cross-site-scripting

name

cross-site-scripting

description

SAST detection methodology for cross-site scripting (CWE-79), including reflected, stored, and DOM-based XSS, mutation XSS (mXSS), framework sink gadgets (dangerouslySetInnerHTML, v-html, bypassSecurityTrust), template auto-escape bypasses, and HTML/attribute/JS/URL/CSS context confusion. Use when reviewing code that emits user data into HTML, the DOM, or a template. Writes confirmed findings to findings/13-cross-site-scripting.md.

Cross-Site Scripting (XSS) — SAST Methodology

Class: Cross-Site Scripting · CWE-79 · OWASP: A03 Injection Findings file: findings/13-cross-site-scripting.md

1. Overview

XSS occurs when attacker-controlled data reaches an HTML/DOM/script rendering sink without encoding appropriate to the output context, letting the attacker inject markup or script that executes in a victim's browser session. The core test: does a source reach a sink that interprets its value as markup/code, and is the encoding on that path (a) present and (b) correct for that specific context? Encoding that is safe in HTML text is unsafe in an attribute, a URL, a <script> block, or a style value — context is everything.

2. Where it lives

• Server-side templates that interpolate values (render, f-strings into HTML, String.format, PHP echo, JSP <%= %>, ERB <%= %>).
• Client-side DOM manipulation in JS/TS (innerHTML, document.write, jQuery).
• SPA framework escape hatches (dangerouslySetInnerHTML, v-html, [innerHTML], bypassSecurityTrustHtml).
• API responses reflected into the DOM by a front-end; JSON injected into inline <script> bootstrap blocks (window.__STATE__ = {{ data }}).
• Error pages, search results, chat/comment rendering, markdown-to-HTML, WYSIWYG/rich-text fields, filenames, Referer/User-Agent echoed into pages.

3. Sources (tainted input)

All HTTP inputs (query/body/path/header/cookie), plus DOM-local sources for DOM XSS: location.href/search/hash/pathname, document.URL, document.referrer, document.cookie, window.name, postMessage data, localStorage/ sessionStorage. Plus stored/second-order: values persisted (DB, cache, file) then rendered later (comments, profile fields, admin dashboards viewing user data). Trace stored values back to their write source.

4. Sinks (dangerous operations)

// DOM XSS — dangerous JS sinks
el.innerHTML = location.hash.slice(1);        // markup parsed
el.outerHTML = userInput;
el.insertAdjacentHTML('beforeend', userInput);
document.write(userInput);                     // and document.writeln
el.setAttribute('href', 'javascript:' + x);   // URL/JS context
new Function(userInput); eval(userInput);      // script context via XSS gadget
$('#x').html(userInput); $('#x').append(userInput);  // jQuery parses HTML
$(userInput);                                  // jQuery( '<img onerror=..>' )

// React — dangerous
<div dangerouslySetInnerHTML={{ __html: userInput }} />
<a href={userInput}>            // javascript: URL not blocked by React

<!-- Vue — dangerous -->
<div v-html="userInput"></div>

// Angular — dangerous (defeats built-in sanitizer)
this.sanitizer.bypassSecurityTrustHtml(userInput);
this.sanitizer.bypassSecurityTrustResourceUrl(userInput);
element.innerHTML = userInput;                 // outside Angular binding

# Jinja2 / Django — dangerous (auto-escape disabled/bypassed)
Markup(user_input)                             # marks safe, no escaping
"{{ user_input | safe }}"                       # Jinja |safe
mark_safe(user_input)                           # Django
render_template_string(user_input)              # SSTI-adjacent, also XSS

// PHP — dangerous
echo $_GET['q'];                                // no htmlspecialchars
print "<div>{$_POST['bio']}</div>";

{{! Handlebars — dangerous: triple-stash disables escaping }}
{{{ userInput }}}

{# Twig — dangerous #}
{{ userInput | raw }}

5. Sanitizers / safe patterns

Safe:

• Context-correct output encoding at the sink: HTML-entity encode for HTML text (htmlspecialchars, escape, template auto-escape ON), attribute encoding for attribute context, JS-string encoding + JSON.stringify for script context, URL encoding for URL context, and URL-scheme allowlisting (https:/http:/ relative) for href/src.
• Assigning via safe DOM APIs: textContent, innerText, setAttribute for non-URL attributes, el.classList, framework text interpolation ({{ }} with escaping, React {value} children, [textContent]).
• HTML sanitizer libraries for rich text with a strict allowlist: DOMPurify (DOMPurify.sanitize), OWASP Java HTML Sanitizer, Bleach (with tight config), Ruby sanitize.

Fails / not a real sanitizer:

• Right encoder, wrong context — HTML-encoding a value placed inside a <script> block, a javascript: URL, an unquoted attribute, or a CSS style.
• Blocklist/regex stripping of <script>, onerror, etc. — bypassed by case, nesting (<scr<script>ipt>), event handlers, SVG/MathML, data:/ javascript: URLs, HTML entities.
• Encode-then-decode — value encoded server-side but a later decodeURIComponent / unescape / .innerHTML re-parses it (mXSS territory).
• DOMPurify misconfig — ALLOWED_ATTR including href/on*, ADD_TAGS re-enabling dangerous elements, RETURN_DOM_FRAGMENT re-inserted via innerHTML after mutation (mXSS), or sanitizing then modifying the string.
• URL/scheme check that misses javascript: variants (java\tscript:, JaVaScRiPt:, data:text/html, leading whitespace/control chars).
• Angular bypassSecurityTrust* — by name it disables the sanitizer; never a fix.
• Escaping only one of several interpolation points on the same sink.

6. Detection methodology

1. Find DOM/HTML sinks:

   rg -n 'innerHTML|outerHTML|insertAdjacentHTML|document\.write(ln)?|\.html\(|\.append\(|\.prepend\(|\.before\(|\.after\('
   rg -n 'dangerouslySetInnerHTML|v-html|\[innerHTML\]|bypassSecurityTrust'
   rg -n 'eval\(|new Function\(|setTimeout\(\s*[\'"]|setInterval\(\s*[\'"]'
   

2. Find template escape bypasses:

   rg -n '\|\s*safe|\|\s*raw|mark_safe|Markup\(|\{\{\{|render_template_string|autoescape\s*(off|false)'
   rg -n 'html\.escape|htmlspecialchars|escapeHtml' -l   # inventory who *does* encode
   

3. Find DOM sources feeding sinks:

   rg -n 'location\.(hash|search|href|pathname)|document\.(URL|referrer|cookie)|window\.name|localStorage|\.data\b.*addEventListener\([\'"]message'
   

4. Determine the output context of each sink hit: HTML text, attribute (quoted/unquoted), URL attribute (href/src/action/formaction), inline JS/<script>, event handler, or CSS. The context dictates which encoder is correct.
5. Trace the operand to a source (direct, DOM-local, or stored/second-order). Stop at a literal/constant.
6. Check for a real, context-correct sanitizer on the path (see §5). Verify it isn't bypassable and matches the context.
7. Confirm reachability & persistence: reflected (needs a crafted request), stored (persists, hits other users/admins), or DOM (no server round-trip).

7. Modern & niche variants

• Reflected vs stored vs DOM: DOM XSS never touches the server response body — a server-side grep will miss it; you must trace client JS source→sink. Stored is highest-impact (fires for every viewer, including admins) and often second-order.
• Mutation XSS (mXSS): markup that is inert as a string but becomes active after the browser's HTML parser re-serializes and re-parses it — e.g. inside <template>, <noscript>, <svg>/<math> foreign-content, or namespace confusion. Bypasses naive sanitizers that inspect the pre-mutation string. Sanitize-then-reinsert-as-HTML and innerHTML round-trips are the classic trigger.
• Sink-specific JS gadgets: innerHTML/outerHTML/insertAdjacentHTML/ document.write parse markup; jQuery .html()/.append()/.after() and even $(selector) parse HTML if the string looks like a tag. React dangerouslySetInnerHTML, Vue v-html, Angular [innerHTML] all bypass the framework's default escaping.
• Framework sink gadgets: Angular bypassSecurityTrust{Html,Url,ResourceUrl, Script,Style} and template-injection into Angular expressions; React href/ src accepting javascript:; Vue render functions with domProps.innerHTML.
• Template auto-escape bypass: Jinja |safe/Markup, Twig |raw, Handlebars {{{ }}}, Django mark_safe/|safe, ERB <%== %> (Rails raw/html_safe), Go template.HTML/template.JS/template.URL type casts, EJS <%- %>.
• Context confusion: the same value is safe in one context and lethal in another. HTML-text-encoded data in an attribute (breaks out via " or space), in a javascript: URL, in an inline <script>/JSON island (breaks out via </script> or 
), or in CSS (expression(), url(javascript:)).
• href="javascript:" / URL-scheme sinks: any sink setting href, src, action, formaction, xlink:href, data, poster where the scheme isn't allowlisted; javascript:, data:text/html, vbscript:.
• SVG/MathML vectors: <svg><script>, <svg onload=…>, <math> with xlink:href, <foreignObject> — foreign-content parsing rules enable mXSS and sanitizer bypass.

8. Common false positives

• Value emitted only via textContent/innerText/escaped interpolation.
• Framework text binding ({value}, {{ }} with auto-escape ON) — safe unless a dangerously*/v-html/|safe escape hatch is used.
• Static/constant strings or hard-coded HTML not reachable from a source.
• Value passed through a correctly-configured DOMPurify/sanitizer for that context.
• href bound to a value that is scheme-allowlisted before assignment.
• innerHTML set from a trusted, developer-authored constant template.

9. Severity & exploitability

Base Medium for reflected/DOM XSS; High for stored XSS in an authenticated context (fires for every viewer, can hit admins → session/account takeover). Raise toward Critical when it yields admin session theft, CSRF-token exfiltration enabling full account takeover, or executes in a privileged origin/extension context. Lower if the sink is behind strict CSP that blocks the payload class, or requires improbable victim interaction. See references/severity-model.md.

10. Remediation

Encode output for the exact context at the sink (HTML, attribute, JS, URL, CSS) — prefer framework auto-escaping and never disable it. Use safe DOM APIs (textContent) instead of innerHTML. For rich text, run a well-configured allowlist sanitizer (DOMPurify) and re-sanitize after any DOM mutation. Allowlist URL schemes for href/src. Deploy a strict Content-Security-Policy (nonce/hash-based, no unsafe-inline) as defense-in-depth. Never use bypassSecurityTrust*, |safe, raw, {{{ }}}, or mark_safe on tainted data.

11. Output

Append each confirmed finding to findings/13-cross-site-scripting.md using references/finding-template.md. Set Class: Cross-Site Scripting, CWE: CWE-79, and in Chain potential note primitives such as script execution in victim origin (→ session/CSRF-token theft, credential capture, keylogging), DOM sink gadget (consumes prototype-pollution or DOM-clobbering primitives), or stored payload delivery (→ privileged-user compromise). XSS frequently consumes an open-redirect or CORS misconfig and provides the execution primitive that turns other flaws into account takeover.

Primitives (controlled): provides JS_EXEC; consumes HTML_OUTPUT,GADGET

References

• CWE-79; OWASP A03:2021 Injection; OWASP XSS Prevention & DOM XSS Prevention Cheat Sheets; DOMPurify documentation; Google mXSS research.