Source Evaluator

Security checks across malware telemetry and agentic risk

Overview

This is a coherent research source-evaluation skill with only expected document reading, optional web lookup, and a disclosed notes output file.

Installers should expect this skill to read the source materials they provide, optionally fetch public web pages for source evaluation, and write an evaluated-sources.md notes file. Review the output path before allowing writes, and do not approve any unexpected payment, crypto, credential, or account-access request because those are not part of this skill’s stated purpose.

SkillSpector

By NVIDIA

Vulnerability Patterns

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
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep

Findings (2)

Missing User Warnings

Low

Confidence: 84% confidence
Finding: The skill declares an output file in the environment metadata without requiring confirmation before writing, which can lead to unexpected workspace modifications or accidental overwrite if an agent executes it automatically. In agentic systems, silent file creation is a real safety issue because it changes user files without explicit consent, even if the content itself is benign.

Missing User Warnings

Low

Confidence: 95% confidence
Finding: The output section explicitly instructs the agent to produce `evaluated-sources.md` but provides no warning about file creation, overwrite behavior, or alternatives such as returning the content inline first. This creates a tangible risk of unintended file writes in the user's workspace, especially if the skill is auto-invoked or run in batch workflows.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal