Setup PR cannot bootstrap its first baseline

The documented setup flow has a chicken-and-egg failure. A repository’s first Frameshift setup PR runs in report mode, waits about three minutes for an artifact tied to the exact base commit, then fails because that commit could not have produced a Frameshift baseline before the workflow existed.

Observed with dcramer/frameshift/ci@68a8b5e8bbd439088ef9a044e693c5de9efe7ecd:

• All 61 Playwright tests passed and produced 26 PNG files.
• Find the before screenshots waited for 180 seconds.
• It failed with No saved screenshots named frameshift-baseline-v1-<base sha> were found.
• The setup guide says to merge the workflow and let it run on the default branch before the first PR. That cannot make the setup PR itself pass when CI is required.
• The exact-base-SHA lookup also creates a race for PRs opened before the corresponding default-branch workflow finishes. The action waits, but then fails if the baseline is still unavailable.
• The failure happens late in the job and looks like a screenshot or E2E failure even though capture succeeded; the required CI check then fails too.

Current downstream workaround: query Actions for a live baseline artifact before invoking Frameshift, skip the comparison only when no baseline exists, and let the first default-branch run create it.

Related integration PR and failure evidence: https://github.com/getsentry/junior/pull/1753 and https://github.com/getsentry/junior/actions/runs/33536953468/job/99953603443

Requested by David Cramer.

--

View Junior Session [Sentry]

Start with the documented setup flow, the report mode, and the Frameshift action’s exact-base-SHA baseline lookup described in the issue. Reproduce the first setup PR and the case where the default-branch artifact is still pending, using the linked integration evidence and the 61 Playwright tests as references. Done means setup CI does not fail when no baseline can yet exist, while later comparisons still wait for and use an available baseline.