Start here — the guide to all four.
Which browser job to reach for, and the four traps that make a clean report lie.
What this guide covers
- Explains which of the three browser jobs fits your task
- Shows what a good, honest report actually looks like
- Names the four traps that make a clean report lie
- Gets you from nothing to a first run
Why it works
Read this before the other four. It covers the one decision that matters (don't automate a browser if the tool has a proper way in for software), which of the three jobs fits what you're trying to do, what a good result actually looks like, and the four traps that cost me something to find.
The traps are the real value here. Each one is a way these jobs can look like they worked while telling you nothing — an empty box that reads as a zero, a form that says "confirm" while failing, a browser that swears it forgot you, and a test run that quietly kills the schedule it was testing.
No prompt to copy. Just the part that stops you trusting a clean-looking report that isn't.
Tips for this guide
- Start with the Dashboard Check. It is the gentlest of the three.
- Fill every bracket before you run. Blanks are what break these jobs.
- Ask for a screenshot with every claim so you can check the work.
- Run it once by hand before you put it on a schedule.
How to use this guide
- 1Read the Framework section and pick the job that fits
- 2Open that prompt in the pack and fill in the brackets
- 3Paste it into Claude and run it once, watching what happens
- 4Check the result against the Results section before you trust it
- 5Only then set it to run on a schedule
More like thisSetup prompts
Browse all →
