Argus — The Hundred-Eyed

• Outcome: Pass
• Latest audit: May 28, 2026 at 1:14 PM UTC
• Version:
  1.0.4

Overview

Argus is a local code scanner whose report files should be understood before use, but the artifacts do not show hidden exfiltration, destructive behavior, or privilege abuse.

Before installing, be aware that this scanner installs the Python rich package and writes a local Markdown report by default, with optional JSON if OUTPUT_JSON=true. Run it from a directory where report files are acceptable, and avoid scanning highly sensitive repositories unless you are comfortable with findings and short code snippets being saved locally.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (4)

Intent-Code Divergence

High

• Confidence: 97% confidence
• Finding: The skill explicitly claims it 'reads local files only', but the embedded script writes a Markdown report and optionally a JSON file to disk. This is a real integrity/transparency issue because users may run it expecting read-only behavior, and the generated reports can persist sensitive findings or code snippets on the filesystem.

Description-Behavior Mismatch

Medium

• Confidence: 91% confidence
• Finding: The top-level description presents the skill as a scanner, but the implementation also creates persistent report files in the current directory. While not code execution or exfiltration, this is a true behavior mismatch that can surprise users and leave behind sensitive artifacts from scanned source code.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: The markdown describes local scanning but omits that the script persists scan results to disk, including findings metadata and snippets from source lines. Because scanned code may contain secrets or proprietary content, undocumented persistence increases the risk of accidental disclosure to other local users, backups, or later tooling.

Ssd 3

Medium

• Confidence: 98% confidence
• Finding: The report-writing logic stores file paths, line numbers, issue descriptions, and source code snippets for findings, especially critical/high ones. This can expose hardcoded credentials, tokens, or other sensitive application data from scanned files into secondary report artifacts, broadening the exposure surface beyond the original repository.

VirusTotal

VirusTotal findings are pending for this skill version.