[Chemaclass]

Summary

The project suite repeated work without testing anything more. This removes the repeats and keeps every assertion.

• Tests that ran bashunit a second time only to read the exit code now reuse the first run's code (about 40 runs).
• bench_reports, list and output_stdout run their shared command once per file and the tests read the cached result.
• summary_output used real test files as fixtures (bashunit_init_test.sh three times). It now uses a small fixture with one skipped and one incomplete test. This also removes the rule that bashunit_init_test.sh must keep exactly one skip.
• total_tests_parity walked every repo test file three times, once per filter. It now walks once and checks all three cases.
• json_skip_label built its no-jq PATH with about 970 ln calls per test. It now builds it once with one ln call.
• The validation docs pointed at BASHUNIT_PARALLEL_JOBS=4 make test, but make test runs sequentially and that variable does not turn parallel on. They now use make test/parallel, which make help also lists.

Timing

Files run one at a time, before and after:

File	Before	After
bashunit_summary_output_test.sh	9.4s	0.8s
total_tests_parity_test.sh	6.2s	4.0s
bashunit_json_skip_label_test.sh	4.5s	1.7s
bashunit_bench_reports_test.sh	2.4s	1.0s
bashunit_output_stdout_test.sh	2.9s	2.2s

Full suite with --parallel --jobs 4: 69s before, 59s after. A sequential local run (what the old docs led to) took 5m46s.

Test plan

• ./bashunit --parallel --jobs 4 tests: 2840 passed, same test count as main
• ./bashunit --suite strict tests: green
• make sa, make lint
• npm run agents:check