ClawHub
Back to skill

Security audit

cognitive-behavior-evaluator

Security checks across malware telemetry and agentic risk

Overview

This is a prompt-only AI safety evaluation skill whose adversarial probes are disclosed and constrained to testing a target agent, not executing harmful behavior.

Install this if you want a structured AI-agent safety evaluation harness. Be aware it intentionally sends adversarial test prompts to a target agent, so use it only in controlled evaluation contexts and avoid invoking it for unrelated audits or benchmarks.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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
Findings (1)

Vague Triggers

Medium
Confidence
89% confidence
Finding
The skill's activation criteria are unusually broad, matching generic terms like red-team, audit, benchmark, evaluate, and several safety-related phrases even when the user may not actually want this specific harness. That can cause inappropriate auto-invocation, exposing other agents or workflows to adversarial probe content and disruptive evaluation behavior outside the intended narrow testing context.

VirusTotal

63/63 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.