ClawHub
Back to skill

Security audit

音乐

Security checks across malware telemetry and agentic risk

Overview

This music-generation skill is not clearly malicious, but it asks for broad read/write/command execution authority that is not well scoped to making music.

Install only if you are comfortable granting this skill broad local file and command-execution abilities. For normal music prompt generation, prefer a version that either removes exec/write access or documents exact allowed commands, file locations, API providers, and user confirmation steps.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Description-Behavior Mismatch

Medium
Confidence
92% confidence
Finding
The skill is presented as a music-generation tool, but its documented capabilities expand into generic file handling, external API use, and command execution. This mismatch increases the chance that users or agents will grant broader privileges than expected, enabling misuse of read/write/exec features under the cover of a benign creative workflow.

Intent-Code Divergence

Medium
Confidence
95% confidence
Finding
The documentation says risky code and external dependencies were removed, yet the skill still advertises command execution and discusses command-execution risk controls. This contradiction is a red flag because it can create false trust and encourage operators to approve a skill that still exposes a powerful execution surface.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.