ClawHub
Back to skill

Security audit

Improvement Orchestrator

Security checks across malware telemetry and agentic risk

Overview

This skill openly runs a local skill-improvement pipeline that can modify a user-selected skill and write state files, and those capabilities match its stated purpose.

Install only if you want an agent to run an end-to-end local improvement workflow. Use it on a copied or backed-up target first, choose a disposable state directory, prefer --eval-mock for initial runs, and review the subordinate generator, evaluator, executor, and gate skills because this orchestrator delegates meaningful authority to them.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • 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 (1)

Lp3

Medium
Category
MCP Least Privilege
Confidence
95% confidence
Finding
The skill explicitly instructs users to run a Python orchestrator that reads and writes files under a target skill and state directory, invokes multiple subordinate scripts, and may execute shell/CLI-driven evaluation flows, yet it declares no permissions metadata. In an orchestration skill that can apply changes, retry automatically, and coordinate end-to-end execution, missing capability declarations materially weakens review and containment because users and policy systems are not warned about its real operational reach.

VirusTotal

67/67 vendors flagged this skill as clean.

View on VirusTotal