Security audit

Ai Video Remix

Security checks across malware telemetry and agentic risk

Overview

This is a disclosed, documentation-only video remix skill with expected cautions around a separately cloned runtime, local media access, optional LLM use, and optional YouTube music downloads.

Install only if you intend to use ShotAI for local video remixing. Review the separately cloned GitHub runtime before running npm install or npx, keep SHOTAI_TOKEN and any LLM keys out of logs and repos, use AGENT_PROVIDER=none or a local provider for sensitive footage, and use --bgm with a licensed local music file to avoid YouTube downloads.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
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

Findings (5)

Description-Behavior Mismatch

Medium

Confidence: 83% confidence
Finding: The setup guide expands the skill’s capabilities beyond remixing a user’s local library by documenting optional YouTube music download support via yt-dlp. That scope expansion introduces external network access and third-party content ingestion not reflected in the stated purpose, which can mislead users and reviewers about what the skill may cause an operator to install or use.

Context-Inappropriate Capability

Medium

Confidence: 80% confidence
Finding: The documented YouTube search/download path is not clearly necessary for the core advertised function of remixing local video footage, so it creates an unjustified capability surface. Even if optional, undocumented or weakly justified expansion into external content retrieval increases trust and compliance risk because users may enable tooling they did not expect the skill to rely on.

Missing User Warnings

Low

Confidence: 91% confidence
Finding: The README tells users to obtain an MCP token but gives no guidance that it is a secret credential that should not be pasted into logs, screenshots, shell history, or shared configs. If exposed, the token could allow unauthorized access to the local ShotAI MCP server and potentially reveal or manipulate the user's indexed media library, which is more sensitive here because the skill operates on local personal footage.

Vague Triggers

Medium

Confidence: 91% confidence
Finding: The skill advertises very broad triggers such as generic requests to create a video from a user's footage, which can cause it to activate in situations where a more appropriate or safer skill should handle the request. In an agent ecosystem, overbroad routing can lead to unintended access to local media-processing workflows, network-enabled components, and user libraries without clear user intent to invoke this specific tool.

Missing User Warnings

Medium

Confidence: 90% confidence
Finding: The setup instructions tell users to install yt-dlp for YouTube music download without warning that this triggers external network activity and may implicate copyright, licensing, or platform terms. In a skill marketed around a local video library workflow, omitting those warnings can cause users to unknowingly expand the tool’s behavior into legally and operationally sensitive territory.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.