Make mobile action logging tolerate Jest server-config mocks and align the generated UI preview manifest with Biome so validation jobs stay green.

Align the CI wait-for-port steps with the actual web and Metro ports used by the local E2E startup scripts so web and mobile E2E jobs can reach their servers.

Reuse the externally started Playwright server in CI and switch Android emulator workflows to a supported device profile so both E2E jobs can launch reliably.

Stop asking the GitHub emulator action to disable Android animations, because that runner-managed settings step is what crashes after boot in CI.

Use a lighter Android 30 x86 emulator in CI so the mobile E2E runner can boot reliably on Linux hosts without hardware acceleration.

Follow the android-emulator-runner Linux guidance by enabling KVM before boot, restoring the default no-snapshot behavior, and giving the emulator more time to come up on hosted runners.

Keep the researched KVM setup, but switch the Android 30 emulator image to x86_64 because the hosted SDK channel does not provide the default x86 image.

Use the official Gradle action in the Android-backed CI jobs so Expo's prebuild and run:android steps spend less time re-downloading dependencies on each workflow run.

Make the Maestro prep script tolerate package-qualified test IDs and slower dev-client startup so it recognizes the sign-in screen instead of failing before flows begin.

Pass the Android device option after the pnpm script separator so Maestro receives a real flow path instead of treating --device as the file to run.

Add a compatibility bootstrap flow that delegates to the shared Expo dev-client setup so older mobile E2E flows can resolve their reusable include again.

Load local Maestro fixture files only for missing variables so checked-in local accounts do not override seeded CI credentials during mobile E2E runs.

Give post-login mobile tRPC requests more time to complete so CI device flows do not surface transient aborted profile loads immediately after sign-in.

Skip the dev-client reset/bootstrap when the app is already on a usable screen so back-to-back mobile E2E flows do not fail during repeated prepare steps.

Wait for the authenticated tab button test ID instead of the visible Plan label so mobile E2E login does not depend on tab text rendering timing.

Increase the mobile tRPC timeout again so cold-start authenticated queries can finish in CI before the client aborts and blocks the post-login tabs flow.

Keep the mobile insight timeline window within the backend's inclusive 365-day limit so the Plan screen does not throw and block Maestro tab navigation in CI.

Split the calendar UI into focused components, expand Maestro journey coverage, and add shared Worktrunk configuration so parallel agent worktrees are easier to bootstrap and merge safely.

Tighten the coordinator around Worktrunk-first delegation and standardize spec-scoped branch naming so parallel worktree sessions are easier to create, route, and merge safely.

…gration

Major refactor to move GradientPeak toward T3 Turbo-style architecture:
- Migrate auth from Supabase to Better Auth for web and mobile
- Move database operations to Drizzle with packages/db as sole relational authority
- Remove ctx.supabase bridge from packages/api, migrate to repository seams
- Add Drizzle repositories for events, integrations, iCal, Wahoo, training plans
- Clean up packages/supabase to infra-only ownership
- Update mobile auth flows to use Better Auth client seam
- Add Nodemailer mailer with env-driven SMTP/log/disabled modes
- Remove legacy Supabase-generated types and supazod schemas
- Update web auth callbacks, integrations, and settings pages
- Simplify web middleware and Supabase server client

Breaks down @repo/supabase to infra-only, removes packages/api/src/routers/auth.ts