How would you architect a large Cypress test suite for a multi-team monorepo?

A monorepo with five product teams running Cypress runs into three problems quickly: duplicated setup, divergent conventions, and a CI bottleneck. The architecture that scales:

1. Single Cypress install at the root. Avoid per-package installs — they explode node_modules size, complicate CI caching, and create version drift. One cypress dependency, one config file, one runner.

2. Per-area folders inside cypress/e2e/ — cypress/e2e/auth/, cypress/e2e/checkout/, cypress/e2e/admin/. Code-owners (CODEOWNERS) per folder so the right team is paged for failures.

3. Shared support package. A @org/cypress-support package (still in the same monorepo) exposes the custom commands, fixtures helpers, page object bases, and CI utilities every team needs. Each team imports from it; nobody re-implements cy.loginAs.

4. Per-area config overrides. A base cypress.config.ts that extends with area-specific settings:

// cypress.config.ts (base)
export default defineConfig({
  e2e: {
    baseUrl: 'http://localhost:3000',
    specPattern: 'cypress/e2e/**/*.cy.ts',
  },
  retries: { runMode: 2, openMode: 0 },
});

// cypress.checkout.config.ts (area override)
import base from './cypress.config';
export default defineConfig({
  ...base,
  e2e: { ...base.e2e, specPattern: 'cypress/e2e/checkout/**/*.cy.ts' },
  env: { ...base.env, paymentSandbox: 'https://sandbox.example.com' },
});

5. Path-based CI selection. GitHub Actions / Buildkite gates: a PR touching /checkout/ only runs cypress/e2e/checkout/. A PR touching shared libs runs everything. Tools like nx affected or Bazel can compute this for you; otherwise hand-write the rules.

6. Sharded execution per area. Even within an area, shard across runners by spec to keep wall time under 10 minutes. Cypress Cloud or duration-balanced sharding both work.

7. Quality bars enforced in CI. No .only in committed specs (ESLint rule). Flaky tests must be quarantined within 7 days or deleted. New specs require code-owner approval. Each area publishes its escape rate quarterly.

The senior signal: you know the technical setup is the easy part — the harder part is enforcing standards across teams and giving each team enough autonomy without divergence.