test: Playwright e2e suite (auth, chat, projects, tabular reviews, workflows)

[amal66]

Summary

Ports the amal66/mike fork's Playwright end-to-end suite onto upstream's backend/ + frontend/ layout: 27 tests in 7 files covering auth flows, the critical path (create project → upload PDF → ask a question → streamed response), chat management, project management, tabular reviews, and workflows/account settings.

Changes

• e2e/ — auth.setup.ts (idempotently creates a confirmed e2e@mike.local user via the Supabase admin API, reading the service-role key from the environment or backend/.env, and saves Playwright storageState), auth-flows, critical-path, chat-management, project-management, tabular-reviews, workflows-account specs, and fixtures/test.pdf.
• playwright.config.ts — single worker (the suite shares one backend and one test user; concurrency races on shared list views), a setup project that runs before a chromium project reusing the saved auth state. The webServer block is adapted from the fork's monorepo command (npm run dev --workspace apps/web) to upstream's layout: backend npm run dev (health-checked on :3001/health) and frontend npm run dev (:3000), both with reuseExistingServer.
• Root package.json (upstream has none) with @playwright/test, typescript dev-deps and test:e2e* scripts, trimmed from the fork's root manifest; root tsconfig.json scoped to e2e/ + playwright.config.ts so npx tsc --noEmit typechecks the suite; .gitignore entries for Playwright artifacts and e2e/.auth/ (session tokens).

No app-code changes: the specs select exclusively by ARIA role/accessible name and placeholder text — none use data-testid, so no test-id attributes needed adding to frontend/.

Why

Upstream has no end-to-end coverage; unit tests can't catch a broken login redirect, an upload pipeline regression, or an SSE stream that stops rendering. This suite pins the user-visible critical paths.

Testing

• npx playwright test --list enumerates 27 tests in 7 files with no collection errors.
• npx tsc --noEmit (root tsconfig covering e2e/ + the config) passes.
• Backend npm install + npm run build green from clean (this branch does not touch backend/ or frontend/).
• NOT run against a live stack — that needs a running Supabase with service-role credentials. Two known gaps for whoever first runs it live, stated plainly:
  1. A string-level audit of every literal selector in the specs against frontend/src shows 8 of 45 accessible-name/placeholder strings do not appear in the current UI source (+ Create New, Add Subfolder, Ask a question about your documents..., New review, Workflow name, Demo (no key needed), Demo mode; test.pdf/test.txt are runtime filenames and fine). The specs were written against an earlier UI iteration — those selectors will need updating to the current accessible names before the affected tests pass.
  2. The critical-path chat test selects the fork's keyless demo model, which does not exist upstream — running it here requires a configured provider API key and a selector update.

Provenance

All added lines are mechanical ports of amal66/mike@origin/main (b3166dd); exceptions: webServer commands rewritten from monorepo workspaces to backend//frontend cwds (backend entry added per upstream's two-process dev setup); apps/api→backend and apps/web→frontend path rewrites in auth.setup.ts and comments; root package.json/tsconfig.json are new files trimmed from the fork's root manifest and tsconfig.base.json because upstream has no repo-root toolchain to host the Playwright dep; three inert { timeout: 60_000 } test-details objects dropped in project-management.spec.ts (not part of Playwright's TestDetails type and ignored by the runner — removing them changes nothing at runtime and lets the suite typecheck); .gitignore entries copied from the fork's.

🤖 Generated with Claude Code

https://claude.ai/code/session_01CEguyEgXa9JjCciXCcVemC

CI: running e2e as a merge gate (added)

This PR now ships the workflow that makes the suite actually run and block, not just exist:

• .github/workflows/e2e.yml — on every pull_request into main (and upstream-main, the fork mirror) plus manual workflow_dispatch, the e2e / playwright job boots a full stack — MinIO object storage, local Supabase (Auth + Postgres, loading backend/schema.sql), the backend API (:3001) and the Next.js web app (:3000) — then runs npx playwright test and uploads the HTML report + traces on pass or fail. playwright.config.ts already sets webServer: undefined when CI=true, so the job owns the stack.
• docs/e2e-ci.md — the required secret and how to make the check merge-blocking.

Optional secret

• ANTHROPIC_API_KEY — the 4 LLM-dependent specs (chat + critical-path) skip without it (e2e/llm.ts), so a keyless run is green on the other ~23. Set it to also run/enforce those 4.

Making it block merges (branch protection — a repo setting, not code)

The workflow reports pass/fail on its own, but GitHub only blocks the merge when the check is required. After it's been green a few times:

1. Settings → Branches → Add/Edit rule for main.
2. Enable Require status checks to pass + Require branches to be up to date.
3. Add e2e / playwright (appears once it has run on a PR); recommended alongside backend and license/cla.

CLI equivalent and local-run instructions are in docs/e2e-ci.md.

Update — suite is green in CI ✅

e2e / playwright now passes: 23 passed, 4 skipped, 0 failed in ~7 min on a keyless run. Getting there surfaced (and fixed) three pre-existing issues, each verified end-to-end against a local stack:

1. schema.sql lags the migrations. GET /workflows/:id queries workflow_open_source_submissions, a table only a migration creates → 500 that broke the workflow specs. The workflow now applies every dated migration on top of schema.sql (idempotent — 0 errors on a fresh DB), grants service_role after (schema.sql revokes client grants assuming a hosted Supabase), and reloads PostgREST.
2. next dev made the suite flaky. On-demand route compilation (slow first hit → waitForResponse timeouts) and the dev hydration-error overlay (injects nextjs__container_errors DOM that pollutes text locators — getByText('All') matched the overlay's "Call Stack") caused intermittent failures. The workflow now serves a production build (next build + next start).
3. LLM specs need a model key. The 4 message-sending specs now test.skip(!process.env.ANTHROPIC_API_KEY) (e2e/llm.ts), so keyless runs (local, fork PRs) stay green and the LLM specs run only when the secret is set.

Merge-blocking still requires enabling branch protection on e2e / playwright (repo setting) — see docs/e2e-ci.md.