—
—

Security audit — example.com (WordPress installation under /pb)

Result35 findings

1. Summary

A real report, with the identity removed. This is an audit that actually ran: the findings, the versions, the CVEs, the numbers and the commands are the ones in the delivered report. The only thing replaced is what identifies the client — the domain appears as example.com, and the hosting account and the panel IP were substituted. Nothing else was changed.
Order#116
Project/opt/.../SourcesCode/example.com — the project name was not supplied with the order, so the audited path and domain are used instead
LevelONE
Date2026-11-13
Executionsession 89017662 · 2026-11-13
Measurements taken2026-11-13, 09:00–10:30 UTC

1. The ground this runs on

This is where the audit starts, because everything below is graded against it.

Hosting type: shared. Measured, not assumed: the account runs under CloudLinux CageFS/LVE, the panel offers 20 PHP directories under /opt/alt/php*, /home/ shows 1 entry, and the 6 visible processes belong to a single owner. There is no sudo. The server is the provider's and hosts other customers.

What the shared model puts out of reach, by name. These are not oversights — they belong to the provider, not to this account, and no audit of this account can produce them: the host's listening ports and firewall, the Apache/nginx version, the operating-system version (only the kernel is visible through LVE), the processes and cron jobs of the other accounts, and the global sshd configuration. ss and netstat are both absent from the account's PATH, so no port measurement was made — that is "not measured", not "no ports are open".

What the account's own domain serves was measured, through the hostname the order authorises: port 443 answers (HTTP/2 200) and port 80 redirects (301 to https://). The only ports the client's own configuration declares are the loopback 127.0.0.1:7777 used by the host's update plugin and DB_HOST = 127.0.0.1 in wp-config.php.

Core and components. WordPress 6.6.11 against 7.1.4 published today. Of 46 installed components (38 plugins, 7 themes, the core), 30 are behind, 4 are current, and 12 could not be measured because they have no reachable public channel. 12 components carry at least one published advisory whose affected range covers the version installed here. The full table is in section 6.

File modes under the published root. All 26,024 files and 5,707 directories were swept, not sampled: every file is 644, every directory is 755 except one, which is 777. No world- or group-writable files, no setuid/setgid, no symlinks at all.

Where the installation lives, and a boundary that matters. The tree audited was collected from /home/uXXXXXXXX/domains/example.com/public_html/pb. The installation declares itself as https://sub.example.com (example.com/wp-config.php:98-99), while the order authorises example.com. Because of that, the browsing surfaces — what a visitor reaches by clicking, and what a page loads on its own — were not enumerated: deriving them from source would return addresses that never reach a visitor. What was measured, through the authorised hostname, is every individual file path listed in this report: the installation is served under example.com/pb/ for direct file requests (a themed WordPress 404 of 216,110 bytes answers at /pb/, against the host's own 404 of 4,511 bytes elsewhere), while rewrite-based routes do not resolve there. No request was ever sent to any other hostname.

🔒 The shared host itself needs its own audit, contracted by whoever administers it — without that, this assessment is incomplete by construction. The data in this account inherits the risk of the machine underneath it, and nothing in this report can speak for that machine.


2. Summary

This audit read the code of the WordPress installation served under example.com/pb, together with the environment the hosting account exposes, and observed the authorised domain from the outside.

What was examined. The tree holds 26,024 files, of which 16,628 are PHP, totalling 2,913,967 lines of PHP. Every one of the 26,024 files was verified by sha256 against the measurement manifest: 26,024 of 26,024 unchanged, 0 altered, 0 missing, 0 new — the tree audited is exactly the tree that was measured, and it is still exactly that now that the audit is finished.

Code read line by line. The code written or maintained for this site — the child theme, the must-use plugin the host injects, and four test plugins — is 1,368 lines across 9 files, and all of it was read line by line, including every one of its 12 points where a request value is read. The 46 off-the-shelf components were judged the way a site owner can act on them: installed version against the version published today, and against every security advisory whose affected range covers it. Their source was not read line by line — a defect in a third-party plugin belongs to its vendor, and a defect found in an outdated copy is usually one the vendor has already fixed.

Every address in this report was checked. 47 distinct file:line addresses: each resolves to exactly one file in the tree, that file has that line, and the line says what the text claims. Where a filename is not unique in the tree — there are 53 files called style.css, 37 called functions.php, 16 called exactly .htaccess, four called zyko-coverage-test.php and two called wp-config.php — the citation carries enough path to single one out, so you open the file we opened.

What was found: 22 findings — 2 high, 7 medium, 13 low. The two high ones share one mechanism: a published, unauthenticated code-execution path in a component installed here, and a server that executes PHP out of its upload directory. The single most useful thing in this report is not any individual finding: it is finding F-06, which explains why eleven of the others exist at all.

One number to hold on to. The installation's own configuration asks WordPress to install minor core updates automatically. A must-use plugin installed by the host overrides that, switches the updater off entirely, deletes WordPress's two built-in update health checks, and replaces them with one that reports "good" without checking anything. The result is measurable: the core is five release branches behind and 30 of 46 components are out of date, while the dashboard's Site Health page says automatic updates are keeping the site current.

3. What this report found, by impact

  1. An unauthenticated visitor has a published route to upload a file, and this server runs PHP from the upload folder — the two together put code execution within reach → F-01, High
  2. The active theme is four-plus releases behind and twelve published advisories reach it, two of which need no login at all → F-02, High
  3. Anything that can write a file into wp-content/uploads can get it executed, and one folder there is writable by everyone → F-03, Medium
  4. The store's customer and order data is reachable through a published SQL-injection flaw by anyone holding the Shop Manager role → F-04, Medium
  5. 1,032 PHP files of an unused, unmaintained theme are still being served and executed by the web server → F-05, Medium
  6. The site cannot update itself, and the page that would warn about it has been replaced by one that always says everything is fine → F-06, Medium
  7. A Contributor can read any file the site's PHP can reach, including the database credentials → F-07, Medium
  8. A customer account (Subscriber) can store script that runs against anyone who opens the page, administrators included → F-08, Medium
  9. A Contributor can store script and read template information they should not see → F-09, Medium
  10. The component whose job is to sanitise uploaded SVG is two releases behind, with two advisories reaching it → F-10, Low
  11. A link sent to a logged-in administrator can trigger a dropshipping action, and the domain can be used to redirect visitors elsewhere → F-11, Low
  12. A link can reflect script back into the site's own domain → F-12, Low
  13. An Editor can read data submitted through the contact forms → F-13, Low
  14. Four copies of a plugin that says of itself "do not use in production" are installed, one of which can write every database query and every requested URL into the uploads folder → F-14, Low
  15. The database credentials are readable beyond their owner, and the dashboard can edit the site's PHP files → F-15, Low
  16. The backup folder is open to the web, so the first backup taken becomes downloadable → F-16, Low
  17. A paid licence key is stored in a PHP file inside the upload tree → F-17, Low
  18. A leftover access rule in the root .htaccess grants nothing, protects nothing, and will take the whole site down with a 500 if the server changes → F-18, Low
  19. Visitors' browsers are given none of the protections the site could ask for → F-19, Low
  20. The published root still serves files that identify the installation, and the legacy XML-RPC endpoint is alive → F-20, Low
  21. Twelve components have no public channel, so nothing will ever tell the owner that they have fallen behind → F-21, Low
  22. The PHP version that actually serves the website could not be confirmed from here, and one question answers it → F-22, Low

4. Result

SeverityCount
🔴 High2
🟠 Medium7
🟡 Low13

By owner

OwnerFindings
Infra/sysadminF-03, F-15, F-16, F-17, F-18, F-19, F-20, F-22
Management/decisionF-01, F-02, F-04, F-05, F-06, F-07, F-08, F-09, F-10, F-11, F-12, F-13, F-14, F-21
Dev backendF-05 (removal), F-14 (removal), F-15 (wp-config.php constants)
ArchitectureF-06, F-21

Most of this report routes to Management/decision for one reason: the fix is updating or removing a purchased component, and that is a decision about the site, not a line of code to write. The owner's job here is to decide the order; section 8 gives it.


5. Findings

F-01 🔴 High — An unauthenticated file-upload flaw is published against the Elementor Pro version installed here, and this server executes PHP from the upload folder [ANALYSED]

Impact: the site owner loses the site. An attacker who needs no account reaches the point where the server runs code they supplied, and from there the database credentials, the salts that sign login cookies, and the store's customer and order data are all readable.

After access: with code execution inside the account, the attacker reads wp-config.php (database user, password and the eight authentication salts), can forge an authenticated session for any user including an administrator, and can write a new PHP file to come back through later. mu-plugins/ is writable by the account owner, which is where a loader survives plugin updates.

What the attacker takes: database credentials, the eight authentication salts, the customer and order tables, and persistent access. What the attacker does not take: anything outside this hosting account — CageFS isolation was measured (one /home entry, one process owner), so no neighbouring account is reachable, and no root privilege is involved. Persistence: high — a .php file written under uploads stays until someone finds it. Attribution: the requests appear as ordinary visitor traffic from the account's own PHP user. Radius: this site and everyone whose data it holds; it does not leave the product.

Precondition: none for the upload itself — the advisory classifies the attacker as Unauthenticated.

Evidence.

example.com/plugins/elementor-pro/elementor-pro.php:7   Version: 3.17.1   (installed)
CVE-2026-32475   Arbitrary File Upload   affected: <= 4.2.1   patched: 4.2.2
                 CVSS 9.0   required privilege: Unauthenticated
current version: 4.4.0, released 2026-11-09 (vendor changelog)

And the corridor, measured from outside through the authorised hostname:

GET https://example.com/pb/wp-content/uploads/ali2woo/pk.php   ->  200,     0 bytes, text/html
GET https://example.com/pb/zz-inexistente.php                  ->  404, 4,511 bytes
   the file on disk is 75 bytes of PHP that defines a function and prints nothing.
   Served as text it would return 75 bytes; absent it would return 404.
   200 with an empty body means the interpreter ran.

Rastro: example.com/plugins/elementor-pro/elementor-pro.php:7 declares Version: 3.17.1 (inside the affected range <= 4.2.1) → the upload lands under wp-content/uploads/ → uploads/ has no .htaccess (17 files match .htaccess* in the tree, none at the root of uploads/) → a .php file placed there is executed (measured above).

Why the version alone is not the whole finding. What the advisory publishes is a door. What this audit measured is the corridor behind it: on this server the upload directory executes PHP, which is what turns "a file was written" into "code ran". On a host that blocked execution under uploads, the same advisory would be a far smaller problem. That is why F-03 is a separate finding — it is the part the site owner controls directly, and fixing it reduces this one even before the update lands.

Recommendation. Update Elementor Pro to 4.4.0. It is a paid component, so the update comes from the Elementor account that holds the licence, not from the WordPress plugin screen. The fix for this specific flaw shipped in 4.2.2, so any version from 4.2.2 upward closes it; 4.4.0 is current. Then apply F-03, which blocks PHP execution under uploads and limits the whole class.

How to verify it was fixed. On the Plugins screen, Elementor Pro reads 4.2.2 or higher. Then re-run the corridor check: requesting /pb/wp-content/uploads/ali2woo/pk.php must return 403 instead of 200.


F-02 🔴 High — Twelve published advisories reach the version of the active theme, and two of them need no login [ANALYSED]

Impact: the site owner loses control of what the site renders and of what the site's PHP can read. Woodmart is the theme that builds every page a visitor sees; a flaw in it is a flaw in the whole site.

After access: the local-file-inclusion and object-injection entries let an attacker read files the site's PHP can reach — wp-config.php among them — and, through the deserialization path, reach code execution. The unauthenticated shortcode-execution entry lets a visitor make the site render content of their choosing, as the site.

What the attacker takes: file contents from inside the account, including database credentials; and, with shortcode execution, the ability to make the site publish arbitrary rendered output. What the attacker does not take: no root, no access to another hosting account, and the two unauthenticated entries do not hand over a user session on their own. Persistence: medium — it lasts only while the theme stays at 8.0.4; one update closes all twelve. Attribution: ordinary visitor requests in the access log. Radius: the entire site, because this is the active theme.

Precondition: none for CVE-2025-6744 and CVE-2026-56072. The local-file-inclusion entry CVE-2025-6746 requires Contributor.

Evidence — installed version, and that this theme is the active one.

themes/woodmart/style.css:7        Version: 8.0.4
uploads/2024/11/xts-theme_settings_default-1732847875.css
uploads/2024/11/xts-header_188698-1730730714.css
uploads/2024/12/xts-theme_settings_default-1733813389.css
   the `xts-` prefix is XTemos/Woodmart; these files are written by the theme when it
   compiles its settings. No comparable artefact exists for any other installed theme.

Advisories whose affected range covers 8.0.4 (read from NVD on 2026-11-13):

CVEclassaffectedneeds login?
CVE-2025-6744arbitrary shortcode execution<= 8.2.3no
CVE-2026-56072cross-site scripting<= 8.5.3no
CVE-2025-49935PHP local file inclusion< 8.3.2—
CVE-2026-23971deserialization → object injection<= 8.3.8—
CVE-2025-6746local file inclusion via layout<= 8.2.3Contributor
CVE-2026-32405exposure of sensitive system information<= 8.3.9—
CVE-2025-47600script-tag injection → code injection<= 8.3.7—
CVE-2025-6743stored XSS via multiple_markers<= 8.2.3—
CVE-2025-6745information exposure via woodmart_get_posts_by_query()<= 8.2.5—
CVE-2025-8097improper input validation of qty<= 8.2.6—
CVE-2025-49936DOM-based XSS< 8.3.2—
CVE-2026-27086DOM-based XSSbefore 8.3.8—

**What does not reach it, and this matters. CVE-2024-12333 — unauthenticated arbitrary shortcode execution via woodmart_instagram_ajax_query — stops at <= 8.0.3. The installed version is 8.0.4, one patch above it, so that one does not** apply here. The same is true of the advisories bounded at <= 7.2.4, <= 7.2.1, <= 7.1.1 and <= 7.0.4. The twelve above are the ones whose ranges genuinely cover 8.0.4, not the vendor's whole history.

Rastro: themes/woodmart/style.css:7 declares Version: 8.0.4 → the artefacts in uploads/2024//xts-.css show this theme is the one rendering pages → twelve published affected ranges include 8.0.4 → two of them require no authentication.

Current version. The vendor's changelog could not be read from here: xtemos.com/docs/woodmart/changelog/, xtemos.com/changelog/woodmart/ and woodmart.xtemos.com/changelog/ all returned 404, and the ThemeForest item page returned 403. A lower bound is still measurable: an advisory published against Woodmart affects <= 8.5.3, so at least 8.5.4 exists. Installed 8.0.4 is therefore at least four minor releases behind.

Recommendation. Update Woodmart to the current release from the XTemos account that holds the licence, and update woodmart-core with it — the two ship together. Because the theme is active, do it on a staging copy first: a major-version theme update changes templates.

How to verify it was fixed. themes/woodmart/style.css reads 8.5.4 or higher. Confirm against the changelog in your XTemos account that the release you install is at or above the highest bound in the table above.


F-03 🟠 Medium — PHP runs from inside wp-content/uploads, and one folder there is writable by everyone [MEASURED]

Impact: the site owner loses the last barrier between "a file got written" and "code ran". On its own this takes nothing; it is what turns every file-writing flaw on the site into code execution.

After access: nothing, by itself — this is a corridor, not a door. Combined with any write path it becomes what F-01 describes.

What the attacker takes: nothing without a write path, and this audit's census found no anonymous write endpoint in this tree. What the attacker does not take: no data, no session. Persistence: permanent until someone adds an .htaccess or changes the mode. Attribution: none — there is no event to log. Radius: all of uploads/, plus the one world-writable directory.

Precondition: some way to write a file into the upload tree. F-01 and F-02 each publish one.

Evidence.

GET https://example.com/pb/wp-content/uploads/ali2woo/pk.php   ->  200,     0 bytes, text/html
GET https://example.com/pb/zz-inexistente.php                  ->  404, 4,511 bytes
   75 bytes of PHP source on disk, 0 bytes returned, status 200  =>  the interpreter ran.

find <root> -name '.htaccess*'      ->  17 files; none at the root of uploads/
                                        (8 subfolders are protected; the root is not)

find /opt/enclave/areas/pedido-116/SourcesCode/example.com -type d -perm -o+w -printf '%m %p\n'
   777 /opt/enclave/areas/pedido-116/SourcesCode/example.com/uploads/complianz/snapshots
   -> one world-writable directory in the whole tree, and it is under uploads/.
find <root> -type d                 ->  5,707 directories; 5,706 are 755
find <root> -type f -perm -o+w      ->  0

Rastro: no .htaccess at wp-content/uploads/ → the server applies its default PHP handler to .php files there → a request for a .php file in that tree is executed (measured) → uploads/complianz/snapshots is additionally 777, so the file-system imposes no restriction on who may create a file in it.

Guard frontier — where the existing protection stops. Eight subfolders under uploads/ do carry a deny rule: wc-logs, woocommerce_uploads, wp-sheet-editor (and four of its children), wpcf7_uploads, mailchimp-for-wp, and zyko-debug. Four of them use deny from all, which is Apache 2.2 syntax; wpcf7_uploads and mailchimp-for-wp correctly carry both the 2.2 and the 2.4 forms. The protection is per-folder and was written by whichever plugin created the folder — the root of uploads/, and therefore uploads/2026/11/ where new media actually lands, has none.

On the 777 directory, said precisely. This report does not claim that another customer on the shared machine can write there. That was measured and it does not hold: CageFS is active, /home/ shows one entry, and the visible processes have a single owner, so no reachable third party was measured. What the 777 does mean is that the file-system's last check is gone for any write path inside the site itself, and it is gratuitous — the other 5,706 directories are 755 and work.

Recommendation.

  1. Create wp-content/uploads/.htaccess containing a rule that stops PHP from being executed in that tree. The two forms, both needed so the rule survives a server change:

``apache <FilesMatch "\.(?i:php|phtml|php3|php4|php5|php7|phps|pht|phar)$"> <IfModule mod_authz_core.c> Require all denied </IfModule> <IfModule !mod_authz_core.c> Order allow,deny Deny from all </IfModule> </FilesMatch> ``

  1. Set uploads/complianz/snapshots to 755, matching every other directory in the tree.
  2. Ask your host whether the same rule can be applied at the server level for this domain. An .htaccess is a file in your site; a server-level rule survives anyone emptying that file.

How to verify it was fixed. GET /pb/wp-content/uploads/ali2woo/pk.php returns 403 instead of 200. And find wp-content -type d -perm -o+w returns nothing.


F-04 🟠 Medium — Four published advisories reach this WooCommerce version, including SQL injection reachable by a Shop Manager [ANALYSED]

Impact: the store owner loses the confidentiality of the customer and order tables — names, e-mail addresses, order history and password hashes — to anyone holding the Shop Manager role, and can have an unexpected administrator account created by a link sent to a logged-in administrator.

After access: an SQL-injection foothold reads any table the database user can reach, and that user owns the whole site database. The CSRF entry creates an administrator, which then has the dashboard file editor available (see F-15) and therefore code execution without needing any further flaw.

What the attacker takes: the store's customer and order data, password hashes, and — through the CSRF path — a persistent administrator account. What the attacker does not take: no code execution from the injection itself, nothing from other accounts on the host. Persistence: the created administrator lasts until someone notices; the injection lasts while the version stays 9.7.3. Attribution: the injection appears as the Shop Manager who sent it; the CSRF appears as the administrator who was tricked, which is the worse half. Radius: the whole store and every customer in it.

Precondition: the SQL injection requires the Shop Manager role. The CSRF requires a logged-in administrator to open a link.

Evidence.

plugins/woocommerce/woocommerce.php:6   Version: 9.7.3   (installed)
current version: 11.3.0  (api.wordpress.org)

CVE-2026-57777  SQL Injection                            <= 10.9.4   patched 11.0    CVSS 7.6
                required privilege: Shop Manager
CVE-2026-3589   Arbitrary admin user creation via CSRF    <  10.5.3   patched 10.5.3  CVSS 4.3
CVE-2025-15033  Sensitive data exposure                   <= 10.4.2   patched 10.4.3  CVSS 6.5
CVE-2025-49042  Cross-site scripting                      <= 10.0.2   patched 10.0.3  CVSS 5.9

What does not reach it. The privilege-escalation entry is bounded 9.8.0 – 11.1.2 and 9.7.3 is below its floor; the denial-of-service entry starts at 11.1.0; and the XSS bounded <= 9.7.0 stops one patch short of the installed 9.7.3. Three advisories that a version-number glance would have counted do not apply.

Rastro: plugins/woocommerce/woocommerce.php:6 declares Version: 9.7.3 → four published affected ranges include it → the most serious needs the Shop Manager role, which is why this is graded Medium and not High.

Recommendation. Update WooCommerce. Going from 9.7.3 to 11.3.0 crosses two major versions, so do it on a staging copy and in steps (9.7.3 → 10.0 → 10.9 → 11.x), checking checkout and payment after each. If the jump has to wait, the minimum that closes the SQL injection is 11.0. Separately, review who holds Shop Manager on this store and remove the ones who no longer need it — that shrinks this finding immediately, at no risk.

How to verify it was fixed. WooCommerce reads 11.0 or higher on the Plugins screen, and Users → All Users filtered by Shop Manager lists only people who still need that role.


F-05 🟠 Medium — 1,032 PHP files of an unused theme and its framework are still served and executed by the web server [MEASURED]

Impact: the site owner is maintaining a second, invisible site. Nothing in the live site reaches this code, nobody updates it, and nobody will be told when an advisory is published against it — but the web server still hands it to anyone who asks, and still runs it.

After access: nothing today. The published Rehub advisories need the theme to be active for its hooks to load, and it is not active. Both files tested returned an empty body. What makes this worth reporting is that the state can change without anyone deciding anything: activating the theme, or the panel switching the PHP version, turns inert code into running code.

What the attacker takes: nothing today — measured, not assumed. What the attacker does not take: no file read, no XSS, no session; without the theme active, those endpoints do not exist. Persistence: the risk is temporal, not immediate. Attribution: not applicable. Radius: 1,419 files, 1,032 of them PHP, published and unmaintained.

Precondition: for the advisories to apply, the theme would have to be activated.

Evidence — unused, and still reachable. Both axes measured.

Axis 1 — nothing references it:
  themes/rehub-theme/style.css    Version: 19.6.3
  plugins/rehub-framework/...     Version: 19.6.3
  no rehub artefact anywhere under uploads/   (the theme artefacts there are xts-*, i.e. Woodmart)

Axis 2 — the server still serves and runs it:
  GET /pb/wp-content/themes/rehub-theme/style.css     -> 200, 139,120 bytes, text/css
  GET /pb/wp-content/themes/rehub-theme/404.php       -> 200,       0 bytes, text/html
  GET /pb/wp-content/plugins/rehub-framework/index.php-> 200,       0 bytes, text/html
     0 bytes with status 200 means PHP executed; the source was not served as text.

  rehub-theme      1,091 files   824 .php
  rehub-framework    328 files   208 .php

Published advisories reaching 19.6.3: CVE-2025-64197 (stored XSS, Rehub < 19.9.9.1), CVE-2025-67565 (exposure of sensitive system information, <= 19.9.9.1), CVE-2025-63050 (stored XSS, REHub Framework < 19.9.9.7), CVE-2025-14358 (missing authorization, <= 19.9.5), CVE-2025-63051 (sensitive information, < 19.9.9.4). Not reaching it: CVE-2024-31233 (SQL injection), CVE-2024-31231 and CVE-2024-31232 (path traversal → PHP local file inclusion), all bounded <= 19.6.1, one patch below the installed version.

Rastro: no artefact in uploads/ attributable to Rehub, while Woodmart's artefacts are present → the theme is not the one rendering pages → its files are nonetheless returned by the web server, and its .php files are executed (measured) → five published advisories cover 19.6.3, and would apply the moment it were activated.

This code was not read line by line, and that is deliberate. Reading code that should be deleted spends effort without changing the recommendation. The recommendation does not rest on finding a defect inside it — it rests on the two axes above plus the fact that nobody is responsible for it.

One part of it was read, because a census pointed at it, and the result belongs here. The Rehub Framework plugin registers an AJAX endpoint that dispatches a function named in the request:

plugins/rehub-framework/vendor/vafpress/bootstrap.php:66   add_action('wp_ajax_vp_ajax_wrapper', 'vp_ajax_wrapper');
plugins/rehub-framework/vendor/vafpress/bootstrap.php:72       $function = $_POST['func'];
plugins/rehub-framework/vendor/vafpress/bootstrap.php:75       if( VP_Security::instance()->is_function_whitelisted($function) )
plugins/rehub-framework/vendor/vafpress/bootstrap.php:81           $result['data'] = call_user_func_array($function, $params);

This is not an arbitrary-function-call flaw, and this report does not claim it is. The guard at :75 is a closed allow-list: plugins/rehub-framework/vendor/vafpress/classes/security.php:12 starts it as an empty array, :32-37 matches with a strict in_array (no pattern, no prefix), and the only things that ever populate it are 44 literal function names registered across data/sources.php and includes/option_helpers.php. Nothing passes a variable into it, so no new name can be injected. If the flaw existed, in_array('system', …) would be true — and system is not among the 44.

What is true, and is the frontier of that guard: the hook is wp_ajax_, not wp_ajax_nopriv_, and there is no capability check and no nonce. So any logged-in user, down to a Subscriber, can invoke those 44 theme-helper functions with parameters of their own choosing — two of which (top_table_shortcode and top_charts_shortcode, option_helpers.php:209 and :216) render shortcodes. Judging all 44 would mean reading a third-party component hunting for a defect nobody has published, which is not what this audit is for. Deleting the plugin closes the question instead of answering it — which is the recommendation anyway, and now for a second reason.

Recommendation. Delete, from the dashboard: the Rehub theme, the Rehub Framework plugin, the Hostinger Affiliate theme and plugin (same situation: 245 + 553 files, no artefact of use), and the two default themes you are not keeping as a fallback — WordPress only needs one. Deleting closes the whole class at once, including the defects nobody has found yet. Keep one default theme (twentytwentyfour) so the site has something to fall back to if Woodmart fails.

How to verify it was fixed. GET /pb/wp-content/themes/rehub-theme/style.css returns 404, and Appearance → Themes lists only Woodmart, Woodmart Child and one default theme.


F-06 🟠 Medium — The site cannot update itself, and the health check that would say so always reports "good" [MEASURED]

Impact: the site owner is told the site is current when it is not. This is the finding that explains eleven of the others: the core is five release branches behind and 30 of 46 components are out of date, while the dashboard reports that automatic updates are keeping everything up to date.

After access: nothing directly — no data moves because of this. What it does is guarantee that every other component keeps ageing, and that nobody is warned.

What the attacker takes: nothing. What the attacker does not take: nothing — this is not a door, it is the reason the other doors stay open. Persistence: total, and quiet: the plugin lives in mu-plugins/, which does not appear on the Plugins screen, so nobody will deactivate it by accident or on purpose. Attribution: not applicable. Radius: all 46 components, plus the Site Health page that an owner would check to reassure themselves.

Precondition: none. This is the installation's current state.

Evidence — the updater is switched off.

mu-plugins/hostinger-auto-updates.php:25   add_filter( 'automatic_updater_disabled', '__return_true' );
mu-plugins/hostinger-auto-updates.php:26   add_filter( 'auto_update_theme', '__return_false' );
mu-plugins/hostinger-auto-updates.php:27   add_filter( 'auto_update_plugin', '__return_false' );

And the two checks that would have reported it are removed and replaced:

mu-plugins/hostinger-auto-updates.php:316  unset( $tests['async']['background_updates'] );
mu-plugins/hostinger-auto-updates.php:317  unset( $tests['direct']['plugin_theme_auto_updates'] );
mu-plugins/hostinger-auto-updates.php:325      'status'      => 'good',
mu-plugins/hostinger-auto-updates.php:330      'description' => __( 'Automatic updates ensure your site is
                                                 always running the latest and most secure versions of
                                                 WordPress, plugins, and themes.' ),

The replacement test's return value is a literal. There is no check anywhere in it — status is the string 'good' on every call, whatever the state of the site.

And the site's own configuration asks for the opposite:

example.com/wp-config.php:97   define( 'WP_AUTO_UPDATE_CORE', 'minor' );

Rastro: example.com/wp-config.php:97 asks WordPress to install minor core updates automatically → mu-plugins/hostinger-auto-updates.php:25 returns true from automatic_updater_disabled, and must-use plugins are loaded before every other plugin → the updater never runs → :316-317 delete the two health checks that would report that → :325 reports good unconditionally.

The part that does not depend on reading WordPress's own source. How WordPress resolves the constant against the filter is core behaviour, and the core is not in the audited tree, so this report does not rest on it. It rests on the outcome, which was measured: core 6.6.11 against 7.1.4 published today, and 30 of 46 components behind, several by many releases. A green indicator that never measures anything is an alarm that is switched off.

This is not the client's code. The file is installed by the hosting provider. That changes who to raise it with, not whether it is a problem.

Recommendation. Two things, in this order.

  1. Do not trust the Site Health page on this installation. Until the plugin changes, that page cannot tell you whether updates are happening. Use section 6 of this report as the baseline and check versions on the Plugins and Themes screens directly.
  2. Ask Hostinger, in writing, what their managed updates actually cover and when they last ran for this site. The measurement to put in front of them is in this report: WordPress 6.6.11 while 7.1.4 is current, and 30 of 46 components behind. If their answer is that they only backport core security releases, then plugin and theme updates are yours to do, and you need to know that. If you would rather do it all yourself, ask them to remove the must-use plugin and turn WordPress's own automatic updates back on.

How to verify it was fixed. Either the automatic_updater_disabled filter is gone and Tools → Site Health shows WordPress's own background_updates test again, or you have a written answer from the host saying what they update and on what schedule. Either outcome ends the ambiguity; the current state is the problem.


F-07 🟠 Medium — A Contributor can read any file the site's PHP can reach, through this Slider Revolution version [ANALYSED]

Impact: the site owner loses the database credentials and the cookie salts to anyone holding a Contributor account, because wp-config.php is one of the files that becomes readable.

After access: with the database credentials and the eight salts, an attacker can forge an authenticated session and act as an administrator. Reading is enough; nothing has to be written.

What the attacker takes: the contents of any file the PHP process can open — wp-config.php first among them. What the attacker does not take: no write access (the upload advisory for this plugin does not reach this version), no root. Persistence: none by itself; the attacker reads and leaves. Attribution: appears as the authenticated Contributor. Radius: every file the PHP process can reach — and open_basedir was measured empty in the shell interpreter, which widens that set (see F-22 for what that measurement does and does not cover).

Precondition: a Contributor account. On a store, that account exists only if someone created it or if open registration grants it.

Evidence.

plugins/revslider/revslider.php:9    Version: 6.7.24   (installed)
current version: 7.2.10   (vendor changelog, sliderrevolution.com/documentation/changelog)

CVE-2025-10249  Missing authorization -> Contributor+ arbitrary file read   <= 6.7.37
                patched 6.7.38   CVSS 6.5   required privilege: Contributor
                Contributor+ arbitrary file read via used SVG / used images  <= 6.7.36
                Contributor+ arbitrary plugin deactivation          6.0.0 - 6.7.55

What does not reach it. The Subscriber+ arbitrary file upload entry is bounded 7.0.0 – 7.0.10; the installed 6.7.24 is outside that range entirely, so the most serious advisory in this plugin's history does not apply here. The entries bounded <= 6.7.18, <= 6.7.13, <= 6.7.11, <= 6.7.10 and <= 6.7.7 are all below the installed version.

Rastro: plugins/revslider/revslider.php:9 declares Version: 6.7.24 → three published affected ranges include it → the two file-read entries require only Contributor → the file read reaches wp-config.php, which holds the database credentials and the eight salts.

Recommendation. Update Slider Revolution to 7.2.10 from the ThemeForest/Slider Revolution account that holds the licence. This crosses a major version, so stage it. The minimum that closes the file-read entry is 6.7.38. Also review who holds Contributor and above on this site.

How to verify it was fixed. plugins/revslider/revslider.php reads 6.7.38 or higher.


F-08 🟠 Medium — A customer account can store script that runs against anyone who opens the page, through this WPBakery version [ANALYSED]

Impact: the site owner has script running in their own domain against their own visitors and staff. On a store, Subscriber is the role a customer gets, so the required privilege is one anybody can obtain if registration is open.

After access: stored script runs with whatever the viewing user's session can do. If an administrator opens the affected page, it acts as the administrator — and an administrator on this site has the dashboard file editor available (F-15).

What the attacker takes: actions and data available to whoever views the page, administrators included. What the attacker does not take: no PHP execution of its own, no arbitrary file read. Persistence: stored — it lasts until the content is cleaned. Attribution: the payload sits in content, traceable to the Subscriber who saved it. Radius: everyone who visits the affected page.

Precondition: a Subscriber account.

Evidence.

plugins/js_composer/js_composer.php:6   Version: 8.1   (installed; this is WPBakery Page Builder,
                                        per `Plugin URI: https://wpbakery.com` in the header)

affected ranges covering 8.1:
  <= 9.0.1   Subscriber+ stored XSS
  <= 8.7.4   Subscriber+ stored XSS via `data` parameter
  <= 8.7.2   broken access control
  <= 8.6.1   stored XSS via custom JS module
  <= 8.6.1   stored XSS via vc_custom_heading shortcode
  <= 8.6     Contributor+ stored XSS
  <= 8.5     Contributor+ stored XSS
  <= 8.4.1   Contributor+ stored XSS via multiple page builder elements
not reaching: <= 7.6, <= 7.5

Current version. Not measured directly: kb.wpbakery.com/docs/preface/changelog/ and wpbakery.com/changelog/ both returned 404 from here — the vendor's changelog sits behind a customer login. A lower bound is measurable from the advisory data: one of the entries above is recorded as patched in 9.0.2, so at least 9.0.2 exists. Installed 8.1 is therefore at least a full minor line behind.

Rastro: plugins/js_composer/js_composer.php:6 declares Version: 8.1 → eight published affected ranges include it → the two most serious need only Subscriber.

Recommendation. Update WPBakery Page Builder from your account on the vendor's site (it is bundled with some themes; if it came with Woodmart, the Woodmart update in F-02 may carry it). Minimum to close the Subscriber-level entries: 9.0.2. Then check whether this builder is still in use at all — the site also runs Elementor, and two page builders is usually one more than a site needs.

How to verify it was fixed. plugins/js_composer/js_composer.php reads 9.0.2 or higher, or the plugin is gone.


F-09 🟠 Medium — Seven published advisories reach this Elementor version, reachable by a Contributor [ANALYSED]

Impact: the site owner has script stored in their own pages by a low-privileged account, and template information exposed to accounts that should not see it. Elementor is one of the two page builders rendering this site.

After access: stored script runs as whoever views the page. The missing-authorization entry exposes information about templates and content to an account that should not be able to read it.

What the attacker takes: script execution in the site's domain against page viewers, and template and content metadata. What the attacker does not take: no PHP execution, no arbitrary file read. Persistence: stored, until the content is cleaned. Attribution: the Contributor account. Radius: pages using the affected widget.

Precondition: a Contributor account.

Evidence.

plugins/elementor/elementor.php:6   Version: 4.1.1   (installed)
current version: 4.4.0   (api.wordpress.org)

affected ranges covering 4.1.1:
  <  4.3.6   CVE-2026-104718  Contributor+ stored XSS            CVSS 6.5
  <= 4.3.5   cross-site scripting   (four separate entries)
  <= 4.2.3   missing authorization -> Contributor+ sensitive information exposure
  <= 4.1.3   sensitive data exposure
not reaching: <= 4.1.0 (broken access control), <= 4.0.4, <= 3.35.7, <= 3.35.5, <= 3.29.0

Rastro: plugins/elementor/elementor.php:6 declares Version: 4.1.1 → seven published affected ranges include it, and the broken-access-control entry bounded <= 4.1.0 misses it by one patch → the reachable ones require Contributor.

Recommendation. Update Elementor to 4.4.0 on the Plugins screen. Do it together with Elementor Pro (F-01) — the two are version-paired, and Pro will not run correctly against a mismatched Elementor.

How to verify it was fixed. Elementor reads 4.4.0 or higher, and Elementor Pro 4.4.0 beside it.


F-10 🟡 Low — The component whose job is to sanitise uploaded SVG is two releases behind, with two advisories reaching it [ANALYSED]

Impact: the site owner has a Contributor-level path to reach objects belonging to other authors and to store script. The reason this is worth writing down even at Low is what the component is for: SVG is the only image format that can carry script, and Safe SVG is the thing standing between an uploaded SVG and the site.

After access: access to an object belonging to another author (the IDOR), and script in the site's domain (the XSS). Neither is a code-execution path.

What the attacker takes: another author's object, and script execution against page viewers. What the attacker does not take: no PHP execution, and not a bypass of the SVG sanitiser — that is not what either advisory describes, and this report does not claim it. Persistence: low. Attribution: the Contributor account. Radius: objects of the same type.

Precondition: a Contributor account.

Evidence.

plugins/safe-svg/safe-svg.php:6   Version: 2.3.1   (installed)
current version: 2.5.2   (api.wordpress.org)

CVE-2026-94672   Insecure Direct Object References (IDOR)   <= 2.5.0   patched 2.5.1   CVSS 4.3
CVE-2026-94077   Cross Site Scripting                       <= 2.5.0   patched 2.5.1   CVSS 6.5
                 both: required privilege Contributor

Rastro: plugins/safe-svg/safe-svg.php:6 declares Version: 2.3.1 → both published ranges cover it → both require Contributor.

Recommendation. Update Safe SVG to 2.5.2 on the Plugins screen — it is a free component from the WordPress repository, so this is a one-click update. While you are there, decide whether this site needs SVG upload at all; if it does not, removing the plugin removes the format and the question.

How to verify it was fixed. Safe SVG reads 2.5.1 or higher.


F-11 🟡 Low — A link sent to a logged-in administrator can trigger a dropshipping action, and the domain can be used to redirect visitors [ANALYSED]

Impact: the site owner has an action on their site that fires because an administrator opened a link, and a page on their own domain that will forward a visitor to an address of the attacker's choosing — which is what makes a phishing link look trustworthy.

After access: the CSRF performs whatever the targeted plugin action does, as the administrator. The open redirect lends the site's domain and its TLS certificate to someone else's landing page.

What the attacker takes: one administrative action in this plugin, and the credibility of the domain. What the attacker does not take: no site data, no session, no code execution. Persistence: none. Attribution: the action appears as the administrator who was tricked, which is the worse half of this. Radius: the dropshipping plugin, and whoever clicks the link.

Precondition: for the CSRF, a logged-in administrator must open a link the attacker supplies.

Evidence.

plugins/ali2woo/ali2woo.php:8   Version: 3.4.5   (installed; `Plugin URI: https://ali2woo.com/pricing/`,
                                i.e. the paid edition)
current version: 3.7.3   (vendor changelog, ali2woo.com/changelog — the AliNext line)

CVE-2024-37213   Cross-Site Request Forgery   AliNext through <= 3.4.6
CVE-2025-30859   Open redirect -> phishing    AliNext through <= 3.5.1
not reaching: the entries bounded at Ali2Woo Lite <= 3.3.5

One qualification, stated rather than hidden. Both advisories are published against the vendor's AliNext / ali2woo-lite line, and the artefact installed here is the paid edition. The version numbering is the same line (3.4.x running up to the current 3.7.3 in the same changelog), so the ranges cover 3.4.5. If the paid build differs, the vendor is the one who can say so.

Rastro: plugins/ali2woo/ali2woo.php:8 declares Version: 3.4.5 → two published ranges on the same version line cover it → the CSRF needs an administrator to click; the open redirect needs nothing.

Recommendation. Update to 3.7.3 from your ali2woo account. When you do, ask the vendor to confirm that CVE-2024-37213 and CVE-2025-30859 were fixed in the paid build as well as the Lite one.

How to verify it was fixed. plugins/ali2woo/ali2woo.php reads 3.5.2 or higher (which clears both ranges); 3.7.3 is current.


F-12 🟡 Low — A link can reflect script back into the site's own domain, through this YellowPencil version [ANALYSED]

Impact: the site owner has a page on their own domain that will run script supplied in the URL against whoever opens that URL. The attacker is anonymous; the victim is whoever clicks.

After access: the script runs with whatever the clicking user's session has. Nothing is stored, so the effect ends with that page view.

What the attacker takes: whatever the clicking user's session can reach. What the attacker does not take: nothing persistent, nothing from the database on its own. Persistence: none — it lives in the link. Attribution: appears as the victim who clicked. Radius: one visitor per link.

Precondition: the victim must open a link the attacker supplies.

Evidence.

plugins/waspthemes-yellow-pencil/yellow-pencil.php:6   Version: 7.6.0   (installed)
current version: 7.6.7   (vendor changelog, yellowpencil.waspthemes.com/changelog)

affected ranges covering 7.6.0:
  <= 7.6.4   reflected cross-site scripting
  <= 7.6.1   cross-site scripting
not reaching: <= 7.5.8, <= 7.5.3, and the unauthenticated arbitrary options update at <= 7.2.0

Rastro: plugins/waspthemes-yellow-pencil/yellow-pencil.php:6 declares Version: 7.6.0 → two published ranges cover it → the unauthenticated options-update entry, which would have been far more serious, stops at <= 7.2.0 and does not apply here.

Recommendation. Update to 7.6.7 from your WaspThemes account. This plugin is a visual CSS editor; if the site's styling is now done in Woodmart or Elementor, check whether it is still used — uploads/yellow-pencil/custom-15.css shows it has run at some point, so confirm before removing.

How to verify it was fixed. plugins/waspthemes-yellow-pencil/yellow-pencil.php reads 7.6.5 or higher.


F-13 🟡 Low — An Editor can read data submitted through the contact forms [ANALYSED]

Impact: the site owner has form submissions — whatever visitors typed into them — readable by accounts holding the Editor role, which is a role given out for publishing, not for handling personal data.

After access: the data is read. There is no further step.

What the attacker takes: user data stored by the form plugin. What the attacker does not take: nothing at all for anyone below Editor; no file, no code execution. Persistence: none. Attribution: the Editor account. Radius: the form submissions.

Precondition: an Editor account.

Evidence.

plugins/contact-form-7/wp-contact-form-7.php:10  Version: 6.0.6   (installed)
current version: 6.2.5   (api.wordpress.org)

CVE-2026-120407   Editor+ sensitive user data disclosure   < 6.2.5   patched 6.2.5   CVSS 4.5

What does not reach it, and it is the one that would have mattered. Contact Form 7 also carries an unauthenticated arbitrary file upload (CVE-2026-114432, CVSS 8.1). Its affected range is 6.2 – 6.2.3, patched in 6.2.4 — and the advisory states why there is a floor: "6.1.7 still gated it — hence the 6.2 lower bound". The installed version is 6.0.6, below that floor, so it is not affected. Had this report counted advisories by the version in their title instead of by their affected range, it would have reported a critical unauthenticated upload flaw on this site that does not exist.

Rastro: plugins/contact-form-7/wp-contact-form-7.php:10 declares Version: 6.0.6 → one published range (< 6.2.5) covers it → it requires the Editor role.

Recommendation. Update Contact Form 7 to 6.2.5 on the Plugins screen — free component, one-click. Note that this moves the site into the range of the upload flaw's predecessor only if you stop at 6.2–6.2.3; go straight to 6.2.5 or later, never to 6.2.x below 6.2.4.

How to verify it was fixed. Contact Form 7 reads 6.2.5 or higher — not 6.2.0 through 6.2.3.


F-14 🟡 Low — Four copies of a plugin that says of itself "do not use in production" are installed on this site, and one can write every database query and every requested URL into the uploads folder [MEASURED]

Impact: the site owner is running diagnostic instrumentation on a live store. One of the four copies, when its logging is active, writes the first 300 characters of every SQL statement and every requested URL into a file inside the upload tree — which means order identifiers, search terms and anything else that travels in a URL, in a folder whose only protection is an .htaccess. All four rewrite every database query the site makes.

After access: from outside, the only thing reachable without a login returns a timestamp, a version and an environment name. If the log files are present, whoever can read them gets a transcript of the site's database traffic.

What the attacker takes: from outside today, ts, ver and env — nothing about any user. If logging is re-enabled, a transcript of SQL and request URLs. What the attacker does not take: the log contents today, because the directory is empty; and not the dump or clear endpoints, which do require manage_options. Persistence: the HMAC key stays in wp_options after deactivation, by the plugin's own design. Attribution: not applicable. Radius: every request the site handles, while the instrumentation is on.

Precondition: none for the public endpoint. The logging requires the instrumented copy to be active.

Evidence — four copies of the same thing, and what each declares about itself.

plugins/zyko-coverage-test/zyko-coverage-test.php                  Version: 0.1.0-poc
plugins/zyko-coverage-test-v2/zyko-coverage-test.php               Version: 2.0.2
plugins/zyko-coverage-test-v21/zyko-coverage-test.php              Version: 2.1.2
plugins/zyko-coverage-test-v2-instrumented/zyko-coverage-test.php  Version: 2.0.2-instrumented

plugins/zyko-coverage-test/zyko-coverage-test.php:5
   "Plugin de TESTE - instala, mede coverage, desinstala. Nao usa para producao."

The logging, and the proof it has run:

plugins/zyko-coverage-test-v2-instrumented/zyko-coverage-test.php:27
   define('ZYKO_V2_LOG_DIR', WP_CONTENT_DIR . '/uploads/zyko-debug');
plugins/zyko-coverage-test-v2-instrumented/zyko-coverage-test.php:56
   $entry['request_uri'] = isset($_SERVER['REQUEST_URI']) ? substr($_SERVER['REQUEST_URI'], 0, 200) : '';
plugins/zyko-coverage-test-v2-instrumented/zyko-coverage-test.php:92
   'sql_first_300' => substr($query, 0, 300),        (and again at :133 and :163)
plugins/zyko-coverage-test-v2-instrumented/zyko-coverage-test.php:58
   @file_put_contents(ZYKO_V2_LOG_DIR . '/' . $file, $line, FILE_APPEND | LOCK_EX);
plugins/zyko-coverage-test-v2-instrumented/zyko-coverage-test.php:44
   file_put_contents($htaccess, "Deny from all\n");

on disk:  uploads/zyko-debug/.htaccess   14 bytes   ->  the directory exists, so the plugin ran.
          uploads/zyko-debug/            contains nothing else — no log files present now.

The one endpoint with no capability check:

plugins/zyko-coverage-test-v21/zyko-coverage-test.php:206   if (!isset($_GET['zyko_v2_time'])) return;
plugins/zyko-coverage-test-v21/zyko-coverage-test.php:210       $ip = $_SERVER['REMOTE_ADDR'] ?? 'unknown';
plugins/zyko-coverage-test-v21/zyko-coverage-test.php:226       echo wp_json_encode(['ts' => ..., 'ver' => ..., 'env' => ...]);

Guard frontier — where the protection stops. Eleven of the twelve request-value reads across these four files are followed immediately by current_user_can('manage_options') (plugins/zyko-coverage-test/zyko-coverage-test.php:47 and :88; -v2:113, :143, :202; -v21:237; -v2-instrumented:183, :218, :234). The twelfth, ?zyko_v2_time at -v21:206, has no capability check — it is public by design, with a 60-per-minute limit keyed on md5($_SERVER['REMOTE_ADDR']). That key is the frontier: the site answers behind a CDN (Server: hcdn, measured), so if the edge terminates the connection the whole site shares one counter rather than one per visitor. Either way the endpoint returns no user data, which is why this is Low and not higher.

And the .htaccess protecting the log directory is the weaker kind. Deny from all is Apache 2.2 syntax. On an Apache 2.4 build without mod_access_compat, or on nginx, that line does not protect anything — and the panel on this account offers twenty PHP directories, so the stack is not fixed.

Rastro: four copies of the same plugin in plugins/ → the instrumented copy defines its log directory under uploads/ (:27) and writes the request URL (:56) and the first 300 characters of each SQL statement (:92, :133, :163) into it, appending at :58 → uploads/zyko-debug/ exists on disk, so that code has executed → the directory's only protection is an Apache-2.2-only .htaccess the plugin wrote itself (:44) → and uploads/ is a tree where PHP executes (F-03).

Recommendation. Delete all four plugins from the dashboard, then delete the wp-content/uploads/zyko-debug/ directory. Because the plugins' deactivation hooks deliberately keep their options, also remove the leftover rows from wp_options: zyko_hmac_secret, zyko_v2_activated_at, zyko_v21_activated_at, zyko_secret_in_wp_options_since, zyko_coverage_activated_at and zyko_v2_show_secret_notice. If the instrumentation is still needed, run it on a staging copy, not on the store.

How to verify it was fixed. GET /pb/?zyko_v2_time no longer returns JSON, the Plugins screen lists none of the four, and SELECT option_name FROM wp_options WHERE option_name LIKE 'zyko%'; returns no rows.


F-15 🟡 Low — The database credentials are readable beyond their owner, and the dashboard can edit the site's PHP files [MEASURED]

Impact: the site owner has two of the cheapest hardening steps left undone. wp-config.php is readable by group and others, so any file-read path reaches the database password and the eight cookie salts. And because the plugin/theme file editor is enabled, an attacker who reaches any administrator account gets code execution without needing a vulnerability at all.

After access: nothing on its own. After a compromise of an administrator account, the file editor is a direct path to writing PHP into the active theme; and the 644 mode means a file read is enough to take the database.

What the attacker takes: nothing by itself. Afterwards: the database credentials and the salts, and through the editor, code execution. What the attacker does not take: nothing for anyone who is not an administrator, and nothing from another account on the host — CageFS isolation was measured. Persistence: a .php file edited through the dashboard stays. Attribution: appears as the administrator. Radius: the whole site, after the door.

Precondition: for the file editor, an administrator account. For the 644, any file-read path — and F-02 and F-07 each publish one.

Evidence.

inventario/permissoes.txt
   644 uXXXXXXXX:oXXXXXXXX /home/uXXXXXXXX/domains/example.com/public_html/pb/wp-config.php
   644 uXXXXXXXX:oXXXXXXXX /home/uXXXXXXXX/domains/example.com/public_html/pb/.htaccess
   755 uXXXXXXXX:oXXXXXXXX /home/uXXXXXXXX/domains/example.com/public_html/pb

wp-config.php            DISALLOW_FILE_EDIT   -> not defined (0 occurrences)
                         DISALLOW_FILE_MODS   -> not defined (0 occurrences)
                         FS_METHOD            -> 'direct'  (example.com/wp-config.php:95)

644 was measured at the source, by the collection, and again in the copy — the two agree, so it is not an artefact of copying.

Said precisely. With CageFS active and a single uid in this account, no reachable third party was measured that the group/other read bit would hand the file to. That is why this is Low. What the mode does change is that any arbitrary-file-read flaw — and this site has two components with published ones — reaches the credentials, where 600 would not help against a read by the PHP user itself but 640/600 closes every other reader.

Controls that passed, and they are worth knowing. wp-config.php is not reachable over HTTP (GET /pb/wp-config.php → 403). WP_DEBUG is false (example.com/wp-config.php:91-92). The eight authentication salts are unique, not the sample defaults. DB_HOST is 127.0.0.1, not a public address.

Rastro: wp-config.php is mode 644 at the source (inventario/permissoes.txt) → group and other can read it → it holds DB_PASSWORD and the eight salts (names and locations only; no value is quoted in this report) → DISALLOW_FILE_EDIT is absent from wp-config.php, so Appearance → Theme File Editor and Plugins → Plugin File Editor are available to any administrator.

Recommendation. Add two lines to wp-config.php, above the require_once ABSPATH . 'wp-settings.php'; line:

define( 'DISALLOW_FILE_EDIT', true );
define( 'FORCE_SSL_ADMIN', true );

and set the file's mode to 640:

chmod 640 ~/domains/example.com/public_html/pb/wp-config.php

If the site breaks after 640, your PHP runs as a different user than the file's owner — revert to 644 and tell us, because that is worth knowing.

How to verify it was fixed. Appearance no longer offers Theme File Editor, and ls -l wp-config.php shows -rw-r-----.


F-16 🟡 Low — The backup folder is open to the web, so the first backup taken becomes downloadable [MEASURED]

Impact: the site owner has a folder where a full copy of the site and its database is meant to be written, and that folder is readable over the internet. There is no backup in it today. The moment one is created, it is downloadable by anyone who learns its name — and a migration archive contains the database, which contains password hashes and every customer record.

After access: with the archive, an attacker has the site offline: the database, the uploads, the configuration. Nothing further is needed.

What the attacker takes: nothing today — there is no archive. On the first backup: the entire site and the entire database. What the attacker does not take: nothing while the folder stays empty. Persistence: the archive stays until somebody deletes it. Attribution: one GET in the access log. Radius: everything the site holds.

Precondition: a backup has to be created into that folder. Taking a backup is a normal thing to do, which is what makes this worth fixing before it happens.

Evidence.

ai1wm-backups/.htaccess   (191 bytes, 9 lines, read in full)
   :1-3  <IfModule mod_mime.c>      AddType application/octet-stream .wpress
   :4-6  <IfModule mod_dir.c>       DirectoryIndex index.php
   :7-9  <IfModule mod_autoindex.c> Options -Indexes
   -> no `Deny from all` and no `Require all denied` anywhere in those 9 lines.

GET https://example.com/pb/wp-content/ai1wm-backups/index.html  ->  200, 26 bytes
   -> files inside the folder are served.

find <root> -name '*.wpress'  ->  nothing. No archive exists right now.

The asymmetry that settles it. The same vendor protects its other directory properly:

plugins/all-in-one-wp-migration/storage/.htaccess   (477 bytes, 25 lines)
   :1  <IfModule mod_authz_core.c>
   :2    <FilesMatch ".*">
   :3      Require all denied

So the absence in ai1wm-backups/ is an exception, not the plugin author's style.

Rastro: ai1wm-backups/.htaccess:2 sets a MIME type and ai1wm-backups/.htaccess:8 disables directory indexing, and no line of the file's nine denies access → a file inside the folder returns 200 (measured) → AddType application/octet-stream .wpress means a .wpress archive would download cleanly rather than being rendered → the only thing standing between the archive and the internet is nobody knowing its filename.

Recommendation. Add the deny rule to wp-content/ai1wm-backups/.htaccess, keeping what is already there:

<IfModule mod_authz_core.c>
    <FilesMatch ".*">
        Require all denied
    </FilesMatch>
</IfModule>
<IfModule !mod_authz_core.c>
    Order allow,deny
    Deny from all
</IfModule>

Better still: configure All-in-One WP Migration to write backups outside public_html, or download each archive and delete it from the server. A rule in a file can be emptied; a folder outside the published root has no URL at all.

How to verify it was fixed. GET /pb/wp-content/ai1wm-backups/index.html returns 403 instead of 200.


F-17 🟡 Low — A paid licence key is stored in a PHP file inside the upload tree [MEASURED]

Impact: the site owner has a purchased licence key sitting in the folder the site writes uploads into, instead of in the configuration file. Any arbitrary-file-read path reaches it — and this site has two components with published file-read advisories.

After access: the key is read. The loss is commercial — someone else can use the subscription — not a loss of the site.

What the attacker takes: the plugin licence key, if they already have a file-read path. What the attacker does not take: nothing over the plain web (see below), nothing from the database, no session. Persistence: the key works until the vendor rotates it. Attribution: none. Radius: the plugin subscription.

Precondition: an arbitrary-file-read path. F-02 (CVE-2025-6746, CVE-2025-49935) and F-07 (CVE-2025-10249) each publish one.

Evidence.

uploads/ali2woo/pk.php:1     75 bytes, a single line, mode 644
   defines one function, `a2w_plugin_key()`, returning a literal key.  The value is
   deliberately not reproduced in this report — the finding is the key and the place,
   not the secret.

What this is not. Requesting the file over the web does not leak the key: because PHP executes in that tree (F-03), GET /pb/wp-content/uploads/ali2woo/pk.php returns 200 with 0 bytes — the function is defined and nothing is printed. There is no direct leak today, and this report does not claim one.

Rastro: uploads/ali2woo/pk.php:1 holds a credential literal, returned by a2w_plugin_key() → the directory has no .htaccess and is the tree the site writes uploads into → a file read of that path returns the literal → two installed components carry published Contributor-level arbitrary-file-read advisories.

Recommendation. Ask ali2woo/AliNext support to rotate the licence key and to tell you whether the key can live in wp-config.php as a constant instead of in a file under uploads. Then delete wp-content/uploads/ali2woo/pk.php. Rotate first, delete second — if the file has been read, deleting it does not undo that.

How to verify it was fixed. wp-content/uploads/ali2woo/pk.php no longer exists, and the plugin still reports its licence as active.


F-18 🟡 Low — A leftover access rule in the root .htaccess grants nothing, protects nothing, and will take the site down with a 500 if the server changes [MEASURED]

Impact: the site owner is one server migration away from the whole site returning HTTP 500, because of three lines that currently do nothing. The same lines publish an IP address associated with administering the site.

After access: nothing. This is not an attack path; it is a latent availability problem plus an unnecessary disclosure.

What the attacker takes: the IP address the directive names, if the .htaccess were ever served — and today it is not. What the attacker does not take: no access at all; the directive grants nothing. Persistence: not applicable. Attribution: not applicable. Radius: the real risk is availability — a legacy directive becomes an unknown directive, and an unknown directive in .htaccess is a 500 for the entire directory.

Precondition: for the 500, the host moving this account to an Apache 2.4 build without mod_access_compat, or to nginx. The panel on this account offers twenty PHP directories, so the stack is not something you control.

Evidence.

example.com/.htaccess:29   # DO NOT REMOVE THIS LINE AND THE LINES BELOW IPALLOWID:REDACTED
example.com/.htaccess:30   allow from 203.0.113.10
example.com/.htaccess:31   # DO NOT REMOVE THIS LINE AND THE LINES ABOVE REDACTED:IPALLOWID

Two facts about those lines:

Controls that passed. The site answers normally today (GET /pb/readme.html → 200), so the server in use accepts the legacy directive — the risk is conditional, not current. And the file is not served: GET /pb/.htaccess → 403, so the IP address is not published over the web right now.

And it is orphaned. The IPALLOWID marker format belongs to a security/firewall plugin's IP allowlist feature. None of the 38 installed plugins is a security or firewall plugin — so the block was left behind when whatever wrote it was removed.

Rastro: example.com/.htaccess:30 carries allow from <IP> → there is no deny directive anywhere in the file, so it grants and restricts nothing → the directive requires mod_access_compat, which the current server provides but a future one need not → the marker that brackets it belongs to a plugin that is no longer installed.

Recommendation. Delete lines 29–31 of wp-content/../.htaccess (the .htaccess in the published root, public_html/pb/.htaccess). Keep a copy of the file first. If you ever do want to restrict administrative access by IP, the correct form for Apache 2.4 is:

<Files wp-login.php>
    <IfModule mod_authz_core.c>
        Require ip 203.0.113.10
    </IfModule>
</Files>

How to verify it was fixed. grep -n 'allow from' public_html/pb/.htaccess returns nothing, and the site still answers 200.


F-19 🟡 Low — Visitors' browsers are given none of the protections the site could ask for [MEASURED]

Impact: the site owner is leaving the browser's own defences switched off. On their own these headers take nothing; what they do is make every cross-site-scripting finding in this report cheaper to exploit and every first visit interceptable.

After access: nothing directly. Without HSTS, someone in a network position can hold a first-time visitor on http://. Without a content-security policy that restricts script origins, an injected script (F-08, F-09, F-12) runs with no friction. Without nosniff, a file the server labels as one type can be interpreted as another.

What the attacker takes: nothing from these headers alone. What the attacker does not take: no data, no session — this is a multiplier, not a door. Persistence: not applicable. Attribution: not applicable. Radius: every visitor.

Precondition: none; this is the current state of every response.

Evidence. curl -sI https://example.com, 2026-11-13 09:01 UTC:

HTTP/2 200
server: hcdn
content-security-policy: upgrade-insecure-requests
content-type: text/html

Absent: Strict-Transport-Security, X-Frame-Options, X-Content-Type-Options, Referrer-Policy, Permissions-Policy.

On the CSP, said honestly. upgrade-insecure-requests is a content-security-policy header, but it restricts no script origin — it only rewrites http:// subresource requests to https://. Reporting "a CSP is present" would be the half-truth that stops a reader from looking further.

Controls that passed, measured in the same breath.

curl -sI http://example.com        ->  301 to https://example.com/     (redirect is in place)
X-Powered-By                        ->  absent (the PHP version is not disclosed in headers)
TLS 1.0 / TLS 1.1                   ->  refused
TLS 1.2 / TLS 1.3                   ->  accepted  (negotiated: TLSv1.3, TLS_AES_256_GCM_SHA384)
certificate                         ->  Let's Encrypt, CN=example.com,
                                        SAN: example.com, sub.example.com
                                        valid 2026-10-29 to 2027-01-27
                                        Verify return code: 0 (ok)

Rastro: the response to GET https://example.com carries content-security-policy: upgrade-insecure-requests and no other security header → a browser receiving it has no HSTS pin, no frame restriction, no MIME-sniffing protection and no script-origin policy → the stored and reflected XSS findings in this report (F-08, F-09, F-12) therefore meet no browser-side obstacle.

Recommendation. Add these headers for the domain. In Hostinger's hPanel this is usually under the domain's advanced settings; otherwise add them to .htaccess in the published root:

<IfModule mod_headers.c>
    Header always set Strict-Transport-Security "max-age=31536000; includeSubDomains"
    Header always set X-Content-Type-Options "nosniff"
    Header always set X-Frame-Options "SAMEORIGIN"
    Header always set Referrer-Policy "strict-origin-when-cross-origin"
    Header always set Permissions-Policy "geolocation=(), microphone=(), camera=()"
</IfModule>

Add HSTS last, and only once you are sure every subdomain serves HTTPS — includeSubDomains with a one-year max-age is hard to undo. A real Content-Security-Policy that restricts script origins is worth doing too, but it needs testing against the page builders: start it in report-only mode.

How to verify it was fixed. curl -sI https://example.com shows the five headers above.


F-20 🟡 Low — The published root still serves files that identify the installation, and the legacy XML-RPC endpoint is alive [MEASURED]

Impact: the site owner is confirming to anyone who asks what the site is built on and where, and leaving an endpoint open that lets many password attempts be bundled into a single request.

After access: the identification itself takes nothing. XML-RPC's system.multicall lets an attacker try many credentials per request, and its pingback feature has historically been used to make the server issue outbound requests on an attacker's behalf.

What the attacker takes: confirmation of the platform and the installation path; and, through XML-RPC, amplification of password guessing. What the attacker does not take: no password, no data — and this audit did not exercise the XML-RPC endpoint, because doing so would change state. Persistence: not applicable. Attribution: requests in the access log. Radius: the login surface.

Precondition: none.

Evidence. Measured one request at a time, through the authorised hostname:

GET /pb/readme.html        ->  200,  7,409 bytes, text/html
GET /pb/license.txt        ->  200, 19,915 bytes, text/plain
GET /pb/default.php        ->  200, 16,395 bytes, text/html   (Hostinger placeholder page)
GET /pb/wp-login.php       ->  200,  7,503 bytes, text/html
GET /pb/xmlrpc.php         ->  405,     42 bytes, text/plain;charset=UTF-8

On the 405, and why it means the endpoint is alive. The host's own blocks on this server answer 403 with a 10-byte text/plain body, and its 404 answers with 4,511 bytes. The 42-byte text/plain;charset=UTF-8 response is WordPress's xmlrpc.php rejecting the wrong HTTP verb — the file is present and responding. No POST was sent.

On readme.html, stated narrowly. It does not print the core version number — a search for a version pattern in the file returns only the licence's "version 2". What it discloses is that this is WordPress and that the installation lives at this path. That is fingerprinting, not a version leak, and this report says so rather than overstating it.

default.php was read, not assumed. It is Hostinger's placeholder page: 16,395 bytes of plain HTML, with zero occurrences of $_GET, $_POST, include, require or eval. It executes and prints a static page. It is clutter in the published root, not a code path.

Controls that passed.

GET /pb/wp-config.php              ->  403      GET /pb/.htaccess.bk    ->  403
GET /pb/wp-content/debug.log       ->  403      GET /pb/.git/HEAD       ->  403
GET /pb/wp-content/plugins/        ->  200, 0 bytes   (no directory listing)
GET /pb/.env                       ->  403      GET /pb/.env.bak        ->  403
GET /pb/.DS_Store                  ->  403

The host does block the names that matter, by pattern rather than by exact name. The frontier is worth knowing: a non-existent dotfile returns 404, not 403 (GET /pb/.sonda-svo-inexistente → 404; GET /pb/.user.ini → 404), so the protection is a curated list of patterns, not a blanket "no dotfiles" rule. Nothing sensitive is currently outside that list.

Rastro: _raiz_publicada/readme.html and license.txt exist in the published root → both return 200 over HTTP → xmlrpc.php exists in the published root and answers 405 to a HEAD, which is its own response, not the host's → default.php exists and executes, serving a placeholder page.

Recommendation.

  1. Delete readme.html, license.txt, default.php and .htaccess.bk from public_html/pb/. None of them is needed for the site to run. .htaccess.bk is a configuration backup — keep a copy off the server if you want it.
  2. Disable XML-RPC. If nothing you use needs it (the Jetpack and WordPress mobile apps do; most sites do not), add to public_html/pb/.htaccess:

``apache <Files xmlrpc.php> <IfModule mod_authz_core.c> Require all denied </IfModule> <IfModule !mod_authz_core.c> Order allow,deny Deny from all </IfModule> </Files> `` If you do need it, ask your host to rate-limit it instead.

How to verify it was fixed. GET /pb/readme.html returns 404 and GET /pb/xmlrpc.php returns 403.


F-21 🟡 Low — Twelve components have no public channel, so nothing will ever tell you they have fallen behind [MEASURED]

Impact: the site owner cannot find out, by any means available to them, whether twelve of their forty-six components are out of date. This is not a vulnerability — it is blindness, and it is permanent by design. One of the twelve, woodmart-core, ships with the active theme.

After access: nothing. There is no attack here.

What the attacker takes: nothing. What the attacker does not take: nothing. Persistence: permanent — a component outside the public channel never announces that it has aged. Attribution: not applicable. Radius: 12 of 46 components.

Precondition: none.

Evidence. For each of the twelve, two sources were tried and the result recorded:

Componentinstalledwhat was tried
plugins/hostinger-affiliate-plugin3.0.43api.wordpress.org → 404; no vendor changelog reachable
plugins/hostinger-ai-assistant3.1.8api.wordpress.org → 404; no vendor changelog reachable
plugins/hostinger-easy-onboarding2.1.17api.wordpress.org → 404; no vendor changelog reachable
plugins/hostinger-auto-updates-plugin-cpOXqJKPaHM1UzOZ1.0.9not published; slug carries a random suffix
plugins/hostinger-auto-updates-plugin-uO8WtFw3RRlhP6tl1.0.10not published; slug carries a random suffix
themes/hostinger-affiliate-theme1.0.8api.wordpress.org/themes/info/1.2/ → 404
plugins/woodmart-core1.1.0xtemos.com/docs/woodmart/changelog/ → 404 · xtemos.com/changelog/woodmart/ → 404 · woodmart.xtemos.com/changelog/ → 404 · ThemeForest item page → 403
plugins/zyko-coverage-test0.1.0-pocno channel by design (test plugin)
plugins/zyko-coverage-test-v22.0.2no channel by design
plugins/zyko-coverage-test-v212.1.2no channel by design
plugins/zyko-coverage-test-v2-instrumented2.0.2-instrumentedno channel by design
themes/woodmart (current version only)8.0.4changelog 404 from here; a lower bound of ≥ 8.5.4 was derived from a published advisory's affected range

Advisory verdict for these twelve, which does not depend on there being an update channel: nothing published reaches any of them, by text search of the NVD on 2026-11-13. The four Hostinger CVEs that search returns belong to other products (Hostinger core <= 1.9.7, Hostinger Reach, Hostinger Migrator), none of which is installed here. The two woodmart-core advisories (CVE-2023-32242 object injection, CVE-2023-32244 privilege escalation) both stop at <= 1.0.36, below the installed 1.1.0.

That verdict is about what is published, not about what exists. There are tens of thousands of WordPress components and an unknown number of them have flaws nobody has published. "Nothing found" means nothing found — it does not mean safe.

Rastro: 46 components were censused from disk, each read from its own version declaration (plugins/woodmart-core/woodmart-core.php:5 declares Version: 1.1.0; the other eleven rows above were read the same way) → 34 had a measurable current version (from api.wordpress.org or the vendor's own changelog) → 12 did not, and for each of those the attempt and its HTTP result is recorded above → therefore the count that matters is 30 behind and 12 not measured, never 30 on its own.

Recommendation. Two asks, both of them questions someone else can answer in a minute.

  1. Ask Hostinger which of the six hostinger-* components they install and maintain on this account, whether they are updated automatically, and how you would know. Five of them are theirs; you did not choose them. If the affiliate plugin and theme are not in use, ask for them to be removed (that also closes part of F-05).
  2. Ask XTemos for the current version of Woodmart and woodmart-core and where you can read the changelog with your licence. You need that number anyway to act on F-02.

How to verify it was fixed. You have, written down, the current version of Woodmart and woodmart-core, and a statement from your host about the six components they install. The fix here is information, not configuration.


F-22 🟡 Low — The PHP version that actually serves this website could not be confirmed, and one question answers it [MEASURED]

Impact: the site owner does not know which PHP version runs their site, and neither do we. That matters because a runtime vulnerability applies to whatever version answers the web, not to the one you get in a shell — and on shared hosting the panel chooses the web version per domain.

After access: nothing today. If the version serving the web is a line that no longer receives fixes, whatever that line's vulnerabilities allow becomes reachable — and that cannot be stated without measuring it.

What the attacker takes: nothing established. What the attacker does not take: no conclusion from this report — the empty disable_functions and empty open_basedir below were measured in the shell interpreter, and this report does not assert them for the web. Persistence: the version is a menu item in the panel; it changes with a click. Attribution: not applicable. Radius: all PHP on the site.

Precondition: none; this is an open question, stated as one.

Evidence — what was measured, and it is the shell interpreter.

inventario/runtime.txt
   PHP 8.0.30 (cli) (built: Oct 21 2026 10:54:40) ( NTS )
   binaries within reach: /usr/bin/php
   directives (CLI):  disable_functions=      open_basedir=
                      allow_url_fopen=1       allow_url_include=      expose_php=1

Three ways were tried to establish the web version, and none of them closes it.

1. the domain's PHP handler, in .htaccess   ->  only `AddType text/vcard .vcf`.
                                                No PHP handler directive at all.
2. find <root> -name '.user.ini' -o -name 'php.ini'   ->  0 files
3. curl -sI https://example.com  ->  no `X-Powered-By` header

So: the PHP version serving this website cannot be confirmed from what was collected. It is not being guessed.

And the number alone would not settle it anyway. PHP 8.0 stopped receiving upstream security support in November 2023 — but the binary measured here was built on 21 October 2026, which is the signature of a provider backporting fixes into an old line. Concluding "8.0, therefore vulnerable" from the number is the mirror image of concluding "high number, therefore safe". Both are assertions about something that was not measured.

Rastro: inventario/runtime.txt records the CLI binary as PHP 8.0.30 built 2026-10-21 → the domain's .htaccess declares no PHP handler → no .user.ini or php.ini exists in the tree → the HTTP response carries no X-Powered-By → therefore the web SAPI's version and its directives are unestablished, and the CLI values are reported as CLI values.

Recommendation. One question, and it takes ten seconds in hPanel:

In hPanel → Advanced → PHP Configuration, which PHP version is selected for example.com (or for sub.example.com), and are disable_functions and open_basedir set there?

Send us the answer and we will tell you whether any runtime advisory reaches it. If the selected version is below 8.3, ask your host what their backporting policy is for that line before moving — on CloudLinux an old line is often patched, and switching versions on a live WooCommerce store needs staging.

How to verify it was fixed. You have the panel's answer written down: the version selected for this domain, and the values of disable_functions and open_basedir for the web.


6. Component inventory — all 46, measured 2026-11-13 09:00–10:30 UTC

Every installed component appears here, including the ones whose current version could not be established. A missing row would read as "does not exist"; NOT MEASURED reads as what it is. The source of this list is the disk, not any API — the lookups fill the Current and Advisory columns, they do not decide who appears.

None of these components' source code was read line by line.

Core

ComponentInstalledCurrentBehindAdvisory reaching installed
WordPress core6.6.117.1.45 release branchesstable-check marks 6.6.11 outdated, not insecure; 6.6.1–6.6.10 are insecure and no 6.6.12 exists, so this is the head of its branch

Plugins (38)

ComponentInstalledCurrentBehindAdvisory reaching installed
ali2woo (AliNext)3.4.53.7.3yes2 — CVE-2024-37213 (CSRF <= 3.4.6), CVE-2025-30859 (open redirect <= 3.5.1) → F-11
all-in-one-wp-migration7.1117.1132 releasesnone — CVE-2026-81810 is < 7.111, fixed in 7.111
bulk-woocommerce-category-creator2.42.4currentnone — all three stop at <= 2.3
complianz-gdpr7.5.67.5.71 releasenone — highest is <= 7.5.5
contact-form-76.0.66.2.5yes1 — CVE-2026-120407 (< 6.2.5, Editor+) → F-13
currency-switcher-woocommerce2.16.32.16.74 releasesnone — highest is <= 2.16.2
elementor4.1.14.4.0yes7 → F-09
elementor-pro3.17.14.4.0 (2026-11-09)yes12, incl. CVE-2026-32475 unauth arbitrary file upload CVSS 9.0 → F-01
google-site-kit1.187.01.191.04 releasesnone — CSRF stops at <= 1.186.0
gtranslate5.0.15.0.21 releasenone — CVE-2026-86604 is < 5.0.1, fixed in 5.0.1
header-footer-elementor2.9.62.9.71 releasenone — highest is <= 2.9.5
hostinger3.0.653.0.82yesnone — CVE-2023-6751 stops at <= 1.9.7
hostinger-affiliate-plugin3.0.43NOT MEASURED (api.wordpress.org 404; no vendor changelog reachable)not measurednone found → F-21
hostinger-ai-assistant3.1.8NOT MEASURED (same)not measurednone found → F-21
hostinger-auto-updates-plugin-cpOXqJKPaHM1UzOZ1.0.9NOT MEASURED (not published)not measurednone found → F-21
hostinger-auto-updates-plugin-uO8WtFw3RRlhP6tl1.0.10NOT MEASURED (not published)not measurednone found → F-21
hostinger-easy-onboarding2.1.17NOT MEASURED (api.wordpress.org 404)not measurednone found → F-21
js_composer (WPBakery Page Builder)8.1≥ 9.0.2 (lower bound; vendor changelog 404 from here)yes8, incl. Subscriber+ stored XSS <= 9.0.1 → F-08
mailchimp-for-wp4.15.04.15.0currentnone — both 2026 unauth entries are < 4.15.0, fixed in 4.15.0
maintenance4.154.3318 releasesnone — the only one stops at <= 4.02
newsletter9.4.59.5.7yesnone — highest is <= 9.4.0
omnisend1.9.11.9.1currentnone — IDOR stops at <= 1.9.0
rehub-framework19.6.3≥ 19.9.9.7 (lower bound; vendor changelog not readable from here)yes3 — CVE-2025-63050, CVE-2025-14358, CVE-2025-63051 → F-05
revslider (Slider Revolution)6.7.247.2.10yes3, incl. CVE-2025-10249 Contributor+ arbitrary file read → F-07
royal-elementor-addons1.7.10701.7.10722 releasesnone — CVE-2026-106648 is < 1.7.1070, fixed in 1.7.1070
safe-svg2.3.12.5.2yes2 — CVE-2026-94672, CVE-2026-94077 → F-10
shortcode-in-menus3.5.13.5.1currentnone published for this component
waspthemes-yellow-pencil7.6.07.6.7yes2 — reflected XSS <= 7.6.4, XSS <= 7.6.1 → F-12
woocommerce9.7.311.3.02 major versions4, incl. CVE-2026-57777 SQL injection (Shop Manager) → F-04
woocommerce-currency-switcher (WOOCS/FOX Pro)2.4.2.32.5.7yesnone — published entries are the free line, <= 1.5.6
woodmart-core1.1.0NOT MEASURED (4 vendor URLs tried: 404/404/404/403)not measurednone — both stop at <= 1.0.36 → F-21
wp-external-links2.622.664 releasesnone — CSRF stops at <= 2.57
wp-sheet-editor-premium2.25.82.27.1 (2026-10-21)yesnone found for this product
wpfront-scroll-top2.23.0.11 major versionnone published for this component
zyko-coverage-test0.1.0-pocno channel (test plugin)n/anone found → F-14
zyko-coverage-test-v22.0.2no channel (test plugin)n/anone found → F-14
zyko-coverage-test-v212.1.2no channel (test plugin)n/anone found → F-14
zyko-coverage-test-v2-instrumented2.0.2-instrumentedno channel (test plugin)n/anone found → F-14

Themes (7)

ComponentInstalledCurrentBehindAdvisory reaching installed
woodmart (active)8.0.4≥ 8.5.4 (lower bound; changelog 404 from here)at least 4 minor releases12, two of them unauthenticated → F-02
woodmart-child1.0.0n/a — this site's own coden/an/a — read line by line, no finding
rehub-theme19.6.3≥ 19.9.9.1 (lower bound)yes2 — CVE-2025-64197, CVE-2025-67565 → F-05
hostinger-affiliate-theme1.0.8NOT MEASURED (api.wordpress.org 404)not measurednone found → F-21
twentytwentyfour1.21.64 releasesnone found
twentytwentythree1.51.72 releasesnone found
twentytwentytwo1.82.2yesnone found

The totals, with an honest denominator

46 components · 30 behind · 4 current · 12 not measured · 12 with an advisory reaching the installed version. The 30 is never to be read on its own: the twelve not-measured components are exactly the ones that sit outside the public channel, which is where premium and host-injected components live.

What was searched, and where. api.wordpress.org (plugin and theme information), patchstack.com/database, services.nvd.nist.gov (keyword search, by text — advisories often reach the NVD without a product identifier, and filtering by product returns zero where there are dozens), and each vendor's own changelog where one is reachable. Nothing published reaches the components marked "none". That means nothing published was found, not that there is nothing to find — and for a component with no public channel, it is a statement about a place that cannot be looked at. For each of those, confirm with the vendor whether the installed version still receives security fixes.

Integrity. Every installed component was also checked as bytes: sha256 of 26,024 of 26,024 files matches the measurement manifest. Nothing in this tree differs from what was measured.


7. Destinations this site's code calls

The code written or maintained for this site makes no outbound HTTP request at all. Measured across the child theme, the must-use plugin and the four test plugins: zero wp_remote_*, zero curl_exec, zero file_get_contents over a URL. The only addresses that code names are:

Where it livesAddress calledWhat it is
mu-plugins/hostinger-auto-updates.php:33http://127.0.0.1:7777/hsub_downloadsloopback — the host's local package cache
mu-plugins/hostinger-auto-updates.php:37wpdownloads.hostinger.iothe host's package mirror
mu-plugins/hostinger-auto-updates.php:41wpapi.hostinger.iothe host's WordPress.org API proxy

Three other addresses appear in these files only inside comments (zyko.pro and two products.best URLs documenting the test endpoints) — nothing calls them. The products.best references confirm these test plugins were copied from another installation.

Across the whole tree, including every off-the-shelf component, the census counted 395 outbound HTTP call sites: wp_remote_get 177, wp_remote_post 103, wp_remote_request 48, curl_exec 34, wp_safe_remote_get 33, wp_safe_remote_post 22, wp_safe_remote_request 19, wp_safe_remote_head 3, wp_remote_head 3. All of them belong to purchased or installed components calling their own vendors.

Coverage of this section, stated so an empty list cannot be mistaken for an unexamined one. Destinations reached by clicking, and destinations a page loads on its own: not measured — the order authorises example.com and the installation declares itself as sub.example.com (example.com/wp-config.php:98-99), so there is no authorised browsing of the host the installation serves. Destinations the code calls: 3 distinct addresses across 1,368 lines of site-specific code, and 395 call sites across the 16,628 PHP files of the whole tree. Liveness was not tested: the scope authorises reading this folder, not sending requests to third parties.


8. Recommendations, in the order they are worth doing

This week — these three change the most for the least risk.

  1. Block PHP execution under wp-content/uploads, and set that one directory back to 755 (F-03). This is the single cheapest change in the report and it reduces F-01, F-02 and anything of that class that has not been published yet. It touches no component and can be reverted by deleting one file.
  2. Update Elementor Pro and Elementor together, to 4.4.0 (F-01, F-09). Pro carries the only unauthenticated CVSS 9.0 entry in this inventory, and the fix has been available since 4.2.2.
  3. Add the two constants to wp-config.php and set it to 640 (F-15). Four minutes, and it removes the dashboard as a code-execution path.

This month.

  1. Update the active theme — Woodmart and woodmart-core, on a staging copy first (F-02). Twelve advisories close at once.
  2. Update WooCommerce toward 11.x in steps, testing checkout at each (F-04); and prune the Shop Manager role list today, which shrinks F-04 immediately at no risk.
  3. Delete what is not used: the Rehub theme and framework, the Hostinger affiliate theme and plugin, two of the three default themes, and all four test plugins with their wp_options rows (F-05, F-14). That is 1,032 + 428 PHP files of published, unmaintained code out of the published root.
  4. Update Slider Revolution, WPBakery, Safe SVG, YellowPencil, ali2woo and Contact Form 7 (F-07, F-08, F-10, F-11, F-12, F-13). For Contact Form 7, go to 6.2.5 or later — never 6.2.0–6.2.3.
  5. Close the backup folder and tidy the published root: the deny rule on ai1wm-backups/, and delete readme.html, license.txt, default.php and .htaccess.bk (F-16, F-20).
  6. Add the browser security headers, HSTS last (F-19). And remove lines 29–31 of the root .htaccess (F-18).
  7. Rotate the ali2woo licence key and delete uploads/ali2woo/pk.php — rotate first (F-17).

Three questions to send, and they unblock the rest.

  1. To Hostinger: what do your managed updates cover for this site, when did they last run, and which of the six hostinger-* components do you install and maintain? Put the measurement in front of them: WordPress 6.6.11 while 7.1.4 is current, and 30 of 46 components behind (F-06, F-21).
  2. To Hostinger, in the panel: which PHP version is selected for this domain, and are disable_functions and open_basedir set for it? (F-22)
  3. To XTemos: the current Woodmart and woodmart-core versions, and where to read the changelog with your licence (F-02, F-21).

And one standing decision. Until F-06 is settled, the dashboard's Site Health page cannot tell you whether this site is current. Use section 6 of this report as the baseline and check the Plugins and Themes screens directly.


9. Coverage by class

What was enumerable was counted in full, not sampled.

ClassIn the treeEvaluatedReportedSet aside
Installed components (plugins, themes, core)464612 (with a reaching advisory)34 (no advisory covers the installed version)
Files verified byte-for-byte against the measurement manifest26,02426,0240 differences—
Files swept for mode and ownership26,024 files + 5,707 dirsall1 (the 777 directory)26,023 files / 5,706 dirs at 644/755
Symlinks000—
Lines of site-specific code read line by line1,368 (9 files)1,3681 finding (F-14)—
Request-value reads in site-specific code12121 unguarded (F-14)11 guarded by manage_options
Outbound HTTP call sites395395 counted; 3 addresses in site-specific code resolved0392 belong to vendor components
File-write sinks reachable by an attacker-controlled path288 classified2880278 discarded · 6 high-role · 4 no reachable entry point · 1 candidate refuted
Deserialization gadgets (magic method reaching a sink)53530no deserialization sink takes an attacker-forgeable argument
Account-takeover candidates (role, capability, option key, callback)3,393 classified3,39301,449 discarded · 1,914 unresolved · 10 require Shop Manager or above · 2 opened and read individually, both refuted · 0 anonymous entry points
(maybe_)unserialize sinks2202200all in off-the-shelf code, reading options/meta/own tables
Dynamic include/require sites45145106 core loaders read individually; the rest in off-the-shelf code
$wpdb queries with string concatenation2112110all in off-the-shelf code; judged by version × advisory
eval( sites20200all in off-the-shelf code
Shell-execution functions59590all in off-the-shelf code
REST routes registered471471 counted0866 permission_callback declarations; 59 use __return_true, all in off-the-shelf code
AJAX actions registered (distinct)636, of which 117 noprivcounted0all in off-the-shelf code
admin_post_ actions (distinct)24, of which 0 noprivcounted0all in off-the-shelf code
Archives, dumps, .bak/.old/.log/~/.wpress files anywhere in the tree0—0nothing to report
Version-control directories (.git, .svn, .hg)0—0nothing to report
AI instruction files inside the audited tree (CLAUDE.md, AGENTS.md, .cursorrules, GEMINI.md, …)0—0none present, so none influenced this audit
Individual URLs measured from outside, through the authorised hostname24248one request per URL, paced, GET/HEAD only

What "evaluated" means here. For the 46 components it means: version read from the artefact, current version looked up in the vendor's published channel, and every published advisory's affected range compared against the installed version. For the 1,368 lines of site-specific code it means read in full, line by line. For the large enumerable classes — the 220 deserialization sinks, the 451 dynamic includes, the 211 concatenated queries — it means counted in full and judged as a class: they live in purchased components, and what a site owner can act on there is the component's version against published advisories, which is section 6. Their source was not read line by line.

The locks on this audit, measured rather than promised. Before any file was read, two probes confirmed that the read-only guards were actually in force: a destructive command was refused by the permission layer, and an attempted write outside the audit folder was blocked by the write guard. Execution mode: not declared for this run.


Before you implement

This report was produced by a language model executing the SVO method.

🔴 Every piece of evidence and every trace in this report must be followed and verified by your development team before any correction goes into production. This is not a disclaimer: it is a step of the work, and it is why each finding carries file:line, Rastro, Evidence and How to verify it was fixed.

Whoever checks it is whoever fixes it. The By owner table in section 4 already says who owns each finding. That same owner walks the trace of their own finding — do not delegate the check to someone who will not apply the fix, because it is during the check that you discover what the trace did not cover.

For each finding, before correcting:

  1. Open the file:line and confirm the line says what the finding claims.

1-bis. 🔴 Reproduce the finding in a laboratory — a copy of your environment, separate from production. The trace shows what the code does; execution proves the behaviour, and it is execution that turns a reading into a fact in your environment. This applies to every finding, not only the ones that look doubtful: it is in the laboratory that the real configuration, the real dependency version and the path your deployment actually takes come to light.

  1. Walk the whole Rastro, step by step, from the entry point to the line that executes. The chain has to close in your environment. If a step does not close — the route is not exposed, the real configuration differs from the repository, the path is dead in your deployment — the reach of the finding changes, and you are the one who knows that, not the report.
  2. Read the evidence label. [ANALYSED] is a reading of code, without execution: nothing was reproduced or observed running. [MEASURED] means the output is in the finding. [ESTIMATED] never supports a severity.
  3. Use How to verify it was fixed as an acceptance test — run it before the correction (it should fail) and after (it should pass). A correction without that pair is applied, not confirmed.

Did any sentence in this report not come across clearly?

Write to contato@svo.com.br saying which sentence and what was ambiguous. We will explain what the method meant — that text is ours, and if it was not clear to you, the defect is ours and the correction goes into the method, not just into your answer.

🔒 Where this report lives. Your code was processed in a dedicated SVO area on SVO's own infrastructure; this report was produced there and sent to you by SVO; and that area, with your code and every working file in it, is destroyed once the report is delivered. To ask about a sentence, just quote the sentence.

Provenance of this document
Audit performed bySVO — Software Verification & Operations
Report— · —
Method— · level — · ruleset —

This document records what was examined, with what method, and what was found. The coverage numbers in the Summary are the measure of the examination.