test: establish vitest suite, CI, and coverage reporting (44%→83%)

[rrbe]

Background

The project previously had no tests and no CI — every change had to be verified by hand-loading the extension against real GitHub pages. This PR stands up the test infrastructure from scratch, wires up CI and coverage reporting, and raises coverage from ~44% to ~83%.

Overview

1. Test infrastructure

• vitest + happy-dom (same toolchain as esbuild; native TS/ESM, no extra transpile)
• happy-dom's setURL (wrapped in src/test-utils/url.ts) lets the page-detection logic switch location without touching source
• Scripts: test / test:watch / test:coverage / typecheck

2. CI

• .github/workflows/ci.yml: runs typecheck + test + build on PR/push (pinned pnpm major to match local; actions bumped to node24-compatible versions)

3. Coverage reporting

• @vitest/coverage-v8 with text/html/lcov reporters; coverage/ is gitignored
• Thresholds are a regression floor (statements 78 / branches 62 / functions 78 / lines 82), enforced only under --coverage so plain pnpm test/CI is unaffected — meant to be ratcheted up as coverage grows

4. Test coverage (173 tests / 28 files)

Mocked at the right data seam per layer, rather than one blunt approach:

Layer	Mock	Modules
feature → github-api	vi.mock(\"../lib/github-api\") with canned data; assert rendered DOM	pr-branch-names, pr-diff-stats, pr-review-status, commit-tags, commit-diff-stats, pr-approve-now, watch-fork-star (hover)
github-api → chrome.runtime	stub chrome.runtime.sendMessage	github-api.ts (9 wrappers + invalidated-context / lastError)
service-worker → fetch	mock fetch + chrome.storage	request coalescing, cache TTL, token-change cache clear, stargazers/watchers/forks, GraphQL reviews/commit-diff, tags REST fallback and GraphQL success path
pure logic / DOM / infra	driven directly	page-detect, commit-dom, diff-stats-badge, info-row(-skeleton), default-sort, release-tab, pr-collapse-expand, better-top-repos, file-age-color, pr-label-position, feature-flags, navigation, early-sort-redirect, options, content (removeFeatureElements)

5. Real bugs found and fixed while writing tests

• Disabling the watch/fork/star popup deleted GitHub's native counters: FEATURE_CLASSES listed bg-wfs-counter-wrap — the class we add to GitHub's own counters — as a removable element, so removeFeatureElements ran remove() on the native watch/fork/star counts on disable. Fixed to only remove our own bg-wfs-popup plus a new cleanupWatchForkStarPopup, guarded by an integration test.
• options.ts FEATURE_KEYS was missing feature-pr-collapse-expand: caught by the registry-sync test that keeps the three feature lists aligned.
• Also fixed pre-existing type errors that surfaced once tsc actually ran (status shadowing window.status, missing generics on chrome.storage.local.get<T>(), etc.).

Quality (not coverage for its own sake)

• Every new test asserts observable behavior (rendered DOM, return values, mock call args), and was validated by a mutation self-check: deliberately break one source line → confirm the test goes red → restore byte-for-byte.
• Fixed cross-test listener leakage + stderr noise caused by navigation.ts's non-teardownable module-level listeners.

Coverage

Metric	Before	After
Statements	44%	83%
Branches	36%	67%
Functions	50%	84%
Lines	47%	87%

Verification

• pnpm typecheck → 0 errors
• pnpm test → 28 files / 173 tests pass
• pnpm test:coverage → meets thresholds (exit 0)
• pnpm build → clean, no test files leak into dist/

Still lower / optional follow-up

watch-fork-star-popup 56%, release-tab 74%, pr-collapse-expand / content ~72% — the remainder is mostly GitHub's "new experience" DOM branches with diminishing returns; not forced in this PR.

🤖 Generated with Claude Code