ClawHub
Back to skill

Security audit

cortex-backtest

Security checks across malware telemetry and agentic risk

Overview

The skill is mostly a Cortex backtesting guide, but it also tells agents to modify persistent workspace records and potentially use sudo to change system-level Cortex configuration.

Install only if you are comfortable with the agent managing Cortex setup records and backtest workspace files. Prefer the workspace or tenant configuration modes, and require explicit confirmation before any sudo, /opt, /etc, TOOLS.md, or experience-document deletion/update action.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (3)

Description-Behavior Mismatch

Medium
Confidence
89% confidence
Finding
The skill’s documented workflow goes beyond running backtests and instructs the agent to read, append, delete, and maintain persistent files such as experience documents and TOOLS.md. That expands the skill’s authority into long-lived workspace state mutation, creating a risk of unintended persistence, tampering, or cross-task contamination if triggered during normal use.

Context-Inappropriate Capability

High
Confidence
97% confidence
Finding
The skill includes system-level configuration management steps using sudo and writes under /opt, which is far beyond what is necessary for ordinary backtest execution. If followed by an agent, this could alter shared host configuration, affect other users or agents, and turn a user request about backtesting into privileged system administration.

Vague Triggers

Medium
Confidence
84% confidence
Finding
The trigger phrases are broad enough to activate on generic quantitative-analysis requests, which increases the chance the skill runs in contexts where its setup and file-management instructions are unnecessary. In this skill, that matters because activation can lead the agent into reading and modifying workspace configuration and persistent documents, amplifying the effect of accidental invocation.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.