[vycdev]

Refs #138. It doesn't close the issue; that's left for review.

What changes

npm run test:e2e:all (scripts/run-e2e-suites.mjs) runs @smoke and then each area config one after another: docs, landing, privacy, terms, not-found, i18n, and i18n build. It runs every suite even after one fails, prints a pass/fail summary, and exits non-zero if any failed. Each suite keeps its own test-results/<suite> and playwright-report/<suite>, so a later suite doesn't delete an earlier one's traces.

Before the suites run, it checks coverage, using Playwright's --list (about 4s, no browsers). The check fails if:

• a playwright*.config.ts isn't in the suite list, or
• a test listed by the default config isn't run by any suite, unless it's tagged @full.

npm run check:e2e runs only that check.

@matrix, @stress, and @full stay out of the default run. Every @full test is one of the 12 kromacut-flow.spec.ts export cases (2 images × 3 profiles × 2 meshers). The 3888px @stress cases allow 90 minutes each, and the @matrix cases 30 minutes each. Running them on every PR would turn a ~10 minute gate into hours. The one cheap flow case (1024 PNG, 2 colors, greedy) is already tagged @smoke, so the export flow is still covered. The scripts test:e2e:matrix, test:e2e:stress, and test:e2e:full are unchanged for running the rest on demand.

Five tests weren't run by any script. The coverage check found them, and all five pass on develop. Following the existing @smoke @matrix convention, they're now tagged @smoke:

• palette-proof.spec.ts: Palette Proof round trip
• settings-legal-links.spec.ts: all 3 tests
• stack-matrix-settings.spec.ts: the adaptive thickness round trip, which was @matrix only

Fresh servers. playwright.config.ts and playwright.i18n.config.ts reuse an existing server outside CI. While testing this, my @smoke run reused a vite preview on port 5173 that belonged to another worktree. It tested that checkout's build and failed when that server restarted. test:e2e:all now sets KROMACUT_E2E_FRESH_SERVERS=1, which turns reuse off, so a port that's already in use fails clearly. Running npm run test:e2e on its own still reuses a server, as before.

Config guard. The docs, i18n, i18n-build, not-found, privacy, and terms specs call requirePlaywrightConfig() (tests/e2e/helpers/playwrightConfig.ts, next to the fake desktop helper from #139). Under any other config, a beforeAll fails with, for example:
This spec only runs with playwright.i18n.config.ts, not playwright.config.ts. Use \npm run test:e2e:i18n`.It fails instead of skipping so that a renamed config can't turn a CI suite into a silent all-skipped pass. The landing specs are tagged@smoke` and also run under the default config, so they aren't guarded.

.github/workflows/ci.yml

• Runs on PRs into develop and main, on pushes to them (base-branch runs create the npm and Chromium caches that PRs restore), and on workflow_dispatch.
• checks job: lint, build, npm test, check:e2e.
• e2e job: a parallel matrix with one job per suite and fail-fast: false. The matrix comes from run-e2e-suites.mjs --list, so CI and test:e2e:all share one suite list.
• Chromium is cached by Playwright version. A cache hit only installs the system deps.
• On failure, each job uploads its Playwright report and traces as an artifact.
• CI passed job: a single check that fails if any job failed, was cancelled, or was skipped. Branch protection can require just this check as suites are added.
• Permissions: contents: read at the top level and none for CI passed. Checkout uses persist-credentials: false. The suite name reaches the shell through env, not ${{ }}. Node 22 matches the deploy and release workflows.
• Fork PRs: it uses pull_request, never pull_request_target, so fork code gets a read-only token and no secrets. Maintainer approval for fork runs is a repository setting, not a workflow setting. I suggest Settings → Actions → General → "Require approval for all external contributors". The default only covers first-time contributors.

AGENTS.md testing guidance and CHANGELOG.md (Unreleased → Added) are updated. The desktop fakes from #139/#134 aren't touched. The only edits to specs are the guard call and the @smoke tags.

Verification (macOS)

• npm run lint, npm run build, and npm test (1105 passing after the rebase) all pass.
• npm run check:e2e passes: every e2e test except @full is covered.
• npm run test:e2e:all took 11 minutes, both before and after rebasing onto develop (which now includes test: share one fake desktop runtime across e2e specs #139 and refactor: share one decodeImage helper for image loading #141). 6 of 8 suites pass: smoke 81, docs 6, landing 19, not-found 10, i18n 17, i18n build 4.
• privacy and terms fail on develop: 4 + 4 tests in the narrow-screen footer checks expect 7 footer links, but the Awards link (Add awards page with Snapmaker Innovation Fund winner badge #126) made it 8. It's the same cause as i18n landing footer e2e tests fail after the Awards link #136, but those suites weren't covered by test: align the i18n footer test with the Awards link #137. I reported it as Privacy and terms e2e footer tests still expect 7 links after the Awards link #143 and didn't fix it here, so the privacy and terms CI jobs (and CI passed) will fail until Privacy and terms e2e footer tests still expect 7 links after the Awards link #143 lands.
• actionlint 1.7.12 reports no problems in ci.yml. shellcheck isn't installed, so the shell snippets were reviewed by hand.
• Linux CI on 13b3ef8 matches macOS: lint/build/unit tests and 6 of 8 e2e suites pass (@smoke takes about 11.5 minutes). privacy and terms fail only on the 8 Privacy and terms e2e footer tests still expect 7 links after the Awards link #143 footer tests, and their traces were uploaded as artifacts. The first run missed the Chromium cache, as expected; pushes to develop will fill it.

🤖 Generated with Claude Code