Back to skill

Security audit

Agent Bom Scan

Security checks across malware telemetry and agentic risk

Overview

This skill appears aimed at dependency vulnerability scanning, but its privacy and network behavior are inconsistent enough that users should review it before installing.

Install only if you are comfortable with dependency names or hashes potentially being sent to public vulnerability databases. Treat scan reports as sensitive, scope scanning to projects you are authorized to assess, and verify whether the actual workflow is offline or network-assisted before use.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (4)

Intent-Code Divergence

Medium
Confidence
97% confidence
Finding
The embedded skill reference states that all scans execute locally, no dependency data is sent to external servers, and scanning is offline. Earlier in the README, the documented requirements and security notes explicitly allow network access to query public CVE/OSV databases and say dependency hashes/names may be sent externally, which directly contradicts the offline/privacy claim.

Description-Behavior Mismatch

Medium
Confidence
95% confidence
Finding
The skill's stated description/capabilities in the reproduced reference include 'Offline scanning' and assert that no dependency data is sent to external servers. However, the actual documented operation in this file includes network access to query CVE/OSV databases and sending dependency names or hashes to public APIs, which exceeds the claimed offline-only behavior.

Natural-Language Policy Violations

Medium
Confidence
94% confidence
Finding
The README's earlier sections disclose network access and transmission of dependency names/hashes to public vulnerability databases, but the reproduced reference claims 'Complete Privacy' and 'No source code or dependency data is sent to external servers' with 'Offline scanning'. This natural-language inconsistency can mislead users about data handling and privacy expectations.

Excessive Permissions

Low
Category
Privilege Escalation
Content
All permissions above are capability requirements, not blanket authorization.
Grant only what the selected workflow needs, scope filesystem access to the
working directory, and do not elevate privileges unless SKILL.md explicitly
requires and explains it.

## ๐Ÿ”’ Security & Privacy
Confidence
80% confidence
Finding
elevate privileges

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.