CI never runs a zsh case: every zsh variant of the handoff-skill tests is skipped on the runners #328

What happens

.github/workflows/tests.yml installs no zsh (grep -i zsh finds nothing at 33ce26d), and the review of #265 found the ubuntu-24.04 runner image has none either. The zsh variants carry skipif(shutil.which("zsh") is None, reason="no zsh here") (tests/test_handoff_skill_72_80.py, tests/test_handoff_skill_98_99_100.py, tests/test_handoff_skill_commit_block.py; tests/test_handoff_skill_80_85.py:27 runs "under zsh -f where one is installed"). In run 39337757910 the review counted 60 skipped zsh cases: 19 from test_handoff_skill_80_85.py, 26 from _72_80, 5 from _98_99_100 and 10 from commit_block. That count is the review's, not re-measured here. Since the macOS and Windows jobs were removed (#227), the zsh variants run only on a workstation that has zsh.

Why it matters

The handoff skill's blocks are run by the harness's shell (zsh on the workstations where #51 and #197 measured it), and the zsh variants exist because the two shells disagree (an unset variable under set -u, word splitting). A regression in one reads green in CI. Not introduced by #265.

Suggested fix

Install zsh in the test jobs that run these files, or add a small zsh job for the handoff-skill files, and make the run fail when CI is set and any zsh case is skipped, so a runner-image change cannot silently drop them again.

Found by: the delta review and polish check of #265 (lane 5, 2026-10-19 and 2026-10-20). Filed on Brett Heap's word of 2026-10-23T18:52:34Z, by multiple choice.

Lane: openRepoTools-5