ClawHub

敏感内容扫描器

Security checks across malware telemetry and agentic risk

Overview

This is a local sensitive-content scanner with no evidence of uploading data or destructive behavior, though its broad trigger phrase and generated reports need care.

Install only if you want a local Python tool to scan selected documents for sensitive content. Be explicit when invoking it, avoid pointing it at broad private directories unless intended, and keep generated reports private because they can contain file paths and sensitive matches. Do not rely on the optional encrypted dictionary format as strong secrecy.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Vague Triggers

Medium
Confidence
93% confidence
Finding
The trigger phrase "扫描" is extremely generic in a document-handling context and can easily match ordinary user requests unrelated to this specific skill. That ambiguity can cause unintended invocation on arbitrary attachments, expanding the skill's reach beyond clear user consent and increasing the chance that sensitive documents are processed when the user did not explicitly mean to invoke this capability.

Vague Triggers

Medium
Confidence
95% confidence
Finding
The description says users can simply attach a document and say "scan," which is overly broad and does not define clear boundaries for when the skill should or should not activate. In an agent environment, such loose invocation guidance can cause accidental selection of this skill for common document-review tasks, potentially exposing sensitive files to unnecessary processing and creating confusing or unsafe routing behavior.

VirusTotal

67/67 vendors flagged this skill as clean.

View on VirusTotal