Free tool. Runs in your browser.
Did your AI agent really finish?
Paste what the agent said, your own test run and the code changes. You get a receipt that shows which claims have proof.
Every claim gets one label
- Proof shownA line you pasted backs it up.
- ContradictedA line you pasted says the opposite.
- No proofIt could be proven, but your paste doesn’t show it.
- Can’t tellA paste can’t prove it, or a rule isn’t sure. Check it yourself. Never counted as proof.
Agent receipt
a free tool by softDev23
Paste your evidence
Its last message, where it says it’s done.
Run your tests (for example npm test) and copy all the output.
Run git diff --stat, then git diff, and copy both outputs.
Found in your paste
This is the made-up example. It is marked EXAMPLE on the receipt.
Paste something first.
Agent Receipt never sends what you paste
Requests your browser listed since you pasted: 0
Works with Wi-Fi off
How to check it yourself
- Load this page, then turn off Wi-Fi (or set your browser’s Network tab to Offline). Paste and make your receipt. It still works.
- Or open your browser’s developer tools, go to the Network tab and make a receipt. No new request shows up.
- The counter above reads the list of requests your browser keeps for this page since you pasted. It can’t see every kind of connection, so the Network tab is the fuller check. Browser extensions can read what you paste, so only use ones you trust.
What is Agent Receipt?
Agent Receipt is a free tool by softDev23. You paste what your AI coding agent said when it finished, your own test run and the code changes. It splits the agent’s message into claims and checks each one against the lines you pasted. Use it to verify AI-generated code before you merge it. It doesn’t run your code, and it isn’t a log of the agent’s actions.
What it reads
- Test runs from Vitest, Jest, pytest, go test, XCTest, Swift Testing and Xcode.
- Code changes from
git diffandgit diff --stat. - Common tricks: skipped tests, a test left on
.only, checks that are always true, deleted tests, placeholders instead of real code, errors hidden in an empty catch, and test commands that never really ran.
Why you’ll see Can’t tell
Can’t tell is the default, and it never counts as proof. You get it when a paste can’t show something, like “the checkout works now”, or when a rule isn’t sure.
To stay careful, it only reads new tests written in common, simple shapes: test(), it() or specify() as a statement of its own in JavaScript and TypeScript, def test_ functions in Python, func test methods and @Test in Swift, and func Test functions in Go. Checks inside a nested function or class aren’t counted as the test’s own. A test written any other way gets Can’t tell, never Proof shown, even when it’s fine.
Proof shown only means your paste shows the part that was checked, for example that a file is in your changes. It can’t confirm your test run used this exact code, and it isn’t a code review.
Questions
Is Agent Receipt free?
Yes. It’s free to use in your browser. There’s no account, no download and no charge.
What happens to what I paste?
Agent Receipt checks it inside this tab. Its code never sends or saves what you paste. The counter counts the requests your browser lists for this page since you pasted. It can’t see every kind of connection, so your browser’s Network tab is the fuller check. Browser extensions can read what you paste, so only use ones you trust. The check also works with Wi-Fi off. Like the rest of softdev23.com, the page counts normal page views. For details, see the privacy policy.
Does it work with any AI coding agent?
Yes. It only reads the text you paste, so it works the same whichever agent wrote it.
What if a label looks wrong?
Read the “Checked” line under the claim. It says exactly what was looked at. If your test runner or diff isn’t one it reads, try pasting the 3 parts separately, and treat the label as a hint. Can’t tell never counts as proof.
Is this a signed log of what the agent did?
No. Some tools use “agent receipts” for signed logs of an agent’s tool calls. This receipt only checks a pasted “done” message against your own test run and code changes.
How is it different from Diffective?
Diffective is a free game for practicing these checks on made-up cases. Agent Receipt runs the checks on your own paste.