ClawHub
Back to skill

Security audit

blog-polish-zhcn

Security checks across malware telemetry and agentic risk

Overview

This skill coherently reads a user-specified technical blog draft, uses an LLM to polish and translate it into Simplified Chinese, and writes the result to a disclosed output path.

Before installing, confirm that you want drafts from the configured path translated into Simplified Chinese and saved under the configured output directory. The publisher should align the documented word-count target, but the observed behavior is purpose-aligned and disclosed.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Rogue AgentSelf-Modification, Session Persistence
  • 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 (4)

Intent-Code Divergence

Low
Confidence
97% confidence
Finding
The manifest description at L003 states the skill should produce a 1200–1400 word article, but the Workflow Summary documentation at L113 says the polishing step targets 1000–1200 words. This is an explicit contradiction in the skill's stated intent, which can mislead users or downstream agents about expected output length.

Vague Triggers

Medium
Confidence
94% confidence
Finding
The listed triggers include phrases like "polish my draft" and "translate blog to Chinese," which are common-language requests and do not clearly constrain activation to this specific skill or environment. The file also does not provide exclusion conditions or negative examples to distinguish when the skill should not run.

Natural-Language Policy Violations

Medium
Confidence
96% confidence
Finding
The description requires translation into Simplified Chinese (zh-CN), and the rest of the document reinforces that fixed locale as the expected behavior. Because no user opt-in or alternative language choice is offered, this is a natural-language locale policy violation under the stated audit rules.

Session Persistence

Medium
Category
Rogue Agent
Content
## Workflow Summary

1. **Resolve paths** + create timestamp (`date +"%y%m%d%H%M"`)
2. **Read draft** from `draftPath`
3. **Polish English**: Fix grammar/spelling, improve clarity, structure into 4-5 sections, target 1000-1200 words
4. **Translate to zh-CN**: Preserve code blocks, inline code, technical terms (`openclaw`, `skill`, `cli`)
Confidence
60% confidence
Finding
Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.