Nightly regression writes a before and an after image for every UI step, which is wonderful when something fails and enormous every other night. The artefact bucket is the problem, not the runtime.
What I want on that one environment is: no step captures, still fail on a baseline drift. Is that expressible, or is it all-or-nothing per policy?
Accepted answer
Expressible, and it is the case the flag was added for. Set it on the environment rather than the project so your local runs keep their narrative:
screenshots:
onlyOnFailure: trueSDODS_SHOTS_ONLY_ON_FAILURE=1 sdods run -p demo-shop -e staging -t @regressionEither form downgrades the capture mode to on-failure — no scenario start and end, no per-step before and after — and deliberately leaves the baseline comparison switched on. The visual assertion is an assertion; it is not a screenshot policy, and turning down the narration should not quietly stop testing something.
Worth knowing the precedence at the same time, because the next question is usually "why does my @regression @visual scenario still compare when I set '@regression': step":
@visual wins outright.@smoke, @regression, @sanity).screenshots.policy.default.So @visual is not a policy you can be talked out of by a broader tag, which is what you want given a scenario carries several tags and only one of them is about pixels.
Have a different question? Ask it.