Security audit

Graph Knowledge Base Free

Security checks across malware telemetry and agentic risk

Overview

This skill appears to be a local knowledge-base helper, but its instructions are inconsistent and request write/exec authority without clear scoping or a bundled script.

Install only if you are comfortable with an agent using local read/write and shell execution for this skill. Before using it, define a specific knowledge-base directory, avoid using it for generic analytics or visualization tasks, and treat delete/import/export wording as unsupported unless the publisher provides corrected documentation and the missing script.

SkillSpector

By NVIDIA

Vulnerability Patterns

Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Behavioral ASTexec() Call, eval() Call, Dynamic Import
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

Findings (5)

Description-Behavior Mismatch

Medium

Confidence: 92% confidence
Finding: The capability sections claim create/query/modify/delete/import/export support even though the free edition elsewhere says facts are never deleted and that import/export are unavailable. In an agent setting, contradictory documentation can cause the model to invoke unintended write or destructive operations, expanding the effective permission scope beyond what users reasonably expect.

Intent-Code Divergence

High

Confidence: 96% confidence
Finding: The document states that facts are never deleted, then later says delete operations are supported. This inconsistency is dangerous because an agent may select a destructive path that irreversibly alters local knowledge files, violating the core retention guarantee and user trust.

Intent-Code Divergence

High

Confidence: 95% confidence
Finding: Earlier sections advertise import/export operations, but the free-version limitations later deny export and batch import. In practice, an agent could attempt unsupported data movement actions, causing accidental disclosure, overwrite attempts, or confusing failure modes around local knowledge data.

Vague Triggers

High

Confidence: 89% confidence
Finding: The trigger condition says to use this skill for broad data analysis, reporting, statistics, and visualization, which exceeds the stated local fact-management scope. Overbroad routing guidance can cause the agent to invoke this skill in inappropriate contexts, leading to unnecessary file writes or misuse of exec-enabled operations.

Missing User Warnings

Medium

Confidence: 87% confidence
Finding: The skill uses write and exec capabilities and the documentation mentions summary overwrites, but it does not prominently warn that local files will be created and overwritten. In an agent environment, missing side-effect warnings increase the risk of silent modification of user data and unsafe execution assumptions.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.