Don't know what you built? Neither do your tests.
AI says "all tests passed" and errors still show upbecause those tests were written without knowing what was built.
AI submitted this list. Only a person can confirm it.
Checkout
Name
Address
Card number
Expiry
CVC
AI called itall passed.
The breakage was still there.
"Just tell Claude Code to test it, right?"
AI did not get it wrong.It understood it differently.
A short request leaves room, and AI builds on its own reading.That reading can be reasonable and still differ from what you pictured.
“Show sold out when stock runs out”
What you pictured: and block the purchase
SOLD OUTWireless earbuds$89.00Out of stockOut of stock
SOLD OUTWireless earbuds$89.00Out of stockBuy nowAI built what it understood, and verified exactly that.Whether it matched what you pictured, nobody looked.
Check the list of what AI built first, and this gap gets caught before any test runs
Confirm, test, and get the result
Confirm, test, and get the result
Confirm what was developed
AI lists what it developed this round. Skim it, drop what is wrong, add what is missing.
Test the confirmed scope
A real browser clicks through those items in order. Green for pass, red for fail.
On failure you get a video and a report
See where it stopped on video, with the cause and the fix written up. Hand it to your AI and the fix begins.
Confirm what was developed
AI lists what it developed this round. Skim it, drop what is wrong, add what is missing.
Test the confirmed scope
A real browser clicks through those items in order. Green for pass, red for fail.
Straight from the replay — AI typed the credentials and clicked through itself.
On failure you get a video and a report
See where it stopped on video, with the cause and the fix written up. Hand it to your AI and the fix begins.
Same pass, different outcome
Both said passed.What counted as passing differed.
Same items, same count. The only difference is where the bar sits.
- Passes if the sold-out badge shows
- Passes if the email is sent on signup
- Passes if it expires after 24 hours
- Passes if the photo appears in the list
- Passes if buying is blocked toocorrected
- Passes only after the address is verifiedcorrected
- Passes if it expires once usedcorrected
- Passes if the photo appears in the list
Not because the tests ran better.Because a person set what passing means first.
MCP CONNECT
One MCP line to connect,then a single sentence.
One MCP line in a config file finishes the setup. After that you just talk to the AI you already use.
Say this to your AI
“Run tests on what I developed this round”
Right awayWhat was developed lands in Specnote as a list to confirm
Confirming and running happens in Specnote from there.
How it connects
One MCP line in config · 1-minute setup
Connect GitHub and changed code carries over on its own
Any service that opens in a browser can be verified, regardless of framework.
Start free
Scenarios are unlimited — credits only when you verify.
- Unlimited scenarios
- 260 credits every month
- AI interviews & auto extraction
- 660 credits every month
- Priority AI processing
- Priority support — replies within 24h
Code access is read-only, and sensitive values like passwords and API keys are automatically masked at collection.
Now a pass means
you can ship on it.
30 seconds to sign up · 5 minutes to your first test