Skip to content
Top Notchh.
Verglos · public alpha · local scan

Scan the code your agent helped write.

Verglos 2.0.0-alpha.1 scans JavaScript and TypeScript projects on the developer's machine and writes local HTML and JSON reports. Hunt, attest, paid plans, agent integrations, and hosted verification are not available in this release.

~ / verglos
$ npx verglos@2.0.0-alpha.1 scan

Local. No account. No repo upload.

Verglos current public language scope: JavaScript and TypeScript
Current language scope · public package 2.0.0-alpha.1

Availability boundary

One command works.
Two shells do not.

Scan

Live

Functional local scan for JavaScript and TypeScript projects with local HTML and JSON reports.

Hunt

Unavailable shell

The command name exists in 2.0.0-alpha.1, but it does not run sandbox verification or produce a verified verdict.

Attest

Unavailable shell

The command name exists in 2.0.0-alpha.1, but it does not sign a bundle or publish a verification URL.

Review workflow

What the
current alpha
actually does.

A useful run records the tool version, repository commit, scan scope, local output, and a human decision about each candidate.

01

Authorized project

JavaScript or TypeScript

02

Pinned command

Local alpha scan

03

Human review

Context and limits

04

Local output

HTML and JSON reports

Capability status

Local scanAvailable
HTML and JSON reportsAvailable
Sandbox verificationUnavailable
Signed or hosted evidenceUnavailable

Alpha boundary

Current release

Scan only

ALPHA
ScanReviewRetain

The scan campaign · ICP-top-300 · TypeScript + JavaScript

300

Repositories in the selected public campaign corpus

57.7%

Produced at least one high-or-critical scanner candidate

v1.5.2

Scanner version recorded by the campaign methodology

Alpha

Findings were observations, not verified vulnerabilities

This was a selected corpus scanned with Verglos v1.5.2. The result describes scanner output for that corpus; it is not an estimate of software prevalence or a count of verified vulnerabilities. Read the full campaign.

The problem

Why AI code
fails differently.

AI coding tools can produce plausible software quickly. Plausible is not the same as reviewed: generated handlers can omit ownership checks, input boundaries, or deployment-safe defaults while still compiling and passing a happy-path test.

Those shapes are not proof of AI authorship and are not always vulnerabilities. They are review prompts: trace user input, verify authorization, inspect package identity, and confirm whether a risky-looking path is reachable in production.

Verglos currently contributes one artifact to that review: a local scanner report. It does not certify the code, prove exploitability, or replace a security assessment.

Three boundaries

What can be evaluated today.

Inspectable scanner

The public package and source repository expose the current local scan path and report artifacts for review.

Local-first

The supported scan path analyzes the project on the developer's machine without requiring a repository upload.

Bounded claims

The product page separates working scan behavior from proposed verification, signing, agent, CI, and commercial surfaces.

Market context

Agentic AppSec got crowded in 2026.

ZeroPath, Corgea, Snyk Agent Fix, Cycode Maestro, XBOW, Semgrep AI Assistant, Pixee, and Ghost Security all point at the same buyer shift: security tools are moving from alerts toward agents that triage and repair code.

Verglos does not claim parity with those platforms. Its current public surface is smaller: local JavaScript and TypeScript scanning with reviewable reports. Broader verification and evidence concepts remain proposals.

The AI-specific catalogue

Six patterns AI code
keeps writing.

  • Pattern 01

    Wildcard CORS on authenticated surfaces

    A permissive origin policy can survive into a credentialed path. Review the actual route, credentials mode, and deployment boundary before assigning impact.

  • Pattern 02

    Math.random for tokens and IDs

    Convenient random sources sometimes appear in token-shaped code. Confirm whether unpredictability is part of the security model and whether the value reaches production.

  • Pattern 03

    Auth without authz

    Authentication can be present while object ownership or role checks are absent. Review every identifier that crosses a user or tenant boundary.

  • Pattern 04

    Slopsquat and hallucinated packages

    A plausible package name can be wrong or maliciously claimed later. Verify package identity, ownership, provenance, and the exact version before installation.

  • Pattern 05

    Verbose error handlers that leak stack traces

    Debug-friendly error output can expose paths, queries, or upstream details. Separate operator diagnostics from responses returned to clients.

  • Pattern 06

    Mass assignment on ORM writes

    Passing an entire request body into a data write can expose fields the caller should not control. Allowlist writable fields and enforce authorization at the operation.

The public repository is the source of truth for the current rule catalogue and supported behavior. Read the pattern breakdown.

The workflow

From
scan to
review.

The current output supports review. It is not signed, independently verified, or hosted by Verglos.

  1. 01

    Pin

    Use a pinned alpha version when the result must be reproducible. The current public package is 2.0.0-alpha.1.

  2. 02

    Scan

    Run npx verglos@2.0.0-alpha.1 scan inside an authorized JavaScript or TypeScript project.

  3. 03

    Review

    Read each candidate in its code context. Scanner severity is a review priority, not proof of reachability or exploitability.

  4. 04

    Retain

    Keep the local HTML and JSON reports with the version, commit, scope, and reviewer decision.

Real scanner surface

Ten security
domains, made
inspectable.

Injection, authentication, authorization, cryptography, misconfiguration, supply chain, API, data exposure, runtime, and monitoring are presented as review domains in the current scanner surface.

The scan report keeps the domain list visible so the coverage claim is verifiable, not marketing.

Verglos ten-domain scanner panel showing injection findings and evidence
Ten-domain panel · one finding with sample evidence

How Verglos differs

Not another
generic scanner.

Verglos is an early local scanner, not a replacement for mature application-security platforms. Compare supported languages, detector coverage, deployment model, policy, and workflow needs.

  • Snyk

    A broader commercial developer-security platform. Evaluate it on supported ecosystems, dependency analysis, policy, and organization workflows.

  • GitHub Advanced Security

    Security capabilities integrated into GitHub repositories and organization workflows. Its deployment model differs from a small local CLI.

  • Semgrep

    A mature rule-based static-analysis platform with broad language and workflow support. Verglos does not claim equivalent coverage.

  • Socket

    A supply-chain security product focused on package behavior and dependency risk. That scope overlaps only part of Verglos scanning.

  • Gitleaks / TruffleHog

    Focused secret-detection tools. They should be evaluated directly when secret discovery and verification are the primary job.

The full comparison, with sourced references: Verglos vs Snyk, GitHub, Semgrep, Socket, Gitleaks.

Shipped · Planned

Honest
roadmap.

What runs in the public alpha today, and which ideas remain explicitly unshipped. Planned items have no launch commitment.

Shipped

  • Public npm package at 2.0.0-alpha.1
  • Functional local scan command for JavaScript and TypeScript projects
  • Local HTML and JSON report output
  • Public source repository and campaign methodology
  • Hunt and attest names present only as unavailable command shells

Planned

  • Proposed local sandbox verification; no launch date
  • Proposed signed summary and hosted verification; no launch date
  • Proposed coding-agent and CI integrations; not supported today
  • Proposed paid plan boundaries; no plans or prices are available

What Verglos is not

Claims we
will not
make.

  • Not an autonomous pentester. Verglos does not choose external targets or run open-ended offensive campaigns.
  • Not a cloud SAST. Verglos does not upload source to run an LLM triage pass.
  • Not a compliance platform. A scanner report does not replace an audit, penetration test, threat model, or secure development lifecycle.
  • Not a signed-evidence product today. Hunt, attest, hosted verification, and paid plans remain unavailable.

Run the
public alpha.

Pin the version, scan an authorized project, and review the report in context. No paid plan or hosted evidence service is available today.