The install-time supply-chain firewall

Stop the next supply-chain zero-day.
Before a CVE exists.

A firewall in front of every package install, on laptops, in CI/CD and in Kubernetes. It reads the code and blocks malicious packages in about 80 ms, before any CVE exists. Now covering Go modules and Hugging Face models too.

Book a 15-min demo Check a package ↓

// median proxy decision

~80ms fast enough to gate every install,
across laptops, CI and runtime.

1.3M+vulnerabilities indexed
207K+known malicious packages
6parallel detection signals

// built on academic research CIT → MTU

Caught in the wild

We don't theorize about supply-chain attacks.
We catch them.

Real campaigns, blocked at install before any CVE existed. Here is the proxy log, the moment each one hit the wire, and the full dissection of each.

0packages scanned

0vulnerabilities indexed

0malicious packages catalogued

0installs blocked at the proxy

0 packages scanned this week

app.cyberxyz.io/proxy/findingsreplay · real attacks, re-run through the engine

May 28, 10:40 AM npm

flatmap-stream@0.1.1BLOCKEDgha/CyberXYZ CI/CD Linux172.184.213.226CRITICAL

BLOCKED[known_malicious_version]Confirmed supply-chain attack: credential_stealer. Safe versions: 0.1.0CRITICAL

BLOCKED[vulnerability_match]GHSA-9x64-5r7x-2q53 · ranges ['= 0.1.1'] · fixed: noneCRITICAL

BLOCKED[vulnerability_match]GHSA-mh6f-8j2x-4483 · ranges ['>= 0'] · fixed: noneCRITICAL

May 26, 09:22 AM npm

axios@1.14.1BLOCKEDMike’s MacBook DEV macOS169.254.169.126CRITICAL

BLOCKED[known_malicious_version]Confirmed supply-chain attack: RAT Dropper via compromised maintainer. Safe versions: 1.7.7, 1.7.6, 1.7.5CRITICAL

ALERT[vulnerability_match]GHSA-q8qp-cvcw-x6jj · ranges ['>=1.0.0', '<1.15.2'] · fixed: noneHIGH

QUARANTINE[version_jump_anomaly]axios jumped from 1.10.0 to 1.14.1 (skipped 4 minor versions overnight)MEDIUM

May 22, 12:15 PM npm

left-pad@1.3.1BLOCKEDgitlab/cyber CI/CD Linux34.139.136.251CRITICAL

May 22, 12:15 PM

go-weather-sdk@v0.4.0BLOCKEDgitlab/cyber CI/CD Linux34.139.136.251CRITICAL

May 22, 12:15 PM NuGet

AeroWizard.Net@2.3.1BLOCKEDgitlab/cyber CI/CD Linux34.139.136.251CRITICAL

May 22, 12:15 PM PyPI

durabletask@1.4.1QUARANTINEgitlab/cyber CI/CD Linux34.139.136.251MEDIUM

fig. 02 · proxy findings. Real attacks replayed through the engine; the live feed is at app.cyberxyz.io

// npm · DPRK RAT dropper

axios@1.14.1 RAT dropper

A compromised maintainer shipped two malicious axios tags that pulled in a North Korean RAT dropper. Blocked at install, before any advisory existed.

9.4 XYZ score~100M weekly downloads0 CVE at block

Read the case study → // npm · worm · credential theft

Mini Shai-Hulud AntV npm worm

One stolen npm token poisoned 323 packages across the AntV namespace in 27 minutes, with a credential harvester wired to run before any user script. Every version blocked on its lifecycle hook.

323 packages639 bad versions27 min window

Read the case study →

See every attack we caught →

Trusted & built with

Trusted by security teams. Built with the industry's best.

Engaged with enterprise security, payments and research teams across three continents, and integrated with the ecosystem your stack already runs on.

// trusted by

// built with

Try it · npm v12 ready

Check any package
before you approve it.

npm v12 makes you approve every install script by hand. Paste a package and see the XYZ verdict, the signals that fired, and whether its scripts are actually safe to approve.

How we catch them

We read the code,
not just the scorecard.

Most supply-chain tools rate a package on project-health signals. CyberXYZ does that and reads the actual change, line by line, commit by commit, so a malicious hook is caught the moment it lands. This is where we catch the zero-day: at the commit and the pull request, before it ever reaches an install, no CVE required.

a3f9c2e chore: tidy build scripts 2 days ago

package.json

31 "scripts": {

32- "build": "tsc"

32+ "build": "tsc",

33+ "postinstall": "node ./.cache/setup.js"

XYZLifecycle hook added in a "tidy" commit. setup.js base64-decodes a payload and POSTs ~/.npmrc + env to 45.32.0.0/16. Verdict: BLOCK, no CVE required.

34 }

fig. · commit-level verdict · caught at the pull request, before any install

// where we operate: editor pull request CI / CD install

How it works

One brain.
Every surface.

Wherever a package is requested, a developer's laptop, a CI runner, a Kubernetes node, CyberXYZ is already there with the same verdict. Here's how it shows up across your stack.

ENFORCING 4/4 registries

ONLINE

ONLINE

⟳

xyz-npm-proxy ● live

npm install axios@1.14.1403 BLOCK

pip install requests==2.32.0200 ALLOW

npm install left-pad@1.3.0451 QUAR

dotnet add package Newtonsoft.Json200 ALLOW

tarball scan · commit-level analysis · ~80 ms median

// the platform · proxy

The firewall that gates every install.

A registry proxy intercepts each install, scans the tarball and reads the commit, then returns allow, alert, quarantine or block in milliseconds, across laptops, CI/CD and Kubernetes.

Explore the Platform →

$ xyz audit

npmpackage-lock.json

1 blocked

pythonrequirements.txt

clean

gogo.sum

clean

nugetpackages.lock.json

1 blocked

npmpackage-lock.json

1 blocked

// the cli · terminal-native

Security in your shell.

One pip install puts the brain in your terminal. Enroll a machine, gate every install through the proxy, and audit any project, across all five ecosystems.

xyz proxy setup · gate every install
xyz audit · scan any lockfile
xyz depalert · fail the CI build

Explore the CLI →

// ci/cd · the gate

A red build beats a shipped exploit.

Jobs from any source flow through the XYZ gate. Clean builds pass; malicious ones fail, before they ever merge.

One uses: line on GitHub. Templates for GitLab and Azure →

// any source

job #4821proj #1180

buildtestxyz gate

✓ passed

job #4822proj #2043

buildtestxyz gate

✓ passed

job #4823proj #1180

buildtestxyz gate

✗ failed · 1 malicious

package.json

"dependencies": {
  "axios": "1.14.1" ⛔ blocked,
  "express": "4.18.2" ⚠ 2 high CVE,
  "lodash": "4.17.21" ✓ clean
}

// vscode · inline

Caught as you type.

The extension underlines risky dependencies on every save, and a hover reveals the full XYZ risk profile, CVEs, exploit status, downloads, dependents and any campaign link.

Explore the extension →

Two supply chains, one firewall

Your models are dependencies too.
So are your Go modules.

Both come through the same door as any npm or PyPI package, and until now neither had a gate on it.

// ai models · hugging face

The first install-time gate for AI models.

A model pulled from Hugging Face is a package tree, a loader stack and sometimes code that runs the moment it loads. CyberXYZ tracks the models your teams pull and every package they declare, scans pickle-format weights opcode by opcode without executing them, and re-checks the whole fleet against fresh threat intel every hour.

Pickle weights scanned before anything loads
trust_remote_code and unsafe loaders flagged
Gated repos reported as unknown, never as clean
Silent weight overwrites caught on the next poll

// in development

Next: a per-model forecast of the most likely attack vector, ranked across eight classes, from instructed dependency confusion and weak-maintainer chokepoints to remote-code trust and silent overwrites.

8risk axes per model

hourlyfleet re-check

0pickles executed

Why models need a firewall →

// go modules · goproxy

The first install-time firewall for Go.

One GOPROXY variable puts CyberXYZ in front of every go get. We tail the Go checksum database, the public append-only log of every module version ever published, watch 15,535 modules including 220 Tier-1, and answer a bad module with a hard 403 so the toolchain never falls through to the public proxy.

Every new module version seen within a minute
Version-jump and new-dependency anomalies
Standard-library impostor detection
Rolls out with one environment variable

15,535modules watched

220Tier-1 modules

403hard block, no fallthrough

First to watch Go, first to block it →

The decision brain

Six signals. One verdict.
~80 milliseconds.

Every install runs six detection signals in parallel: new dependency, version jump, known malware, advisory match, commit-level code review and vendor breach. The strongest signal decides, in about 80 ms: allow, alert, quarantine or block. See the full methodology →

Where we fit

We don't replace your scanners.
We add the layer they can't.

SCA tools find known vulnerabilities after they're already in your tree. CyberXYZ blocks malicious installs before they land. Different jobs on the same supply chain, run them together.

// scanners & SCA

Detect after install

✓ Find known CVEs in your dependency tree
✓ License & policy compliance
✓ SBOM reporting
✗ Can't stop day-0 / unknown malware (no CVE yet)
✗ See it only after it is installed
✗ No block at install time

// cyberxyz

Prevent at install

the install-time supply-chain firewall

✓ Blocks malicious installs before they reach disk
✓ Catches day-0 / unknown, no CVE required
✓ Reads the actual code at commit level
✓ One verdict across IDE, CLI, CI/CD & runtime
✓ Centrality / blast-radius scoring + MITRE ATT&CK
✓ Feeds your SIEM / SOAR, ingests SBOMs

We add the prevention layer your scanners can't, run CyberXYZ alongside what you already own.

See it catch a real attack → · The methodology →

Coverage

Every major registry.
Every tool your SOC runs.

npm, PyPI, Maven, NuGet, Go, Cargo, RubyGems and Composer, with every verdict delivered to your SIEM, CI/CD and IDE.

npm

PyPI

Maven

NuGet

Cargo

RubyGems

Composer

JavaScript

GHSA

npm

PyPI

Maven

NuGet

Cargo

RubyGems

Composer

JavaScript

GHSA

IBM QRadar

Splunk

SentinelOne

CrowdStrike

Elastic SIEM

GitHub

GitHub Actions

GitLab CI

Azure DevOps

Jenkins

Docker

Kubernetes

VS Code

Syslog / CEF

Webhooks

JFrog

IBM QRadar

Splunk

SentinelOne

CrowdStrike

Elastic SIEM

GitHub

GitHub Actions

GitLab CI

Azure DevOps

Jenkins

Docker

Kubernetes

VS Code

Syslog / CEF

Webhooks

JFrog

Compliance & Standards

SOC 2

ISO 27001

NIST

PCI DSS

CMMC

In their words

Trusted by the people
who ship the code.

“

I really liked the product, and that you’re ahead of the market: commit and PR-level review that catches zero-days before a CVE. We invited the team to present to our incident response and offensive security groups.

Mansoor HaqaneeCybersecurity Manager · KPMGLinkedIn →

// the ask

Stop shipping
malware
you didn't write.

Get a 15-minute walkthrough of the proxy, the dashboard and the live attack feed. We'll plug it into your CI in under an hour.

✓ 15-min live demo
✓ Free proof-of-concept
✓ CI integration in < 1 hour

✓

Thanks! We'll be in touch.

Check your inbox. We'll reach out within 24 hours.

Stop the next supply-chain zero-day. Before a CVE exists.

We don't theorize about supply-chain attacks.We catch them.

axios@1.14.1 RAT dropper

Mini Shai-Hulud AntV npm worm

Check any packagebefore you approve it.

We read the code,not just the scorecard.

One brain.Every surface.

The firewall that gates every install.

Security in your shell.

A red build beats a shipped exploit.

Caught as you type.

Your models are dependencies too.So are your Go modules.

The first install-time gate for AI models.

The first install-time firewall for Go.

Six signals. One verdict.~80 milliseconds.

We don't replace your scanners.We add the layer they can't.

Detect after install

Prevent at install

Every major registry.Every tool your SOC runs.

Trusted by the peoplewho ship the code.

Stop shippingmalwareyou didn't write.

Thanks! We'll be in touch.

Get a demo

Stop the next supply-chain zero-day.
Before a CVE exists.

We don't theorize about supply-chain attacks.
We catch them.

Check any package
before you approve it.

We read the code,
not just the scorecard.

One brain.
Every surface.

Your models are dependencies too.
So are your Go modules.

Six signals. One verdict.
~80 milliseconds.

We don't replace your scanners.
We add the layer they can't.

Every major registry.
Every tool your SOC runs.

Trusted by the people
who ship the code.

Stop shipping
malware
you didn't write.