工作量拆分-Cosmic

Security checks across malware telemetry and agentic risk

Overview

The available evidence shows a document-analysis skill with only a low-impact shell fallback for counting lines, not hidden or destructive behavior.

Install only if you are comfortable with the skill potentially asking the agent to run a simple local shell command as part of output validation. Review the prompt before use in restricted environments, but the supplied evidence does not justify holding it for Review.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (1)

Context-Inappropriate Capability

Medium
Confidence
92% confidence
Finding
The prompt instructs the agent to execute Windows PowerShell commands as a fallback for counting lines in a generated file. Even though the command is simple, it expands the skill's operational scope from document analysis into shell execution, which increases attack surface and can normalize command execution in a workflow that does not inherently require it. The surrounding skill context is document decomposition, so this capability is less justified and therefore more concerning.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal