Doubao API Toolkit

• Outcome: Review
• Latest audit: May 29, 2026, 9:27 PM
• Version:
  1.0.0

Overview

This appears to be a real Doubao/Volcengine media toolkit, but it can upload user-selected local video files and other prompts/media to a third-party API without a clear privacy warning or point-of-use confirmation.

Install only if you are comfortable sending prompts, image URLs, and any local video files you choose to analyze to Volcengine/Doubao. Avoid confidential, regulated, proprietary, or third-party media unless you have permission, and prefer an updated version that adds explicit upload warnings, consent before local file transmission, and corrected image-to-image documentation.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• MCP Least Privilege: Underdeclared Capability, Wildcard Permission, Missing Permission Declaration
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access

Findings (5)

Lp3

Medium

• Category: MCP Least Privilege
• Confidence: 95% confidence
• Finding: The skill documents use of environment variables and outbound API calls, but it does not declare corresponding permissions. Missing permission declarations undermine least-privilege review and can cause users or platforms to underestimate that the skill can read secrets from the environment and send data over the network. In a toolkit that processes prompts, URLs, and local media for a third-party API, that gap materially affects trust and reviewability.

Tp4

High

• Category: MCP Tool Poisoning
• Confidence: 90% confidence
• Finding: The declared purpose does not accurately match the described functionality: undocumented video analysis, image-to-video generation, and status-checking expand the operational surface, while claimed image-to-image support appears unimplemented. Description-behavior mismatches are dangerous because reviewers and users may approve or invoke a skill under incorrect assumptions about what data it handles and what actions it can perform. Here, the hidden/omitted behaviors all involve external processing of user data and media, increasing the risk of unintended disclosure or misuse.

Description-Behavior Mismatch

Medium

• Confidence: 92% confidence
• Finding: The skill metadata emphasizes image/video generation and vision analysis, but the code also implements local video analysis by reading a local file, base64-encoding it, and sending the full contents to a remote API. This hidden capability creates a transparency and consent problem because users may not realize local files are being uploaded off-device.

Missing User Warnings

Medium

• Confidence: 97% confidence
• Finding: The skill explains API-key setup and output storage but does not clearly disclose that prompts, image URLs, local video files, and generated content are sent to the external Doubao/Volcengine ARK service. This is a real security and privacy issue because users may submit sensitive text or local media without understanding that third-party transmission and processing will occur. The context makes this more dangerous because the skill explicitly supports local video analysis, so potentially private local files may be uploaded off-host.

Missing User Warnings

Medium

• Confidence: 97% confidence
• Finding: The analyze_video function reads an arbitrary local video file and transmits its entire contents to the remote chat/completions API without any warning, consent flow, or data-minimization controls. This can expose sensitive local media, including personal, proprietary, or regulated content, to a third party unexpectedly.

VirusTotal

64/64 vendors flagged this skill as clean.