Security audit

电商带货数字人 Ecommerce Avatar

Security checks for vulnerabilities and agentic risk

Overview

The skill’s cloud processing, saved login key, and paid generation steps fit its stated purpose of creating AI sales videos.

Prompts and selected images, audio, and videos are sent to dLazy for processing. Submit only material you are authorized to share, especially portraits and confidential product photos. Protect the saved API key and check estimated charges before generation. Local ffmpeg avoids the additional hosted-editing upload.

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

Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
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 (3)

The reference-image workflow sends a local portrait to an external service without a privacy warning.

Medium

Category: Output Handling
Confidence: 90% confidence
Finding: The skill instructs the agent to use gpt-image-2 --images base.png to redraw a reference portrait, which submits the image to the external generation service. Although API endpoints are listed in the metadata, the instructions never disclose that portrait contents leave the user's system or warn against submitting sensitive or unauthorized images.
Content: No source excerpt is available for this finding.

The workflow sends local video assets to a hosted editing service without a privacy warning.

Medium

Category: Output Handling
Confidence: 94% confidence
Finding: The recommended assembly step instructs the agent to submit local video files through dlazy chat --files for sandbox processing. The file does not warn that these assets leave the local system or advise users to check for private likenesses, confidential product information, or other sensitive content before submission.
Content: No source excerpt is available for this finding.

The recipes instruct users to submit local reference images to an external image-generation service without a privacy disclosure.

Medium

Category: Output Handling
Confidence: 94% confidence
Finding: The executable recipe passes ./product.jpg through --images, and L294 explicitly requires supplying a real product image. The file does not warn that reference images are transmitted to an external service, which matters when product photos contain confidential packaging, unreleased designs, or other private information.
Content: No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.