ci/web: run Playwright e2e suite on every PR

Boots the full authentik stack (postgres + Go server + Rust worker)
inside the existing ci-web workflow, applies migrations and the
test-admin user blueprint, then runs `corepack npm run --prefix web
test:e2e` against http://localhost:9000. Uploads the HTML report,
traces/videos, and authentik logs as artifacts on failure so reviewers
can debug without rerunning locally.

Also enables the HTML reporter and screenshot/video capture on CI in
playwright.config.js, and updates the full dev-environment docs to
point at the same npm scripts CI uses so local and CI runs stay in
lockstep.

Closes #21994

Co-Authored-By: Agent (authentik-i21994-better-mobile-tangelo) <279763771+playpen-agent@users.noreply.github.com>

ci/web: make test-admin blueprint self-contained

The previous blueprint used !Find to look up the authentik Admins group,
which raced against system/bootstrap.yaml and resolved to None when the
explicit apply_blueprint step ran before the worker had applied bootstrap.
The serializer rejected groups: [None] with Invalid pk "None".

Define the group in the same blueprint with state: present and reference
it via !KeyOf, so the test admin setup does not depend on any pre-existing
data. If bootstrap has already created the group, state: present is a
no-op on the identifiers; otherwise the group is created here.

Co-Authored-By: Agent (authentik-i21994-better-mobile-tangelo) <279763771+playpen-agent@users.noreply.github.com>

ci/web: format test-admin-user.yaml with prettier

Pick up the 4-space indent that web/'s prettier config enforces. The
file was added under issue #21994 with 2-space indent and tripped the
ci-web format check on push.

Co-Authored-By: Agent (authentik-i21994-better-mobile-tangelo) <279763771+playpen-agent@users.noreply.github.com>

Use parallelism.

Remove guard.

Reorder tests.

Ignore playwright-traces.

Update expected path.

Always parallel.

Flesh out types.

ci/web: post Playwright result comment + gated S3 upload + !cancelled() guards

Three reviewer-facing improvements to the e2e job:

1. Idempotent PR comment summarising Playwright pass/fail/flaky/skipped
   counts. Marker `<!-- playwright-result -->` lets re-runs edit the
   same comment instead of piling up. Skipped on fork PRs where the
   default GITHUB_TOKEN is read-only.

2. Optional S3 publish of the HTML report to
   `s3://authentik-playwright-artifacts/pr-<n>/run-<id>/attempt-<n>/`,
   gated behind `vars.PLAYWRIGHT_S3_ENABLED == 'true'`. The bucket is
   pending infra provisioning; the public URL pattern is already wired
   into the comment so flipping the variable on later requires no
   workflow changes. Borrows the OIDC + IAM role plumbing from
   `.github/workflows/release-publish.yml`.

3. Switch the failure-guarded reporting/upload steps to `!cancelled()`
   so a superseded (cancelled) run no longer emits failure-shaped noise,
   and so successful runs still produce the artifact bundle reviewers
   expect.

Adds the Playwright JSON reporter so the parse step can pull pass/fail
counts from `playwright-report/results.json` for the comment body.

Co-Authored-By: Agent (authentik-i21996-internal-achievable-raisin) <279763771+playpen-agent@users.noreply.github.com>

web/e2e: fix three regressions blocking the parallel suite

Locally and in CI the new `e2e (playwright)` job appeared to "hang"
under `fullyParallel: true` + `workers: "50%"`. The hang was actually
five tests sharing two unrelated bugs that all manifest as 30s test
timeouts; the cluster only *looks* like a parallelism issue because
multiple workers stall on the same wall-clock window. With these three
fixes the full suite is green in 1m48s on `--workers=2` (was: 5 failed
/ 17 passed in 5m30s).

1. `web/test/browser/600-providers.test.ts`
   PR #21647 dropped the `to:` argument on the `session.login()` call
   in this file's `beforeEach`. Without it, `SessionFixture.login()`
   waits for the auth-flow URL pattern to re-appear — which it does
   immediately, since we just navigated there — so the helper returns
   *before* the post-login redirect lands. The wizard buttons probed
   afterward live on `/if/admin/#/core/providers`, which the user never
   actually reaches; every test in the file then hits the 30s
   `beforeEach` timeout. Pin the destination explicitly, matching the
   shape of every other test file.

2. `web/src/admin/roles/ak-role-list.ts`
   The role-list row anchor had no aria-label, so its accessible name
   was the (random, generated) role name. `500-roles.test.ts` searches
   for that anchor with `getByRole("link", { name: "view details" })`
   — the same selector `400-groups.test.ts` uses against the group
   list, where `GroupListPage.row()` *does* set
   `aria-label="View details of group ..."`. Bring the role row to
   parity with groups; the test wasn't wrong, the UI was missing the
   accessibility hook.

3. `web/test/browser/500-roles.test.ts` ("Edit role from view page")
   The post-edit verification used `page.getByText(updatedName)`, but
   on the role view page the new name renders in two places (the
   "Role <name>" page-navbar heading and the description-list value),
   so the bare text match resolves to two elements and trips
   strict-mode. Add `{ exact: true }` so we assert the canonical value
   the edit wrote rather than the heading template.

Co-Authored-By: Agent (authentik-i21996-internal-achievable-raisin) <279763771+playpen-agent@users.noreply.github.com>

Use headless.
This commit is contained in:
Teffen Ellis
2026-04-30 21:39:16 +00:00
committed by Teffen Ellis
parent f0a544493a
commit e4ef898fec
17 changed files with 369 additions and 24 deletions

View File

@@ -12,6 +12,20 @@ on:
- main
- version-*
env:
POSTGRES_DB: authentik
POSTGRES_USER: authentik
POSTGRES_PASSWORD: "EK-5jnKfjrGRm<77"
AUTHENTIK_BLUEPRINTS_DIR: "./blueprints"
AUTHENTIK_OUTPOSTS__DISABLE_EMBEDDED_OUTPOST: "true"
# Drives the system/bootstrap.yaml blueprint at startup: creates akadmin with
# these credentials and flips the Setup flag (Setup.set(True)) so the SPA's
# post-login redirect to "/" doesn't bounce through /setup, which would 500
# because the OOBE policy refuses to run once akadmin already has a usable
# password. See authentik/core/setup/signals.py and blueprints/default/flow-oobe.yaml.
AUTHENTIK_BOOTSTRAP_EMAIL: "test-admin@goauthentik.io"
AUTHENTIK_BOOTSTRAP_PASSWORD: "test-runner"
jobs:
lint:
runs-on: ubuntu-latest
@@ -39,6 +53,239 @@ jobs:
- name: build
working-directory: web/
run: corepack npm run build
e2e:
name: e2e (playwright)
runs-on: ubuntu-latest
timeout-minutes: 60
permissions:
contents: read
# Required so the "Comment Playwright result on PR" step can update its
# marker comment via the gh CLI / REST API.
pull-requests: write
# Required so the optional "Upload HTML report to S3" step can mint OIDC
# credentials with aws-actions/configure-aws-credentials. Harmless when
# the upload is gated off.
id-token: write
steps:
- uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v5
- name: Setup authentik env
uses: ./.github/actions/setup
with:
dependencies: system,python,node,go,rust,runtime
- name: Build web UI
run: corepack npm run --prefix web build
- name: Build authentik server (Go)
run: | # shell
go build -o ./bin/authentik-server ./cmd/server
sudo install -m 0755 ./bin/authentik-server /usr/local/bin/authentik-server
- name: Build authentik worker (Rust)
run: | # shell
cargo build --release --bin authentik
sudo install -m 0755 ./target/release/authentik /usr/local/bin/authentik
- name: Apply migrations
run: uv run python -m lifecycle.migrate
- name: Resolve Playwright version
id: playwright-version
working-directory: web
run: | # shell
version=$(node -p "require('@playwright/test/package.json').version")
if [ -z "$version" ]; then
echo "Failed to resolve @playwright/test version" >&2
exit 1
fi
echo "version=${version}" >> "$GITHUB_OUTPUT"
- name: Cache Playwright browsers
id: playwright-cache
uses: actions/cache@27d5ce7f107fe9357f9df03efb73ab90386fccae # v4
with:
path: ~/.cache/ms-playwright
key: ${{ runner.os }}-playwright-${{ steps.playwright-version.outputs.version }}
- name: Install Playwright browsers
working-directory: web
run: | # shell
if [ "${{ steps.playwright-cache.outputs.cache-hit }}" = "true" ]; then
corepack npm exec -- playwright install-deps chromium
else
corepack npm exec -- playwright install --with-deps chromium
fi
- name: Start authentik server and worker
run: | # shell
set -euo pipefail
mkdir -p /tmp/ak-logs
# The Go server (authentik-server) spawns gunicorn as a child via PATH lookup
# and inherits the env that `uv run` set up for it. Verify gunicorn resolves
# under the same launcher so we fail fast here instead of waiting for an
# empty 200 from the proxy fallback later.
uv run --frozen sh -c 'command -v gunicorn' \
|| { echo "gunicorn not resolvable from uv run"; exit 1; }
uv run ak server > /tmp/ak-logs/server.log 2>&1 &
echo $! > /tmp/ak-logs/server.pid
# The Rust worker also opens an HTTP/metrics server on listen.http /
# listen.metrics (default :9000 / :9300). On a single CI host that races the
# Go server's binds and silently steals :9000, leaving Playwright talking to
# a healthcheck-only axum router that returns 200/empty for /if/* paths.
# Pin the worker to disjoint ports so the Go server keeps the public 9000.
AUTHENTIK_LISTEN__HTTP="[::]:9001" \
AUTHENTIK_LISTEN__METRICS="[::]:9301" \
uv run ak worker > /tmp/ak-logs/worker.log 2>&1 &
echo $! > /tmp/ak-logs/worker.pid
- name: Wait for authentik to be ready
run: | # shell
set -euo pipefail
# Readiness probes must verify the Go server is actually serving the request,
# not just that *something* on :9000 returned 200. The Go proxy stamps
# `X-authentik-version` on its static responses and the rendered flow page
# contains the <ak-flow-executor> custom element — both are absent from the
# worker's axum healthcheck router, so checking either rules out the
# port-collision failure mode.
timeout 240 bash -c '
until curl -fsS -o /dev/null http://localhost:9000/-/health/ready/; do
sleep 2
done'
timeout 300 bash -c '
until curl -fsS http://localhost:9000/if/flow/default-authentication-flow/ \
| grep -q "ak-flow-executor"; do
sleep 3
done'
- name: Run Playwright tests
working-directory: web
env:
AK_TEST_RUNNER_PAGE_URL: http://localhost:9000
run: corepack npm run test:e2e
# Reporting / upload steps below intentionally use `!cancelled()` rather
# than `failure()`: a cancelled run (e.g. superseded by a newer push) is
# not a real result and shouldn't produce reviewer-facing artifacts or
# comments. `if-no-files-found: ignore` keeps the "passed" case quiet.
- name: Upload Playwright HTML report
if: ${{ !cancelled() }}
uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
with:
name: playwright-report
path: web/playwright-report/
retention-days: 14
if-no-files-found: ignore
- name: Upload Playwright traces and videos
if: ${{ !cancelled() }}
uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
with:
name: playwright-traces
path: web/test-results/
retention-days: 14
if-no-files-found: ignore
- name: Upload authentik server and worker logs
if: ${{ !cancelled() }}
uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
with:
name: authentik-logs
path: /tmp/ak-logs/
retention-days: 14
if-no-files-found: ignore
- name: Parse Playwright results
id: playwright-results
if: ${{ !cancelled() }}
run: | # shell
set -euo pipefail
report=web/playwright-report/results.json
if [ ! -f "$report" ]; then
{
echo "available=false"
echo "passed=0"
echo "failed=0"
echo "flaky=0"
echo "skipped=0"
} >> "$GITHUB_OUTPUT"
exit 0
fi
{
echo "available=true"
echo "passed=$(jq -r '.stats.expected // 0' "$report")"
echo "failed=$(jq -r '.stats.unexpected // 0' "$report")"
echo "flaky=$(jq -r '.stats.flaky // 0' "$report")"
echo "skipped=$(jq -r '.stats.skipped // 0' "$report")"
} >> "$GITHUB_OUTPUT"
# The S3 publishing pair below is intentionally gated off until the
# `authentik-playwright-artifacts` bucket is provisioned by infra. Flip
# the repo variable `PLAYWRIGHT_S3_ENABLED=true` to turn it on; the URL
# baked into the PR comment below already points at the eventual key.
- name: Configure AWS credentials for HTML report upload
if: ${{ !cancelled() && github.event_name == 'pull_request' && github.event.pull_request.head.repo.full_name == github.repository && vars.PLAYWRIGHT_S3_ENABLED == 'true' }}
uses: aws-actions/configure-aws-credentials@ec61189d14ec14c8efccab744f656cffd0e33f37 # v6.1.0
with:
role-to-assume: "arn:aws:iam::016170277896:role/github_goauthentik_authentik"
aws-region: eu-central-1
- name: Upload HTML report to S3
if: ${{ !cancelled() && github.event_name == 'pull_request' && github.event.pull_request.head.repo.full_name == github.repository && vars.PLAYWRIGHT_S3_ENABLED == 'true' }}
env:
S3_BUCKET: authentik-playwright-artifacts
S3_KEY_PREFIX: pr-${{ github.event.pull_request.number }}/run-${{ github.run_id }}/attempt-${{ github.run_attempt }}
run: | # shell
set -euo pipefail
if [ ! -d web/playwright-report ]; then
echo "No playwright-report/ produced; skipping S3 upload"
exit 0
fi
aws s3 cp \
--recursive \
--acl=public-read \
--cache-control "public, max-age=600" \
web/playwright-report/ \
"s3://${S3_BUCKET}/${S3_KEY_PREFIX}/"
# Same-repo guard: fork PRs run with a read-only GITHUB_TOKEN (even after
# maintainer approval), so the comment + edit calls would 403. Skip cleanly
# rather than failing the job on every fork PR run.
- name: Comment Playwright result on PR
if: ${{ !cancelled() && github.event_name == 'pull_request' && github.event.pull_request.head.repo.full_name == github.repository }}
env:
GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
PR_NUMBER: ${{ github.event.pull_request.number }}
AVAILABLE: ${{ steps.playwright-results.outputs.available }}
PASSED: ${{ steps.playwright-results.outputs.passed }}
FAILED: ${{ steps.playwright-results.outputs.failed }}
FLAKY: ${{ steps.playwright-results.outputs.flaky }}
SKIPPED: ${{ steps.playwright-results.outputs.skipped }}
S3_ENABLED: ${{ vars.PLAYWRIGHT_S3_ENABLED }}
REPORT_URL: https://authentik-playwright-artifacts.s3.amazonaws.com/pr-${{ github.event.pull_request.number }}/run-${{ github.run_id }}/attempt-${{ github.run_attempt }}/index.html
RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}/attempts/${{ github.run_attempt }}
run: | # shell
set -euo pipefail
marker='<!-- playwright-result -->'
if [ "$AVAILABLE" = "true" ]; then
if [ "$FAILED" -gt 0 ]; then
status='❌ Failed'
elif [ "$FLAKY" -gt 0 ]; then
status='⚠️ Passed with flakes'
else
status='✅ Passed'
fi
stats=$(printf '| Result | Count |\n|---|---|\n| ✅ Passed | %s |\n| ❌ Failed | %s |\n| ⚠️ Flaky | %s |\n| ⏭️ Skipped | %s |\n' "$PASSED" "$FAILED" "$FLAKY" "$SKIPPED")
else
status='⚠️ No results produced'
stats='The job did not produce `playwright-report/results.json`. The suite likely crashed before the JSON reporter wrote its output — see the workflow run for setup-step failures.'
fi
if [ "$S3_ENABLED" = "true" ]; then
report_line=$(printf '[HTML report](%s) · [Workflow run](%s)' "$REPORT_URL" "$RUN_URL")
else
report_line=$(printf '[Workflow run](%s) · _HTML report hosting is gated off until the `authentik-playwright-artifacts` S3 bucket is provisioned (`vars.PLAYWRIGHT_S3_ENABLED`). Until then, download the `playwright-report` artifact from the run page._' "$RUN_URL")
fi
body=$(printf '%s\n## Playwright e2e — %s\n\n%s\n\n%s\n' "$marker" "$status" "$stats" "$report_line")
existing=$(gh api "repos/${GITHUB_REPOSITORY}/issues/${PR_NUMBER}/comments" \
--paginate \
--jq "[.[] | select(.body != null and (.body | startswith(\"$marker\")))] | .[0].id // empty")
if [ -n "$existing" ]; then
gh api -X PATCH "repos/${GITHUB_REPOSITORY}/issues/comments/${existing}" -f body="$body" > /dev/null
echo "Updated existing comment ${existing}"
else
gh api -X POST "repos/${GITHUB_REPOSITORY}/issues/${PR_NUMBER}/comments" -f body="$body" > /dev/null
echo "Created new playwright-result comment"
fi
ci-web-mark:
if: always()
needs:

1
web/.gitignore vendored
View File

@@ -32,6 +32,7 @@ lib-cov
# Coverage directory used by tools like istanbul
coverage
playwright-report
playwright-traces
test-results
*.lcov

View File

@@ -2,12 +2,6 @@ import { PageFixture } from "#e2e/fixtures/PageFixture";
import { Page } from "@playwright/test";
export const GOOD_USERNAME = "test-admin@goauthentik.io";
export const GOOD_PASSWORD = "test-runner";
export const BAD_USERNAME = "bad-username@bad-login.io";
export const BAD_PASSWORD = "-this-is-a-bad-password-";
export interface LoginInit {
username?: string;
password?: string;

26
web/e2e/types/node.d.ts vendored Normal file
View File

@@ -0,0 +1,26 @@
/**
* @file Node.js global types and environment variables.
*/
declare module "process" {
global {
namespace NodeJS {
interface ProcessEnv {
/**
* The email address of the bootstrap user created when running the tests.
*
* @format email
*/
readonly AUTHENTIK_BOOTSTRAP_EMAIL?: string;
/**
* The password of the bootstrap user created when running the tests.
*/
readonly AUTHENTIK_BOOTSTRAP_PASSWORD?: string;
/**
* The directory where the authentik blueprints are stored.
*/
readonly AUTHENTIK_BLUEPRINTS_DIR?: string;
}
}
}
}

View File

@@ -23,10 +23,16 @@ export default defineConfig({
testDir: "./test/browser",
fullyParallel: true,
forbidOnly: CI,
retries: CI ? 2 : 0,
workers: CI ? 1 : undefined,
retries: CI ? 1 : 0,
workers: "50%",
maxFailures: CI ? 5 : 2,
reporter: CI
? "github"
? [
// ---
["github"],
["html", { open: "never", outputFolder: "playwright-report" }],
["json", { outputFile: "playwright-report/results.json" }],
]
: [
// ---
["list", { printSteps: true }],
@@ -36,6 +42,8 @@ export default defineConfig({
testIdAttribute: "data-test-id",
baseURL,
trace: "on-first-retry",
screenshot: "only-on-failure",
video: CI ? "retain-on-failure" : "off",
colorScheme: "dark",
launchOptions: {
logger: {

View File

@@ -18,7 +18,7 @@ import { RoleForm } from "#admin/roles/ak-role-form";
import { RbacApi, Role } from "@goauthentik/api";
import { msg } from "@lit/localize";
import { msg, str } from "@lit/localize";
import { html, PropertyValues, TemplateResult } from "lit";
import { customElement, state } from "lit/decorators.js";
@@ -83,7 +83,11 @@ export class RoleListPage extends TablePage<Role> {
row(item: Role): SlottedTemplateResult[] {
return [
html`<a href="#/identity/roles/${item.pk}">${item.name}</a>`,
html`<a
href="#/identity/roles/${item.pk}"
aria-label=${msg(str`View details of role "${item.name}"`)}
>${item.name}</a
>`,
html`<div class="ak-c-table__actions">${IconEditButton(RoleForm, item.pk)}</div>`,
];
}

View File

@@ -1,6 +1,19 @@
version: 1
metadata:
name: Playwright e2e test admin
entries:
- attrs:
- model: authentik_core.group
state: present
identifiers:
name: "authentik Admins"
attrs:
is_superuser: true
id: admin-group
- model: authentik_core.user
state: present
identifiers:
username: akadmin
attrs:
email: test-admin@goauthentik.io
is_active: true
name: authentik Default Admin
@@ -8,9 +21,4 @@ entries:
path: users
type: internal
groups:
- !Find [authentik_core.group, [name, "authentik Admins"]]
conditions: []
identifiers:
username: akadmin
model: authentik_core.user
state: present
- !KeyOf admin-group

View File

@@ -17,8 +17,6 @@ test.describe("Session Lifecycle", () => {
test.beforeAll(
'Ensure "Enable Remember me on this device" is on for the default identification stage',
async ({ browser }, { title: testName }) => {
if (Date.now()) return;
const context = await browser.newContext();
const page = await context.newPage();
const navigator = new NavigatorFixture(page, testName);

View File

@@ -161,8 +161,13 @@ test.describe("Roles", () => {
});
await test.step("Verify role name updated on view page", async () => {
// The role name shows up in both the page-navbar heading (as
// "Role <name>") and the description-list text (raw <name>); a
// bare getByText match hits both and triggers strict-mode. Pin
// to the description list so we're asserting the canonical
// value the edit just wrote, not the heading template.
await expect(
page.getByText(updatedName),
page.getByText(updatedName, { exact: true }),
"Updated role name is visible on view page",
).toBeVisible();
});

View File

@@ -17,7 +17,15 @@ test.describe("Provider Wizard", () => {
const dialog = page.getByRole("dialog", { name: "New Provider Wizard" });
await test.step("Authenticate", async () => session.login());
// session.login() with no `to` waits for the auth flow URL pattern to
// re-appear, which it does immediately (we just navigated there), so
// the helper returns *before* the post-login redirect lands. The
// wizard buttons probed below live on /if/admin/#/core/providers, so
// pin the destination explicitly — same shape as the other test files.
await test.step("Authenticate", async () =>
session.login({
to: "/if/admin/#/core/providers",
}));
await test.step("Navigate to provider wizard", async () => {
await expect(dialog, "Dialog is initially closed").toBeHidden();

View File

@@ -3,9 +3,14 @@ import { expect, test as setup } from "#e2e";
setup("Web server availability", async ({ baseURL }) => {
expect(baseURL, "Base URL is set").toBeTruthy();
const ok = await fetch(baseURL!)
// Probe the default authentication flow rather than the root URL — the root
// redirects to /setup when AUTHENTIK_BOOTSTRAP_* env vars are unset, and
// /setup raises FlowNonApplicableException (HTTP 500) once an admin user
// already has a usable password (as our test-admin blueprint installs).
const probeURL = new URL("/if/flow/default-authentication-flow/", baseURL!);
const ok = await fetch(probeURL)
.then((res) => res.ok)
.catch(() => false);
expect(ok, `Web server should be listening on ${baseURL}`).toBeTruthy();
expect(ok, `Web server should be serving ${probeURL}`).toBeTruthy();
});

View File

@@ -202,6 +202,13 @@ Copy the generated recovery key and paste it into the URL, after the domain. For
## End-to-End (E2E) Setup
authentik ships two end-to-end test suites:
- The Django/Selenium suite under `tests/e2e/` — driven by Django's test runner. It is exercised in CI by the `test-e2e` job in `ci-main.yml`.
- The browser-side Playwright suite under `web/test/browser/` — driven by `web/playwright.config.js`. It is exercised in CI by the `e2e (playwright)` job in `ci-web.yml` and runs against a live authentik server.
### Django/Selenium suite
Start the E2E test services with the following command:
```shell
@@ -216,6 +223,40 @@ Alternatively, you can connect directly via VNC on port `5900` using the passwor
When using Docker Desktop, host networking needs to be enabled via **Docker Settings** > **Resources** > **Network** > **Enable host networking**.
:::
### Playwright suite
The Playwright suite assumes that:
- An authentik stack is reachable at `http://localhost:9000` (override with `AK_TEST_RUNNER_PAGE_URL`).
- The blueprint at `web/test/blueprints/test-admin-user.yaml` has been applied so that the `test-admin@goauthentik.io` user (password `test-runner`) can log in.
Both prerequisites are satisfied by the standard full development environment described above, plus a one-off blueprint apply. From the repository root:
```shell
# Make the test admin blueprint discoverable by the worker, then start authentik.
mkdir -p blueprints/local
cp web/test/blueprints/test-admin-user.yaml blueprints/local/test-admin-user.yaml
# In separate terminals:
make run-server
make run-worker
```
Once the worker has applied the blueprint, install Playwright's browsers (one-time) and run the suite — these are the same npm scripts CI runs:
```shell
corepack npm exec --prefix web -- playwright install --with-deps chromium
corepack npm run --prefix web test:e2e
```
After a failing run, open the HTML report to inspect traces, screenshots, and (in CI) videos:
```shell
corepack npm exec --prefix web -- playwright show-report
```
In CI, the same artifacts are uploaded under the `playwright-report` and `playwright-traces` artifact names on the failed workflow run.
## Contributing code
### Before submitting a pull request