Security audit
first-order-model-fitting
Security checks across malware telemetry and agentic risk
Overview
This is a simple instructional skill for fitting first-order response models and it does not request system access, credentials, persistence, or network activity.
This appears safe to install as a reference skill. Treat the Python snippets as examples rather than a complete program, and avoid sharing sensitive experimental data with any external agent or service unless you trust where that data will be processed.
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
66/66 vendors flagged this skill as clean.