Morning
After 3 fixesTop fixDead video player
Do not touchBranding and section order
An open tool from Studio Haim
עבריתA small agent skill for one question: is this homepage a product people can understand and act on, or a polished demo?
One verdict. Exactly three fixes. And what not to touch.
01
Each card shows only the part preserved from the original output.
Both cases received the same verdict — one of the skill’s 3 defined output states.
Morning
After 3 fixesTop fixDead video player
Do not touchBranding and section order
iamfinance
After 3 fixesTop fixPrimary CTA is invisible
Do not touchTestimonials


02
No form. No discovery workshop. Give your agent a homepage and read the answer.
Add the skill to the agent you already use.
Send the homepage or primary landing page.
One verdict, three ranked fixes, and elements that should stay locked.
03
Reality Check is for the moment a homepage looks ready but still needs a narrow ship decision.
04
No brief required. Give the agent a URL, a screenshot, or the first-screen copy.
Live URL
Run a Reality Check on this homepage: [URL]Screenshot
Reality-check the attached first screen. What is broken before this ships?Copy only
Here are the hero and navigation. Give me one verdict, 3 fixes, and what not to touch: [copy]05
Four short answers about the boundaries of Reality Check.
It checks whether the offer is understandable, the primary action is clear, claims rest on evidence, and anything looks live without actually working.
Yes. Attach a screenshot or paste the hero, navigation, and first-screen copy.
No. Those are outside v1 unless a specific issue blocks understanding, action, or trust on the homepage.
No. It returns one verdict, exactly 3 fixes, and what not to touch. It does not invent strategy or a new design direction.
Open source
The skill is free and MIT licensed. Tested with Codex and Grok, and designed to work with Claude.
Download the skill files and add the reality-check folder to the agent you use.
Studio Haim
Studio Haim builds websites, systems, AI video and automations from the problem and the user - not from a feature list.