Automated Threat Modeling

Overview of CVE-2026-125033: Unauthenticated Stored XSS in Smash Balloon Social Post Feed via Facebook Attachment Descriptions

A brief summary of an unauthenticated stored cross-site scripting flaw in the Smash Balloon Social Post Feed WordPress plugin, reachable through Facebook attachment descriptions in versions up to 4.15.1. The post reviews the root cause and the code changes shipped in the 4.16.0 patch.

CVE Analysis

7 min read

Gaurav Sarraf
Gaurav Sarraf

2026-11-18

Overview of CVE-2026-125033: Unauthenticated Stored XSS in Smash Balloon Social Post Feed via Facebook Attachment Descriptions

Introduction

A stored cross-site scripting flaw in the Smash Balloon Social Post Feed plugin allows text written by outside Facebook users to execute as script on WordPress pages that embed a Facebook feed. Because the payload arrives through Facebook attachment descriptions rather than through WordPress itself, the attacker needs no WordPress account, and every anonymous visitor to a page carrying the feed shortcode is in scope.

Smash Balloon Social Post Feed (listed on WordPress.org as Simple Social Feeds for WordPress) is a plugin that displays Facebook posts on WordPress sites. The plugin listing reports more than 200,000 active installs, and Smash Balloon, which began in 2013 and is now part of Awesome Motive, reported more than 1.3 million active installs across its social feed plugins at the time of that announcement. Plugins like this sit on a trust boundary, rendering third-party social content inside first-party pages, and this bug lives on that boundary.

The issue is tracked as CVE-2026-125033 (CWE 79). Wordfence, acting as CNA, scores it CVSS 3.1 7.2 High with vector AV:N/AC:L/PR:N/UI:N/S:C/C:L/I:L/A:N. NVD has not published its own assessment. The available sources do not report exploitation in the wild, and Rapid7 notes the issue is not in the CISA Known Exploited Vulnerabilities catalog. Wordfence credits researcher Filip Kowalski (Agentunio) with the finding. The fix shipped in version 4.16.0.

Technical information

Where the attacker input comes from

The vulnerable field is the Facebook Graph attachment description, attachments.data[].description. In 4.15.1, inc/CFF_Shortcode.php requests attachment descriptions from Facebook and assigns the first attachment description to the post description property. Any party able to contribute an attachment description to a post that the configured feed ingests (visitor posts, shared posts, or attached content) therefore controls a string that the plugin later renders into HTML on the WordPress site.

Root cause: escaping, autolinking, then parsing

In 4.15.1, the description was built in this order: escape it, autolink it, then expand mention tokens.

$cff_description_text = CFF_Autolink::cff_autolink(htmlspecialchars($description_text), $link_color = $cff_posttext_link_color); $cff_description_tagged = CFF_Utils::cff_desc_tags($cff_description_text);

The mention parser CFF_Utils::cff_desc_tags() matched @[...] tokens with a lazy pattern that accepts any character, then joined the pieces straight into HTML:

preg_match_all("/@\[(.*?)\]/", $description, $cff_tag_matches); ... $cff_tag_parts = explode(':', $cff_tag_match); $replace_strings_arr[] = '<a href="https://facebook.com/' . $cff_tag_parts[0] . '">' . $cff_tag_parts[2] . '</a>';

Two separate defects combine here:

  1. Wrong processing order. htmlspecialchars() did convert the attacker's own quotes into &quot;. The autolinker then inserted real markup, including literal " characters. Because the mention parser ran afterwards, a crafted @[...] token could stretch across that generated markup, swallow part of an <a href="..."> tag, and leave an attribute quote open.
  2. No escaping or validation of token parts. The token id and name were concatenated into the anchor exactly as received, and (.*?) accepted any character.

The net effect is that the parser could not distinguish attacker text from markup the plugin itself had just produced. Once an attribute quote is left dangling, the browser interprets the text that follows as attributes of an existing tag.

Why the existing sanitizer did not help

Both 4.15.1 and 4.16.0 pass every incoming Facebook post field through CFF_Utils::strip_untrusted_html(), added in 4.13.1. That routine removes anything that looks like a tag, such as < followed by a letter, /, ! or ?. As the WPSec description explains, this payload does not need any tags of its own. The token reuses a quote generated by the autolinker, so later text becomes attribute content of a tag the plugin already emitted. An input filter that looks for tags cannot see this, which is why the fix targets the HTML assembly step rather than the input filter.

Template output

templates/item/post-text.php passes text to a formatting function and visibly escapes some link and label values, but does not visibly escape the post text at line 33. That template is identical in 4.15.1 and 4.16.0, so the flaw and the fix are entirely in how $post_text is built before the template outputs it.

Attack flow

Based on the advisory and the patch analysis, exploitation proceeds as follows:

  1. A WordPress site configures the plugin to display a Facebook feed and places the feed shortcode on one or more pages.
  2. An outside Facebook user contributes content with an attachment description to a post the feed ingests, such as a visitor post, a shared post, or attached content. The description contains a crafted mention token.
  3. The plugin fetches the post from Facebook and assigns the attachment description to the post description.
  4. During rendering, the description is escaped and autolinked, and then the mention parser runs over the resulting HTML. The crafted token spans the autolinker markup and leaves an attribute quote open.
  5. Every visitor to any page carrying the shortcode, including anonymous visitors, receives the injected markup, and the script executes in their browser in the context of the WordPress site.

The advisory does not publish a payload, and none is reproduced here.

Patch information

The flaw is fixed in version 4.16.0. Wordfence lists 4.15.1 and earlier as affected and gives the remediation as "Update to version 4.16.0, or a newer patched version". Both the NVD and Wordfence records link to the WordPress.org SVN changeset between tags/4.15.1 (r3779993) and tags/4.16.0 (r3799153). WPSec, which tracks the same bug as WPSEC-2026-3108, dates the fixed release to November 12, 2026. The code below was compared directly from the two tagged releases in SVN.

1. Strict token format and context-aware escaping

cff_desc_tags() was rewritten:

$tagged = preg_replace_callback( '/@\[(\d+):\d+:([^\[\]<>"]+)\]/', function ($matches) { $href = esc_attr('https://facebook.com/' . $matches[1]); return '<a href="' . $href . '">' . esc_html($matches[2]) . '</a>'; }, (string) $description ); return null === $tagged ? (string) $description : $tagged;
  • A token now matches only if the id and type are digits and the name contains no [, ], <, > or ". The docblock states that this means a token "can never span markup and break out of an attribute".
  • Anything that does not fit the format stays as plain text instead of becoming a link.
  • The href goes through esc_attr() and the visible name through esc_html(), so even a valid token is escaped for its output context.
  • If preg_replace_callback fails and returns null, the function returns the original escaped text rather than an empty or partially built string.

2. Processing order enforced in one helper

A new helper fixes the order as escape first, then mentions, then autolinking:

public static function cff_desc_tags_autolink($text, $link_color = '') { return CFF_Autolink::cff_autolink(self::cff_desc_tags(htmlspecialchars($text)), $link_color); }

Its docblock, which references an internal Smash Balloon ticket, says mentions are parsed on the escaped text before the autolinker adds its markup, so a token cannot overlap an autolinked URL. The description code in CFF_Shortcode.php now calls this helper. The 4.15.1 code:

$cff_description_text = CFF_Autolink::cff_autolink(htmlspecialchars($description_text), $link_color = $cff_posttext_link_color); $cff_description_tagged = CFF_Utils::cff_desc_tags($cff_description_text);

becomes, in 4.16.0:

$cff_description_tagged = CFF_Utils::cff_desc_tags_autolink( $description_text, $cff_posttext_link_color );

The rule at the center of the patch is that parsing runs on plain, escaped text, never on HTML that an earlier step generated.

The story tag and message tag code in the same render loop also built anchor tags by joining Graph API id and name values directly. In 4.16.0 these go through a new CFF_Utils::cff_tag_link(), which applies rawurlencode() to the id, esc_attr() to the href, and esc_html() to the name. The offset-based replacement paths were changed the same way. The story tag path still passes its fixed target="_blank" attribute as a trusted third argument.

Unchanged code paths

  • templates/item/post-text.php is identical in both releases.
  • When the link post text option is enabled ($cff_title_link), the description still goes through CFF_Utils::cff_wrap_span(htmlspecialchars(...)). That path never parses mention tokens.

Verifying the fix

To confirm a site is running patched code, look in inc/CFF_Utils.php for the cff_desc_tags_autolink() method and the digit-only @\[(\d+):\d+:...\] mention pattern. The reviewed sources do not document a temporary workaround or a cleanup procedure for sites that may already have rendered malicious feed content.

Affected systems and versions

  • Product: Smash Balloon Social Post Feed (Simple Social Feeds for WordPress), WordPress.org plugin custom-facebook-feed
  • Affected: all versions up to and including 4.15.1
  • Fixed: 4.16.0 and later
  • Exposed configuration: WordPress pages that render a configured Facebook feed via the plugin shortcode, where the feed ingests posts to which outside users can contribute attachment descriptions (visitor posts, shared posts, or attached content). The mention parsing path analyzed above applies to descriptions rendered without the link post text option. With that option enabled, the description is wrapped through cff_wrap_span(htmlspecialchars(...)) and mention tokens are not parsed.

Vendor security history

The Wordfence vulnerability index lists ten other reported vulnerabilities for this plugin, including stored cross-site scripting findings in both authenticated and unauthenticated variants, along with other issue types. The existence of strip_untrusted_html(), introduced in 4.13.1, shows earlier hardening of incoming Facebook content. The reviewed material does not provide a complete remediation history for the earlier entries, so this record should be read as a list of reported issues rather than a measure of vendor response quality.

References

Detect & fix
what others miss

Works with
  • GitHub
  • GitLab
  • Bitbucket
  • Azure DevOps Services
  • Gerrit
  • CVS
  • Jira
  • Linear
  • Slack
  • Security Compass
Explore integrations
Security magnifying glass visualization