Write it the way you would to a teammate. A vision-grounded examiner runs it on your live app, looking at the screen and clicking like a real user. No selectors, no scripts.
The vision model is the locator and a verifier confirms each step. It adapts when a screen is not what it expected.
Your flow, written in plain English, is broken into steps.
A vision model looks at the screen and locates the target. No CSS selectors.
It clicks, types, scrolls and drags in a real browser.
A second model confirms each step actually worked, before and after.
Verified steps are saved to app memory, so repeat runs get faster and steadier.
On a miss it re-grounds, re-verifies, and replans the rest of the flow.
As the examiner walks your flow, specialists run in parallel on every page it visits, and one slow check never holds up the rest.
Each step with the plain-English action and the screen it was looking at, so you can see exactly how the verdict was reached.
no stack traces. a walkthrough.
App memory recalls the steps that worked, so repeat runs get faster and steadier. Your review guidance and muted findings steer what the examiners flag, applied automatically on every run.
Point it at a live site and pick a profile, Quick, Standard or Thorough, to balance speed against depth.
Run a flow on a cron schedule and get told when something breaks.
Comment @shipguarde verify checkout works on https://preview… to test a preview deploy.
Provide credentials, encrypted at rest, and it signs in to test authenticated flows.
Code review checks the diff. ShipGuarde also checks the running product, so a release does not ship on the hope that nothing broke.
NO CREDIT CARD · CONNECT GITHUB IN UNDER FIVE MINUTES · CANCEL ANY TIME
RELEASE CLEARANCE BUREAU
ISSUING AUTHORITY FOR SOFTWARE RELEASES
[email protected]