Nothing changed.
The build broke anyway.
TestInspector re-runs your suite until the flake shows itself, then an AI agent names the cause and proposes the fix. You review it like any other diff.
Async assertion races the toast animation. The test asserts on the discount total before the re-render settles. It passes whenever the runner is slow enough to let the frame land. That is most of the time, and never when you need it.
Three passes. No quarantine.
A skipped test is a deleted test with extra steps. TestInspector works the problem instead of hiding it.
Make it contradict itself
Re-runs the same commit until the mask slips. No new code, no new inputs. just enough runs to catch a test disagreeing with itself.
Name the actual cause
The model reads the failing runs against the passing ones and identifies what differs: the race, the shared fixture, the wall-clock assumption, the state left behind by the test before it.
Patch the test, not the product
Proposes a diff scoped to the test file. It waits on the condition instead of the clock. You read it, you merge it, or you don't.
What it won't do
Quarantine and forget
Auto-skipping a flaky test buys a green build and loses the coverage. That trade is never made for you.
Touch your product code
Every proposed diff stays inside the test file. If the flake is a real bug, it says so and stops.
Merge anything
Nothing lands without you. The agent opens the diff and gets out of the way.
Get it before your next red build.
Early access is going out in small batches to individual developers and small teams.