ClawHub
Back to skill

Security audit

Ab Test Runner

Security checks across malware telemetry and agentic risk

Overview

This is an instruction-only A/B testing workflow that can run subagent comparisons and save experiment notes, so users should avoid sensitive data unless they are comfortable with that storage.

Install this if you want a structured agent workflow for prompt or content A/B tests. Before running it on confidential prompts, customer data, or proprietary outputs, ask the agent to confirm before spawning subagents or writing files, and consider requesting a planning-only or no-save run.

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 (3)

Vague Triggers

Medium
Confidence
90% confidence
Finding
The activation phrases include broad patterns like "compare X vs Y" and "design an experiment," which can match ordinary user requests that do not intend to invoke an A/B-testing workflow. This can cause unintended skill activation and lead the agent to run structured experiment logic, spawn subagents, or archive data when the user only wanted a simple comparison.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The skill instructs automatic aggregation of experiment results into a memory file without requiring a user-facing notice or consent. If experiment inputs or outputs contain sensitive prompts, proprietary content, or personal data, this creates an unnecessary persistence channel and increases privacy and data-retention risk.

Missing User Warnings

Medium
Confidence
97% confidence
Finding
The archival steps direct the agent to append/update hypothesis registries, write detailed analyses, and modify a template with new findings and pitfalls, but there is no warning that detailed outputs and reasoning may be stored long-term. This broad persistence of experimental artifacts can capture sensitive model outputs, internal reasoning summaries, or customer data and make later unintended disclosure more likely.

VirusTotal

66/66 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.