ClawHub
Back to skill

Security audit

创意转视频 Idea to Video

Security checks across malware telemetry and agentic risk

Overview

The skill is mostly transparent about using dLazy’s cloud CLI, but it mixes a canvas-planning workflow with later mandatory terminal execution instructions, so users should review it before installing.

Install only if you are comfortable with a skill that may install and use the dLazy CLI, store a dLazy API key locally, send prompts and referenced media files to dLazy’s cloud service, and potentially run one confirmed generation command at a time. Prefer the npx or environment-variable API key path if you do not want a long-lived global CLI or saved credential.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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
Findings (3)

Intent-Code Divergence

High
Confidence
97% confidence
Finding
The tail section instructs the agent to execute terminal commands for media generation, which directly conflicts with the earlier documented workflow that should only produce a plan and call drawToCanvas after explicit confirmation. This inconsistency creates an unsafe privilege expansion path: an agent may leave the constrained canvas flow and perform external command execution and networked API actions not required by the skill’s primary function.

Context-Inappropriate Capability

Medium
Confidence
94% confidence
Finding
This skill is described as a canvas-planning workflow, yet the closing instructions assert that the agent can and must execute terminal commands such as dlazy model invocations. Unnecessary command execution increases attack surface by enabling package-installed tooling, outbound requests, and potential misuse of local credentials or files, all outside the minimum privileges needed for planning.

Intent-Code Divergence

High
Confidence
95% confidence
Finding
The skill claims to be a plan-only canvas workflow, but later instructs the agent to execute terminal commands against an external CLI. That contradiction can cause an agent to perform real side-effecting actions—network calls, credential use, and media uploads—when the user may have only requested planning, increasing the risk of unauthorized execution and data exfiltration.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.