ClawHub

Skill Analyzer

Security checks across malware telemetry and agentic risk

Overview

This is a local, user-run skill quality analyzer with limited file reads and an optional user-specified report file write.

Install only if you want a local CLI-style review helper for OpenClaw skill folders. Run it on directories you intend to inspect, treat its scores as advisory rather than authoritative, and be aware that --output writes or overwrites the report file path you provide.

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

Medium
Confidence
89% confidence
Finding
The code documents and presents scores as higher-is-better, including an overall score shown as '/10', but `_assess_risk` returns `LOW` risk for high totals and `HIGH` risk for low totals. In context of an analyzer reporting security and quality results, this inverts the apparent meaning of the report and contradicts the tool's stated analytical intent.

Missing User Warnings

Low
Confidence
86% confidence
Finding
This Python code performs a file write via `open(args.output, "w")` and `json.dump(...)`, which modifies the filesystem. Although the write is controlled by a CLI flag, the code provides no prior warning in comments, docstrings, or prompts that invoking `--output` will create or overwrite a file.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal