Security audit

Ad Creative

Security checks across malware telemetry and agentic risk

Overview

This is mostly a legitimate ad-creative guide, but it needs review because it reaches into campaign-management commands and voice-cloning workflows without clear authorization and consent boundaries.

Review this skill before installing if your agent has access to ad-platform accounts, API keys, media-generation services, or voice-cloning tools. Use it for drafting and iterating creative, but require explicit approval before it creates or manages campaigns, spends ad budget, uploads assets, installs production tools, or clones any real person's voice.

SkillSpector

By NVIDIA

Vulnerability Patterns

Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
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: 91% confidence
Finding: The skill’s declared purpose is ad-copy generation, but this section expands into image/video generation workflows and scaled production guidance. That broadens the operational scope beyond copywriting and can cause an agent to invoke external media tooling or production workflows the user did not request, increasing the chance of unnecessary tool use and unintended side effects.

Description-Behavior Mismatch

Medium

Confidence: 93% confidence
Finding: The tool integration section includes campaign-management operations such as campaign creation/listing even though the skill is scoped as ad-creative generation. In an agentic environment, documenting adjacent management actions can encourage over-privileged behavior, where the agent transitions from drafting copy to pulling account data or modifying campaigns without a tightly bounded purpose.

Context-Inappropriate Capability

Medium

Confidence: 89% confidence
Finding: Code-based video rendering and external generative media tools are unnecessary for a skill whose primary function is ad-copy generation. Including these capabilities increases attack surface and encourages escalation into file generation, external service use, or automated media production that is not justified by the skill’s core task.

Vague Triggers

Medium

Confidence: 83% confidence
Finding: The activation description is very broad and includes many common phrasings, which can cause this skill to trigger for loosely related marketing requests. Over-broad invocation criteria increase the risk of inappropriate skill selection, unnecessary context access, and execution of guidance outside the user’s actual intent.

Missing User Warnings

Medium

Confidence: 92% confidence
Finding: This section explicitly recommends voice cloning for ad creative and brand spokesperson replication but provides no warning about consent, authorization, disclosure, or impersonation risks. In a marketing-focused skill, this omission can normalize deceptive or unauthorized cloning of real people’s voices, enabling fraud, brand impersonation, or misleading ads at scale.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.