gh-aw quality suite silently skips 13 tests (28 assertions) on Windows via existsSync('/bin/sh') probe #1833

Summary

test/gh-aw-quality.test.ts:30 gates two describe blocks on a POSIX shell that never exists on Windows:

const HAS_POSIX_SHELL = existsSync('/bin/sh');

// :1515
describe.skipIf(!HAS_POSIX_SHELL)('gh-aw: Team Guard roster-row detection — committed-HEAD git repo coverage (#1689 revision 3)', …)
// :1687
describe.skipIf(!HAS_POSIX_SHELL)('gh-aw: Cast PR dedup jq filter behavioral coverage (#1689 revision)', …)

On Windows, /bin/sh does not exist, so both blocks skip. Measured: 13 tests containing 28 expect() calls never execute — and the suite reports green.

This is the same defect class as #1824 / #1812 / #1801 / #1822 / #1827: a gate that structurally cannot observe the failure it exists to catch. Per the standing decision in .squad/decisions.md (2026-08-20): a permanently green — or permanently red — gate is equivalent to no gate.

Why this one stings

The skipped blocks are #1689's own behavioral coverage — assertions written specifically to prove gh-aw Team Guard and Cast PR dedup behavior. A Windows contributor sees a green suite and reasonably concludes that behavior is verified. It isn't.

Git Bash ships a POSIX shell on essentially every Windows dev machine that has Git installed, so the skip isn't even buying portability — it's discarding coverage that could have run.

Prior art in this repo — the fix shape already exists

test/gh-aw-command-parse.test.ts (added in #1832 for #1824) hit the identical problem and solved it: it resolves Git Bash rather than probing /bin/sh, and then hard-asserts that a shell was found. If no shell can be resolved, the suite goes red, not green.

FIDO verified this by force during review of #1832 — forcing the no-shell path produced 13 RED, a loud guard failure rather than a silent skip.

That is the correct shape and it is already in the tree. This issue is to apply it to gh-aw-quality.test.ts.

Proposed fix

1. Replace the existsSync('/bin/sh') probe with the Git Bash resolution helper used by gh-aw-command-parse.test.ts (extract it to a shared test util rather than copying — a third parallel implementation is how this class of bug spreads).
2. Remove skipIf entirely. Hard-assert a shell was resolved so absence reports red.
3. Confirm the 13 tests actually pass on Windows once they run. They may not — they have never executed in this environment, so treat any failure as a real, previously-hidden defect rather than a porting nuisance.

Acceptance criteria

1. gh-aw-quality.test.ts executes all 13 previously-skipped tests on Windows, or fails loudly if no POSIX shell can be resolved. No silent skip path remains.
2. Proven capable of failing: with the shell resolution forced to fail, the suite reports red and the message names the missing dependency. A status-only assertion is not sufficient — per fix(gh-aw): fail loudly when /squad parses no command (#1824) #1832's finding, a generic-diagnostic mutation passes a status-only check clean.
3. Reported test count for the file increases by 13 on Windows. State the before/after counts as measured, not estimated.
4. No remaining skipIf-on-environment-probe in the gh-aw test suites — or, for any that remain, an explicit comment justifying why absence is genuinely not a failure.

Provenance

Found by Procedures during #1824 (PR #1832), out of scope there and deliberately left alone. Original report said "102 skipped assertions"; FIDO re-measured it as 13 tests / 28 expects — a counting-convention discrepancy, not a mechanism error. The mechanism is confirmed real. Numbers in this issue are FIDO's measured figures.