README.md

Athena E2E checks

Playwright smoke tests. These exist because curl proves the server responds but cannot see client-side failures — a broken frontend can sit behind a stack where every endpoint returns 200.

cd e2e
npm install          # first time only
node smoke.mjs       # loads /, /auth/signup, /auth/login; reports console errors + HTTP >=400
node peek.mjs        # screenshot + dump page text. Use this FIRST when something looks broken.
node routes.mjs      # enumerates real admin nav routes (needs an account)
node theme.mjs <tag> # screenshots light+dark and samples the live colour tokens
node devshot.mjs <tag> [url]  # same, against `next dev` (default :3010)
node landing.mjs [url]        # /app/home: card list + both click targets, against `next dev`
node flow.mjs        # ⚠️ REGISTERS AN ACCOUNT. First account becomes admin.

Credentials in .env.local (gitignored). Screenshots land in shots/.

These all target http://localhost:3003, resolved once in base.mjs. If the stack was brought up on another port, pass the same ATHENA_PORT you gave Compose — or BASE_URL for a host that isn't localhost:

ATHENA_PORT=3007 node smoke.mjs
BASE_URL=http://box.local:3007 node smoke.mjs

devshot.mjs and landing.mjs are the exception: they point at the next dev server on :3010 and take the URL as an argument.

What to look for

• 404 /api/enterprise-settings → the frontend hard-depends on this route. Without it every page renders "problem loading your Onyx settings."
• 502 + text/html on /api/* → nginx holds a stale upstream IP. Restart nginx.
• 403 on /api/me, /api/settings, /api/notifications while logged out → normal, ignore.

Check the actual rendered page with peek.mjs before reporting a regression. Three "outages" during the enterprise-code removal were all bad diagnostics — a grep matching a "Create an Account" link, and a selector assuming input[type="email"] when the login inputs aren't that type.