ClawHub

better-skill-audit (deprecated)

Security checks across malware telemetry and agentic risk

Overview

The skill appears aimed at auditing and fixing issues, but its instructions conflict about when it may write or edit files.

Review this skill carefully before installing. Use it in read-only audit mode unless you explicitly want it to create reports, backups, or file edits, and only run any fix mode on a scoped workspace with version control or separate backups.

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
97% confidence
Finding
The skill's safety model is internally contradictory: it states a red-line of 'no write operations' during audit, yet elsewhere mandates writing audit reports and allows file modification in fix mode. This ambiguity can cause an agent to perform filesystem writes when the user expected read-only behavior, undermining least-privilege and increasing the chance of unintended file changes.

Intent-Code Divergence

High
Confidence
99% confidence
Finding
In the fix workflow, the document instructs the agent to create backups and edit files, then immediately restates a red-line of 'do not execute any write operations.' This creates a direct policy conflict that an autonomous agent may resolve inconsistently, leading either to unsafe writes or unreliable behavior around destructive operations.

VirusTotal

63/63 vendors flagged this skill as clean.

View on VirusTotal