Security audit

Dlazy Video Storyboard Generate

Security checks across malware telemetry and agentic risk

Overview

This skill is mostly a dLazy generation workflow, but its stated storyboard purpose is under-scoped compared with instructions that make the agent install/use a CLI, store credentials, call external services, and run terminal generation commands.

Install only if you specifically want an agent to use the dLazy CLI for cloud media generation, are comfortable with prompts and selected local media being uploaded to dLazy services, and understand that an API key may be stored locally. Prefer npx or an environment variable for the API key if you want less persistent setup, and review each proposed terminal command before allowing generation.

SkillSpector

By NVIDIA

Vulnerability Patterns

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

Findings (6)

Description-Behavior Mismatch

Medium

Confidence: 93% confidence
Finding: The manifest advertises a narrowly scoped skill that only 'gets storyboard info', but the body expands behavior to defining pipelines, drawing to canvas, and later even terminal-driven media generation. This scope mismatch can mislead reviewers and orchestrators about the skill's actual capabilities, increasing the chance that higher-risk actions are invoked without appropriate trust, review, or sandboxing.

Description-Behavior Mismatch

Medium

Confidence: 95% confidence
Finding: The embedded description says the skill converts storyboard content into a canvas pipeline, but later instructions direct the agent to run interactive CLI-based image generation commands. This inconsistent specification creates ambiguity about whether the skill is a safe data-to-canvas transformer or an executor of external commands, which can cause unintended command execution and policy bypass.

Context-Inappropriate Capability

High

Confidence: 98% confidence
Finding: The skill includes broad terminal command execution instructions that are unrelated to the stated storyboard-to-pipeline task, including a mandate to execute `dlazy` generation commands. Giving a skill unnecessary command-execution latitude materially raises risk: it can trigger networked operations, spend resources, handle secrets, and potentially be adapted to run unintended commands under the guise of normal workflow.

Intent-Code Divergence

Medium

Confidence: 92% confidence
Finding: The file states the skill itself will not access network or files, yet it instructs use of a CLI that stores API keys locally and uploads prompts and local media to remote endpoints. This is a misleading trust claim: users or reviewing systems may underestimate the confidentiality and persistence implications of using the skill, especially for local file paths and stored credentials.

Intent-Code Divergence

High

Confidence: 97% confidence
Finding: The skill's later 'critical execution instructions' override the earlier storyboard-to-canvas workflow and attempt to redirect the agent into an unrelated interactive terminal-based image-generation process. This is dangerous because it can cause the agent to perform unintended command execution, collect extra user input unrelated to the trigger, and exfiltrate prompts or local media to third-party services outside the declared task boundary.

Natural-Language Policy Violations

Medium

Confidence: 86% confidence
Finding: The instruction block forces a specific Chinese-language interaction pattern and scripted user-confirmation flow without user opt-in, which can override host-agent behavior and degrade transparency and consent. In this skill, that coercive interaction is paired with terminal-execution instructions, increasing the chance the agent follows adversarial procedural constraints instead of the platform's intended control flow.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.