ClawHub
Back to skill

Security audit

skill-trust-auditor

Security checks across malware telemetry and agentic risk

Overview

This skill is a coherent security-auditing tool, but users should treat its scores as advisory and know that optional LLM mode sends selected audit context to Anthropic.

Install and run this only in a virtual environment as documented. Treat audit results as one input, not as proof a skill is safe. Avoid --llm for private or sensitive skill material unless you are comfortable sending selected excerpts and flagged context to Anthropic.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Lp3

Medium
Category
MCP Least Privilege
Confidence
90% confidence
Finding
The skill invokes shell scripts, uses networked verification commands, and may access environment variables (including an optional API key), yet it declares no explicit permissions or capability boundaries. This can mislead users about the effective privileges of the skill and increase the chance that a user runs code with broader access than expected.

Natural-Language Policy Violations

Medium
Confidence
85% confidence
Finding
The trust score table labels 90-100 as 'SAFE' and says 'Install freely,' which can cause users to over-trust a heuristic score despite the skill elsewhere stating that scores are not proof of safety. This contradiction can reduce user caution and lead to installation of a malicious or risky skill that happened to evade the bounded checks.

Missing User Warnings

Medium
Confidence
93% confidence
Finding
When --llm is enabled, the analyzer sends flagged network-call context and up to 1500 characters of SKILL.md to Anthropic. That is a real confidentiality issue because third-party skill content, and potentially sensitive text embedded in it, is disclosed to an external API without an explicit user-facing consent prompt or strong minimization/default-off notice beyond the CLI flag itself.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

Detected: suspicious.dynamic_code_execution

Dynamic code execution detected.

Critical
Code
suspicious.dynamic_code_execution
Location
scripts/test_analyzer.py:15