Security audit

飞影数字人 V2

Security checks for vulnerabilities and agentic risk

Overview

The skill is a mostly transparent digital-human video guide, but it blurs into emotion-recognition and companion-style emotional support while also claiming it is not an emotional companionship tool.

Review this carefully before installing if you intend to use it in interactive or companion-like settings. It appears safe from a code-execution perspective, but users should keep it to one-way video generation, avoid emotional dependency or counseling use cases, and only upload voice or face samples with clear consent.

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

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 (3)

Intent-Code Divergence

High

Category: Not specified by scanner
Confidence: 96% confidence
Finding: The document claims the tool is exempt because it does not involve emotional companionship or virtual intimacy, yet elsewhere it explicitly defines emotion recognition, emotional response, and a 'companion scenario' focused on emotional support. This contradiction can mislead operators, compliance reviewers, or downstream agents into enabling regulated affective-interaction use cases under a false assumption of safety.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium

Category: Not specified by scanner
Confidence: 90% confidence
Finding: The file’s natural-language instructions, configuration guidance, and compliance notes are all presented in Chinese, and there is no indication that users may choose another language or that the skill is intentionally limited to a Chinese-speaking region. Under the policy, forcing a specific language without user opt-in can constitute a locale/language policy violation.
Content: No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium

Category: Not specified by scanner
Confidence: 88% confidence
Finding: Adding 'companion scenario' and 'emotional support' behavior expands the skill from passive video generation into affect-sensitive interaction design. In an agent ecosystem, that broader framing may encourage deployment for psychologically influential or dependency-forming use cases that exceed the stated product boundary and may trigger legal, trust, and safety risks.
Content: No source excerpt is available for this finding.