pkgscan.dev · Comparison overview
One malware engine, five ways to check it.
Argus is our in-house malware engine, and it scans any artifact — packages, extensions, agent skills, or a raw file. These pages compare it honestly with the engines we run alongside it. The point isn't who wins; it's which engine answers the question you've got.
- 6 engines in a pkgscan scan
- 16+ registries covered
- 0 accounts needed to read a report
Choose a comparison
GuardDog's a fast, free rule engine, though its results can be noisy. Argus reads what the package actually does — which is how it catches a campaign before a rule exists for it.
Best for: Teams deciding whether a hosted content engine is worth it next to an open-source rule set.
Read the comparison → 02 Argus vs SkillSpector NVIDIA Generalist vs skill specialistArgus scans any artifact, skills included. SkillSpector only focuses on skills, so it's the one that names the skill-specific risk.
Best for: Anyone scanning AI agent skills alongside npm, PyPI and VS Code packages.
Read the comparison → 03 Argus vs OSV Google / OpenSSF Advisories vs behaviourOSV is the open advisory database — great for what's already known. Argus looks at the code in front of you, including the window before disclosure.
Best for: Teams that already run OSV and want coverage for the window before disclosure.
Read the comparison → 04 Argus vs Grype Anchore CVEs vs malwareGrype's an offline CVE scanner built for scale. Argus looks for malicious intent, which isn't a CVE.
Best for: Container and SBOM teams separating vulnerability management from malware detection.
Read the comparison → 05 Argus vs deps.dev Google Metadata vs bytesdeps.dev is a fast, free PURL-only metadata API. Argus sees the bytes and reads them, so it sees behaviour metadata can't describe.
Best for: Identity-only pipelines that want to add content analysis where bytes exist.
Read the comparison →What every pkgscan scan includes
pkgscan doesn't ask you to pick an engine. A submission runs every provider that accepts its input — bytes, package URL or digest — and the report shows each verdict separately. Argus, SkillSpector and the advisory engines aren't competitors in the product; they're layers.
| Engine | Class | Question it answers |
|---|---|---|
| Argus | Malware | Does this code behave like an attack? |
| SkillSpector | Malware | Does this agent skill try to hijack, exfiltrate or escalate? |
| GuardDog | Heuristic | Does this package match a known bad pattern? |
| OSV | Vulnerability | Is this version covered by a published advisory? |
| Grype | Vulnerability | Does this version match a known CVE? |
| deps.dev | Vulnerability + advisories | What does the package metadata say? |
Why Argus leads the malware layer
- It reads first-party code. A clean package name, a copied README and a fresh publisher account won't hide behaviour from content analysis.
- It scans any artifact. Packages, extensions, agent skills and raw files go through the same engine — skills don't need a separate scanner to be read.
- It's trained on live campaigns. Detections come from attacks our researchers track and take down, not only from public feeds an attacker can read.
- It runs inside Open VSX at publish time. Argus is deployed as a registry gate, so its findings reflect real publish and takedown history.
- It catches what no one else does. Argus has flagged artifacts that every other engine that ran passed, and a VS Code Marketplace backdoor that only 3 of 71 antimalware engines detected.
Machine-readable copy: /comparison.json.