ClawHub
Back to skill

Security audit

Background Removal

Security checks across malware telemetry and agentic risk

Overview

The skill is a disclosed inference.sh image background-removal helper with some install and scope cautions, but no artifact-backed evidence of malicious behavior.

Before installing, prefer the documented manual install or inspect the inference.sh install script instead of piping it directly to `sh`. Install only if you trust inference.sh, and use the skill for user-directed image work because some examples go beyond background removal into image generation or background replacement.

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
91% confidence
Finding
The skill is presented as a narrowly scoped background-removal capability using BiRefNet, but its instructions also direct the agent to use unrelated image generation and general editing apps such as Reve and FLUX. This scope expansion can cause an agent to perform materially different actions than the user intended, including generating or altering content instead of only removing a background, which is a real security/safety issue in tool selection and least-privilege terms.

Intent-Code Divergence

Medium
Confidence
89% confidence
Finding
The skill claims its output is a transparent PNG, but examples show broader operations like changing the background and generating new images. This mismatch can mislead an agent or user about expected behavior and output guarantees, increasing the chance of unintended edits, fabricated content, or incorrect downstream handling of outputs under the assumption that only background removal occurred.

VirusTotal

66/66 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.