Password Reset Test Cases Beyond the Email Link
TL;DR: A password reset has five links in the chain: the request, the email, the link, the new password, and the next sign-in. Most test lists stop at the email. Check what the browser shows at each step, then prove the reset worked by signing in with the new password and being refused with the old one.
You click “Forgot password,” get the email, set a new password, and see “Password updated.” That feels like a tested feature.
Then a customer clicks the same link twice because the first tab looked stuck. Another opens the email the next morning. A third asks for three emails in a row and uses the first one. None of this is exotic; it is a normal Tuesday for any app with a login form. And the reset page is where someone who is already locked out finds out whether your app can help them.
Which password reset test cases are worth running first?
Public checklists often run to 15 or 26 cases, which is fine if you have a QA team and a spreadsheet. If you ship alone, rank by what would cost you a customer. Here is a first pass of eight, using a test account you control.
| Case | What you do | A pass looks like |
|---|---|---|
| Normal reset | Request a reset, follow the link, set a new password | You land somewhere sensible, and the new password works |
| Unknown email | Request a reset for an address with no account | The page says the same thing it says for a real account |
| Expired link | Open a link after your app’s stated lifetime | A plain message and a way to request a new link |
| Used link | Open the same link again after a successful reset | Refused, with the same way forward |
| Older link | Request two emails, then use the first one | Whatever rule you chose; many apps retire the older link |
| Bad new password | Enter a weak password, or two that do not match | A readable error, and the form still works |
| Old password | After the reset, sign in with the old password | Refused |
| Other session | Stay signed in in a second browser, reset, go back to it | Whatever you decided should happen |
Some rows end with “whatever you decided” on purpose. Nobody can tell you whether an older link should still work; they can only tell you to pick a rule and write it down. A test needs something to be checked against, and a missing rule is a gap in the product before it is a gap in the test.
Two things stay out of this list. How long and unguessable the secret inside the link is, and whether your server really deletes it after use, are server questions. Your browser can show a link being refused; it cannot read your database. For those, hand an engineer the OWASP forgot password cheat sheet. The failure-path items in the pre-launch QA checklist cover the login side of the same area.
What can go wrong before the email arrives?
Start with the request page, because it can leak something without anyone noticing.
Type a real address, then a made-up one, and compare the two screens word for word. OWASP says to return a consistent message for existing and non-existing accounts. Supabase’s docs say the same about their reset call: it does not reveal whether an account exists. A message like “No account found with that email” feels helpful, and it also tells a stranger which addresses are customers.
Next, look at where the link points. Builders report reset emails that carry a localhost address or an old preview domain, as in this Indie Hackers write-up. If your app uses Supabase Auth, its docs say the redirect address you pass must be on the project’s list of allowed redirect URLs. So request the reset from your published app, not the builder’s preview, and read the domain in the email before you click anything.
One practical trap: email providers cap how many messages they send. Supabase’s docs say the default email service is limited to 2 emails per hour and is best effort, and they recommend your own SMTP server for production. Check the current limit before you conclude that a missing email is a bug. If your third test run produces nothing, it may simply be the quota.
How do you test the link itself?
You need an inbox you control, tied to a test account. Do not use a real customer’s address. Then work through the link cases one at a time.
Expired link. Find out your app’s link lifetime first; a lifetime nobody wrote down cannot be tested. If it is an hour, request a link now and open it after lunch.
Used link. After a successful reset, click the same link from the email again. Use the email, not the back button; the back button may just show you the page you already loaded.
Older link. Request two emails, then open the first. Check that your app does what you decided, not what you assumed.
For all three, look at the refusal itself. A pass is a message that names the problem plainly and offers “request a new link.” A blank page or a raw error code is a fail, even if the secret behaved perfectly. A refused link that leaves the user stuck is a bug the customer feels.
This is the line between what the browser shows and what the server holds. You can watch the link get refused; you cannot see whether the server deleted the secret or just hid the form. The end-to-end testing tools guide goes into what a browser pass does and does not prove.
How do you know the password actually changed?
The success message proves very little. The real proof is behavior, and the order matters:
- Sign out.
- Try the old password. It should be refused.
- Try the new password. It should land you in the app.
Use a throwaway account for this. After a reset, whatever stored the old password, whether a password manager or a testing tool, is now wrong. If you forget that, your next run fails at sign-in and looks like a regression when it is just stale data.
Then check the sessions. A session is the signed-in state your browser keeps. Sign in on a second browser or profile, reset the password from the first, and refresh the second. OWASP suggests either asking the user whether to end all existing sessions or ending them automatically, and it advises against logging the user in automatically after a reset; they should sign in the usual way. Whatever your policy, confirm the second browser does that, rather than assuming.
Last, note where the user lands after the reset: a sign-in page with a notice, or straight into the app. Both can be right. A dead end is not.
Where does WayRunner fit in a password reset test?
WayRunner works in the browser, from a plain-English instruction against your connected site. It does not open your inbox. The emailed link needs a starting state you arrange yourself, so the middle of this flow, opening the email and getting the link, is yours.
The parts at either end fit well. You can have it request resets for two addresses and compare what the page says, and you can run the sign-in checks: the old password refused, the new one accepted. Those are the steps you will want to repeat after every release; the smoke testing guide shows how to keep that kind of check small.
One detail from the build changes how you should phrase these checks. WayRunner’s Planner cannot read a password field back. Before an assert or extract step runs, the orchestrator screens the Planner’s stated intent, and if it mentions a password, a one-time code, or another credential, the step is rejected. The rejection tells the Planner to verify success a different way, for example by asserting a non-sensitive confirmation on screen. The screen matches words in the stated expectation; it does not look at the page. So an expectation like “the page confirms the password was updated” can get bounced, and the Planner has to try another route.
That is a real quirk, and it points at a good habit for any reset test: prove the change by outcome, not by reading a field. New password works, old password refused.
The limits are plain. It is web only, in Chromium. It will not check your database or confirm that a secret was invalidated on your server. And a human-verification challenge it cannot clear on its own pauses the run for a person, so if your reset page sits behind one, expect to step in.
A reset flow is done when a locked-out person can get back in and the wrong person cannot. Run the eight rows against your published app with a test account, write down your three decisions (link lifetime, older links, other sessions), and repeat the sign-in checks after each release.