Security audit

Miao Vision

Security checks across malware telemetry and agentic risk

Overview

This skill is a local report and infographic generator that uses a disclosed helper CLI and does not show hidden exfiltration or destructive behavior.

Install only if you are comfortable approving a local CLI executable from the Miao Vision GitHub release. Use explicit $miao-vision invocation, review any install prompt before accepting, and delete ~/.miao-vision if you want to remove the shared CLI later. Treat generated reports as local artifacts unless you separately choose to share them.

SkillSpector

By NVIDIA

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 (4)

Tp4

High

Category: MCP Tool Poisoning
Confidence: 88% confidence
Finding: The skill description presents a content-generation/validation tool, but the body also authorizes installer execution, platform detection, and binary installation into the user's home directory. That hidden expansion of behavior increases supply-chain and unexpected code-execution risk, especially if a user invokes the skill expecting only document processing rather than software installation.

Description-Behavior Mismatch

Medium

Confidence: 95% confidence
Finding: The skill metadata says activation should occur only when the user explicitly invokes `$miao-vision`, but the installation guidance demonstrates a generic natural-language trigger: 'Use miao-vision to analyze ...'. This broadens invocation semantics and can cause the agent to activate on ordinary prose requests, increasing the chance of unintended local file access or shell-assisted artifact generation in contexts the user did not explicitly authorize.

Description-Behavior Mismatch

Medium

Confidence: 91% confidence
Finding: The documentation states that on first use the tool may request approval to download a matching release binary into a shared user directory. That introduces code acquisition and persistence behavior beyond the manifest's narrowly described artifact-generation scope, creating supply-chain and user-expectation risks if the agent facilitates installation of executables not clearly disclosed in the skill contract.

Natural-Language Policy Violations

Medium

Confidence: 89% confidence
Finding: The example confirmation prompt is hard-coded in Chinese, which can steer the agent to respond in a language the user did not request or understand. In a planning and approval workflow, this can undermine informed consent by making safety/privacy confirmation less intelligible to some users.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

Detected: suspicious.dangerous_exec

Shell command execution detected (child_process).

Critical

Code: suspicious.dangerous_exec
Location: scripts/check-miao-viz.mjs:14