Verification Before Completion
Security checks across malware telemetry and agentic risk
Overview
This is an instruction-only quality-control skill that tells agents to verify real results before claiming work is complete.
This appears safe to install as a quality-control aid. Expect it to make the agent verify outcomes before reporting completion; for production systems, real accounts, or irreversible actions, keep verification read-only, use dry-runs where possible, or explicitly approve any state-changing check.
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.