ClawHub
Back to skill

Security audit

humanize-zh-writing

Security checks across malware telemetry and agentic risk

Overview

This text-rewriting skill appears non-technical, but its guidance can make AI-written or rewritten content look authentically human while adding unsupported personal details or stronger opinions.

Review this skill carefully before installing. Use it only for style and clarity edits where the facts, viewpoint, and authorship disclosure remain intact. Avoid using it for testimonials, reviews, personal stories, academic work, job materials, or public comments unless every personal detail and claim is genuinely supplied by the user.

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)

Intent-Code Divergence

High
Confidence
94% confidence
Finding
The skill sets a hard rule not to fabricate facts or personal experiences, but later instructs the model to insert first-person cognition and 'real' details that may not exist in the source. That contradiction can cause the assistant to generate false attributions, invented perspective, or misleading narrative elements while presenting them as authentic rewriting.

Intent-Code Divergence

High
Confidence
91% confidence
Finding
The skill says the rewrite must not change the original core stance, but the comment-writing section tells the model to make views more one-sided, emotional, and less balanced. In practice this can distort the author's meaning, strengthen claims beyond the source, and generate persuasive content that misrepresents the original position.

Intent-Code Divergence

High
Confidence
95% confidence
Finding
The experience-writing guidance explicitly encourages adding extra 'real' details, structural jumps, and contradictory feelings, despite the earlier prohibition on inventing personal experiences. This can fabricate autobiographical content and emotional states, which is especially risky when users are rewriting testimonials, reviews, or first-person accounts that may be relied on as genuine.

Ssd 4

Medium
Confidence
88% confidence
Finding
The skill is explicitly designed to make AI-generated text appear human-authored by simulating human imperfections and removing AI markers. In context, that increases deception risk: users can use it to obscure machine generation, mislead audiences about authorship authenticity, or polish synthetic testimony, commentary, or reviews so they appear personal and organic.

VirusTotal

66/66 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.