ClawHub
Back to skill

Security audit

Decision-Grade Reasoning (DGR)

Security checks across malware telemetry and agentic risk

Overview

This is an instruction-only reasoning skill that produces structured JSON decision records, with a caution that one incident-response example under-gates a rollback recommendation.

Installers can treat this as a low-risk documentation and reasoning aid, not an automation tool. Before storing generated artifacts, remove secrets, personal data, and confidential incident details, and require qualified human review for production rollbacks, access-control, financial, legal, medical, safety, or other high-impact decisions even if an example output says review is not required.

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
Findings (1)

Missing User Warnings

Medium
Confidence
87% confidence
Finding
The example recommends an immediate production rollback for a live incident affecting purchases, but it does not include a prominent user-facing caution that rollback is an operationally destructive action with possible side effects such as data inconsistency, additional downtime, or incomplete recovery. Because this skill is designed to generate audit-ready decision artifacts, users may over-trust the structured recommendation and act quickly without adequate operational safeguards.

VirusTotal

59/59 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.