At which rung did your agent verify?
Your agent says it verified the change. But "verified" hides the difference: did it run the real thing, or only read the diff? One command grades the evidence, then ships it or blocks it.
pip install rung-ai
One word hides how real the check was
"The tests pass" and "I ran the real thing and watched it work" reach you as the same word: verified. So does "I checked it" versus "someone independent checked it." When an agent reports done, you cannot tell which one you got.
"I reviewed the fix."
Read the diff. Reasoned about it. Never ran the code. It might be right, but nothing was observed.
"I ran it and watched it pass."
Exercised the real surface and captured what happened. A different level of confidence, under the same word.
It grades the claim, then enforces your bar
rung asks two things about every claim your agent makes: how real the check was, and who did it. Those two answers place every claim on this grid.
-
How real
Ran it, or read it?
Did the agent run the real surface, or only reason about the code? A green test suite alone does not count as running it.
-
Who checked
The author, or someone else?
A change checked by whoever wrote it is not the same as one an independent party ran and confirmed.
The aim: ran the real thing, and confirmed by someone independent. rung holds every claim to the bar you set, and never lets one read higher than its evidence.
| By the author | Independent | |
|---|---|---|
| Ran the real thing | It ran, self-checkedrung 1 · author | ★ the aimRan, independently confirmedrung 1 · independent |
| Only read it | You have its wordrung 0 · author | A second readrung 0 · independent |
Three steps, one command
Your agent records the run
As it works, the agent captures what it did: what ran, on what surface, and what came out. That becomes the evidence file rung reads.
You set the bar
Say what each risk level demands. Routine changes can pass on the author's word. Higher stakes require a real run and an independent check.
rung ships it or blocks it
One command reads the evidence against your bar and returns a verdict. Drop it into CI, or run it on your machine before you merge.
Trust "done" when you can see how it was checked
rung does not run your tests or replace review. It makes "verified" accountable: every claim carries the evidence that earned it, and nothing ships on its own say-so. The agent keeps moving fast, and you still know what was checked.
"I verified the fix to the auth check."