ClawHub
Back to skill

Security audit

Humanizer Academic

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed academic/popsci prose rewriting workflow with a local diagnostic detector, not evidence of hidden or harmful behavior.

Install only if you want an agent to process and potentially edit user-provided academic or serious nonfiction drafts. Be aware it may run a local Python detector and may use a subagent for judging rewrite quality, but the artifacts do not show hidden network use, credential access, persistence, or destructive behavior.

SkillSpector

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 (2)

Description-Behavior Mismatch

High
Confidence
97% confidence
Finding
The file’s top-level documentation states that it is only a detector for AI-writing signals and explicitly says it never rewrites text, which materially conflicts with the skill’s manifest claiming a humanizing rewrite capability. In an agent skill, this kind of behavior/manifest mismatch is dangerous because downstream orchestrators or users may invoke the skill expecting transformation but instead get classification output, causing policy bypasses, broken control flow, or unsafe reliance on the wrong component.

Intent-Code Divergence

High
Confidence
96% confidence
Finding
The documentation explicitly says the module must never be described as a 'humanizer', directly contradicting the skill branding and intended purpose. In this skill context, that contradiction increases risk because it suggests the shipped behavior may be misrepresented, enabling deceptive invocation semantics and making it harder to reason about whether user content is being rewritten, scored, or abstained on.

Static analysis

No suspicious patterns detected.