Skip to main content
A procedure is a saved test: a fixed, ordered list of steps that replays exactly as recorded. Replays do not call the planner. They are cheaper, faster and deterministic: the same steps every time, so a change in the result means a change in your site, not a change in how the test was interpreted. Procedures are what schedules run, what CI triggers, what the GitHub PR bot fires, and what fuzzing and visual baselines attach to.

​
Saving one

Press Save as test on a finished run. What gets saved is what actually worked:
  • An Enter key the page ignored, recovered by clicking the field’s own submit button, saves as that click.
  • A typed value a combobox committed as a selection saves as the selection.
  • A click skipped because the page was already there saves as nothing at all.
  • Coverage-gap placeholders are stripped.
So the procedure you replay tomorrow is the sequence that worked today, not the first guess at it.

​
Running one

Each of those accepts an environment and a base-URL override, so the same procedure can run against staging, production, or a PR preview.

​
Login and auth mode

Every procedure has an auth mode that decides how its browser starts. New procedures get a conservative guess from their name and steps; change it in the saved-tests panel if the guess is wrong.

​
Capturing a login state

Mark a sign-in procedure as the project’s login-setup test. When it passes, the browser’s storage state is captured, encrypted and stored on the project. Every later run that reuses login state starts already signed in. Re-run the login-setup procedure whenever the session expires.

​
Exporting

GET /api/procedures/{id}/export returns the procedure as a standalone Playwright spec, useful for taking a flow into your own repository, or for seeing exactly what a step does.

​
Maintenance

A procedure is only as good as the page it was recorded against. When a redesign moves things:
  • The flakiness dashboard on the project overview sorts your tests worst-first, so the ones drifting show up before they fail outright.
  • A failed replay is attributed: test means the step no longer matches the page, app means the site is genuinely broken.
  • Re-record by describing the flow in chat again and saving the new run.