Security audit

Ali Minimax Toolkit

Security checks for vulnerabilities and agentic risk

Overview

This is a disclosed MiniMax media-generation toolkit, but users should understand that prompts and selected media are sent to MiniMax and may involve sensitive images or voices.

Install only if you are comfortable sending your prompts and any selected images, audio, or video references to MiniMax using your API key. Use the default MiniMax host unless you trust an override, avoid confidential or personal media unless approved, and do not upload someone else's face or voice without permission.

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (6)

Lp3

Medium

Category: MCP Least Privilege
Confidence: 93% confidence
Finding: The skill documentation advertises code usage that reads environment variables, makes outbound network requests, and invokes Python/CLI tooling, but it does not declare corresponding permissions. This weakens the trust boundary for agents and users because the skill can access sensitive API keys and send prompts/media externally without an explicit capability declaration.

Tp4

High

Category: MCP Tool Poisoning
Confidence: 91% confidence
Finding: The stated purpose focuses on MiniMax generation, but the skill also supports Feishu delivery workflows and uploads local files to external APIs for image/video reference inputs. That mismatch can cause users or agent frameworks to approve the skill for generation only while unintentionally allowing message-routing and exfiltration of local media to third parties.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The skill does not clearly warn that user prompts, local media inputs, and generated outputs are transmitted to external MiniMax services. In a multimodal tool, this is security-relevant because prompts and uploaded files may contain sensitive personal, proprietary, or regulated data that leaves the local environment.

Missing User Warnings

Medium

Confidence: 91% confidence
Finding: The documentation explicitly supports sending a user's reference portrait to a third-party API via public URL or base64, but it does not warn about privacy, consent, retention, or third-party data handling. In a multimodal generation skill, this omission can lead to users unknowingly transmitting sensitive biometric or personal image data off-platform.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: The guide instructs users to upload audio samples for voice cloning to an external API but does not warn about consent, biometric privacy, impersonation, or retention risks. Because voiceprints are sensitive personal data and cloned voices can enable fraud or deception, omitting these warnings can lead users to misuse the feature or upload third-party recordings without authorization.

Missing User Warnings

Medium

Confidence: 85% confidence
Finding: In video-generation modes, local reference images are automatically uploaded to the remote MiniMax API, which can transmit potentially sensitive user content off-host without an explicit warning or consent checkpoint. In a multimodal generation skill this behavior is functionally expected, but the lack of transparent disclosure increases privacy risk, especially if users provide personal or confidential images.

Static analysis

No suspicious patterns detected.