ClawHub
Back to skill

Security audit

PaperRead-Skills

Security checks across malware telemetry and agentic risk

Overview

This is a narrow PDF-review skill that only guides an evidence-based two-pass analysis of a user-uploaded paper.

Install this if you want structured, evidence-grounded reviews of CVPR-style paper PDFs. Be aware that the second pass defaults to Chinese output, and only upload papers you are comfortable sharing with the agent environment, especially if they contain unpublished or private research.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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 (3)

Vague Triggers

Medium
Confidence
86% confidence
Finding
The default prompt describes an automatic multi-step workflow triggered by simply uploading a PDF, but it does not define scope limits, confirmation requirements, or safety constraints. Ambiguous trigger text can cause the agent to run unexpectedly or process untrusted documents without clear user intent, increasing the chance of unintended data handling or prompt-injection exposure from PDF contents.

Natural-Language Policy Violations

Medium
Confidence
79% confidence
Finding
The skill metadata and prompt text appear to be written in Chinese, which may effectively force a locale or language preference without explicit user opt-in. While not directly a code-execution issue, this can cause misinterpretation of user intent, reduce transparency about what the skill does, and increase operational risk if users cannot clearly understand prompts or consent to the workflow.

Hidden Instructions

High
Category
Prompt Injection
Content
You are doing FACT EXTRACTION ONLY.
Do not evaluate quality, do not conclude impact, and do not use external knowledge.
Use only the uploaded PDF.
Confidence
96% confidence
Finding
The file begins with a hidden Unicode character/BOM before the visible instructions, which qualifies as hidden content and can interfere with deterministic parsing, policy checks, or line-accurate inspection. Although the visible prompt appears to be a normal fact-extraction template, hidden characters at the start of instruction files are risky because they can conceal prompt content differences or cause downstream tooling to mis-handle the file.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.