chore/playwright mode test plan (#778)

* ci: make standalone playwright manual and document e2e gating

* docs(test): refine playwright plan with implemented CI strategy

---------

Co-authored-by: Stefan Stidl <stefan.stidl@ffg.at>

Author: sstidl

.github/workflows/docker-publish.yml

@@ -24,6 +24,8 @@ env:
   IMAGE_NAME: ${{ github.repository }}
 
 jobs:
+  # Keep e2e in this workflow so Docker image build/push is hard-gated by test success.
+  # Do not move automatic e2e back to playwright.yml, otherwise master pushes run duplicate e2e jobs.
   e2e:
     runs-on: ubuntu-latest
     timeout-minutes: 45

.github/workflows/playwright.yml

@@ -1,9 +1,9 @@
 name: Playwright E2E
 
+# Manual-only by design:
+# - Automatic e2e runs that gate Docker image builds live in docker-publish.yml.
+# - Keeping this workflow manual avoids running the same e2e suite twice on master pushes.
 on:
-  push:
-    branches:
-      - master
   workflow_dispatch:
 
 jobs:

tests/PLAYWRIGHT_MODES_PLAN.md

@@ -3,6 +3,12 @@
 ## Objective
 Build a deterministic Playwright test suite that validates LibreSpeed behavior across all supported deployment modes and UI design modes, without asserting real network throughput values.
 
+## Current Status
+- Phase 1 is implemented and passing in Chromium.
+- Added regression coverage for the classic standalone "No servers available" issue path.
+- Docker image workflow is hard-gated by e2e (`build` depends on `e2e`).
+- Standalone Playwright workflow is manual-only to avoid duplicate e2e runs.
+
 ## Modes to Cover
 
 ### Docker runtime modes
@@ -68,11 +74,12 @@ Expectations:
 ## Playwright Architecture
 
 ### Files
-- `playwright.config.ts`
-- `tests/e2e/modes.spec.ts` (runtime-mode smoke)
-- `tests/e2e/design-switch.spec.ts` (classic/modern/switch overrides)
-- `tests/e2e/helpers/env.ts` (base URLs + mode metadata)
-- `tests/e2e/helpers/ui.ts` (shared selectors, start/abort helpers)
+- `playwright.config.js`
+- `tests/e2e/modes.spec.js` (runtime-mode smoke)
+- `tests/e2e/design-switch.spec.js` (classic/modern/switch overrides)
+- `tests/e2e/classic-standalone-regression.spec.js` (revert regression guard)
+- `tests/e2e/helpers/env.js` (base URLs + mode metadata)
+- `tests/e2e/helpers/ui.js` (shared selectors, start/abort helpers)
 
 ### Environment boot
 - `docker compose -f tests/docker-compose-playwright.yml up -d --build`
@@ -108,8 +115,9 @@ Use role/text selectors anchored on stable labels and IDs already in pages; avoi
   - Mitigation: maintain per-design helper selectors with minimal coupling.
 
 ## CI Proposal
-- Trigger on PR + main branch
-- Build test image once, run mode services in parallel ports
+- Docker workflow runs e2e first, then build/push only if e2e passes
+- Standalone Playwright workflow is `workflow_dispatch` only for manual branch runs
+- Keep a single automatic e2e path to avoid duplicate runs
 - Playwright retries: `1` in CI, `0` locally
 - Upload traces/screenshots on failure only
 - Browser scope for v1: Chromium only
@@ -118,3 +126,4 @@ Use role/text selectors anchored on stable labels and IDs already in pages; avoi
 1. Browser scope for v1: Chromium only.
 2. Telemetry checks are deferred to Phase 3.
 3. `backend` mode tests assert backend endpoint contracts only.
+4. Automatic e2e gating lives in Docker workflow; standalone Playwright workflow is manual.