WP-19: the Files section still described the originally-planned CI approach
(background servers with & + wait-on); the actual implementation (documented in
ci.yml's own comment and playwright.config.ts) moved server lifecycle into
Playwright's webServer to fix the 2-hour e2e hang. Recorded as a Deviation entry
and corrected the Files section to match reality. Spot-checked the rest of the WP
(retryClicked wiring, the currentScenario() claim, the still-open retry gaps in
registration-detail/aanvraag-detail/brief) against current code — all still hold.
Session protocol: replace the removed model mandate with per-step guidance —
Opus for planning, Sonnet for implementation, Haiku subagents for simple
read-only CLI checks — so cost matches the step's difficulty, not the whole WP.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
One behavior per test across FE machine/store specs and backend endpoint
tests, so a failure names exactly what broke.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>