playwright tests

Situation

The repo had Playwright e2e specs and a store app already instrumented with PostHog (@rimo/ui provider, session replay). CI against Vercel previews/production was not wired end-to-end: no automated runs on deployment URLs, no durable link between a failing test and analytics/replay, and no Linear triage—so repeated failures on every commit would either be ignored or create noisy duplicate tickets.

Task

1. Run Playwright reliably against Vercel (preview and production), not only localhost.
2. Tie each Playwright run to a relevant PostHog identity so failures can be traced to session replay (and related person/search URLs in CI).
3. Integrate Linear so failures become trackable work, without opening a new issue on every commit when the same test keeps failing.
4. Orchestrate reporting and integrations in one place—global teardown after tests finish.

Action

CI / Vercel + GitHub Actions

• Taught Playwright to skip the local dev server when PLAYWRIGHT_BASE_URL is a non-local origin (Vercel URL), while keeping optional PLAYWRIGHT_WEB_SERVER=0.
• Tuned CI behavior: longer timeouts for cold previews, GitHub + HTML reporters, JSON output for downstream processing.
• Added two workflows triggered by deployment_status: preview (non-Production) vs Production, checkout at deployment.sha, optional vercel inspect on the deployment URL, then bun run e2e with secrets; artifacts include HTML report and e2e/output/.

PostHog correlation

• Global setup generates a per-run runId + distinctId and writes e2e/.e2e-run-context.json.
• Shared fixtures inject window.__PLAYWRIGHT_E2E__ via browser.addInitScript so every browser context bootstraps the same run identity before React.
• PostHog provider uses bootstrap.distinctID + posthog.register for e2e_run_id / e2e_source so events and replays are filterable in PostHog.
• After each failed / timed-out / interrupted test, capture posthog.get_session_id() (when available) and append JSON lines to e2e/output/failure-sessions.jsonl keyed by Playwright testId.