and sales-enablement, which is where this stops being a marketing repo and starts being interesting to anyone who has ever filed a ticket and waited.

Installation is one command if you already have a coding agent open. npx skills add coreyhaines31/marketingskills pulls the whole set, or you add the repo as a plugin marketplace inside Claude Code and install it from there. The README warns about one real trap: if you ask your agent to run the install for itself, the command runs without prompting and may drop the files into a shared directory Claude Code does not read, so you pass the agent name explicitly and it works. That is roughly the level of technical difficulty involved. If you have installed a browser extension, you can do this.

The architecture is the part worth understanding. One skill, product-marketing, is the foundation. It writes a file to your project called product-marketing.md that holds who you are, who you sell to, what you actually do, and how you say it. Every other skill checks for that file before it does anything else. Ask for a landing page teardown and the agent reads your positioning first, then applies the CRO framework to your page with your audience in mind, rather than giving you the generic advice that makes most AI marketing output worthless.

The one a RevOps person should read first

Open revops and you are looking at a document somebody would normally charge you to produce.

It defines the lifecycle stages, subscriber through evangelist, with entry criteria, exit criteria, and a named owner for each. It defines an MQL as requiring both fit and engagement and explains why either one alone is garbage: a perfect-fit company that never engages is not a lead, and a student downloading every ebook is not one either. It sets the handoff SLA at four business hours to first contact and 48 hours to qualify or reject, with rejected leads going to a recycling nurture with a reason code attached rather than into a hole. It gives you the scoring model construction in six steps, including the one everybody skips, which is testing the model against your own closed-won history before you turn it on. It names the four routing methods and tells you to always set a fallback owner, because unassigned leads go cold faster than anything else in the funnel.

Then it hands over the benchmark table. Lead to MQL at 5 to 15 percent. MQL to SQL at 30 to 50 percent. SQL to opportunity at 50 to 70. Pipeline coverage at three to four times quota. LTV to CAC above three to one.

None of that is secret. All of it is the sort of thing a RevOps manager either learned over six years or is currently guessing at, and it is now sitting in a file on their laptop that an agent will apply to their actual CRM configuration on request. That is the engineering ticket that does not get filed. That is the six-week wait that does not happen. For the practitioner reader, the displaced cost was never an invoice. It was the queue.

What it actually costs

Free is doing work in that first sentence, so here is the rest of it.

You need an agent to run the skills in. Claude Code is included in the Pro plan at $20 a month billed monthly, $17 if you pay for the year. If you use it hard enough to hit limits, Max starts at $100. The skills also work in Codex, Cursor and Windsurf, which have their own bills. So the floor is twenty dollars a month for a tool most people reading this are already paying for, and the honest ceiling is a hundred if this becomes how you work.

Setup is twenty minutes for the install and the better part of a day for the thing nobody warns you about, which is writing product-marketing.md honestly. Positioning, ICP, the actual differentiation. If you already have that written down, you are fine. If you do not, you are about to discover that you do not, at which point you are staring at the exact deliverable the $1,000 audit exists to produce. The free thing leans on the expensive thing. That is not a trick, it is just true, and it is the single biggest predictor of whether these skills work for you or produce confident nonsense.

There is also a version tax. The library went to 2.0 recently and renamed seventeen skills, merging what used to be separate page and form CRO skills into one. The upgrade does not clean up after itself. The new skills install alongside the old ones and you get both sets in your directory until you manually delete nineteen stale folders, which the README lists for you. Anyone who has managed software knows what that means: at some point the same thing will happen again, and it will happen on a morning when you have other problems.

The honest take

Three things, and none of them are reasons to skip it.

The first is that a skill is a checklist, not a consultant. Read the CRO one closely and it asks you five questions at the end: what is your current conversion rate, where is traffic coming from, what does the flow look like after this page, do you have heatmaps or session recordings, what have you already tried. If you cannot answer those, the agent will produce a teardown anyway, and it will be a plausible-sounding guess with your logo on it. The framework was never the scarce part of that engagement. Getting the data, and being told which of your beliefs about your own funnel are wrong, was the scarce part. A file on your disk will do the first half of that job well and cannot do the second half at all.

The second is that the agency does not just tell you. It ships. The $6,000-a-month embedded team writes the page, designs it, builds it, and runs the test. The skill hands you a document called Quick Wins, High-Impact Changes, Test Ideas, and Copy Alternatives, and then somebody at your company still has to go do all of it. If the reason you were going to hire an agency was capacity rather than expertise, this repo does not solve your problem. It moves the bottleneck from knowing to doing, which is progress only if doing was the part you had covered.

The third one is the one nobody installing this has read, and it is genuinely new. The tools registry underneath the skills, the file your agent consults when a skill needs to name an actual product, is sponsored. There is a published rate card. Two hundred dollars a month gets a logo. Five hundred a month, six month minimum, gets your tool a dedicated integration guide inside the repository. Fifteen hundred a month buys the only such slot in your category, and twenty three categories are currently open, from CRM to cold outbound to session replay. Two tools hold verified partner slots today, Converly and Ploy, and both are marked as such wherever they appear.

I want to be fair about this, because the disclosure here is better than the industry deserves. The rules document is explicit that money buys presence and never a recommendation, that partner entries are additive and never demote a competitor, that the badge means paid and vetted rather than best, and that the disclosure header is baked into the file itself so it survives being forked and re-indexed. One category is held off the market entirely because the maintainer owns a tool in it, and the repo says so in writing. Most sponsored content on the internet does not come with a governance doc.

But look at what the sponsorship page is actually selling, in its own words. It pitches placement inside the workflow rather than a banner beside it, and it argues that sponsorships depending on humans visiting websites are losing reach to AI assistants, while a library file installed into an agent travels with the workflow wherever it goes. That is correct, and it is a description of a brand new advertising surface: the files on your machine that your agent reads before it answers you. Nobody is hiding it. Almost nobody installing fifty skills in one command is reading it either.

And all of it sits on top of a funnel. Every outbound link in that README carries a referral tag. The paths out lead to the agency, a training program, a newsletter, and a separate product that packages these same skills into an autonomous marketing agent. That is not a scandal, it is a business model, and it is also the reason the skills are as good as they are. Somebody with a $3,000-a-month advisory practice has a strong incentive to make the free version genuinely useful, because the free version is the demo. Install it with your eyes open about what it is the free tier of.

Maintenance looks healthy for now. Just under 500 commits, and the most recent one last Friday updated the AI search skill for a format change in how ChatGPT presents answers. That update cadence is the tell. These files are not evergreen. They are tracking platforms that move monthly, and an SEO playbook that stopped being maintained eighteen months ago is worse than no playbook, because it will be confidently wrong in a way you cannot detect from inside the agent. MIT means you can fork it and keep it alive yourself. Whether you would is a different question.

One thing that is straightforwardly good: nothing phones home. The skills are text files, the work happens inside whatever agent you already trust with your other work, and your customer list and pricing and CRM export go wherever your existing plan sends them and nowhere else. Your data governance question here is the same one you already answered when you installed the agent.

Marketing advice has been free for twenty years, and none of it ever did the work. The change worth noticing is not that fifty playbooks landed on GitHub. It is that the playbooks now sit next to a machine willing to act on them, and twenty three categories of vendor have already figured out what that file is worth.