Refinement Loop (Opus 4.8 Edition)
Security checks across malware telemetry and agentic risk
Overview
This is a plain instructional skill for running AI refinement loops, with no executable code or hidden behavior found.
Installers should treat this as guidance for using external AI models. Set token and cost limits, and avoid sending sensitive documents or private data to model providers unless that matches your account settings, provider terms, and data-handling expectations.
SkillSpector
By NVIDIA
Vulnerability Patterns
- 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
- Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
VirusTotal
64/64 vendors flagged this skill as clean.