ARGUS · Software composition analysis

Dependency scanning that shows which CVEs your code actually reaches.

ARGUS software composition analysis reads your lockfiles, matches every installed version to OSV advisories and checks whether your code imports the vulnerable package. Reachable, actively exploited CVEs rise to the top. Unreachable ones sink without disappearing. Malicious and lookalike packages are caught too.

Coverage

What it checks

Every scan reads your lockfiles, not just manifests, so it sees the exact versions you install.

  1. 01

    Known vulnerabilities

    Each installed version matched against OSV.dev advisories, enriched with EPSS and the CISA Known Exploited Vulnerabilities flag.

  2. 02

    Reachability, with evidence

    Each finding marked reachable, unreachable or unknown, with the reason, such as the file that imports it.

  3. 03

    Malicious and lookalike packages

    Known malicious packages are flagged critical. Names one edit away from a popular package are flagged as possible typosquats, including names an AI assistant made up.

  4. 04

    A check before install

    Your coding agent can ask ARGUS over MCP for an allow, warn or block verdict before it installs a package.

How it works

From scan to fix

  1. 01

    Lockfiles are parsed

    npm, Yarn and pnpm, PyPI, Go, RubyGems, Cargo and Composer lockfiles.

  2. 02

    Advisories are matched

    Exact versions against OSV, with CVSS from the advisory and EPSS and KEV added.

  3. 03

    Imports are traced

    For JavaScript, TypeScript and Python, ARGUS checks whether your source imports the package. Any doubt resolves to reachable.

  4. 04

    One ranked list

    A 0 to 100 priority blends severity, exploitation signals and reachability. Group by package and a long list becomes a few upgrade decisions.

Why teams use it

Signal over noise

Marked, never deleted

Unreachable findings are ranked lower and labelled with the evidence. Nothing is silently hidden.

Exploitation first

A reachable CVE on CISA's known-exploited list ranks above a theoretical one.

Supply chain, not just CVEs

Malicious packages, typosquats and AI-invented package names are checked in the same scan.

Limits, stated plainly

What it does not do yet

We would rather you hear it here than find it in week two.

  • Reachability is import-level, not call-graph, and covers JavaScript, TypeScript and Python. Go, Maven, Cargo and RubyGems findings are always treated as reachable.
  • Java and .NET lockfiles are not yet matched to advisories. Those packages still appear in your SBOM and container results.
  • A package.json without a lockfile produces no dependency findings, because ARGUS reads installed versions.
  • Pull request reviews do not run full advisory matching; full scans do.
FAQ

SCA questions, answered

Straight answers about SCA and how ARGUS does it. Updated 17 November 2026.

Still have questions? Contact us

What is software composition analysis (SCA)?

Software composition analysis (SCA) identifies the open-source and third-party packages in your application and checks them against vulnerability databases. ARGUS reads your lockfiles, matches every installed version to OSV advisories, adds EPSS and CISA KEV signals, and checks whether your code actually imports each vulnerable package.

What is the purpose of software composition analysis?

SCA answers three questions: what open source do we ship, which of those versions have known vulnerabilities or are malicious, and which should we fix first. ARGUS adds the third answer by ranking each finding with a 0 to 100 priority that blends severity, exploitation signals and reachability.

What is the difference between SAST and SCA?

SAST analyses the code your team writes. SCA analyses the third-party packages that code depends on. You need both, because a vulnerability can sit in either place. ARGUS runs SAST and SCA in the same scan and puts every finding on one ranked list.

What is reachability analysis in SCA?

Reachability analysis checks whether your code actually imports a vulnerable package. ARGUS marks each finding reachable, unreachable or unknown, with an evidence string such as the file that imports it. Unreachable findings are ranked lower, never deleted, and any doubt resolves to reachable.

Why does ARGUS read lockfiles instead of package.json?

Lockfiles record the exact versions you install; manifests such as package.json usually list version ranges. Matching exact versions avoids guessing. The trade-off: a project with a manifest but no lockfile produces no dependency findings, so commit your lockfile.

Which package ecosystems does ARGUS support?

Advisory matching covers npm, Yarn and pnpm, PyPI, Go, RubyGems, Cargo and Composer lockfiles. Reachability covers JavaScript, TypeScript and Python. Java and .NET lockfiles are not matched to advisories yet, though those packages still appear in your SBOM.

How do EPSS and CISA KEV change the priority of a CVE?

EPSS, from FIRST, estimates the probability that a vulnerability will be exploited soon. CISA's Known Exploited Vulnerabilities catalog lists flaws already exploited in the wild. ARGUS blends both with CVSS and reachability, so a reachable, actively exploited CVE rises to the top.

What are typosquatting and slopsquatting?

Typosquatting is publishing a malicious package under a name one typo away from a popular one. Slopsquatting is registering a package name that AI coding assistants tend to invent. ARGUS flags names one edit away from popular npm, PyPI, Go and RubyGems packages, and flags known malicious packages as critical.

Can ARGUS stop a coding agent installing a bad package?

Yes. Over MCP, an assistant such as Claude Code or Cursor can ask ARGUS for an allow, warn or block verdict before installing. The check covers known vulnerabilities, malicious advisories, lookalike names and whether the package exists at all.

Can I scan dependencies without giving ARGUS code access?

Yes. Upload a CycloneDX or SPDX SBOM and ARGUS runs the same vulnerability and malicious-package checks. Checks that need source code, such as SAST, secrets and reachability, are reported as not assessed, never as zero.

Try it on one repository.

Connect a GitHub repo and see the findings, ranked, in your first scan.