brainstorm-experiments-existing helps a product team validate assumptions before building. After clarifying the feature idea and the assumptions behind it, the agent suggests experiments for each one, drawing on first-click and prototype tests, feature stubs, fake door tests, technical spikes, production A/B tests, Wizard of Oz setups and behavioral surveys.
Every experiment is written down as an assumption, the experiment, a metric and a success threshold. The skill tells the agent to measure real behavior, protect users and the business, and explain risk mitigation for production tests.
What You Get
- One or more experiments for each assumption you want to validate
- A stated assumption, experiment, metric and success threshold for every experiment
- Risk mitigation notes for experiments that run in production
- A table or structured list, saved as markdown if it is long
Common Use Cases
- Validating a feature idea with a fake door before engineering starts
- Planning an A/B test with guardrails for a checkout change
- Choosing between a prototype test and a technical spike
- Turning a ranked assumption list into a test plan
What Makes It Unique
It is built for existing products, where production tests such as A/B experiments are available and carry risk, so it requires mitigation notes. Its principles are to measure behavior and not opinions, test responsibly and aim for maximum validated learning with minimal effort. The new-product counterpart, brainstorm-experiments-new, uses pretotypes and XYZ hypotheses instead.
Under The Hood
A single SKILL.md of instructions with no scripts. The agent reads any PRD, assumption list or design files you attach with its own file tools and writes the experiment plan in the conversation. The skill says to save substantial output as markdown, so it can write a file. It does not tell the agent to fetch web pages, call APIs or run commands. Your idea replaces $ARGUMENTS in the prompt.