kandev/docs/plans/kubernetes-executor-ux-polish/task-03-responsive-e2e.md

id, title, status, wave, depends_on, plan, spec

id	title	status	wave	depends_on	plan	spec
03-responsive-e2e	Responsive E2E	complete	3	01-task-disclosure 02-profile-status-hierarchy	plan.md	../../specs/kubernetes-executor/spec.md

Task 03: Responsive E2E

Acceptance

• Desktop E2E proves a Kubernetes task disclosure is structured, a manual refresh stays busy until its exact read settles, updated Pod facts publish, and no duplicate read/reset control appears.
• Desktop settings E2E proves direct profile navigation exposes cluster test, executor-wide sessions, and Edit cluster connection before profile fields.
• Pixel 5 E2E proves Drawer capability parity, 44 px actions, content-sized YAML, one vertical page scroll owner, and no document horizontal overflow.

Files likely touched

• apps/web/e2e/tests/settings/kubernetes-task-environment.spec.ts (new)
• apps/web/e2e/tests/settings/mobile-kubernetes-task-environment.spec.ts
• apps/web/e2e/tests/settings/kubernetes-executor.spec.ts
• apps/web/e2e/tests/settings/mobile-kubernetes-executor.spec.ts
• docs/plans/kubernetes-executor-ux-polish/task-03-responsive-e2e.md
• docs/plans/kubernetes-executor-ux-polish/plan.md

Dependencies

Tasks 01 and 02.

Parallelism

sequential. These scenarios validate the integrated UI and must run only after both production surfaces settle.

Inputs

• All user-visible scenarios added to the Kubernetes executor spec on 2026-08-31.
• Existing isolated backend fixture, mobile Pixel 5 project, assertNoDocumentHorizontalOverflow, and causal request-wait patterns.
• Existing mobile task-environment seeding helper and Kubernetes settings specs; no real cluster is required for these frontend behavior checks.

TDD sequence

1. Add the desktop disclosure spec with a deferred exact status route and confirm the current Refresh busy/update assertions fail for the diagnosed reason.
2. Extend desktop/mobile profile specs to assert first-section order, executor-wide sessions, current-draft diagnostic payload, connection route, YAML grow/shrink geometry, touch targets, and overflow.
3. Run fresh built artifacts with retries disabled in both projects. Preserve Playwright artifacts for any failure; remove only task-created temporary capture files after recording evidence.
4. Record exact discovery/pass counts, per-spec artifacts, backend/frontend teardown evidence, and synchronize task/plan status.

Verification

cd apps && pnpm install --frozen-lockfile && pnpm --filter @kandev/web build:e2e && pnpm --filter @kandev/web e2e:raw --project=chromium e2e/tests/settings/kubernetes-task-environment.spec.ts e2e/tests/settings/kubernetes-executor.spec.ts --retries=0 && pnpm --filter @kandev/web e2e:raw --project=mobile-chrome e2e/tests/settings/mobile-kubernetes-task-environment.spec.ts e2e/tests/settings/mobile-kubernetes-executor.spec.ts --retries=0 && cd web && pnpm run e2e:sleep-ratchet && pnpm run typecheck && git diff --check

Output contract

Report initial RED behavior, final desktop/mobile discovery and pass counts, artifact paths, exact commands, cleanup/teardown evidence, blockers, risks, and synchronized task/plan status. Confirm the isolated E2E backend used random ports and that the main Kandev instance on :9998 was untouched.

Results

• RED: the first managed desktop refresh run reached the updated restart value but could not select it because the summary lacked a stable restart-count hook. After adding that semantic hook, the integrated desktop run exposed a test-order mistake: the spec dirtied the profile before exercising the connection action, so the real navigation guard correctly kept it on the profile. The action assertion was moved before the draft edit; production navigation behavior did not change. Failure evidence was captured under apps/web/e2e/test-results/settings-kubernetes-task-e-6d1eb-s-the-refreshed-Pod-summary-chromium/ and apps/web/e2e/test-results/settings-kubernetes-execut-eefac-ection-and-profile-settings-chromium/ before the managed runner replaced the artifacts on the final green runs.
• GREEN desktop: pnpm e2e:run --no-build tests/settings/kubernetes-task-environment.spec.ts tests/settings/kubernetes-executor.spec.ts -- --retries=0 discovered and passed 3/3 Chromium tests in 21.6 seconds. It proves the structured Pod summary, in-flight refresh join with no duplicate request, updated restart fact, no reset action, leading profile cluster section, current-draft diagnostic payload, executor-wide sessions, member boundary, and canonical connection navigation.
• GREEN mobile: pnpm e2e:run --no-build --project mobile-chrome tests/settings/mobile-kubernetes-task-environment.spec.ts tests/settings/mobile-kubernetes-executor.spec.ts -- --retries=0 discovered and passed 4/4 Pixel 5 tests in 22.4 seconds. It proves Drawer parity, 44 px Pod copy/Refresh/settings and profile controls, member-readable sessions, YAML grow/shrink, no vertical textarea scrollbar, contained horizontal YAML scrolling, and no document overflow.
• pnpm run build:e2e completed from the final production sources in 8.10 seconds. pnpm run e2e:sleep-ratchet, focused ESLint, pnpm run typecheck, and git diff --check passed. The final Playwright result is { "status": "passed", "failedTests": [] }.
• The managed runners used random isolated worker ports. No current-run backend, Vite, Playwright, or worker-port listener remained after completion. The five-day-old disposable seed instance on 47646/48646 was deliberately preserved, and the main Kandev process listening on :9998 was never stopped, restarted, or reconfigured. No real Kind cluster was provisioned.
• Blockers: none. External side effects: only disposable E2E fixture data, cleaned by the managed runner.