Security audit

Web Crawler Engine Free

Security checks across malware telemetry and agentic risk

Overview

The skill is mostly a local web and community archive guide, but it asks for broad execution and storage authority while under-scoping private message collection and general database use.

Review this before installing if you plan to archive Discord, Slack, or similar communities. Use it only on data you are authorized to collect, prefer official bot APIs and public channels, avoid user tokens, set clear retention limits, and keep exports free of private messages, credentials, and unnecessary user identifiers. Treat it as a write-capable local archive tool, not a read-only SQL helper.

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)

Intent-Code Divergence

High

Confidence: 93% confidence
Finding: The skill metadata and description frame the SQL capability as read-only, but the body broadens behavior to support create/query/export operations and later discusses deletion and unsafe modification workflows. This mismatch can cause an agent to grant the skill write-capable authority in contexts where users or orchestrators expected only non-destructive access, increasing the risk of unintended data mutation or destructive execution via exec-enabled commands.

Description-Behavior Mismatch

High

Confidence: 91% confidence
Finding: The documented behavior exceeds the manifest's stated scope by introducing create operations and lifecycle management actions such as deletion/cleanup. In an agent ecosystem, this kind of scope expansion is dangerous because routing, permissioning, and user trust may be based on the manifest, while the actual instructions encourage broader and potentially destructive behavior.

Vague Triggers

Medium

Confidence: 77% confidence
Finding: The invocation guidance says to use the skill for general database operations, SQL queries, and storage management, which is much broader than a web/archive crawler with local search. This can lead an agent to invoke the skill for unrelated database tasks, exposing exec and write tooling in contexts where a narrower, safer tool should have been selected.

Vague Triggers

Medium

Confidence: 82% confidence
Finding: The keyword coverage section enumerates an overly broad trigger surface, effectively advertising the skill for many loosely related requests. In an automated agent-selection pipeline, this raises the chance of over-invocation, causing the agent to use exec-capable crawling/archive instructions in situations outside the intended scope.

Missing User Warnings

Medium

Confidence: 88% confidence
Finding: The skill encourages archiving community messages including direct messages, user IDs, timestamps, and attachment URLs, but does not present a clear upfront privacy notice, consent boundary, or retention warning before describing collection workflows. Because the skill is designed for local archival and search, this creates a meaningful risk of over-collection, retention of sensitive communications, and privacy violations, especially when paired with exec tooling and broad invocation language.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.