Security audit

Dlazy Search Video

Security checks across malware telemetry and agentic risk

Overview

The skill is a disclosed dLazy-hosted Pixabay video search wrapper, with some confusing generic documentation but no evidence of hidden, destructive, or deceptive behavior.

Before installing, be comfortable sending search queries to dLazy and storing a dLazy API key locally. Use --query for searches, prefer DLAZY_API_KEY if you do not want a saved key, and only use --save with paths you explicitly choose.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (6)

Description-Behavior Mismatch

High

Confidence: 95% confidence
Finding: The skill is presented as a simple Pixabay keyword video search tool, but the documentation states that prompts and parameters are sent to dLazy-hosted APIs and may involve generic media upload and hosted result delivery. This mismatch is dangerous because it can mislead users and agents about the true data flow, trust boundary, and external service being invoked, causing unintended disclosure of prompts, metadata, or local media to a third-party SaaS.

Context-Inappropriate Capability

Medium

Confidence: 90% confidence
Finding: For a tool whose stated purpose is keyword-based video search, support for local file uploads, saving remote assets to arbitrary local paths, and pipe-based references expands capabilities beyond what users would reasonably expect. This increases the risk of unnecessary local file exposure to remote infrastructure and unintended writes to the local filesystem if an agent follows the documented options without strong validation.

Intent-Code Divergence

Medium

Confidence: 82% confidence
Finding: The command help documents a `--query` parameter, but examples and error handling refer to `--prompt`, indicating inconsistent or possibly templated documentation. This is dangerous because agents may invoke the wrong interface, hit fallback behaviors, or send unintended free-form prompt data to a backend that appears more general-purpose than a constrained search API.

Description-Behavior Mismatch

Medium

Confidence: 95% confidence
Finding: The skill metadata claims a narrow Pixabay keyword video search tool, but the body documents a generic hosted dLazy API client with authentication, remote inference, task handling, uploads, and saved credentials. This mismatch can mislead an agent or user about the actual trust boundary and capabilities, causing unintended data exposure or execution of a broader remote service than expected.

Description-Behavior Mismatch

Medium

Confidence: 94% confidence
Finding: The usage and behavior sections describe prompt submission to a hosted inference API plus uploading local image/video/audio paths, which materially exceeds a simple keyword-based stock video search function. A user invoking what appears to be a search skill could unintentionally transmit sensitive prompts or local files to third-party infrastructure.

Missing User Warnings

Low

Confidence: 87% confidence
Finding: The skill discloses elsewhere that local file paths may be uploaded to files.dlazy.com and that --save writes results locally, but the invocation section does not foreground these side effects where agents are most likely to copy commands. This weak notice increases the chance of accidental exfiltration of local media or unintended filesystem writes during routine use.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.