What is the caveman marketing skill?
Caveman marketing is an agent skill by lodar that rewrites a page, ad or signup flow so a stranger understands it in three seconds. It sets seven rules, from a picture in the top half to one filled button, and ships a script that counts words and buttons on each screen.
Caveman marketing is a skill for cutting any ad, landing page or signup down to what a stranger can grasp in three seconds. The SKILL.md frames it as a test: look for three seconds, cover the screen, then answer three questions.
The three questions are what this is, whether you want it, and where to tap. If one answer is missing, the screen fails. The skill adds hard numbers to that test and applies it to ads and funnels too.
What changes in your agent's output
- Each screen gets a picture in the top half, a headline of at most six words and fifteen words in total.
- There is exactly one filled button, and its label says what happens next.
- Signups show a step bar and ask one thing per screen, ending when the person gets the result.
- Industry words and hype terms such as world-class or end-to-end are replaced with plain, concrete language.
- The agent reports before and after counts: words, ways out and steps to the first win.
Who it is for
Best for
- Pages and ads that get views but few clicks, or that have no obvious next click.
- Founders redesigning a homepage for a phone first view, 390 by 844 pixels.
- Signup flows with long forms that could be split into one question per screen.
Not for
- Long persuasive copy written from scratch. The skill description says it is not for that.
- Pages where the list is the answer. App screens keep their rows and only the surrounding chrome is cut.
- Compliance-heavy pages that need extra text, though a required legal line is allowed as one short grey line.
Use it with UiChemy
After a page is simplified, you still need to ship it. UiChemy Composer writes HTML, CSS and JS inside Elementor or Gutenberg, so a one-button layout can become an editable WordPress page. Click Get Started on uichemy.com.
How to install Caveman Marketing Skill: One Picture, 15 Words, One Button
git clone https://github.com/lodar/caveman-marketing.gitThe checker needs Python 3. The author also supplies a test script that checks the checker itself.Clone the repository
Run the clone command above. The skill is the SKILL.md at the repository root, with scripts and examples folders beside it.
Copy the folder
Copy the whole folder into your agent's skills directory so the checker script and examples stay with the SKILL.md.
Use the skills CLI instead
The README says its own install command works for Claude Code, Codex and Cursor, tested with skills version 1.7.4.
Run it
Tell your agent to caveman your website, and give it a URL, a screenshot or pasted copy.
Check that it loaded
Give the agent a wordy homepage. It should report the before counts, rewrite each screen in the screen format, then run the check script until every screen passes.
Key features
- The three-grunt testA screen must answer what this is, what you get and where to tap within three seconds, or it fails.
- Picture firstA picture fills the top half and shows the result happening, not a dashboard; words only label it.
- Fifteen word capEvery word on the screen counts, including buttons, placeholders, small links and text inside the picture.
- One button ruleExactly one filled button, with a pointer emoji above it; everything else is small grey text.
- Honest button labelsThe label says what happens next and the next screen must keep that promise.
- Step bar for flowsEvery step of a signup shows a trail, with three steps or fewer before the first win.
- No tribe words or hypeIndustry jargon and hype terms are cut, and a company swap test checks the headline says something specific.
- Checker scriptA Python script counts words and buttons per screen, prints PASS or FAIL and exits non-zero on failure.
- Per-surface rulesSeparate guidance for ads, landing pages on phone and desktop, funnels and app screens.
- Before and after reportThe output lists counts for the old and new version, with one line for each rule the old version broke.
Frequently asked questions
What is the caveman marketing skill?
It is an agent skill that simplifies an ad, landing page or signup flow until a stranger gets it in three seconds. It enforces one picture, at most fifteen words and one filled button, then verifies the result with a word-counting script.
How do I use caveman marketing?
Install the skill, then tell your agent to caveman your website or ad. Provide a URL, a screenshot or the copy. The agent counts the old version, names the one job, rewrites each screen, runs the checker and shows before and after.
Does it work with Claude Code, Codex and Cursor?
Yes, the README says the install command installs it into Claude Code, Codex and Cursor, and into any other agent the skills CLI supports. It notes the command was tested with skills version 1.7.4.
Why only one button and fifteen words?
The README cites Hick's law, choice overload studies and progressive disclosure. Its argument is that each extra choice slows a decision, so one screen should show one clear action with minimum decisions before the visitor gets value.
Will it work for long sales pages?
No. The skill says it is not for long persuasive copy written from scratch. It targets screens people glance at for three seconds, such as ads, landing page first views, signups, onboarding and app screens.
What does the checker script do?
It reads the screen format, counts words and buttons for each screen, prints PASS or FAIL with a reason, and exits non-zero on failure. The author added it because models miscount, so it checks the numbers.
How does it differ from other conversion skills?
It applies fixed numeric limits rather than general advice, which makes the result easy to check with a script. For broader review see landing-page-conversion-audit or page-cro, and for layout rules and call to action placement see landing-page-design.
Can it write the ad images?
It describes the picture in a one-line brief and, if your agent can render images, renders before and after. Otherwise the picture line is the brief for whoever draws it. The repository's example mocks are labelled as not the companies' own pages.


