Experiment Designer
Experiment Designer is an ecommerce AI skill for Alireza Rezvani, built for teams working with Codex, Claude Code, OpenClaw. Use it to the Experiment Designer…
What this skill helps you do
Use the Experiment Designer playbook when you need to plan campaigns, produce channel-ready material, and control advertising work. It gives the operator a repeatable set of checkpoints.
Use the Experiment Designer playbook when you need to plan campaigns, produce channel-ready material, and control advertising work. It gives the operator a repeatable set of checkpoints.
Before you start
- Choose one bounded, reversible campaign brief or marketing asset to test.
- Prepare the offer, audience, approved claims, brand voice, channel limits, and budget, removing sensitive fields the trial does not need.
- Record the current conversion rate, cost per acquisition, contribution margin, and unsubscribe rate, then name the approver and stop conditions.
- Map the data path from source to destination, then review read, write, and administrator scopes separately.
How to test it safely
- Test one normal case, one edge case, and one case with a deliberately missing critical field.
- Compare the result with the pre-test baseline for conversion rate, cost per acquisition, contribution margin, and unsubscribe rate; do not record time saved alone.
- Review errors, human edits, permissions used, and unresolved exceptions before expanding scope.
- Simulate a timeout, a duplicate event, and a partial destination failure to verify alerts and recovery.
Operator-ready setup plan
Start with one real task
Do not begin with a store-wide rollout. Pick one reversible task where Experiment Designer can help you plan campaigns, produce channel-ready material, and control advertising work.
Checkpoint: The input boundary, owner, and one primary measure from conversion rate, cost per acquisition, contribution margin, and unsubscribe rate are written down.
Prepare the input and guardrails
Collect only the the offer, audience, approved claims, brand voice, channel limits, and budget needed for this test. Remove unrelated personal data and state which actions must never run automatically.
Checkpoint: Every input has a known source, sensitive fields are minimized, and the approver knows what the trial can read or change.
Inspect the source Skill, then run it
Read the source, installation method, and permission notes before adding Experiment Designer to a separate test project. Keep commands and Skill text exactly as published.
Checkpoint: You have a campaign brief or marketing asset that a responsible operator can inspect, and it stayed inside the approved boundary.
Review it against a baseline
Do not judge the result by fluency. Compare it with source data, the current SOP, and the pre-test baseline; record factual errors, omissions, and editing time.
Checkpoint: conversion rate, cost per acquisition, contribution margin, and unsubscribe rate has a pre-test baseline, and errors and exceptions are logged separately.
Add monitoring, approval, and recovery
Alert on failures, timeouts, duplicate runs, and permission changes. Keep human approval, idempotency checks, an action log, and a recovery path you have rehearsed.
Checkpoint: A failed run can be traced in logs, bad writes can be reversed, and ownership of recovery is explicit.
Limits to account for
- We checked the public source and resource identity on 2026-07-19. That review does not cover every workflow result, and vendor performance claims are not treated as EcomAgentTools tests.
- Generated copy can invent claims, miss channel rules, or make weak performance look successful.
- This page reflects the review completed on 2026-07-19, not a permanent guarantee. Recheck the current documentation, pricing, and contract terms before production use.
Questions at this experience level
Which permissions should Experiment Designer receive?
Grant the smallest scope required for this workflow. Separate read, draft, production-write, and administrator access, and require human approval for high-risk writes.
How should failures be rolled back?
Keep source records, request IDs, versions, before-values, and action logs. Rehearse timeouts, duplicate runs, partial success, and third-party API failure outside production.
What should be monitored after launch?
Monitor success, exceptions, latency, execution cost, unauthorized writes, and conversion rate, cost per acquisition, contribution margin, and unsubscribe rate. A completed run is not proof of a safe result.
Creator, original source, and platform proof
Alireza Rezvani
HealthTech CTO and open-source maintainer focused on applied AI, agentic coding, and practical skills for product, research, growth, and operations teams.