ClawHub
Back to skill

Security audit

deAPI AI Media Suite (Community)

Security checks across malware telemetry and agentic risk

Overview

This is a documentation-only deAPI media API skill with disclosed API-key use and media processing behavior, though users should handle inputs and voice/media content carefully.

Install only if you are comfortable giving the agent a deAPI API key and sending chosen prompts, media files, voice samples, and URLs to deAPI. Use a dedicated or low-balance key, avoid confidential or third-party media unless you understand provider handling, get consent before voice cloning, and validate URLs and file paths before running the curl 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
98% confidence
Finding
The security note makes materially inaccurate safety claims: several examples do not use jq and instead interpolate variables directly into curl -F arguments and shell constructs. Misstating the safety properties of the skill can cause downstream agents or users to overtrust the examples and reuse unsafe patterns, increasing the chance of shell argument injection or accidental mishandling of untrusted input.

Intent-Code Divergence

Medium
Confidence
96% confidence
Finding
The skill claims it does not download content or binaries, but multiple examples fetch remote media with curl -o before uploading it for OCR, img2video, background removal, upscale, and img2img. This discrepancy is security-relevant because downloading attacker-controlled remote files to local storage changes the trust boundary and can expose agents to SSRF-like fetches, oversized files, or unsafe content handling.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal