Code Refactor for Reproducibility

Security checks across malware telemetry and agentic risk

Overview

This skill is a coherent local research-code refactoring helper, with expected file-generation behavior but some setup hygiene users should review.

Install/use this only for research-code reproducibility work. Run it in a sandbox or disposable working directory, choose a fresh output directory, inspect requirements.txt before installing dependencies, and review generated reproducibility_info.json before sharing results because it may include hostname and package inventory.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (10)

Vague Triggers

Medium

Confidence: 89% confidence
Finding: The activation description is broad enough to match many ordinary coding and research-assistance requests, which can cause the skill to trigger in contexts beyond its intended workflow. Because the skill includes command execution and file modification guidance, overbroad invocation expands the chance of applying powerful actions in unsuitable or insufficiently reviewed situations.

Vague Triggers

Medium

Confidence: 88% confidence
Finding: The 'When to Use' section repeatedly describes positive matches but does not define clear stop conditions or exclusions, making it easy for the skill to be used on adjacent tasks. In a skill that recommends editing code, running scripts, and generating environment setup, ambiguous scope increases the risk of unsafe execution or unintended repository changes.

Missing User Warnings

Medium

Confidence: 83% confidence
Finding: The script recursively reads an attacker- or user-selected input tree and writes a large set of files into a user-selected output directory with no overwrite protection, dry-run mode, or explicit confirmation. In practice, this can destroy or pollute existing directories, especially if pointed at an important path, causing integrity loss and unintended filesystem modification.

Missing User Warnings

Medium

Confidence: 89% confidence
Finding: The generated code collects host metadata such as hostname, platform, Python version, and installed packages and writes it to disk without meaningful user notice or consent. In research and enterprise settings, that inventory can leak sensitive environment details, aiding fingerprinting, internal host identification, or unintentional disclosure when results are shared.

Missing User Warnings

Medium

Confidence: 86% confidence
Finding: The generated code executes `pip freeze` automatically to collect environment information without disclosing that behavior in the skill. While the command is fixed, it still triggers external process execution and creates a detailed software inventory that may be sensitive or unexpected in restricted environments.

Unpinned Dependencies

Low

Category: Supply Chain
Content: numpy pandas pytest scipy
Confidence: 95% confidence
Finding: numpy

Unpinned Dependencies

Low

Category: Supply Chain
Content: numpy pandas pytest scipy src
Confidence: 95% confidence
Finding: pandas

Unpinned Dependencies

Low

Category: Supply Chain
Content: numpy pandas pytest scipy src
Confidence: 94% confidence
Finding: pytest

Unpinned Dependencies

Low

Category: Supply Chain
Content: numpy pandas pytest scipy src
Confidence: 95% confidence
Finding: scipy

Unpinned Dependencies

Low

Category: Supply Chain
Content: pandas pytest scipy src
Confidence: 89% confidence
Finding: src

VirusTotal

66/66 vendors flagged this skill as clean.

View on VirusTotal