Publisher: Corey Haines · Editorial profile by AI Tools Ratings · Review method
Not runtime tested; source-document review only. Referenced scripts were not executed or comprehensively audited. No security certification.
What task is this skill designed for?
Specify a two-variant test of directory-card copy with downstream quality guardrails.
What input and output should you expect?
Input: Closed-window baseline measurements, available traffic, proposed difference, randomization unit, duration and implementation constraints.
Output: An experiment specification and analysis plan; expected, not independently tested. No winner is inferred before results exist.
What does it need to run?
An Agent Skills-capable client and trustworthy baseline measurements. Experiment software and statistical computation are separate dependencies.
Publisher README declares support for Agent Skills clients including Claude Code, Codex and Cursor. This is author-declared, not independently runtime tested.
How should you approach installation?
- Inspect the pinned source, license and task-specific reference files before installing anything.
- Follow the publisher README guidance for your actual client; no installation command or cross-client compatibility was independently tested here.
- Try the illustrative prompt with synthetic input in a separate folder, inspect the output against the checklist and approve external actions separately.
Installation guide: discovery, dependencies and runtime checks
What files, accounts or data can it access?
Reads approved metrics and may draft variant code or configuration. Planning requires no live traffic allocation changes; deployment needs separate authorization.
What does the skill cost?
The inspected source license is MIT. Agent subscriptions, model usage, dependencies and external services have separate costs and terms. No paid runtime was tested.
What are its limits and alternatives?
Low traffic, sample-ratio mismatch, overlapping experiments and stopping on favorable interim results can undermine conclusions. Sample-size suggestions need explicit assumptions.
Use CRO for initial page diagnosis. When traffic is too low, moderated task observation can be more useful than an inconclusive experiment.
What could you ask in a small trial?
Try this with a small sample:
Plan a test of two synthetic card descriptions. State assumptions, allocation unit, primary metric, guardrails, stopping rule and the meaning of an inconclusive result.
How would you verify the result?
- Review power assumptions against actual available traffic.
- Validate allocation and event definitions using synthetic records.
- Apply the agreed rule and report uncertainty, including no clear result.
Use a baseline and record acceptance evidence before treating a successful task as a reliable workflow.
Which source and license were checked?
Read the reviewed SKILL.md at its pinned commit · Complete reviewed folder
Repository and current publisher guidance
Checked:
MIT — identified in LICENSE at the reviewed commit. Dependencies and third-party assets may have separate terms. License evidence
Version details and review scope
Source: coreyhaines31/marketingskills/skills/ab-testing
Reviewed commit: e85f3cbc2e4a79bf33b2a12e9bdd9087d414735b
SKILL.md SHA-256: fba679be14a9b21a06cde009caf17a652b92040c6f78a520a5d4baec0a5bd5da
Source-document review. Source text and stated dependencies were inspected for this profile. Referenced code was not comprehensively audited, no runtime test was performed, and source files are not mirrored here.