Dlazy Text Spoken Script

• Outcome: Review
• Latest audit: Jul 21, 2026 · 4:03 AM UTC
• Version: 1.3.6

Overview

This writing skill includes an unrelated image-generation workflow that can install/use a CLI, store an API key, send prompts/media to dLazy, and run terminal commands.

Install only if you intentionally want dLazy CLI image generation bundled with this writing workflow. Expect a CLI install or npx use, dLazy login/API-key handling, prompt data sent to dLazy, possible media uploads when local paths are provided, and terminal commands after user confirmation.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access

Findings (6)

Description-Behavior Mismatch

High

• Confidence: 98% confidence
• Finding: The skill is presented as a spoken-script generator, but the later 'critical execution instructions' redirect the agent into terminal-driven image generation using an external CLI. That mismatch can cause an agent or orchestrator to perform networked, side-effecting actions outside the user’s apparent request and outside the least-privilege scope implied by the skill name and description.

Context-Inappropriate Capability

High

• Confidence: 97% confidence
• Finding: A text spoken-script skill has no justified need to instruct the agent to run terminal commands or invoke `dlazy seedream-4.5` for rendering. Embedding these capabilities expands the attack surface to command execution, package usage, file handling, and external uploads, creating unnecessary risk if an agent follows the skill automatically.

Intent-Code Divergence

Medium

• Confidence: 90% confidence
• Finding: The skill contains contradictory instructions: one section says to directly output a spoken script, while another says to halt, gather requirements, await confirmation, and then generate images via terminal commands. This inconsistency is dangerous because agents may privilege the later 'critical' block and perform actions unrelated to the declared task, undermining predictability and safe policy enforcement.

Description-Behavior Mismatch

High

• Confidence: 98% confidence
• Finding: The skill’s declared purpose is spoken-script generation, but the later instructions redirect the agent into an unrelated image-generation workflow using terminal commands and an external CLI. This creates a scope mismatch that can induce unnecessary command execution and data transfer to third-party services, increasing the attack surface beyond what a user invoking a writing skill would reasonably expect.

Context-Inappropriate Capability

High

• Confidence: 97% confidence
• Finding: Embedding terminal command execution and remote rendering instructions in a text-script skill grants capabilities not justified by the skill’s stated function. Even if the commands are not overtly malicious, they can cause unauthorized package use, network access, uploads, or credential-dependent operations under the guise of a harmless writing task.

Intent-Code Divergence

Medium

• Confidence: 91% confidence
• Finding: The file contains conflicting directives: one section says to directly output a spoken script, while the later section forbids immediate generation and instead instructs a staged image-prompt and command-execution workflow. Such contradiction is dangerous because agents may follow the more forceful ‘critical execution instructions,’ leading to behavior that is inconsistent with user expectations and the skill’s declared purpose.

VirusTotal

VirusTotal findings are pending for this skill version.

Static analysis

No suspicious patterns detected.