ClawHub
Back to skill

Security audit

多人对话配音 ElevenLabs Dialogue

Security checks across malware telemetry and agentic risk

Overview

This skill is a disclosed cloud audio-generation wrapper, but its examples contain documentation mistakes users should verify before automating it.

Install only if you are comfortable sending dialogue text and any user-selected media files to dLazy's hosted service. Prefer per-invocation credentials or protect ~/.dlazy/config.json if you use saved login. Before building automation around this skill, confirm the actual CLI help and output type because the artifact has stale image-oriented and --prompt examples.

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)

Intent-Code Divergence

Medium
Confidence
94% confidence
Finding
The documented output schema shows an image (`type: "image"`, PNG URL) for a tool advertised as audio dialogue synthesis. This mismatch can cause downstream agents or users to mishandle results, trust the wrong artifact type, or build unsafe automation around incorrect assumptions about generated files and MIME types. In an agent setting, incorrect tool contracts are security-relevant because they can lead to unsafe file handling and confused-deputy behavior.

Intent-Code Divergence

Medium
Confidence
92% confidence
Finding
The examples and error documentation reference a `--prompt` parameter even though the command interface is documented around `--inputs` for multi-speaker dialogue. This inconsistency can cause agents to invoke the tool incorrectly, produce unexpected behavior, or route user content into the wrong parameter format, undermining reliability and potentially exposing unintended data in malformed requests.

VirusTotal

62/62 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.