Humanizer Rewrite

Security checks across malware telemetry and agentic risk

Overview

This writing skill is mostly coherent, but it can overwrite a user-provided file without confirmation or backup.

Install only if you are comfortable with the skill changing files you point it at. Prefer pasting text, or explicitly tell the agent to show the rewritten text, create a backup, or write to a new file instead of overwriting the original.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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
Findings (3)

Description-Behavior Mismatch

Medium
Confidence
92% confidence
Finding
The skill’s manifest presents it as a text-rewriting tool, but the documented behavior expands scope to reading from arbitrary user-supplied file paths and writing changes back. That mismatch is security-relevant because it enables file-system side effects beyond simple content transformation and can cause unintended modification of local files.

Context-Inappropriate Capability

Medium
Confidence
95% confidence
Finding
Automatically overwriting the original file is not necessary to achieve the stated goal of humanizing text and creates avoidable integrity risk. If the wrong path is supplied or the rewrite degrades the content, the original data may be lost or silently corrupted.

Missing User Warnings

High
Confidence
98% confidence
Finding
The skill instructs the agent to overwrite the original file immediately, without warning or confirmation. This is dangerous because a single prompt containing a path can trigger destructive modification of user content, making accidental data loss or abuse through prompt-driven file edits more likely.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal