ClawHub
Back to skill

Security audit

Ask Claude

Security checks across malware telemetry and agentic risk

Overview

The skill’s purpose is clear, but it under-discloses important data sharing and execution authority when delegating work to Claude Code.

Review before installing. Use this only if you are comfortable letting an agent run Claude Code on selected project directories, potentially transmit prompts and file context to Claude, bypass normal Claude Code permission prompts, and retain session history locally. Verify the wrapper script before use and avoid sensitive, regulated, credential-heavy, or trading/account data unless you have explicitly accepted that exposure.

SkillSpector

By NVIDIA
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)

Intent-Code Divergence

High
Confidence
99% confidence
Finding
The skill claims 'no external calls' and 'never send data to external servers,' yet it delegates prompts and workspace content to the Claude CLI, which inherently transmits data to Anthropic's service for processing. This is dangerous because it can mislead operators into sending sensitive code or files under a false privacy assumption, undermining informed consent and data-handling controls.

Intent-Code Divergence

Medium
Confidence
96% confidence
Finding
The file states operations are confined to the specified WORKDIR, but also documents persistent session storage in ~/.claude/projects/, which is outside the workspace boundary. This weakens the claimed isolation model and can cause sensitive context, filenames, or task history to persist in a broader user directory than expected.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.