Security audit
Hallucination Guard — 4-Layer AI Fabrication Defense
Security checks across malware telemetry and agentic risk
Overview
This is a documentation-only skill for checking AI agent claims against evidence, with no hidden code or install behavior found.
Before installing, confirm that your environment allows session-history review and optional cross-model audits. Avoid sending secrets, private logs, customer data, or sensitive source files to spawned auditors unless your data policy permits it.
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.