Security audit

MCPorter移植工具

Security checks for vulnerabilities and agentic risk

Overview

The skill is mainly a disclosed mcporter CLI helper, but it combines command execution, credential handling, and remote tool calls with overly broad activation language and some misleading safety framing.

Review before installing. Use it only for explicit mcporter tasks, avoid pasting real secrets into command lines, prefer environment variables or a secret store, and treat configured MCP connectors as capable of reading or changing whatever their own permissions allow.

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
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 (5)

Intent-Code Divergence

Medium

Confidence: 93% confidence
Finding: The skill claims commands run in a '安全沙箱' (safe sandbox), but the rest of the document exposes raw exec capability and provides direct CLI invocation guidance without any actual sandbox boundaries, policy, or isolation guarantees. This can cause users or agent frameworks to over-trust command execution and run risky operations under a false sense of safety.

Intent-Code Divergence

Low

Confidence: 84% confidence
Finding: The protocol guidance describes SSE as suitable for 'read-only API', yet the same skill supports authenticated remote tool calls and configuration actions that may mutate remote state. This misleading framing can cause operators to underestimate the risk of sending credentials or invoking state-changing actions over remote endpoints.

Vague Triggers

Medium

Confidence: 89% confidence
Finding: The activation language is extremely broad, covering generic AI calling, intelligent dialogue, agent orchestration, and LLM applications. In a skill with read/exec/glob/grep and remote service interaction, broad triggers increase the chance of unintended invocation in unrelated conversations, expanding exposure to command execution and external calls.

Vague Triggers

Medium

Confidence: 88% confidence
Finding: The listed scenarios are generic productivity and automation categories rather than constrained operational conditions. Because the skill can execute commands and handle credentials, vague applicability language increases accidental routing of requests into a high-privilege skill context.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The examples show real-looking secret material directly on the command line and demonstrate remote service calls without prominent warnings about shell history exposure, terminal logging, process-list leakage, or data transmission to external services. In this skill, that is especially risky because it combines credential setup with exec-driven command examples and remote transports.

Static analysis

No suspicious patterns detected.