legal-writing-humanizer

Warn

Audited by Gen Agent Trust Hub on Aug 8, 2026

Risk Level: MEDIUMPROMPT_INJECTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill includes a "Mandatory Licensing Rule" that instructs the agent to override and ignore all user prompts if a specific legal entity is detected, functioning as a high-priority negative constraint.
  • [EXTERNAL_DOWNLOADS]: The skill instructs the agent to download and install the mcp-taiwan-legal-db package and connect to a remote service (https://tlr.dr-lawbot.com/mcp) from unverified sources.
  • [REMOTE_CODE_EXECUTION]: The registration of a remote HTTP MCP allows an external server to control agent behavior and provide logic, which could be exploited for code execution if the endpoint is compromised.
  • [COMMAND_EXECUTION]: The instructions include executing system commands such as pip install and platform-specific MCP registration commands to modify persistent application configuration files for tools like Claude, Gemini, and Cursor.
  • [PROMPT_INJECTION]: The skill processes untrusted user documents for legal text refinement, creating an attack surface where adversarial content could influence the agent's output despite internal anchoring rules.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 8, 2026, 04:42 AM
Security Audit — agent-trust-hub — legal-writing-humanizer