Fix #85: sum per-suite test counts instead of reporting the last one

[laird]

Closes #85

Stacked on #78 (feature/issue-73-master). Based on that branch so the diff shows only this work; merge #78 first.

Root Cause

count_before() ended in tail -1, taking the last "<N> passed" in the log. That is correct for Jest/Playwright, which emit one summary per run. This repo's unit command is a loop over a dozen independent bash scripts, each printing its own Results: N passed, M failed — so a green run of 178 assertions reported the last suite's 7. The denominator was worse: no bash suite prints the word total, so UNIT_TOTAL was always the literal 0.

Console read Unit Tests: ✅ PASSED (7/0 passed).

Scope — reporting only, not a false green

Worth stating so this is not read as a repeat of #73:

• UNIT_EXIT comes from PIPESTATUS[0], not from these counts, so a red run still goes red. The regression-test.sh reports "All tests passed" after running zero tests, and writes reports into a directory named tests/test_*.sh #73 fix holds.
• The failure path is unaffected: if [ "$UNIT_FAILED" -gt 0 ] reads the same parser, but || exit 1 in the suite command makes a failing script the last summary in the log, so the count was already right exactly when it mattered.

What was wrong is the number a human reads, in the console and in docs/test/regression-reports/*.md — understated ~25x, over a nonsensical /0.

Changes

• count_before() sums matches with awk instead of tail -1. It still returns empty when nothing matched, which the no-parseable-summary guard depends on to distinguish "the runner never ran" from a genuine zero.
• UNIT_TOTAL is derived from passed + failed only when the runner reports no total, and only after that guard — otherwise an unparseable run would synthesise 0 + 0 = 0 and read as a valid green. A total the runner did report is never overwritten, since Jest's legitimately counts skipped tests.
• Mirrored both into .agent/scripts/regression-test.sh per the parallel maintenance requirement in CLAUDE.md.
• E2E needed no separate change: it already derives its own total, so it inherits the summing fix.

Verification

Two new fixtures in tests/test_regression_test_config.sh pin both directions:

• three green suites (6 + 13 + 7) must report 26/26 passed, not 7/0
• two failing suites (4+2, 5+3) must report 9/14 passed and still exit non-zero

$ bash tests/test_regression_test_config.sh
=== 21 passed, 0 failed ===        # 19/2 red before the fix

$ bash -n plugins/autocoder/scripts/*.sh && bash -n .agent/scripts/*.sh \
    && python3 -m py_compile plugins/autocoder/scripts/*.py
✅ build OK

$ bash plugins/autocoder/scripts/regression-test.sh   # exit 0
  Unit Tests:  ✅ PASSED (178/178 passed)             # was (7/0 passed)
  E2E Tests:   ⏭️  SKIPPED (not configured) (0/0 passed)

Noted, not fixed here

.agent/scripts/regression-test.sh is missing the whole of #73 — no PIPESTATUS read, no all-skipped guard, no report-dir scoping — so the mirror can still report a false green. That is #78's / #71's territory, deliberately left alone to keep this diff to one defect.

🤖 Generated with Claude Code