The question behind every audit
The skill turns the usual order around. Instead of starting with a change someone wants and justifying a test for it, it asks what the most important reason is that visitors with the right intent do not convert. Traffic numbers alone cannot answer that, so it pairs analytics with recordings, heatmaps, and what visitors say.
It also refuses to average across sources. A page that converts well for branded search and poorly for cold social traffic tells a different story than its blended rate.
Before the page: traffic, message, and speed
Some conversion problems start before the visitor arrives. The skill checks whether the ad’s promise matches the page, such as an ad that says no credit card is needed above a form that asks for one. It looks for a shift in traffic quality before blaming the copy, and for slow loads, broken mobile buttons, or form errors that no rewrite would fix.
Friction in five places
- Copy: a headline disconnected from why the visitor came, a buried value proposition, or claims without proof.
- Design: a primary button that blends into the page, competing calls to action, or a weaker mobile layout.
- Form: too many fields, questions visitors are not ready to answer, and no word on what happens next.
- Offer: an unclear offer, or a commitment that feels too large for a first step.
- Trust: no visible proof, and no refund or cancellation policy where money is involved.
Evidence before opinions
The skill has you watch 20 to 30 recordings of visitors who did not convert, read scroll and click maps, and run a one question exit survey. It adds interviews with three to five recent customers about what worried them before they signed up, plus any chat transcripts. Their words often become the page’s words.
Scoring, then one hypothesis per test
Each opportunity gets a score from 1 to 10 for impact, confidence, and ease, and the three are averaged. The top ten go into a ranked list. Every test then starts from a written hypothesis: because of this evidence, we believe this change will produce this result for this audience.
A test changes one variable, fixes its primary metric and end date before launch, and never stops early because the variant looks ahead.
The CRO method, step by step, in the cro file
The funnel checks, friction categories, scoring, and sample size rules above come from the cro skill file on GitHub. It is released under the MIT License, “Copyright (c) 2026 ScayverGraphix.” The skills follow the Agent Skills spec published by vercel-labs/skills.
Low traffic changes the plan.
The skill’s own example shows why sizing comes first. A page converting at 3% needs roughly 8,000 to 12,000 visitors per version to detect a 20% relative lift. At 1,000 visitors a week, that test would run 16 to 24 weeks. When traffic cannot support a clean test, the skill recommends making the change as a best bet and comparing before and after.
Start with the example request from the repository.
Use cro to identify funnel friction and create an ICE-scored experiment backlog.
The readme connects it with copywriting and ab-testing, so you can add all three with npx skills add scayver/marketing-skills --skill cro copywriting ab-testing.
A clear path from question to action
Our Conversion Optimization service helps identify where a visitor’s path to a meaningful action breaks, defines a specific change, and reviews what the evidence supports afterward.