test: cover useFocusTrap scroll lock, tab cycling and focus restore

[kubestellar-hive]

Test Improvement

src/components/hooks/useFocusTrap.js is the shared accessibility primitive behind every modal dialog on the site — used by src/components/MemberDirectory/index.js and src/components/CommunityPeople/index.js. Before this PR it had no coverage of any kind: src/ was absent from the node --test --experimental-test-coverage report entirely, because no test could load anything under src/.

Every behaviour it owns is a silent failure mode — a broken scroll restore leaves the page unscrollable after a dialog closes, a broken focus wrap lets keyboard users tab out of a modal into inert content, a broken focus restore strands screen-reader users. Nothing in the build, prettier, or the validator scripts would catch any of it.

Files claimed by this PR (all new, all test-only — no production code, no dependency, no package.json, no .github/workflows/**):

• tests/tools/react-hook-driver.mjs — a renderHook() that swaps React's per-render dispatcher slot so useRef/useEffect resolve against a small local implementation, then restores the slot synchronously. It honours effect dependency arrays: rerender() re-runs an effect, previous cleanup first, only when deps actually change. Because the slot is restored immediately and no module specifier is aliased process-wide, this cannot disturb other tests that import React for real — notably [quality] test: cover src/components/ArchitectureFilters filterArchitectures and add a JSX import path (tests/tools/jsx-hooks.mjs + tests/helpers-jsx.mjs + tests/architecture-filters.test.mjs) #229's.
• tests/tools/fake-dom.mjs — a stand-in for only the DOM surface this hook touches: body.style.overflow, activeElement, add/removeEventListener('keydown'), element.focus() and querySelectorAll('button, a[href]'). Deliberately tiny, so the hook's real contract with the DOM is explicit rather than buried in a DOM library.
• tests/use-focus-trap.test.mjs — 17 tests: stable ref identity across renders; scroll lock and restore of the previous overflow value; focus moved to the close button on mount; unattached closeRef tolerated; Escape invoking onClose; other keys ignored; Shift-Tab wrapping first→last; Tab wrapping last→first; Tab mid-dialog left to the browser; a lone focusable element trapping onto itself both ways; Tab ignored with no focusable children and with an unattached dialogRef; non-focusable descendants (<a> without href, <span>) excluded from the cycle; listener removal on unmount; focus restored to the trigger; fallback to the previously focused element; and unmount not throwing with nothing to restore to.

Not a dependency on #229

#228 deferred this hook on the grounds that it "needs a DOM environment and a React renderer." It turns out neither is required, and this PR does not depend on #229 landing: useFocusTrap.js contains no JSX and imports no CSS module, so Node imports it as-is. This adds no package and touches neither package.json nor package-lock.json, so it cannot conflict with #229 or #225 there. File sets are disjoint from every open PR: #229 owns tests/tools/jsx-hooks.mjs, #225 owns tests/tools/coverage-report.mjs, #231 owns tests/helpers.mjs — none of which is touched here. MemberDirectory, under refactor in #118, is likewise untouched.

Verification

node --test: 72 tests pass, 0 fail (55 before). With --experimental-test-coverage:

src/components/hooks/useFocusTrap.js | 100.00 | 100.00 | 100.00

Repo totals move 73.51 → 82.93 line and 47.12 → 63.53 branch. npx prettier --check is clean on all three added files.

The suite was mutation-checked rather than assumed effective. Seven independent mutations of the hook each produced failures (1–3 tests apiece): overflow not restored, preventDefault dropped, keydown listener never removed, first/last focusable swapped, Escape no longer closing, close button not focused on mount, focus not restored on unmount. src/ was restored to a clean tree afterwards and is unmodified in this PR.

Coverage evidence

• Unit: node --test --experimental-test-coverage (node v26.8.2), local clone of cncf/endusers at parent revision 00b44df after npm ci, 2026-09-17. Baseline: 55 tests, src/ absent from the report (0%, never loaded). After: 72 tests, figures above.
• End-to-end: unavailable — this repository has no end-to-end or browser suite and publishes no coverage artifact from any suite (tracked in [quality] CI publishes no coverage evidence, so coverage findings cannot be verified #186). The two sources cannot be combined, and no claim is made that this path lacks end-to-end coverage.

Deliberately out of scope

The rendered components (AwardsTimeline, CNCFProjectCard, CommunityPeople, MetricsDashboard, ProjectsBorn, ReferenceArchitectures, MemberDirectory, theme/Footer) still have no coverage. Those genuinely do need a renderer and the JSX import path from #229, and belong in their own issue.

Related Issue

Closes #267

---

Filed by quality agent (hold-gated mode). Human review required.

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

[kubestellar-hive]

Important

Held for human review by the hive's ACMM level gate.

This PR was opened by the "quality" agent while Hive policy required a human checkpoint for that agent. Non-outreach agents are held at ACMM L3–L5; the outreach agent is always held because it publishes project-facing communication.

Hive will automatically remove the hold label once current policy no longer requires a level hold for "quality". If this is an outreach PR, a human must review it and remove the label.