Both flags survived the cascade that gave them meaning. --held marked a
cheap pass withholding a strong review, and --fix-first let the blocking
count follow the fix-first set on such a run, because block-on need not
contain those severities. There is no withholding and no tier line any
more, so the count has followed block-on alone for some time — while the
two flags went on being parsed, and --fix-first went on being validated,
with neither value read by anything downstream.
fix-first REMAINS an action input. It stops the exhaustive loop early
(action.yml), which is a different question from what the summary counts;
the header now says so, so the name cannot argue its way back into this
count later.
Removing a flag the script no longer understands is safe to ship on its
own: the argument loop has no else branch, so an unrecognised flag and the
value after it are skipped one token at a time without shifting the flags
that follow. The surviving block-on test now passes --fix-first in the
MIDDLE of the argv to hold that property down.
Also scopes the action-inputs reference to this action: the hosted App runs
a different review engine, so fix-first and exhaustive can behave
differently there, and the file now says to answer App questions from the
console instead.
Test suite is unchanged by this: 558 pass / 9 fail before and after, the
failures being a pre-existing CRLF mismatch in the action.yml input tests
on a Windows checkout.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-09-20 18:44:41 +08:00
3 changed files with 30 additions and 16 deletions
