dorfteich/apps/web/e2e/README.md
Claude Fable 5 0875e2a087
All checks were successful
CI / Build container images (push) Has been skipped
CI / Lint, typecheck, test (push) Successful in 2m57s
CI / Import/export fidelity gate (push) Successful in 46s
CD / Build and push images (push) Successful in 3m16s
CD / Deploy to Test (push) Successful in 8s
CI / Auth e2e pack (push) Successful in 4m8s
CD / Smoke tests against Test (push) Successful in 1m9s
CD / Promote to Int (push) Successful in 10s
Add the sandbox host runtime for plugin iframes (#73)
Implements the security core of the plugin system: code-plugin surfaces
run in opaque-origin iframes (sandbox="allow-scripts", never
allow-same-origin) with a capability-filtered RPC bridge.

- api: serve a per-plugin sandbox frame document at
  /plugins/:id/:version/frame with a CSP that pins every load to the
  plugin's own asset path (built from APP_BASE_URL, not the request Host,
  so a Host-rewriting proxy cannot break it) and forbids network access
  (connect-src 'none'). Plugin assets get Access-Control-Allow-Origin: *
  so the null-origin frame can load its own module bundle.
- web: sandbox-host creates the frame, wires the SDK host bridge over a
  source-filtered postMessage transport, drives render under a 5 s
  deadline (hung/failed plugin -> placeholder, never a frozen page), and
  tears down on unmount. PluginFrame/PluginPreviewPage surface it; the
  built-in ui.resize handler clamps plugin-requested heights.
- plugin-sdk: host bridge reports gate violations via onViolation and
  registers a gated handler for every v1 method, so an undeclared
  capability is rejected with capability_not_permitted (not
  unknown_method).
- tests: SDK gate unit test; web sandbox unit tests (opaque origin,
  source filtering, timeout); and the e2e security pack with a permanent
  malicious fixture plugin proving no escape (DOM/cookies/storage/fetch/
  undeclared capability all blocked) plus well-behaved and hung cases.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EwZ4jR4KFAPvpjWevfUGX1
2026-07-11 09:16:30 +02:00

9.2 KiB
Raw Permalink Blame History

End-to-end tests

Playwright suites, most local-only — three run in CI/CD:

Suite Target Where it runs
smoke.spec.ts any deployed stage CD pipeline against https://test.dorfteich.cloud after every deploy
auth.spec.ts full local stack with Mailpit CI job auth-e2e on every PR/push; locally against the dev stack
content.spec.ts full local stack same CI job auth-e2e (a second step), right after the auth pack
everything else full local stack locally only — editor/sidebar/image/link/markdown/trash.spec.ts

content.spec.ts is the M2 content regression pack (issue #32): page lifecycle, editor basics, image paste, Markdown round-trip, and trash — enough to catch a regression across the whole content model without re-running every edge case the feature-specific packs above already cover. Its Markdown round-trip test is a real regression pin, not just a smoke check: it compares the seeded "Every Element" fixture page's exported Markdown byte-for-byte against the checked-in content-page.md (see "Content fixtures" below) — any schema/serializer change that alters how a node round-trips fails it, once the seed has re-run against the changed code (build → migrate → seed → test, exactly CI's order).

Running locally

# 1. Stack: database + Mailpit, api (3001), web dev server (5173)
docker compose -f deploy/compose/docker-compose.yml -f deploy/compose/compose.dev.yml up -d db mailpit
DATABASE_URL=postgresql://dorfteich:dorfteich@localhost:5434/dorfteich pnpm --filter @dorfteich/api db:seed
DATABASE_URL=postgresql://dorfteich:dorfteich@localhost:5434/dorfteich PORT=3001 pnpm --filter @dorfteich/api start:dev &
pnpm --filter @dorfteich/web dev &

# 2. Tests
E2E_BASE_URL=http://localhost:5173 E2E_MAILPIT_URL=http://localhost:8025 pnpm --filter @dorfteich/web e2e

auth.spec.ts skips itself when E2E_MAILPIT_URL is unset, so the CD smoke run never trips over it.

Fixture matrix

Seeded by pnpm --filter @dorfteich/api db:seed (idempotent — re-running never duplicates). Shared password: fixture passwort 123. Fixtures exist only on dev machines and disposable CI/Test databases.

Username State Purpose
fixture-admin active, Site Admin admin UI/permissions cases
fixture-user active regular journeys, settings, sessions
fixture-editor active second regular account for the collab-permissions pack (reader/editor of another's pond)
fixture-viewer active signed-in non-member for authenticated/public access-rule cases (issue #55)
fixture-outsider active the "foreign user" of the permission matrix — member of nothing (issue #60)
fixture-pending e-mail not verified unverified-login cases

Permission matrix (permission-matrix.spec.ts, issue #60)

The cross-feature permission hardening pack pins the security-relevant subject × surface combinations against regressions. It is API-level (the UI adds nothing over the resolved status code) and enforces the 404-vs-403 policy: an unauthorized read is 404 (existence hidden), an unauthorized write on something readable is 403.

  • Subjects: site admin (fixture-admin), pond admin / owner (fixture-user), editor (fixture-editor), the same editor label-restricted by a secret-label deny, reader (fixture-viewer), public (anonymous), and the foreign user (fixture-outsider, a member of nothing).
  • Surfaces: page read, edit (collab-token rw/ro), sidebar list, search, versions (history = write), media, and the public HTML endpoint.
  • Extending it: a new permission-touching feature adds a surface here (one expect row per subject) rather than a bespoke test, so the matrix stays the one place the policy is pinned. A weakened guard is caught here — verified by temporarily loosening a route decorator and watching the pack go red.

Plugin sandbox (plugins.spec.ts, issue #73)

The sandbox security pack drives the Site-Admin plugin preview page (/admin/plugins/:id/preview) — the exact sandbox runtime pages embed — with three fixture plugins built in plugin-fixtures.ts and installed through the admin API:

  • well-behaved: answers render and resizes its own frame via the declared ui capability;
  • malicious (the permanent security regression asset, ADR 0008): probes the parent DOM, cookies, localStorage, same-origin and external fetch, and an undeclared capability — every probe must report blocked. A LEAKED verdict is a sandbox escape and fails the build;
  • hung: never answers, so the host's 5 s deadline must collapse it to the failure placeholder while the surrounding page stays responsive.

Runs in the auth-e2e CI job (needs the api's writable PLUGINS_DIR, satisfied by the default ./data/plugins).

Attachments (attachments.spec.ts, issue #61)

Non-image attachments: a page's attachments section uploads an allowlisted file, lists it, and inserts it into the document as a download link (verified to serve with Content-Disposition: attachment + nosniff, never inline); a disallowed extension is rejected with the localized allowlist error; the Pond Admin file manager reports storage usage and flags an orphan (a pond-level upload with no embedding page). The SVG sanitize/reject policy is covered at the api level in files.e2e.db.test.ts.

Import (import.spec.ts, issue #64)

The sidebar "import document" action: pick a file, upload, watch progress, open the new page. .md imports directly (the response is already succeeded); .docx/.odt poll a conversion job. Cases: a .docx corpus fixture (#63) opens the converted page, a .md opens directly, an unsupported .txt shows the localized error and creates no page, and two concurrent .md imports both complete. The .docx case self-skips unless E2E_PANDOC is set — CI's e2e stack has no reachable pandoc sidecar (jobs are container-networked; same reason the api's real-pandoc fixtures test skips in CI, #63), so it runs locally / on a stage. Run it locally with a sidecar reachable at the api's PANDOC_URL:

docker run -d -p 3030:3030 pandoc/core:3.6 server   # api PANDOC_URL → this
E2E_PANDOC=1 E2E_BASE_URL=http://localhost:5990 \
  pnpm --filter @dorfteich/web exec playwright test e2e/import.spec.ts

Export (export.spec.ts, issue #65)

The pond-settings "Download pond as ZIP" link (a Markdown ZIP of the readable pages) and the page-menu office export. The ZIP download needs no sidecar; the .docx export runs a conversion job and, like the import .docx case, self-skips unless E2E_PANDOC is set. The permission-filtered ZIP contents, the docx pandoc output, and the 500-page streaming path are covered at the api level in export.service.db.test.ts (+ export-markdown.test.ts).

Content fixtures

db:seed also creates a shared pond content-fixtures (owned by fixture-user) with two pages, for the content regression pack and manual QA:

  • Every Element (every-element) — every editor schema node and mark (issue #24: headings 14, all list types, table, blockquote, code block, horizontal rule, hard break, and all five marks). Loaded from the checked-in apps/api/prisma/fixtures/content-page.yjs, a Yjs snapshot generated from the human-readable content-page.md next to it — content-page.md is the thing to read or edit; the .yjs file is a build artifact of it, not source.
  • Fixture Image (fixture-image) — one real, servable uploaded image (the placeholder fileId inside the Markdown fixture above is not a real attachment; this page's image is).

Regenerating after editing content-page.md:

pnpm --filter @dorfteich/api fixtures:regenerate

This is deterministic — re-running without editing the Markdown produces a byte-identical .yjs file (the script pins the Yjs document's clientID, which is otherwise randomized per Y.Doc instance) — and it refuses to write a snapshot that isn't a fixed point of the Markdown round-trip (docToMarkdown(markdownToDoc(x)) === x), so a stale fixture can't get checked in silently.

Conventions

  • New feature packs get their own <feature>.spec.ts next to these and extend the fixture matrix here (permission matrix arrives with M5, issue #60).
  • Use contextForUser() from helpers.ts for signed-in tests — it logs in through the api and hands you a browser context with the session cookie, no UI login repetition.
  • Flaky tests are defects (ADR 0014): fix or quarantine immediately.