Skip to content
KitploitKITPLOIT
ToolsExploitsBlog
Log in
Submit
ToolsExploitsBlog
Submit

Hacking, PenTest, and Cybersecurity Tools for Your Security Arsenal!

Kitploit is a directory of hacking, cybersecurity, and pentesting tools. Discover the latest project updates to find vulnerabilities, analyze systems, automate testing, and strengthen your security.

FeedsContactPrivacy© 2026 Kitploit

Tool Directory

Categories

View all categories
Loading categories
CVE-2026-126274 — Differential PoC lab for CVE-2026-126274, a stored XSS in NocoDB's table-deletion dialog, verifying the innerHTML sink in served bundles against vulnerable and fixed images. | Kitploit
Tools/GitHubGitHub/googlesky/cve-2026-126274
Static AnalysisVulnerability AnalysisExploitationWeb Application ExploitationSecurity VirtualizationWeb Security
GitHubgooglesky/cve-2026-126274

CVE-2026-126274

Differential PoC lab for CVE-2026-126274, a stored XSS in NocoDB's table-deletion dialog, verifying the innerHTML sink in served bundles against vulnerable and fixed images.

View Repository
5 days agoNot yet reviewed
Share

CVE-2026-126274 — NocoDB Stored XSS: Table/Column Titles Rendered Through innerHTML in the Deletion Dialog

License: MIT Research status: validated

CVECVE-2026-126274 (GHSA-638g-xw92-xq9x, High, CWE-79)
ComponentNocoDB nc-gui — components/dlg/Table/Delete.vue
Affected< 2026.08.0 (sink verified present through the 2026.08.0 source and bundle)
Fixed in2026.08.1 (verified: the 2026.08.1 bundle ships the plain-content builder; 2026.08.0's still ships the sink)
CWECWE-79 (Cross-site Scripting)
ImpactA Creator embeds markup in a table or column title; when another user attempts to delete a table whose deletion is blocked by link columns, the dependency message interpolates those titles into an innerHTML binding and the markup executes in the victim's browser — able to steal the session token from localStorage["nocodb-gui-v2"]
VectorStored title (Creator+) → victim confirms a table deletion → same-origin script execution
TriggerinnerHTML: binding interpolating ${column.title} / ${relatedTable.title}
CreditReported upstream (NVD CVE-2026-126274); advisory GHSA-638g-xw92-xq9x

Delete.vue built the deletion-blocked dependency message by interpolating the blocking column's title and the related table's title into a template string and rendering it as a VNode created with h('div', { innerHTML }). Vue does not escape innerHTML bindings, so markup authored inside a title — stored verbatim by the meta API — becomes live HTML in the confirm dialog. The lab proves both halves of the stored-XSS equation against the official images: the vulnerable build stores markup titles verbatim and ships the exact innerHTML sink (extracted from the bundle the server itself hands to browsers, interpolation intact); the fixed build stores the identical titles but ships a plain-content renderer in which the same titles produce zero HTML elements. Feeding the attacker titles through the exact vulnerable interpolation yields an `` element — the execution condition — while the fixed rendering path yields plain text.

1. Root cause

packages/nc-gui/components/dlg/Table/Delete.vue (2026.07.0 source, and the identical compiled sink in the 2026.08.0/2026.07.0 bundles):

message.info(
  h('div', {
    innerHTML: `<div style="padding:10px 4px">Unable to delete tables because of the following.
        <br><br>${refColMsgs.join('<br>')}<br><br>
        Delete them & try again</div>`,
  }),
)

with each dependency line built as

return `${i + 1}. ${c.title} is a LinkToAnotherRecord of ${(refMeta && refMeta.title) || c.title}`

c.title and the related table's title are user-controlled (Creator-authored table/column titles). The message renders only after the user confirms deleting a table whose deletion is blocked by link columns. Vue escapes interpolations in templates but not innerHTML bindings, so the joined string is parsed as HTML.

The compiled sink, extracted verbatim from the bundle the vulnerable image serves to browsers (_nuxt/BGeeiuXx.js):

ht.info(Se("div",{innerHTML:`<div style="padding:10px 4px">Unable to delete tables
because of the following. <br><br>${F.join("<br>")}<br><br> Delete them & try again</div>`}))

The fix (2026.08.1) replaces the binding with escaped child VNodes:

rRn = t => xe("div", {style:{padding:"10px 4px"}},
  ["Unable to delete tables because of the following.", xe("br"), xe("br"),
   ...t.flatMap(e => [e, xe("br")]), xe("br"), "Delete them & try again"])

— titles become text children that Vue escapes, so no HTML parsing of user input occurs.

2. Reproduction environment

serviceimagerole
nocodb-vuln2026.07.0vulnerable side — bundle ships the innerHTML sink
nocodb-fixed2026.08.1fixed side — bundle ships the plain-content builder
poc—stdlib Python driver

All containers are CPU/memory bounded on the isolated cve126274lab network. No browser is required: the two halves of the stored-XSS equation are proven from the server side and from the exact shipped code.

3. Running the environment

./run-lab.sh build     # PoC runner image
./run-lab.sh sweep     # exploit / probe / mitigation / check
./run-lab.sh down

Single mode:

podman run --rm --network=cve126274lab cve-2026-126274-poc:local exploit

4. The PoC

poc/poc.py — stdlib-only, four modes:

modepurpose
checkstore a markup title via the meta API, verify it is served verbatim, scan the bundles the target serves for the innerHTML sink, and classify
exploitfull differential: markup stored verbatim + sink shipped on the vulnerable build; identical storage + plain-content builder on the fixed build; plus the DOM execution condition of the exact vulnerable interpolation
probecompanion primitives: the table meta API returns the payload verbatim, the title is stored raw, and the sink inventory across all served chunks
mitigationthe fixed bundle ships no sink chunk and the fixed rendering yields zero handler attributes — upgrade is the fix; no client-side compensating control exists on the vulnerable build

Exit codes: 0 safe / mapped, 1 vulnerable, 2 target unreachable, 3 inconclusive, 4 internal error.

The DOM execution condition feeds the attacker titles through the exact vulnerable interpolation (same template, same join) and parses the result as HTML: the vulnerable output contains an `` element — what a browser executes inside an innerHTML binding — while the fixed path's escaped content contains none.

5. Expected results

reproduction pathvulnerable 2026.07.0fixed 2026.08.1evidence
markup title stored via meta APIserved verbatim to the UIserved verbatim (storage is not the defect)evidence/poc-exploit.txt
bundle the server ships to browserssink chunk BGeeiuXx.js — innerHTML: with ${F.join("<br>")} interpolationplain-builder chunk B5CMIqqa.js — escaped child VNodes, no innerHTML near the messageevidence/poc-exploit.txt
DOM execution condition of the exact sink template4 handler attributes incl. ('img','onerror','alert(document.domain)')0evidence/poc-exploit.txt
sink inventory across served chunks153 scanned, sink present166 scanned, no sinkevidence/poc-probe.txt
check canarystored + sink → exit 1—evidence/poc-check.txt

6. Environment preconditions

  • Podman with podman compose; --cpus 2 --memory 4g per container; isolated cve126274lab network.
  • Outbound network once, to pull the two official nocodb/nocodb images.
  • The bundle scan fetches every /_nuxt/*.js chunk the entry page preloads (~150–170 assets per side); the sink discriminator is local to the dependency-message context, because the shared chunk contains unrelated innerHTML uses.

7. Additional findings (0-day sweep)

Download Tool

Most Popular

View all →

Discover the most used tools by our community.

  • 1

    OsintHunter

    letox4/osinthunter

    🎯 The ultimate OSINT toolkit: investigation graph, confidence scoring, alias families, map, timeline and reports. 49 modules, 400+ sites, no API keys required.

    936
  • 2

    ghostlock-app

    yukonga/ghostlock-app

    GhostLock One-Tap Execution App (CVE-2026-43499)

    555
  • 3

    redforge

    cybersecurity91/redforge

    Browser-side red team toolkit with 102 offline tools for recon, payload generation, credential attacks, crypto, encoding and engagement reporting.

    353
  • 4

    ARTEX

    baarai/artex

    LLM multi-agent autonomous penetration testing platform with planner/worker agents, asset and exploration graphs, MITM traffic recording, and vulnerability retesting.

    348
  • 5

    looking-glass

    anonghosty/looking-glass

    CSIRT platform that ingests Shadowserver reports into MongoDB and replays them as interactive maps, analysis, findings, advisories and executive dashboards.

    336
  • 6

    CyberStrikeAI

    ed1s0nz/cyberstrikeai

    The system of action for AI-native cybersecurity—where intent becomes governed execution, evidence becomes operational memory, and every operation improves the next.

    288

Explore all tools

Browse our collection of tools

View all tools →