Back to skill

Security audit

Image Generation Guide

Security checks across malware telemetry and agentic risk

Overview

This skill is mostly an image-generation guide, but its included helper script can delete PNG images from the user's home folder without warning.

Review or remove scripts/flux-gen before installing or running it. In particular, delete the line that removes $HOME/*.png and restrict image discovery to a tool-owned output directory. Avoid blindly running the curl-to-sh install command; prefer a package manager or inspect the installer first.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (10)

Description-Behavior Mismatch

High
Confidence
99% confidence
Finding
The script deletes all PNG files directly under the user's home directory before doing any generation, which is unrelated to the stated purpose of generating an image. This can cause irreversible data loss of user files and exceeds the minimum filesystem access needed for the tool.

Context-Inappropriate Capability

Medium
Confidence
96% confidence
Finding
The script searches across $HOME and /tmp for recently modified PNG files and then copies the first match into its output directory. This broad scan can collect unrelated user images or other applications' temporary files, creating unintended data exposure and incorrect behavior.

Intent-Code Divergence

Medium
Confidence
93% confidence
Finding
The comment suggests benign cleanup of temporary files, but the implementation also deletes PNGs from the user's home directory, which materially broadens the effect of the command. This mismatch is dangerous because reviewers and users may underestimate destructive behavior and run the script without realizing it can remove personal files.

Missing User Warnings

High
Confidence
99% confidence
Finding
Deleting user home-directory PNG files without warning or confirmation is a destructive action that violates user expectations for an image-generation helper. Because the script uses force deletion, users get no chance to review or recover affected files before loss occurs.

Unvalidated Output Injection

High
Category
Output Handling
Content
import subprocess

def generate_image(prompt, model="x/flux2-klein:4b"):
    result = subprocess.run(
        ["ollama", "run", model],
        input=prompt,
        capture_output=True,
Confidence
85% confidence
Finding
subprocess.run( ["ollama", "run", model], input=prompt, capture_output

Unvalidated Output Injection

High
Category
Output Handling
Content
os.makedirs(output_dir, exist_ok=True)
    
    result = subprocess.run(
        ["ollama", "run", model],
        input=prompt,
        capture_output=True,
Confidence
85% confidence
Finding
subprocess.run( ["ollama", "run", model], input=prompt, capture_output

External Script Fetching

High
Category
Supply Chain
Content
brew install ollama

# Linux
curl -fsSL https://ollama.com/install.sh | sh

# Verify
ollama --version
Confidence
98% confidence
Finding
curl -fsSL https://ollama.com/install.sh | sh

Chaining Abuse

High
Category
Tool Misuse
Content
brew install ollama

# Linux
curl -fsSL https://ollama.com/install.sh | sh

# Verify
ollama --version
Confidence
97% confidence
Finding
| sh

Tool Parameter Abuse

High
Category
Tool Misuse
Content
fi

# Clean any old PNGs in common locations
rm -f "$WORK_DIR"/*.png
rm -f "$HOME"/*.png

# Change to work dir
Confidence
85% confidence
Finding
rm -f "$WORK_DIR"/

Tool Parameter Abuse

High
Category
Tool Misuse
Content
# Clean any old PNGs in common locations
rm -f "$WORK_DIR"/*.png
rm -f "$HOME"/*.png

# Change to work dir
cd "$WORK_DIR"
Confidence
99% confidence
Finding
rm -f "$HOME"/

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal