How to use A/B Test Planner in your AI agent

What you'll get

This skill helps product managers and growth teams plan rigorous A/B tests for features, pricing, UI changes, and onboarding flows. It calculates sample sizes, estimates test duration, defines success metrics, and provides a decision framework for interpreting results. You get a complete test plan that prevents common pitfalls like peeking early or shipping inconclusive results.

Step 1 — Get the skill source URL

The A/B Test Planner skill lives at:

Step 2 — Add it to your AI agent

Copy one of these prompts and paste it into your AI tool. The agent will fetch and install the skill for you.

Add the agent skill from https://github.com/mohitagw15856/pm-claude-skills/tree/main/skills/ab-test-planner to my project.

Install this agent skill: https://github.com/mohitagw15856/pm-claude-skills/tree/main/skills/ab-test-planner

Add this skill to .codex/skills/: https://github.com/mohitagw15856/pm-claude-skills/tree/main/skills/ab-test-planner

Add the skill at https://github.com/mohitagw15856/pm-claude-skills/tree/main/skills/ab-test-planner to my .gemini/skills/ directory.

Install this agent skill from https://github.com/mohitagw15856/pm-claude-skills/tree/main/skills/ab-test-planner

Add this skill: https://github.com/mohitagw15856/pm-claude-skills/tree/main/skills/ab-test-planner

Step 3 — Try these prompts

Once the skill is installed, ask your AI things like:

• Planning to test a new feature, pricing change, or UI redesign
• Know how many users and how long a test requires
• You want a clear decision framework before launching an experiment
• Unsure whether your test setup will produce trustworthy results

What this skill does

• Creates structured test plans with hypothesis, variants, metrics, and decision criteria
• Calculates required sample size and estimates how long your test needs to run
• Defines primary success metrics and guardrail metrics to protect key business outcomes
• Provides a results interpretation guide so you know when to ship, iterate, or stop
• Flags risks like low traffic, multiple comparisons, and premature result-checking