Browse docs
Docs

Video proof

Implementation work comes with a recording of the feature actually working in a real browser.

"It works" should mean there's a recording that proves it works. So implementation tasks don't just end with a green checkmark — a UI-driving agent demonstrates the change in your real app.

Before and after

A UI driver is used at two moments:

  • Before a change — to confirm a reported bug is real. It reproduces the problem and captures a screenshot as evidence of the starting state.
  • After a change — to prove the fix works. It walks the whole flow and records it.

How it drives

The UI driver opens Chrome on the sandbox desktop and controls it like a person — screenshotting, clicking, typing, navigating, and checking that the expected result is actually on screen. It runs inside the thread's sandbox, where the dev server is already listening and any login/auth state is already set up, so it tests the real application, not a mock.

It records the session to a .webm and ends with a plain-language verdict — confirmed, works, partially, or not reproduced — plus exactly what it saw.

What you get

The recording and screenshots surface directly in the chat as the evidence you review. Instead of reading a diff and hoping, you watch the feature run and merge in minutes.

Turning it on

Validation is a per-turn toggle. When it's on, after the work is done the lead spawns a single UI driver to prove the flows that matter for the change — kept to the few that count, recorded one at a time.