WF10 / Build / Check something

Ask an agent to check a real browser journey

Turn a user task into an observed result with evidence of the clicks, output, and failure.

THE WAY THROUGH

From your material to a useful result.

01
Specify the entry URL, user goal, expected result, and forbidden side effects.
02
Have the agent actually navigate and interact. Capture the tested URL or revision and evidence at meaningful steps.
03
Check the final result, including empty states, persistence, and whether an action happened once.
04
Report observed failures separately from suspected causes; repeat the smallest failing path after a fix.

A FICTIONAL TASK TO PRACTICE ON

Here’s the job.

Fictional shop preview: search “green mug”, open the matching product, choose quantity 2, and add it to a test basket. Expected basket: one product line with quantity 2. Stop before checkout.

What a careful result should contain

Required evidence: search result, correct product page, basket line, quantity 2, preview URL, and tested time.
If the button is disabled, record “blocked at add to basket”; do not claim a successful basket.
This sample is a test specification, not a claim that a shop was visited.

This is an authored reference, not a captured Grok Bot, Instinct, or other product run.

Download sample prompt Download reference

BEFORE YOU RELY ON IT

Give the result a proper check.

A written test plan is not a browser test.
The report identifies what was and was not exercised.
A disabled or intercepted control cannot be reported as a passed action.

Try to break it

A page shows a success toast but the basket remains empty. Pass: inspect the basket and report failure. Fail: rely on the toast alone.

TAKE IT FOR A SPIN

Bring your own details.

Describe your task and paste the source material you want to use. This page prepares a prompt; your chosen AI tool does the work.

Your material and constraints

These details stay in this page until you copy them. They are not saved or sent to a model here.

Read the full starting instructions

Execute this journey in an authorized browser environment if browser control is available. First confirm the URL and side-effect boundaries. Record actual observations, screenshots or DOM evidence, and the final state. Check that the expected output exists, not merely that a success message appeared. If browser access is unavailable, say not run and return the test plan only. Do not claim unobserved steps passed.