ClawHub
Back to skill

Security audit

AI Text Humanizer — Remove AI Writing Patterns

Security checks across malware telemetry and agentic risk

Overview

This is a document-editing skill that openly rewrites text to sound less AI-generated, with no evidence of hidden execution, data theft, persistence, or privilege escalation.

Use this as a style-editing aid, not to misrepresent authorship or bypass disclosure rules. Review rewritten text for invented opinions, feelings, first-person claims, or unsupported specifics before accepting changes, and confirm the package identity because the displayed slug/version metadata is inconsistent.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • 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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (2)

Natural-Language Policy Violations

High
Confidence
95% confidence
Finding
The skill’s stated purpose is to make text sound more natural and human-written and to remove signs of AI generation. That directly enables concealment of AI authorship and can be used to evade disclosure policies, academic integrity controls, moderation heuristics, or provenance checks, even if the stated use is ordinary editing.

Natural-Language Policy Violations

Medium
Confidence
91% confidence
Finding
The guidance goes beyond copyediting and tells the agent to add opinions, first-person perspective, uncertainty, humor, and personality. That can fabricate a human persona or subjective experience not present in the source material, creating deceptive authorship signals and potentially introducing false statements about beliefs or feelings.

VirusTotal

62/62 vendors flagged this skill as clean.

View on VirusTotal