Lauren Tan explains verification loops to make agent-written code more correct

Lauren Tan, writing as @poteto on X, says agents should verify their own work first, running and interacting with the app like a real user.

She says the goal is an artifact, such as a video or screenshots, that shows the work is correct, matches the intent, and is easier to review than the full PR.

Tan points to the /create-verification-skill and /maintain-verification-skill commands in pstack, a Cursor marketplace plugin, for building and keeping up these checks. She adds that verification is only an approximation and that not every domain can be verified yet.