Security audit

Edit Video

Security checks for vulnerabilities and agentic risk

Overview

This skill is a documented video-editing guide that sends user-provided video to generation models, with some routing ambiguity but no hidden execution, persistence, or destructive behavior.

Install only if you want agents to edit videos through external generation providers. Avoid supplying confidential, biometric, unreleased, or rights-restricted footage unless you are comfortable with the provider processing it, and be aware that this skill may handle both global restyles and localized visual edits.

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (4)

Description-Behavior Mismatch

Medium

Confidence: 95% confidence
Finding: The manifest frames this as a whole-frame restyling skill, but the body expands scope to localized surgical edits using a different model and workflow. That mismatch can cause an orchestrator or user to invoke the wrong capability, bypass intended skill routing, and perform edits that belong to a separate, more constrained skill such as replace-in-video.

Description-Behavior Mismatch

Medium

Confidence: 91% confidence
Finding: The documentation tells users to use this skill for localized changes like wardrobe, environment, and grade edits, which overlaps with other specialized video-editing skills. Overlapping routing guidance increases the chance that an agent chooses this skill instead of a more appropriate one, leading to unintended modifications, weaker guardrails, or policy bypass through ambiguous capability boundaries.

Description-Behavior Mismatch

Medium

Confidence: 95% confidence
Finding: The examples expand from the declared whole-frame restyle scope into surgical/object-specific edits using Runway Aleph, including color swaps, environment replacement, and branded object modification. This can misroute users or downstream agents into invoking capabilities outside the skill’s stated boundaries, weakening safety assumptions, review coverage, and policy enforcement tied to the manifest.

Missing User Warnings

Medium

Confidence: 88% confidence
Finding: The examples instruct users to provide source videos, frame anchors, and externally hosted media URLs without warning that this transmits potentially sensitive user media to third-party services and may expose public URLs in logs or service infrastructure. In a media-editing skill, users are especially likely to upload proprietary, personal, or biometric content, so the omission creates a meaningful privacy and data-handling risk.

Static analysis

No suspicious patterns detected.