Build the testable prototype
Prototype only enough reality to answer one costly product question.
Use this when
A polished prototype with no written decision it informs
A facilitator explaining the interface during every task
Success declared from positive comments without behavioral thresholds
The quality gate
- The prototype names one decision and the uncertainty it is built to reduce
- A participant can complete one realistic end-to-end task without a guided tour
- Only evidence-critical behavior is real; simulated behavior is clearly identified
- Success, failure, and ambiguous thresholds are written before sessions begin
- The test plan defines participants, tasks, observations, capture, and the decision after results
Prove it: Remove each prototype screen in turn; any screen whose removal cannot affect the named decision should stay removed.
Install or copy
Save the download as testable-prototype/SKILL.md in your agent's supported skills directory, or copy the complete package below. Load it only when its description matches the work.
--- name: testable-prototype description: "Scopes a minimal prototype and test plan that answers one costly product question. Use when a prototype is needed for evidence rather than presentation or general exploration." --- # Build the testable prototype Apply this skill during **Spec** work. It supports Any stack. Prototype only enough reality to answer one costly product question. ## Instructions Design the smallest prototype that can answer one named product decision. State the uncertainty first, then define one realistic end-to-end task a representative participant can attempt without coaching. Make only the interactions that affect evidence behave realistically; label simulated data, manual operations, and unavailable paths instead of pretending they work. Write success, failure, and ambiguous thresholds before testing. Specify participant criteria, task script, what to observe rather than ask, how evidence will be captured, and what decision follows each threshold. Cut every screen and feature that cannot change the decision. ## Detect the problem - A polished prototype with no written decision it informs - A facilitator explaining the interface during every task - Success declared from positive comments without behavioral thresholds ## Hold the gate Do not declare this phase complete until every item passes: - [ ] The prototype names one decision and the uncertainty it is built to reduce - [ ] A participant can complete one realistic end-to-end task without a guided tour - [ ] Only evidence-critical behavior is real; simulated behavior is clearly identified - [ ] Success, failure, and ambiguous thresholds are written before sessions begin - [ ] The test plan defines participants, tasks, observations, capture, and the decision after results ## Verify the result Remove each prototype screen in turn; any screen whose removal cannot affect the named decision should stay removed.
More spec skills
Cut to a shippable slice
Scope is a knife: one thin, end-to-end slice a user can finish.
Specs written for agents
An AI agent is your fastest junior — brief it like one, verify it like one.
Write the agent contract
Give an agent one bounded outcome, the context to act, and evidence that proves it is done.
Want this quality gate applied to a real product decision?
Start a First Loop