Skip to content
Marketing

A/B Testing

coreyhaines31

Design an experiment with a falsifiable hypothesis, assignment unit, guardrails, and stopping rules.

SKILL.md

---
name: knoveya-ab-testing
description: "Design an experiment with a falsifiable hypothesis, assignment unit, guardrails, and stopping rules. Use for matching marketing work."
---

# A/B Testing

## Scope

Apply this guidance when the requested task matches the description. Honor the user's chosen outcome, repository rules, and existing authorization.

## Workflow

Design an experiment with a falsifiable hypothesis, assignment unit, guardrails, and stopping rules.

Before acting, inspect the relevant artifact and ask only for information that materially changes the result. Use existing tools and conventions; do not install dependencies, publish, send messages, deploy, or mutate external data unless the user has authorized that operation.

## Acceptance check

The analysis follows agreed criteria instead of selecting a favorable metric afterward.

## Example task

Test a revised signup explanation while tracking activation quality.

## Report

State what changed, the evidence checked, and what remains unverified. Do not claim a successful operation or completed test without observing it.

How to use

Copy the full block, including its frontmatter, into a new folder named knoveya-ab-testing as SKILL.md. Review the instructions before enabling them. For Codex, a user skill normally lives in ~/.codex/skills/knoveya-ab-testing/SKILL.md; check the current documentation for another agent's supported location. Do not overwrite an existing skill.

This is an independent Knoveya skill. No files are installed by this website.

Example

Test a revised signup explanation while tracking activation quality.

Verify

The analysis follows agreed criteria instead of selecting a favorable metric afterward.

Test the skill on a small, authorized example. File validation alone does not prove that an agent will make good decisions.

Limits

The skill supplies instructions, not tool access or expertise guarantees. Inspect generated work and preserve the user's approval boundaries.