ClawHub
Back to skill

Security audit

OpenAI助手专业版

Security checks across malware telemetry and agentic risk

Overview

This skill is mostly a documented API-management helper, but its provider identity and execution scope are broad enough that users should review it before installing.

Install only if you are comfortable letting an agent use local command execution and provider API credentials for account-changing LLM workflows. Verify that api.llm-provider.com is the provider you intend to trust before exposing OPENAI_API_KEY-like credentials, and require explicit previews before create, update, delete, upload, fine-tune, batch, or vector-store operations.

SkillSpector

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 (2)

Description-Behavior Mismatch

Medium
Confidence
95% confidence
Finding
The skill is presented as a scoped llm-provider management tool, but it also advertises generic file handling, search, and especially local command execution as core capabilities. That scope expansion is dangerous because it creates a mismatch between user/operator expectations and the actual authority granted to the agent, enabling broader host interaction than is necessary for the stated API-management purpose.

Intent-Code Divergence

Medium
Confidence
97% confidence
Finding
The documentation claims commands run in a 'safe sandbox', but the skill file only declares an unrestricted exec tool and provides no verifiable enforcement mechanism. This can mislead users and reviewers into trusting operations that may actually run with the agent's ambient permissions, increasing the risk of arbitrary command execution, data access, or system modification.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.