Security audit

Ai Humanizer.Bak

Security checks across malware telemetry and agentic risk

Overview

This is a local writing-analysis and rewriting skill with some overconfident style guidance, but no hidden access, exfiltration, destructive behavior, or automatic persistence was found.

Install only if you want AI-writing detection and humanization behavior available to your agent. Treat its scores as heuristic, avoid using them as sole evidence in academic, employment, moderation, or compliance decisions, and do not enable the always-on prompt templates unless you intentionally want persistent style influence. Preserve or rewrite factual limitation disclosures when they matter instead of blindly deleting them.

SkillSpector

By NVIDIA

Vulnerability Patterns

Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration

Findings (6)

Natural-Language Policy Violations

Medium

Confidence: 93% confidence
Finding: The README instructs users to bake a persistent anti-AI writing policy into an agent's system prompt so the agent always enforces a single communication style, even when not requested. In an agent setting, this can override user intent, suppress locale-appropriate or accessibility-friendly phrasing, and create hidden behavioral constraints that affect downstream outputs without explicit consent.

Vague Triggers

Medium

Confidence: 83% confidence
Finding: The trigger language is broad enough to match many ordinary editing or rewriting requests, which can cause the skill to activate when the user did not explicitly ask for AI-humanization. In context, that can silently rewrite user content, alter tone, and optimize text to evade AI-writing detection, creating both integrity and policy-evasion risk.

Natural-Language Policy Violations

Medium

Confidence: 95% confidence
Finding: The 'Always-on mode' section explicitly instructs operators to embed these style rules into system or personality prompts, overriding normal user-directed writing preferences. That is dangerous because it promotes persistent, non-consensual behavior shaping and can systematically suppress transparency markers or safety-oriented phrasing across unrelated tasks.

Natural-Language Policy Violations

Medium

Confidence: 95% confidence
Finding: The code emits definitive claims such as 'The text looks human-written' and tiered labels like 'heavily AI-generated' despite only performing heuristic pattern matching and stylometric scoring. In a skill explicitly designed to 'humanize' text and score AI-likeness, this can mislead users into treating uncertain inferences as factual provenance judgments, creating reputational, moderation, or compliance harms when used in high-stakes contexts.

Natural-Language Policy Violations

Medium

Confidence: 93% confidence
Finding: Helper labels describe individual metrics as 'human-like' or 'AI-like,' which overstates what burstiness and vocabulary measures can establish. Because this skill's purpose is AI-detection and de-AI rewriting, those labels materially increase the risk that users over-rely on noisy stylometric indicators as definitive authorship signals.

Natural-Language Policy Violations

Medium

Confidence: 89% confidence
Finding: The guidance explicitly tells users to delete knowledge-cutoff disclaimers, which can remove an important transparency signal about uncertainty or model limitations. In a skill designed to make AI text appear more human, this increases the risk of concealing that content is AI-generated or based on incomplete knowledge, potentially misleading readers.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.