Security audit

AI KEY·口播剪辑

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed video-editing workflow with scoped project edits and user-confirmed paid generation/export steps.

Before installing, be aware that it can edit a ChatCut timeline, read configured vault rules/memory, and write editing feedback memory. Review the account/default settings, and only approve generated video or export steps when the proposed cost, clips, and output match your project.

A.I.G

Vulnerability Patterns

Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs

SkillSpector

Vulnerability Patterns

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
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (2)

Natural-Language Policy Violations

Medium

Confidence: 89% confidence
Finding: The description states that the workflow criteria are written for Chinese talking-head content and will still be applied when the user uses another language, with only an explanatory note. This is a natural-language locale policy constraint without explicit user choice or opt-in, which can impose language-specific behavior on users outside that locale.

Intent-Code Divergence

Medium

Confidence: 93% confidence
Finding: The skill contains conflicting instructions about when account-specific defaults may be applied. One section says not to ask again and to use defaults directly, while other sections require confirming the account or reading config first; this can cause the agent to apply settings intended for a specific account to the wrong user or project. In a media-editing workflow, that risks unauthorized use of account-specific presets, wrong export behavior, and privacy or branding mistakes.

Static analysis

No suspicious patterns detected.