All postsWinning the argument on a duplicate
Convinced triage to upgrade an Open Redirect to DOM XSS one minute before they closed it. Payout: EUR 0.
Process · 4 min read
A private program for a global luxury retail brand. An international e-commerce
platform with localized storefronts across thirty countries.
I found an unauthenticated DOM-based sink, chained it through the brand's own
passwordless login emails, argued the triage team into upgrading the finding from
an Open Redirect to DOM XSS, and had the ticket retitled to CWE-79.
One minute later, they closed it as a duplicate.
The technical work was sound. The escalation persuaded them completely. And the
bounty was €0, because I broke my own reconnaissance gate to hunt on a saturated
surface.
The sink#
On the storefront's SPA, /{locale}/alternative-login handled fallback magic-link
authentication.
In the minified route chunk served to every visitor:
alternativeLogin() {
try {
this.trackAlternativeLoginSuccess(),
this.redirectUrl
? window.location.assign(this.redirectUrl)
: window.location.assign(this.overviewUrl)
} catch(e) {
return this.trackAlternativeLoginFailure(),
this.$pageErrorHandler(new f(e, this.getTargetUrl()))
}
}
Where this.redirectUrl came from:
redirectUrl() {
return this.isVerifiedContext
? this.response?.redirect_url
: (this._.provides[e] || this.$route).query['redirect-url']
}
If the user lands from a verified authentication callback, the destination comes
from the API response. Otherwise, it pulls redirect-url straight off the query
string with no origin check and no scheme check.
Feed it an https:// destination:
https://shop.example.com/en-us/alternative-login?redirect-url=https://attacker.example/
The browser hits the page, hydrates, and immediately navigates away. An open
redirect that needs no session, no token, and no valid credentials.
Escalating to script execution#
An open redirect alone is usually a low-severity finding that programs triage
reluctantly. But window.location.assign() is an execution sink:
https://shop.example.com/en-us/alternative-login?redirect-url=javascript:alert(document.domain)
No Content Security Policy (script-src) was configured on the route. The
parameter passed straight into assign(), and the script executed in the origin
of the e-commerce store.
To make the impact undeniable, I chained it to the passwordless login flow. The
storefront's unauthenticated API exposed an endpoint:
POST /api/v1/login/send_magic_link HTTP/2
Host: api.example.com
{
"username": "victim@example.com",
"redirect_url": "https://shop.example.com/en-us/alternative-login?redirect-url=javascript:..."
}
The backend API validated that redirect_url belonged to the retailer's own
domain. Because shop.example.com is legitimate, the API accepted it and
dispatched a real login email from the retailer's mail server.
The lure was a genuine, cryptographic magic link with no visible external
destination. When the victim clicked it, authenticated, and landed on the store,
the unvalidated parameter triggered script execution in their authenticated
session.
The triage escalation#
I filed the initial report as an Open Redirect (CWE-601, CVSS 6.1) while I
verified the execution chain, then followed up with evidence for DOM XSS
(CWE-79):
- Quote their shipped code. Instead of pasting generic scanner traces, I
located the exact unminified expressions in their client bundle and walked
triage through the control flow.
- Correct bad remediation advice early. In my first comment, I had suggested
their API validation was weak. I spent a day running 72 normalisation test
cases against the backend validator (IDNA mappings, authority terminators,
JSON type confusion, duplicate keys) and proved the API parser was conformant.
The bug was purely in the frontend chunk. I told them not to waste engineering
hours patching the backend.
- Argue against your own proposal when the evidence demands it. I initially
proposed CVSS 9.3. But when I audited what was actually reachable in
localStorage, there were only analytics tokens and device IDs—no customer
PII or session secrets. Edge filters blocked cross-origin reads. I explicitly
withdrew the 9.3 rating and told triage the original 6.1 score was accurate,
asking only for the vulnerability class to be updated to Cross-Site
Scripting.
It worked. The security team accepted the analysis, retitled the ticket to
"DOM XSS + Open Redirect", and updated the vulnerability classification to
Cross-site Scripting (XSS) - DOM (CWE-79).
Sixty seconds later, an automated note closed the ticket:
Closed: Duplicate of #329.
Payout: €0. Points: 6.
The gate that called it#
The technical argument was right. The failure happened days before I typed the
first payload.
When I set up the engagement, my own scope gate evaluated the storefront asset:
Row 1 (Storefront Country Locales): 342 prior reports. FAIL: Saturated. Do
not hunt here.
I had a working rule: when an asset has several hundred resolved reports on a
mature public program, commodity issues—especially unauthenticated query-string
parameters on main navigation routes—are almost certainly already claimed.
I overruled the gate anyway with a piece of clever rationalisation:
"The target had an explicit scope carve-out for legacy .jsp endpoints.
Everyone else must be avoiding the storefront, and our finding is on a modern
Nuxt SPA route. The SPA is the uncrowded sibling."
That narrative was completely false.
The .jsp carve-out hadn't kept other hunters away from the SPA; at least one
other researcher had found the exact same parameter on the exact same route weeks
earlier.
A count of 342 prior reports isn't an indicator of difficulty. It is a direct
probability that someone has already stood where you are standing.
What was actually learned#
- Triage listens to code. Quoting the specific minified functions and showing
the execution path won the reclassification without an argument.
- Withdraw bad advice immediately. If you tell a team their API parser is
broken and then discover it isn't, post the retraction. It builds credibility
that lasts past the ticket.
- Never override a saturation gate with a clever story. The raw report count
was the honest instrument. The explanation I invented to dismiss it was the
mistake that cost three days of unpaid work.