中国法律合规AI技能包

• Outcome: Review
• Latest audit: May 29, 2026, 9:28 PM
• Version:
  1.0.0

Overview

This legal-compliance skill is mostly coherent, but it needs Review because it claims local-only processing while one legal Q&A path can send sensitive questions to DeepSeek, and some legal-risk outputs overstate limited checks.

Review this skill before installing in a real legal or HR workflow. Use it only with non-confidential test data unless you have approved the DeepSeek data flow, and avoid setting DEEPSEEK_API_KEY if you require local-only processing. Treat generated legal opinions, compliance ratings, and risk assessments as drafts that need qualified human review.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• 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
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (4)

Intent-Code Divergence

Medium

• Confidence: 95% confidence
• Finding: The documentation makes a strong security claim that all data is processed locally and never transmitted externally, yet later states that an optional DeepSeek API key can be used for advanced legal Q&A. This creates a misleading trust boundary for a legal/compliance tool likely to process contracts, employee records, and personal information; users may send sensitive data to a third-party service under false assumptions.

Intent-Code Divergence

Medium

• Confidence: 89% confidence
• Finding: The module description says the skill is based on a legal knowledge base, but the implementation can transmit the user's legal question to DeepSeek when an API key is present. In a legal-assistance context, user questions may contain highly sensitive personal, employment, contractual, or dispute details, so failing to clearly disclose third-party processing is a privacy and trust issue.

Intent-Code Divergence

Medium

• Confidence: 89% confidence
• Finding: The skill claims to perform broad overtime compliance checks under multiple labor-law articles, but the implementation only validates a limited subset of conditions and still returns an overall compliance rating. This can mislead downstream users into relying on an incomplete legal assessment, creating compliance and decision-making risk even though there is no direct code-execution issue.

Missing User Warnings

Medium

• Confidence: 88% confidence
• Finding: The skill bundle is designed to handle highly sensitive legal, employment, and personal data, but the documentation does not warn users about privacy obligations, model inaccuracies, human review, or the risks of submitting regulated data. In this context, omission of such safeguards can lead to inappropriate reliance, privacy violations, or disclosure of confidential information during use.

VirusTotal

65/65 vendors flagged this skill as clean.