ClawHub
Back to skill

Security audit

AI-review

Security checks across malware telemetry and agentic risk

Overview

This is a markdown-only review helper that reads user-provided content and produces styled commentary, with a real safety caution around its shell-based PDF extraction example.

Install only if you want an agent to fetch or read content you provide and generate Chinese-leaning analytical reviews. Prefer safer PDF/file-reading tools when available, and avoid untrusted PDF URLs or sensitive local file paths unless the agent will quote and constrain shell arguments.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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 (4)

Context-Inappropriate Capability

Medium
Confidence
94% confidence
Finding
The skill explicitly instructs use of shell commands (`curl` and `pdftotext`) to process user-supplied PDFs, which unnecessarily expands capability from content review into command execution. Even though the example is simple, shelling out on attacker-controlled URLs or paths increases the risk of unsafe command construction, unexpected file access, and misuse of a higher-privilege tool than is needed for summarization.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The skill directs the agent to use `shell` and `curl` on user-provided URLs and files without warning about command execution or the risks of accessing local resources. In context, this is more dangerous because the skill accepts both remote and local inputs, so an attacker could induce retrieval of sensitive local files or interaction with unintended network locations under the guise of content processing.

Natural-Language Policy Violations

Medium
Confidence
90% confidence
Finding
The style guide strongly prescribes Chinese as the output language and does not indicate that the agent should respect the user's requested language. In a review assistant that processes arbitrary user-provided content, this can cause instruction-priority conflicts, degrade usability, and lead to unauthorized transformation of outputs away from the user's expectations or downstream system requirements.

Natural-Language Policy Violations

Medium
Confidence
92% confidence
Finding
The file is a prescriptive style guide that explicitly defines Chinese-language output and does not indicate that this behavior depends on user preference or locale. In a review assistant that reads arbitrary URLs/files and generates summaries, this can override user intent, reduce usability, and create prompt-level policy conflicts where the agent responds in an unexpected language.

VirusTotal

63/63 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.