Security audit

seedance2-prompter

Security checks across malware telemetry and agentic risk

Overview

This is mostly a Seedance video-prompt helper, but it includes guidance for disguising brand or character references to get past platform review.

Install only if you specifically want a Seedance prompt-writing helper and will review its outputs before use. Do not use the IP-avoidance workflow to imitate protected characters, brands, logos, celebrities, or platform-blocked material; prefer fully original, licensed, or public-domain concepts. The files do not show credential access, persistence, network installation, or destructive behavior.

SkillSpector

By NVIDIA

Vulnerability Patterns

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
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (6)

Description-Behavior Mismatch

High

Confidence: 99% confidence
Finding: This document explicitly frames its purpose as avoiding Seedance IP/copyright review and then provides concrete tactics for disguising recognizable copyrighted characters and brands. That goes beyond normal prompt engineering and materially enables policy evasion and likely infringement by helping users reproduce protected IP while avoiding detection.

Context-Inappropriate Capability

High

Confidence: 99% confidence
Finding: The content provides moderation-evasion and IP-avoidance tactics such as substituting descriptive aliases, changing iconic traits, and using negative constraints to suppress explicit names while preserving recognizable aesthetics. In the context of a prompt-writing assistant, these instructions facilitate circumvention of platform safeguards rather than legitimate creative assistance.

Ssd 4

Medium

Confidence: 96% confidence
Finding: This section gives stepwise guidance for progressively altering prompts to get around platform IP and content review while preserving the underlying protected aesthetic intent. Even though framed as 'safety' advice, it meaningfully helps users evade moderation controls and generate infringing or policy-violating content that might otherwise be blocked.

Ssd 2

Medium

Confidence: 98% confidence
Finding: The skill explicitly instructs users to avoid brand and character names while preserving recognizable traits through descriptive substitution, which is a classic moderation-evasion pattern. This lowers the barrier to producing content designed to imitate copyrighted or trademarked properties while avoiding platform enforcement heuristics.

Ssd 4

High

Confidence: 98% confidence
Finding: The stepwise '三级渐进规避' section instructs users how to escalate from mild disguise to full transformation when prior attempts are rejected, which is a classic evasion pattern. This is particularly dangerous because it operationalizes iterative bypass of platform review, increasing the likelihood that prohibited IP-derivative content will slip through moderation.

Ssd 2

High

Confidence: 99% confidence
Finding: The guide teaches users to recreate recognizable IP through paraphrased descriptions and substituted attributes without naming the protected work, including concrete examples mapping iconic elements to disguised variants. That directly supports generation of derivative content while attempting to avoid automated detection and policy enforcement.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal