Tatsh / wiswa / 34019027150

100%

push

github

claude/skills/make-release: check tests, coverage, and CI before bumping

Nothing between the changelog edit and the version bump ran the test
suite or looked at the remote. pre-commit covers formatting, linting, and
spelling, none of which fail on a broken test, so a release could be cut
from a working tree whose tests were red and pushed on top of a default
branch whose last CI run had failed.

Two steps now sit before the pre-commit run. The first runs the suite
with coverage and holds it to 100%, or to fail_under when
[tool.coverage.report] sets one, since a project that has chosen a lower
number has chosen it deliberately. The second reads the newest run of
every push-triggered workflow, which is where the packaging jobs live:
Flatpak, Snap, AppImage, and PyInstaller each break in ways only their
own workflow sees, and a tag push is a poor time to find out.

The workflow list comes from gh rather than from the skill, because each
of those files is written under a different setting and a list kept here
would drift. The tests run is also compared against HEAD, so a release
built on commits CI never saw is reported rather than assumed green.

The accompanying rule closes the obvious way out: a coverage shortfall
is answered with tests, not by lowering the threshold or omitting the
file.

Signed-off-by: Andrew Udvare <audvare@gmail.com>

788 of 788 branches covered (100.0%)

Branch coverage included in aggregate %.

2040 of 2040 relevant lines covered (100.0%)

5.0 hits per line