ab-testing
A/B Testing
A/B testing answers one question: does this specific change actually make things better for real users, or does it just feel like it should? This skill covers the full loop — deciding whether a test is even the right tool, forming a real hypothesis, designing variants, implementing the split, running it without poisoning the data, and reading the results honestly — plus the parts of that loop an agent can do that a human running tests by hand usually can't.
This is a general skill for agent-built projects. It does not assume a specific framework, audience, or analytics vendor. Read the repo first, adapt patterns to the stack, and reuse whatever instrumentation already exists.
Works with any agent-built project
Use this skill on projects the agent (or team) created or maintains — greenfield or legacy:
| Project type | Example tests |
|---|---|
| Marketing / landing page | Hero headline, CTA label, social proof layout, pricing table |
| Web app / SaaS | Onboarding steps, dashboard empty states, settings UX, upgrade prompts |
| E-commerce / checkout | Cart layout, trust badges, shipping copy, payment step order |
| Mobile app | Tab labels, permission dialogs, paywall copy, signup flow |
| Internal / admin tools | Default filters, table density, bulk-action placement |
| API / backend (when UI-adjacent) | Error message wording surfaced to users, email template variants |
| Edge / serverless | Middleware-based splits, cached page variants, worker routing |