[quality] CI gate scripts generate_skill_index.py and update_coverage_snapshot.py had zero unit coverage

[kubestellar-hive]

Finding

scripts/ contains 10 Python entry points. Eight have a matching tests/unit/test_*.py. Two did not:

script	lines	CI gate
scripts/generate_skill_index.py	165	.github/workflows/docs-validate.yml runs it with --check
scripts/update_coverage_snapshot.py	186	.github/workflows/pr-validate.yml runs it with --check

Both are merge gates. Their failure modes are quiet:

• generate_skill_index.collect() enforces nine front-matter validations (required fields, category/status enums, entry_point self-match, id == name). A regression that drops a check lets a malformed skill into index.json with a green build.
• generate_skill_index.main(--check) compares index.md against render_md(skills, current['generated_at']) — deliberately reusing the committed date rather than date.today(). Nothing pinned that behaviour; switching it to today's date would make the docs gate fail every day after the catalog is committed (the same class of failure tracked in dakota-iso#137).
• update_coverage_snapshot.parse_scenarios() implements non-obvious Gherkin tag semantics: feature tags inherit, scenario tags reset after each scenario, and Rule:/Background:/Examples: discard pending tags. classify() layers a precedence order on top (quarantine > hardware_blocked/future/pending > active). A silent miscount here corrupts the suite-map coverage table that this repo uses as its coverage source of truth — and the whole point of that script is to stop humans hand-editing those numbers.
• update_file() returns three distinct exit codes (0 ok, 1 stale, 2 markers missing) that CI branches on; none were pinned.

Recommendation

Add unit tests for both scripts. Done in PR #766 (hold-gated):

• tests/unit/test_generate_skill_index.py — 36 tests
• tests/unit/test_update_coverage_snapshot.py — 49 tests

85 tests, all passing. No production code changed.

Priority

• Impact: high (both scripts are required merge gates with silent failure modes)
• Effort: low (both are pure functions plus a --repo-root seam; no mocking of external services needed)

---

Filed by quality agent (hold-gated mode)

🐝 Hive Agent: quality | Instance: hosted-projectbluefin-knuckle-gjvq | SHA: f368ee5

— hive: agent=quality backend=copilot model=claude-opus-5